Home

Data Display, Acquisition and Feedback System for Biomedical

1. Section Section Title Primary Author s Abstract Pat Acknowledgements Pat Authorship Pat Executive Summary Pat 1 Introduction Pat 2 Background Pat 2 1 Introduction to EMG Pat 2 1 1 Sources and Characteristics of EMG Tony 2 1 2 Methods of EMG Amplitude Estimation Pat 2 1 2 1 Noise Rejection Pat 2 1 2 2 Whitening Pat 2 1 2 3 Multiple Channel Combination and Gain Pat Normalization 2 1 2 4 Detection Smoothing and Relinearization Pat 2 2 Uses of EMG Pat 2 2 1 Prosthetic Limb Development Pat 2 2 2 Gait Rehabilitation Pat 2 2 3 Other Control Systems Pat 2 2 4 Discussion Pat 2 3 Biomedical Signal Acquisition System Pat 2 3 1 Embedded System Pat 2 3 2 PC based System Pat 2 3 2 1 Data Acquisition Board Fundamentals Pat 2 3 2 2 Benefits of a PC based System Pat 2 4 PC Operating System Pat 2 4 1 Operating System Functions Pat 2 4 1 1 Extended Machine Pat 2 4 1 2 Resource Manager Pat 2 4 2 Types of Operating Systems Pat 2 4 3 Scheduling Pat 2 4 3 1 General Scheduling Approaches Pat 2 4 3 2 Scheduling in Real Time Systems Pat 2 4 3 3 Scheduling in Windows XP Pat 2 5 Existing System Pat 2 5 1 Hardware Overview Pat 2 5 2 Software Overview Pat 2 5 3 Discussion Pat 2 6 Programming in LabVIEW Pat Tony 2 6 1 Basics Pat Tony iv
2. 2 6 2 Calling Code from Other Languages Pat Tony 2 6 3 Linking MATLAB with LabVIEW Pat Tony 2 6 4 Linking C code with LabVIEW Pat Tony 2 6 5 LabVIEW Analysis Methods Pat Tony 2 7 Summary Pat 3 Problem Statement Pat 3 1 Problem Statement and Project Goals Pat 3 2 Hard Real Time Requirements Pat 3 3 Objectives and Tasks Pat 3 4 Summary Pat 4 DAQ Board Performance Analysis Pat 4 1 Measurement Computing PCI DAS6402 16 Pat 4 1 1 Hardware Overview Pat 4 1 2 LabVIEW Universal Library Functions Pat 4 2 NI PCI 6229 Pat 4 2 1 Hardware Overview Pat 4 2 2 NI DAQmX LabVIEW Function Pat 4 3 DAQ Board Comparisons Pat 4 3 1 Data Transfer Pat 4 3 2 Simultaneous Input and Output Pat 4 4 Discussion Pat 4 5 Chapter Summary Pat 5 LabVIEW Analysis Method Comparison Pat 5 1 EMG Amplitude Estimation VIs Jesse 5 2 Latency Test Procedure Tony 5 2 1 Timed Loop Vi Pat 5 2 2 Analysis Method Comparison Procedure Tony 5 3 Timed Loop Test Results Tony Pat 5 4 Discussion Tony Pat 5 5 Chapter Summary Tony Pat 6 Hard Real Time LabVIEW Performance Pat 6 1 Final Testing Procedure MTBF Tony 6 2 Results Tony Pat 6 3 Discussion of Results Tony Pat 6 4 Limitation of Windows Operating System Tony 6 5 Chapter Summary Pat 7 Using LabVIEW in a Data Logging Application Tony 8 Future Recommendations Jesse 8 1 Other LabVIEW Design Options Jesse 8 2 Other Alternatives Jesse 8 3 Chapter Summary Jesse 9 Conclusion Pat Appen
3. 12 2 1 2 4 Detection Smoothing and Relinearization 0 0 ccceeeeees 13 2 2 Uses of EMG ean E E a E A a A E eS 14 2 2 1 Prosthetic Limb Development cecccccesccessceceseeeceseeecsseeecssseeesaes 14 2 2 2 Gait RR GTi ca POUNCE AEN ONS 3 5058s soos ashew sv eskss ous cashes ss tacos sw eunosaodae dev see esa cengntetees 15 2 2 3 Other Control Systems 55550 sh ys cinch vases Wy pessoa a cas ee des Biv oucoareaes PGR 17 2 24 DISCUSSION scaisveccassvesgscecssiaeaasa ays cdasauaide a ead sac lua dad E wean Mebane 17 2 3 Biomedical Signal Acquisition Systems 0 ccccccccccceescecesscecesseecsseeeeeseeeees 18 2 3 1 Embedded Systema sccetsiess diego esse we aes Ae 18 2 3 2 PC Based System 30 oc secloc p35 dosicenslesosesssu gsi ws tw does dvsness eden osaced oa tontions 19 2 3 2 1 Data Acquisition Board Fundamentals c ccccccccessceesseeesseees 20 2 3 2 2 Benefits of a PC Based System cccccccccescccssscecssececsseeeceseeeesseees 21 2 4 Operating System Information 0 ccc ceecceeeceececeaeeeeeeeeneeeaeeenaeeneeeaes 21 2 4 1 Operating System Functions 0 ccccccccesccesececesececeseeecsseeeceseeeesseeeesaes 22 2 4 1 1 Extended Machine i 0 icsiidoscscasaciviiveeiin niente naetivtieainteans 22 2 4 1 2 Resource Manager scone ccusecte satis vieessaeaieredncacdavig iasieaeatiecs beabeeceetiens 22 2 4 2 Types of Operating Systems
4. For small sampling frequencies it was expected that the MTBF would be quite large Table 6 4 displays the results garnered from calculating the MTBF at the small sampling frequency of 200 Hz These results were garnered when LabVIEW was operating with normal priority The FIR filter used to induce processing was a point by point FIR windowed filter set to operate as a lowpass filter with no window and an arbitrary cutoff frequency value The raw results for all of these tests can be viewed in Appendix G Sampling Frequency Number of FIR MTBF Standard Deviation Hz Filter Coefficient sec None 17 Hours Without N A Failure 200 500 13 7 14 59 5000 14 01 13 77 50000 1 24 34 Table 6 4 MTBF Results for 200 Hz Sampling Frequency using Varying Computation Loads at Normal Priority with One Analog Input Channel 90 6 3 Discussion of Results The previous section displayed the results that were garnered from our hard real time testing of standard LabVIEW running on the Windows XP operating system These results establish several interesting trends Figures 6 2 graphically display the MTBF results obtained when running the previously described timing test with no computational load for one analog input and output channel As the graph indicates the MTBF is dependent on the sampling frequency utilized For low sampling frequencies the MTBF is approximately a maximum of eight seconds However as th
5. 79 Figure 5 4 Latency of EMG Array Based and Point by Point Algorithm Real Time Priority 79 Figure 5 5 Average Latency for Buffer Size of One 0 cece eee e cence eee e ene ee ee ene en ens 80 Figure 5 6 Average Point by Point Latency for Different Priority Levels 055 81 Figure 5 7 Raw Data Histogram Point By Point 2048 Real Time Priority 008 82 Figure 5 8 Raw Data Histogram Array Based Buffer Size 1 2048 Real Time Priority 82 Figure 6 1 MTBF Test Vle nense fesse cee don eevee een EEEE beds Ae nE TEA AA AE EE 86 Figure 6 2 MTBF 1 Analog Input Output Channel sssseesssereeessserrrrerressrrrererereree 91 Figure 6 3 MTBF 2 Analog Input Output Channels ccc cece ece eee eee ee ene ene ene ene eeeees 92 Figure 6 4 MTBF 4 Analog Input Output Channels 0 cece cee ee cece eee ene ene eneeneeneees 93 Figure 6 5 MTBF w Computation Load at 200 Hz Sampling Frequency 0080 94 Figure 8 1 Impact of FPGA based system 18 0 ccc eee sees eee e eee ee eee eee eee enaens 100 Figure EA 1 VI to Test Latency of MATLAB Script Node Adding Two Real Numbers 119 Figure EJ 1 Block Diagram of Array Based VI 0 cce cece cece eee e eee eee eae eneneenens 169 Figure EJ 2 EMG Array Based Amplitude Estimation VI cece eece scene ence ee eeennenes 171 Figure EJ 3 Module that Averages Data from Multip
6. 0 4 a5 0 6 Mean 0 35 sec Standard Deviation 0 19 sec 162 Priority High Moving Histogram a en N an o i fi ni Number of Occurrence n f 20 if 0 1 0 2 0 3 0 4 0 5 0 6 Time sec o a Mean 0 36 sec Standard Deviation 0 19 sec Priority Real Time Moving Histogram 100 Number of Occurrence Time sec Mean 0 56 sec Standard Deviation 0 04 sec 163 Sampling Frequency 2000 Hz Priority Normal Moving Histogram a Number of Occurrence J 0 1 0 2 0 3 0 4 0 5 0 6 Time sec Mean 0 21 Standard Deviation 0 17 Priority High Moving Histogram 355 Q a O i i e N N ao i i o i Number of Occurrence uo 1 o it i 0 1 0 2 0 3 0 4 0 5 0 6 Time sec o o Mean 0 35 Standard Deviation 0 18 164 Priority Real Time Moving Histogram 80 60 N o I v im amp o a pad 6 4047 s o Oo 3 3 2 o D o 0 1 0 2 i 0 3 Time sec 0 4 0 5 0 6 Mean 0 50 sec Standard Deviation 0 14 sec Sampling Frequency 2500 Hz Priority Normal Moving Histogram 80 gis N I pi y D lt u yu e pas 5 ei f k i amp oO oO on i 0 1 Time sec 0 2 Mean 0 01 sec Std Dev 0 02 sec 165 Priority High Moving Histogram Pa o Z S 50
7. Array Size Taps for all Windows Mean ms Standard Three Filters Priority Deviation Setting ms Normal 0 70 0 78 512 High 0 61 0 79 Real time 0 67 0 92 Normal 1 18 0 77 1024 High 1 10 0 84 1 Real time 1 30 1 94 Normal 225 1 49 1536 High 1 92 1 25 Real time 1 91 1 28 Normal 2 87 1 51 2048 High 2 69 1 39 Real time 2 88 1 41 Normal 0 94 1 17 512 High 0 81 0 95 Real time 0 80 0 92 Normal 1 65 1 19 1024 High 1 50 1 20 10 Real time 1 48 1 04 Normal 2 25 1 38 1536 High 2 07 1 34 Real time 2 04 1 34 Normal 2 96 1 76 2048 High 2 62 1 49 Real time 2 59 1 42 Table 5 3 Experimental Results of Array Based EMG Function 76 5 4 Discussion Upon analyzing the results from the previous section several observations can be made regarding the latency associated with utilizing each analysis method In particular one can see that the point by point analysis version of the EMG algorithm executes at a significantly lower average latency when compared to the two array based method even when an array buffer size of one is utilized By looking at the average latency and standard deviation of these data collected during the point by point analysis routine preliminary results suggest LabVIEW may be suitable for low latency applications First it is rather apparent that as background research initially suggested that the EMG amplitude estimation VI utilizing point by point analysis exhibits the lowest la
8. 22 23 24 25 DeLuca Carlo Surface Electromyography Detection and Recording DelSys Incorporated 2002 Description of Performance Options in Windows lt http support microsoft com kbid 259025 gt Duwairi B Combined Scheduling of Hard and Soft Real Time Tasks in Microprocessor Systems Department of Electrical and Computer Engineering Iowa State University lt http www informatik uni trier de ley db conf hipc hipc2003 html gt Earnshaw Garret National Instruments Applications Engineer Interview 2 18 2006 Englehart Kevin A Robust Real Time Control Scheme for Multifunction Myoelectric Control IEEE Transactions of Biomedical Engineering Vol 50 No 7 July 2003 Fuld Matthew Design of a System for Data Acquistion and Computer Control of a Pulmonary Physiology Lab lt http orchid hosts jhmi edu lunglab CVs mfuld_senior_design pdf gt Getting Started With LabVIEW Apr 2003 lt http ni com gt Jovanov Emil Pedro Gelabert Reza Adhami Bryan Wheelock and Robert Adams Real Time Holter Monitoring of Biomedical Signals 1999 LabVIEW Analysis Concepts March 2004 lt http ni com gt LabVIEW User Manual Apr 2003 lt http ni com gt Lyons E C Development of a Hybrid Hands off Human Computer Interface Based On Electromyogram Signals and Eye Gaze Tracking 2001 Proceedings of the 23 Annual EMBS International Conference
9. 3 49 msec 143 Priority High Moving Histogram Number of Occurrence 100 2 0 6 0 10 0 12 0 14 0 16 0 18 0 20 0 22 0 24 0 26 0 26 0 30 0 32 0 34 0 36 0 36 0 Value Mean 2 18 msec Standard Deviation 3 62 msec Priority Real Time Moving Histogram 100 Number of Occurrence I 1 i 1 1 l 1 1 i 1 I i 1 1 1 00 20 40 60 80 100 120 140 160 180 200 220 240 260 28 0 30 0 320 34 0 36 0 Value Mean 2 13 msec Standard Deviation 3 33 msec NOTE Can not use more then 512 Taps when using Express VIs 144 Sampling Frequency 500 Hz Number of Samples 10 Whitening Filter Coefficients 512 Priority Normal Moving Histogram 100 Number of Occurrence 2 0 6 0 10 0 12 0 14 0 16 0 18 0 20 0 22 0 24 0 26 0 26 0 30 0 32 0 34 0 36 0 36 0 Time msec Mean 2 39 msec Std Dev 3 60 msec Priority High Moving Histogram 100 Number of Occurrence 1 i 1 I 1 1 1 1 1 i 1 l 1 1 1 1 1 1 1 1 1 00 20 40 60 80 100 12 0 14 0 160 180 20 0 22 0 24 0 26 0 28 0 30 0 32 0 34 0 36 0 38 0 40 0 Time msec Mean 2 71 msec Standard Deviation 3 72 msec 145 Priority Real Time Moving Histogram 100 80 60 40 20 Number of Occurrence 0 j I I 1 L I I j I 1 1 1 00 62 0 40 60 60 10 0 12 0 14 0 16 0 160 20 0 22 0 24 0 26 0 26 0 30 0 32 0 34 0 36 0 36 0 Time msec Mean 2 40 msec Standard Deviation 3 60 msec NOTE C
10. Moving Histogram 100 Number of Occurrence 1 0 0 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 1 0 Time msec Mean 0 19 msec Standard Deviation 0 39 msec 126 Priority High Moving Histogram 100 Number of Occurrence 0 1 0 2 0 3 0 4 1 0 5 Time msec 0 6 0 7 0 8 0 9 1 0 Mean 0 16 msec Standard Deviation 0 37 msec Priority Real time Moving Histogram 100 Number of Occurrence 0 1 0 2 0 3 0 4 I 0 5 Time msec 0 6 0 7 0 8 0 9 1 0 Mean 0 18 msec Std Dev 0 38 msec 127 Whitening Filter Coefficients 2048 Priority Normal Moving Histogram N i Number of Occurrence 80 0 0 0 I i i j i 1 1 i j ii ji 1 li i KUTUT aby N UA sure G OA OF ti att E a LA S hy sy lhe eG NA Time msec Mean 0 23 msec Standard Deviation 0 44 msec Priority High Moving Histogram 100 Number of Occurrence I 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 1 0 Time msec Mean 0 18 msec Standard Deviation 0 38 msec 128 Priority Real time Moving Histogram 80 60 o uy o l pa 6 40 he V oO oO Si 0 0 0 1 0 2 0 3 i 0 5 Time msec 0 6 0 7 0 8 0 9 1 0 Mean 0 22 msec Standard Deviation 0 41 msec 129 Iteration 100 Intervals 25 Channels 2 Array Based Method Smoothing
11. ad 4 2 0 i 0 0 0 1 0 2 0 3 0 4 0 5 0 6 Time sec Number of Occurrence Mean 0 25 sec Standard Deviation 0 20 sec Priority Real Time Moving Histogram 357 302 v 25 1 3 20 4 ISe 102 5 0 1 0 0 0 1 0 2 0 3 0 4 0S 0 6 Time sec rrencs Number of Oc Mean 0 33 sec Std Dev 0 22 sec 160 Iteration 100 Intervals 25 Channels 4 Computation None Sampling Frequency 500 Hz Priority High Moving Histogram ww an oO i I n i 2 20 1 1 a tt D k pui peg e3 Pas 5 pe G f gt uo I 0 ij i j i I i I i i i i i 000S GIO AS oped Paste mie aS aM Me yl SS Tea GS eal 7A S0 sieges Hal belay dan pin Eas Time sec Mean 4 19 sec Standard Deviation 3 23 sec Priority Real Time Moving Histogram Number of Occurrence 1 i i 0 5 1 0 ie 2 0 2 5 3 0 oi 4 0 45 5 0 5 5 6 0 6 5 7 0 tide 8 0 6 5 9 0 Time sec Mean 5 84 sec Standard Deviation 2 27 sec 161 Priority Normal Moving Histogram 30 N on I 20 Number of Occurrence 0 0 1 0 2 0 3 0 4 0 5 0 i 6 0 Time sec 8 0 10 0 12 0 Mean 3 15 sec Standard Deviation 3 40 sec Sampling Frequency 1000 Hz Priority Normal Moving Histogram 20 on I Number of Occurrence m n o I 1 o ol ca 0 1 0 2 i 0 3 Time sec
12. ccccccccecceeeceeeceeceeeaeceseeeseeeeaeecnaeeneenaes 23 2 4 3 SCO UMN iiine oni weeds Fades ERE E yaa ER T Ws SS 24 2 4 3 1 General Scheduling Approaches cccccccccesscecessceesseceeeseeeesseees 25 2 4 3 2 Scheduling in Real Time Systems 0 ccccccessccceseecsseceesseeeeseees 25 2 4 3 3 Scheduling in Windows XP u00 cccccccccceeseecessceceseeeceneeecsseeeesseeeesseees 26 2 5 EXisting Syste 2556 cs nanara a a a aa Nee 28 2 5 1 Hardware Oyeryie Wenus nn ornis asa E E TARS 28 2 5 2 Software OVERVIEW 3 565565 35 ota iaesh BAS UB AeA BHR 30 2 5 3 DISCUSSION a n hahha in dena Re A OR AS 33 2 6 Programming in LabVIEW 55 235 cess soe Sises gn aee step ees esses 33 2 0 1 BASICS niun e e e a a as aay sande shod eh doce aaia 34 2 6 2 Calling Code from Other Languages cccccccscccessceceseeeesseeeesseeeesaes 36 2 6 3 Linking MATLAB with LabVIEW 0 ccccccccccsseesceeseeseeesecnseeeeeseeneees 36 2 6 4 Linking C code with LabVIEW 00 0 ccccecccceccceceesceceteceeeeeeeeesaeecnaeenteenaes 38 2 6 5 LabVIEW Analysis Methods cccccccessccssseecesccecesseecsseeecsseeeesseeeesaes 39 2 7 Chapter SamMMaey is oiiire a eee 43 3 Problem CALEMNC ING 5 855 255 Seay Raya Aeneas Gave 44 vii 3 1 Problem Statement and Project Goals 0 cccccccccccccesceceseeeceseeecsseeeesseeeess 44 3 3 Objectives and Tasks ioc cc s sccsssaonesseadtt ga certiwestaaisees dan tan e
13. 0 0 0 1 0 2 0 3 0 4 0 5 0 6 Time sec Mean 0 48 sec Standard Deviation 0 15 sec 152 Sampling Frequency 2500 Hz Priority Normal Moving Histogram Time sec Mean 0 01 sec Standard Deviation 0 02 sec Priority High Moving Histogram rari v 20 2 5 15 paj o Teh les Oo 5S 2 0 1 0 0 0 1 0 2 0 3 0 4 0 5 0 6 Time sec Mean 0 22 sec Standard Deviation 0 19 sec 153 Priority Real Time Moving Histogram a5 30 o Cc 25 currenc N I Number of Oc i 0 0 0 1 0 2 0 3 0 4 0 5 0 6 Time sec Mean 0 34 sec Standard Deviation 0 22 sec 154 Iteration 100 Intervals 25 Channels 2 Computation None Sampling Frequency 500 Hz Priority Normal Moving Histogram 35 7 Q an O I I 2 a gt 1 1 v J E o ot yu O iie 5 o fe a 1 i 1 1 0 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 3 0 10 0 11 0 12 0 Time sec Mean 4 54 sec Standard Deviation 4 22 sec Priority High Moving Histogram 30 N N on i I I Number of Occurrence n uo I Li 1 i i 1 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 10 0 11 0 12 0 Time sec o oa Mean 3 37 sec Standard Deviation 2 64 sec 155 Priority Real Time Moving Histogram t L v pai T 20 fa 815 S 3 10 E 2 5 0 i i 0 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 10 0 11 0 12
14. 3 S 5 25 O 2 0 T 0 0 D 0 2 0 3 0 4 OS 0 6 Time sec Mean 0 34 Standard Deviation 0 21 Priority Real Time Moving Histogram Number of Oc i 0 0 0 1 0 2 0 3 Time sec 0 4 0 5 0 6 Mean 0 34 Std Dev 0 21 166 Iteration 100 Intervals 25 Channels 1 Computation 1 FIR LP Windowed Filter no window cutoff frequency 0 125 Sampling Frequency 200 Hz Taps 500 Priority Normal Moving Histogram Penia 20 n i i pi y 5 u y o pas 5 D P n I Ji i i i i i 1 1 i 1 I 5 0 10 0 15 0 20 0 25 0 30 0 35 0 40 0 45 0 50 0 Time sec 9 Mean 13 70 sec Standard Deviation 14 59 sec 167 Taps 5000 Priority Normal Moving Histogram Number of Occurrence 1 1 i 1 1 I i 1 i 1 1 I 0 0 5 0 10 0 15 0 20 0 25 0 30 0 3 0 400 45 0 500 55 0 Time sec Mean 14 01 Standard Deviation 13 77 Taps 50000 Priority Normal Moving Histogram 70 60 o 5 50 3 40 45 30 V ael 2 10 0 0 0 1 0 2 0 30 4 0 5 0 6 0 7 0 8 0910111213141516 1718192021 22232425 2627 28 2 9 3 0 Time sec Mean 1 24 sec Standard Deviation 54 sec 168 Appendix H Detailed Description of EMG Amplitude Estimation VIs In section 5 1 of the report a high level overview of the three different EMG amplitude estimation VIs which each util
15. if isempty varargin 1 0 Order varargin 1l end if isstr Order error Fourth argument Order not numeric end if length Order gt 1 error Fourth argument Order not a scalor end if Order lt 0 error Fourth argument Order must be non negative end end sCalVec 2 sCal 100 2 sCal 100 2 sCal Default if length sCalVec 100 error Internal sCalVec length error end if ArgsVar gt 1 amp isempty varargin 2 sSpec varargin 2 if isstr sSpec error Fifth argument sSpec not numeric end SizeRes size sSpec if length SizeRes gt 2 SizeRes 1 1 amp SizeRes 2 1 error Fifth argument sSpec not a vector end if length sSpec 3 error Fifth argument sSpec not length 3 end if sSpec 1 lt 0 0 error sSpec 1 less than or equal to zero end if sSpec 2 lt 0 0 error sSpec 2 less than or equal to zero end if sSpec 3 lt sSpec 1 error sSpec 3 less than sSpec 1 end sCalVec sSpec 1 sSpec 2 sSpec 3 end Compute magnitude response of the fixed whitener Estimate PSD s of SandN and Noise Also get frequency bins Svv f psd Noise NNfft 2 NNfft round 0 75 NNfft Smm_sCal psd SandN NNfft 2 NNfft round 0 75 NNfft Estimate PSD of JUST signal Snn max Smm_sCal Svv eps sCal sCal Snn ClearEPS Snn Interpolate over the eps values Fixed whitener is 1 SQRT of power spe
16. msec Mean 1 65 msec Standard Deviation 1 19 msec Priority High Moving Histogram 80 a O I Number of Occurrence O 1 I i 3 5 4 0 4 5 5 0 5 5 6 0 6 5 7 0 time msec o N 5 hn Le a N in wo Mean 1 50 msec Standard Deviation 1 20 msec 138 Priority Real time Moving Histogram 80 60 N l o uv V 3 o 6 40 Ge V E 0 I I I 1 1 I I I I I i i iy i I i I t ro r2 tA 16 18 2022 24 26 2030 32 34 ib did 4 0 424 4 4G 4 6 19 0 5 2 0 9 5 0 9 00 0 time msec Mean 1 48 msec Standard Deviation 1 04 msec Whitening Filter Coefficients 1536 Priority Normal Moving Histogram 60 mber of Occurrence N on oO o o I I li 1 o I I i I 1 1 I I I i I 1 I I 1 1 1 1 I I I 1 1 1 LU hie It L6 ra 20 2a at ea eae ol spare isis eshes CNW Ceres Ce G Ae S0 S SA 56 50 60 time msec Mean 2 25 msec Std Dev 1 38 msec 139 Priority High Histogram v 40 D 5 30 uy 3 Size 5 amp 10 2 02 1 0 12 14 16 18 2 0 22 2 4 2 6 2 8 3 0 3 2 3 4 3 6 3 8 4 0 4 2 4 4 4 6 48 50 5 2 5 4 5 6 5 8 6 0 time msec Mean 2 07 msec Standard Deviation 1 34 msec Priority Real Time pace Histogram TE 40 ww T N rt S pi y D lt pui u e t fa x f o UAT ale a Gh aia gern rel mae eae athe Hol shes Por Sha
17. 0 Time sec Mean 4 24 sec Standard Deviation 2 45 sec Sampling Frequency 1000 Hz Priority Normal Moving Histogram on Number of Occurrence oO EN i 0 1 0 2 0 3 0 4 0 5 0 6 Time sec o oa Mean 0 32 sec Standard Deviation 0 20 sec 156 Priority High Moving Histogram a5 30 Cc 25 currence N o I Number of Oc j 0 0 G Lehre 0 3 0 4 0 5 0 6 Time sec Mean 0 36 sec Standard Deviation 0 19 sec Priority Real Time Moving Histogram 100 Number of Occurrence i 0 4 0 5 0 6 Time sec Mean 56 sec Standard Deviation 0 02 sec 157 Sampling Frequency 2000 Hz Priority Normal Moving Histogram Number of Occurrence i 0 0 0 1 0 2 0 3 0 4 0 5 0 6 Time sec Mean 0 15 sec Standard Deviation 0 15 sec Priority High Moving Histogram 20 n I Number of Occurrence 0 0 0 1 0 2 0 3 0 4 0 5 0 6 Time sec Mean 0 31 sec Standard Deviation 0 20 sec 158 Priority Real Time Moving Histogram 100 Number of Occurrence I 0 0 0 1 0 2 0 3 0 4 0 5 0 6 Time sec Mean 0 51 sec Standard Deviation 0 14 sec Sampling Frequency 2500 Priority Normal Moving Histogram 100 Number of Occurrence 0 0 0 1 Time sec Mean 01 sec Standard Deviation 01 sec 159 Priority High Moving Histogram i 125 10 8
18. 5 0 10 0 15 0 20 0 25 0 30 0 Time msec Mean 8 66 msec Standard Deviation 2 26 msec 120 Priority High Moving Histogram 60 50 I 1 1 1 i I I 1 60 9 0 10 0 11 0 12 0 13 0 14 0 15 0 16 0 Time msec Mean 8 62 msec Standard Deviation 1 19 msec Priority Real Time Moving Histogram F 7 Tal 8 0 9 0 10 0 11 0 12 0 13 0 14 0 Time msec Mean 8 53 msec Standard Deviation 95 msec 121 Appendix F Latency Test Results for EMG Algorithm In Chapter 5 of the report information regarding how each EMG amplitude estimation algorithm each utilizing one the three different LabVIEW analysis methods was tested to determine which algorithm had the least latency While summary plots were presented this Appendix contains all of the raw data collected Each algorithm implementing a different analysis method was respectively tested at the Normal High and Real time Windows XP priorities using 512 1024 1536 and 2048 highpass smoothing and whitening filter coefficients Details of the parameters set are shown with the histograms below Furthermore it should be noted that the data that were collected to test these functions was first obtained from the PCI 6229 and then stored in the buffer Then once this data was stored the testing began 122 Raw Data Analysis Method Point by Point Iteration 100 Intervals 25 Channels 2 MVC 6 Smoothing Filter Coefficients High Pass
19. D 0 05 o gt lt 0 512 1024 1536 2048 Filter Taps Figure 5 6 Average Point by Point Latency for Different Priority Levels Therefore it is apparent that the average latency for the point by point algorithm and the standard array based algorithm utilizing a buffer size of one are quite small and indicate that LabVIEW may be an acceptable option to conduct the hard real time biomedical experiments that were previously described However the average latency is not the only statistical figure that must be configured in analyzing these data In particular the standard deviation is also a statistical measurement that is used to analyze data By definition 68 percent of data are captured one standard deviation away from the mean in either direction Moreover between 95 and 99 percent of data are respectively captured two and three standard deviations away from the mean 81 Moving Histogram 80 60 40 20 o i i 0 0 0 1 0 2 0 3 0 4 0 6 0 7 0 8 0 9 1 0 Time msec Number of Occurrence Figure 5 7 Raw Data Histogram Point By Point 2048 Real Time Priority Moving Histogram a O O Q I i I Number of Occurrence Oo 1 5 4 30 2 1 oO O fein 70 2 0 2 2 24 2 6 2 8 3 0 3 2 3 4 3 6 3 8 4 0 4 2 4 4 46 48 5 0 5 2 5 4 5 6 5 8 6 0 6 2 6 4 6 6 6 8 7 0 time msec Figure 5 8 Raw Data Histogram Array Based Buffer Size 1 2048 Real Time Priority The standard deviation for all
20. EMG signals are currently being utilized in the development of artificial limbs and in the gait rehabilitation of stroke and brain injury patients 4 5 16 23 Hence a workstation that can be used to acquire analyze and display EMG and other similar signals can be beneficial to the continuation of research related to biomedical signal analysis A workstation that will be used to acquire and analyze the electrical activity of human muscle may need to operate in hard real time for certain control applications Operation in hard real time refers to the notion that the system must not only perform logically correct computations but also operate under critical time constraints as well 14 The reason that a hard real time system is required is that most of the previously described applications for this system involve feedback in a timely manner For instance signals analyzed by the workstation might be used to control the movement of a prosthesis In order to accurately work like a human limb the prosthesis must move within a certain window of time when it receives the appropriate stimulation and consequently not react after a long delay There are at least two fundamental ways that a system that can be used to conduct various biomedical experiments can be designed The first option is to design an embedded system that can only perform the acquisition task While this is a feasible approach there are several drawbacks to such an approach A s
21. Filter Coefficients Whitening Coefficients Sampling Frequency 500 Hz Whitening Filter Coefficients 512 Priority Normal Moving Histogram 100 80 60 40 Number of Occurrence 20 0s 1 0 0 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 1 0 Time msec Mean 0 12 msec Standard Deviation 0 33 msec 123 Priority High Moving Histogram 100 Number of Occurrence 0 1 0 2 0 3 0 4 1 0 5 Time msec 0 6 0 7 0 8 0 9 1 0 Mean 0 11 msec Standard Deviation 0 31 msec Priority Real Time Moving Histogram 100 Number of Occurrence 0 1 0 2 0 3 0 4 0 5 Time msec 0 6 0 7 0 8 0 9 1 0 Mean 0 10 msec Standard Deviation 0 30 msec 124 Whitening Filter Coefficients 1024 Priority Normal Moving Histogram 100 Number of Occurrence 1 0 0 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 1 0 Time msec Mean 0 13 msec Standard Deviation 0 34 msec Priority High Moving Histogram 100 Number of Occurrence 1 0 0 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 1 0 Time msec Mean 0 14 msec Standard Deviation 0 35 msec 125 Priority Real time Moving Histogram 100 Number of Occurrence 1 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 1 0 Time msec Mean 0 16 msec Standard Deviation 0 37 msec Whitening Filter Coefficients 1536 Priority Normal
22. Input Output Channels 92 Mean Time Until Failure Using 4 Analog Input Channels 7 2 E 5 e Normal co 4 3 8 High o 2 3 Real Time a 2 0 500 1000 2000 2500 Frequency Hz Figure 6 4 MTBF 4 Analog Input Output Channels Consequently for frequencies above 500 Hz it is apparent that LabVIEW is not appropriate for conducting hard real time control experiments as a result of the frequency of data points being missed even without a computational load Moreover increasing priority as shown on these three previous graphs does not guarantee an increase in MTBF but instead decreases the standard deviation from the MTBF While it is apparent that LabVIEW can not conduct hard real time control experiments with sampling frequencies greater then approximately 500 Hz without losing numerous data points preliminary evidence suggested that a PC based system might be able to perform hard real time control applications at low frequencies In particular when the MTBF test was run with no computation at a sampling frequency of 200 Hz with LabVIEW operating with normal priority seventeen hours elapsed and no failures occurred As a result this would seem to indicate that a hard real time control experiment could be conducted at low frequencies However a FIR filter was added as computation 93 to see if these impressive results could remain constant
23. Oct 25 28 Istanbul Turkey Malik Anna Girlie Legacion Roni Estein and Saiful Huq Development of a Prosthetic Arm for Rehabilitation of Arm Muscles Using EMG Signals 2005 lt http www ee umanitoba ca NI 622X Specification lt http www ni com pdf products us 20044546301101dlr pdf gt Perry J Bekey GA EMG Force relationships in skeletal muscle Critical Reviews in Biomedical Engineering 1981 107 26 27 28 29 30 31 32 33 Salini Tranquilli and Prakash Design of Instrumentation for Recording the Electromyogram MQP Project EXC 0302 2003 Stallings William Computer Organization amp Architecture Designing for Performance Pearson Prentice Hall Upper Saddle River NJ 2006 Strom J Real Time Tracking and Modeling of Faces An EKF Analysis by Synthesis Approach lt http www cs ucsb edu cs28 1 b winter2002 papers Strom pdf gt Tanenbaum Andrew S Modern Operating Systems Prentice Hall Inc Upper Saddle River NJ User Manual PCI DAS6402 16 lt www measurementcomputing com gt Using External Code In LabVIEW April 2003 Edition lt http www ni com gt Wall A E and Hidler J M Alterations in EMG Patterns During Robotic Assisted Walking IEEE Engineering in Medicine and Biology April 2004 Xiang Feng Towards Real time Enabled Microsoft Windows Proceedings of the 5 ACM International Conference on Embedded S
24. Time Systems Hard real time systems are characterized by having deadlines that must be met while soft real time operating systems are characterized by having deadlines that should 25 be met Moreover process scheduling in real time systems must be highly predictable and regular In real time systems priority scheduling is the type of scheduling algorithm that is implemented The basic idea of priority scheduling is that each process is assigned a priority After a process is assigned a priority the process with the highest priority will be allowed to execute There are variations regarding how this priority scheduling algorithm is implemented Most real time operating systems utilize a non preemptive priority scheduling algorithm In this scheduling algorithm the highest priority process continually runs until it blocks or voluntarily releases the CPU Then of the available processes the process with the highest priority executes when the CPU becomes available In the case when two processes have the same priority one process is randomly scheduled to execute and the other process has to wait for the other process to finish execution before it can run 29 2 4 3 3 Scheduling in Windows XP The Windows operating system is currently the most widely used general purpose operating system The Windows operating system does not have the same operational criteria as real time operating systems As a general operating system Windows is not
25. allows for a workstation to execute more then one application is the operating system An operating system has two basic functions extending the machine and managing resources In the following sections these two seemingly unrelated functions will be discussed 2 4 1 1 Extended Machine At the machine language level the architecture of most computers is primitive and awkward to program While the designer of an embedded computer system needs to be aware of the low level details of how to program and interact with all of the hardware components of an embedded system an individual using an operating system does not need to be fully aware of these details Thus the function of the operating system is to present the user with the equivalent of an extended machine that is easier to program than the underlying hardware The operating system achieves this goal by providing a variety of services that programs can obtain using special instructions called system calls 29 2 4 1 2 Resource Manager An alternative but equally important function of an operating system is to manage all of the pieces of a complex system Modern computers consist of wide array of devices including processors memories and I O devices The job of the operating system 22 is to provide for an orderly allocation of these various devices among the various programs competing for them Resource management includes multiplexing resources in two ways in time and in spac
26. an input signal was directly passed to an output channel with no processing involved Sampling Frequency Windows Priority MTBF sec Standard Deviation Hz Setting sec Normal 4 09 4 11 500 High 5 21 4 26 Real Time 7 74 3 62 Normal 0 36 0 18 1000 High 0 35 0 19 Real Time 0 54 0 04 Normal 0 20 0 15 2000 High 0 30 0 20 Real Time 0 48 0 15 Normal 0 01 0 02 2500 High 0 22 0 19 Real Time 0 34 0 22 Table 6 1 MTBF Results Obtained Using 1 Analog Input Output Channel 88 Sampling Frequency Windows Priority MTBF sec Standard Deviation Hz Setting sec Normal 4 54 4 22 500 High 3 37 2 64 Real Time 4 24 2 45 Normal 0 32 0 20 1000 High 0 36 0 19 Real Time 0 56 0 02 Normal 0 15 0 15 2000 High 0 31 0 20 Real Time 0 51 0 14 Normal 0 01 0 01 2500 High 0 25 0 20 Real Time 0 33 0 22 Table 6 2 MTBF Results Obtained Using 2 Analog Input Output Channels Sampling Frequency Windows Priority MTBF sec Standard Deviation Hz Setting sec Normal 4 19 3 23 500 High 5 84 221 Real Time 3 15 3 40 Normal 0 35 0 19 1000 High 0 36 0 19 Real Time 0 56 0 04 Normal 0 21 0 17 2000 High 0 35 0 18 Real Time 0 50 0 14 Normal 0 01 0 02 2500 High 0 34 0 21 Real Time 0 34 0 21 Table 6 3 MTBF Results Obtained Using 4 Analog Input Output Channels 89
27. boards do not follow the same conventions Hence the DAQ board chosen to create a PC based system for conducting biomedical experiments does contribute to system performance Consequently we have decided to utilize the PCI 6229 device to perform our latency analysis measurements in order to determine to what extent a workstation can be created to conduct biomedical experiments using LabVIEW and a commodity PC 65 4 5 Chapter Summary In this chapter a comparison between two different DAQ boards was conducted to determine if the choice of DAQ board played an impact on system performance when creating a PC based system to conduct biomedical experiments As our comparison displays all DAQ boards do not follow the same conventions For instance the manner in which DAQ boards trigger a request to transfer data between the DAQ board and PC is unique for different boards Moreover every DAQ board does not support simultaneous input and output functionality As a result the DAQ board chosen plays an impact on system performance For our application to create a workstation to conduct biomedical experiments that may involve control experiments the PCI 6229 DAQ board was chosen 66 5 LabVIEW Analysis Methods Comparison In section 2 6 the notion that LabVIEW has different analysis methods was discussed In particular Lab VIEW has array based analysis and point by point analysis Furthermore array based analysis can be further subdivided int
28. can be seen in the following sections 24 2 4 3 1 General Scheduling Approaches A key issue related to scheduling is when to make scheduling decisions There are various instances when a schedule decision needs to be made For instance a scheduling decision needs to be made when a new process is created when a process exits when a process is blocked and when an I O interrupt occurs Moreover scheduling algorithms need to make a decision regarding how they deal with clock interrupts Generally scheduling algorithms can be divided into two categories regarding how they deal with clock interrupts A non preemptive scheduling algorithm picks a process to run and then lets it run until it blocks or until it voluntarily gives up the CPU To the contrary a preemptive scheduling algorithm picks a process and lets it run for a maximum of some fixed time If the process is still running at the end of the time interval the process is suspended and the scheduler picks another available process to execute 29 Now with an understanding of when scheduling decisions need to be made scheduling algorithms can now be discussed There a wide variety of scheduling algorithms The most common scheduling algorithms include round robin scheduling first come first served and priority scheduling In the following sections the scheduling algorithms that are utilized in real time operating systems and Windows XP will be discussed 2 4 3 2 Scheduling in Real
29. eens 30 Figure 2 8 Software Flowchart for EMG Workstation 7 cece eee eee ence eee eeeeeeeeeenes 32 Figure 2 9 Example Front Panel and Corresponding Block Diagram 21 ceeeeeeee 35 Figure 2 10 Example of Linking LabVIEW with MATLAB 20 ce ceeeee ee eee eee eene een 37 Figure 2 11 Example of CIN Configuration 31 cece eee eeeeee eee eee eee ee ee ee eee eenes 38 Figure 2 12 Array based Express VI and Point by Point Butterworth Filters 43 Figure 4 1 VI That Acquires Data in Background Continuous Mode eceeee eee eens 52 Figure 4 2 VI That Acquires Data in Background Continuous Mode for Use with Point By Point Functions sc iese9 2 ei edennacy to na sev bile Gave scbeuled sau EE OE OAS 54 Figure 4 3 NI DAQ Assistant 5 000 aei doce cb ecedie bend on gots aria E E E A 57 Figure 4 4 NI DAQ Assistant Configuration Palette 0c ccc eee eee eee eee ee ene ene ee ences 58 Figure 4 5 Simultaneous Input and Output Using PCI 622 oe eee e eee ence een enee 64 Figure 5 1 EMG Amplitude Estimation Block Diagram Utilizing Adaptive Whitening 68 Figure 5 1 VI to Test Latency of LabVIEW Functions c ce eceeee scence neces eee eeeneees 72 Figure 5 2 Latency of EMG Array Based and Point by Point Algorithm Normal Priority 78 Figure 5 3 Latency of EMG Array Based and Point by Point Algorithm High Priority
30. if first value in Snn is minval replace with first non minval value in Snn Snn 1 first 1 end idx length Snn if epsl end length Snn if last value in Snn is minval while Snn idx minval idx idx 1 end s find last good value num_eps length Snn idx for p 1 num_eps 1 Snn idx p Snn idx Snn idx num_eps p Interpolate between last good point and 0 end end for z 1 length eps1 x epsl z Get index of location of minval o9 a x 1 Get index of point ahead b xtl1 Get index of point before if a 0 amp amp b length Snn amp amp b Length Snn 1 if Snn a minval amp amp Snn b minval Snn x mean Snn a Snn b interpolate if Snn a and Snn b are good values else if b length Snn 1 while a 0 amp amp Snn a minval amp amp a 1 a a 1 end find last good value while Snn b minval 117 ol if b length Snn 1 If last value is minval then break break else b b 1 loop to find next value end end temp Snn b Snn a interpolation num_eps b a for p 1 num_eps 1 Snn at tp Snn a temp num_eps p end end end end end return 118 Appendix E MATLAB Script Node Latency Test As stated in the main body of the report it was originally desired to use MATLAB code to implement the gain normalization step of the adaptive whitening procedure However results indic
31. is a graphical programming language it is fairly easy to learn and operate Similarly since the workstation will be run on a Windows PC it can readily be used and adopted by a wide audience Furthermore drivers for data acquisition boards are readily available for use with LabVIEW As a result by using LabVIEW on a Windows PC the need for writing customized hardware drivers was avoided While we used LabVIEW for a hard real time application LabVIEW was originally intended for soft real time applications Moreover while Windows XP allows for the availability of data acquisition board drivers it is not a real time operating system Therefore the objective of this project was to determine the extent to which our created system can be used for a hard real time application In particular our goal was to determine if biomedical experiments can be conducted with LabVIEW with latencies no greater then five milliseconds This five millisecond worst case latency was chosen to ensure that a computer monitor used to display graphical data will have a real time screen refresh rate In order to make this determination a comparison of the various LabVIEW analysis methods was conducted LabVIEW has two standard analysis methods array xiii based and point by point With point by point analysis a datum point is operated on as soon as it becomes available On the contrary with array based analysis all computations are array oriented As a result wh
32. msec Standard Deviation 84 msec 132 Priority Real Time Moving Histogram 6 z E 3 g g f I I 1 I 1 j 20 30 40 50 60 70 60 90 10 0 110 120 130 14 0 15 0 16 0 17 0 180 Te ao gt o o Mean 1 30 msec Standard Deviation 1 94 msec Whitening Filter Coefficients 1536 Priority Normal Moving Histogram 60 g 50 S 40 I 1 5 2 0 2 5 3 0 3 5 4 0 4 5 5 0 5 5 6 0 6 5 7 0 time msec Mean 2 25 msec Std Dev 1 49 msec 133 Priority High Moving Histogram 0 Number of Occurrence N wo 7 o Oo a Z a m D ae D a 3 So b time msec Mean 1 92 msec Standard Deviation 1 25 msec Priority Real Time Moving Histogram N o Q I I I Number of Occurrence o I 50 o F 1 0 1 2 14 16 18 2 0 22 2 4 2 6 2 8 3 0 3 2 3 4 3 6 3 8 4 0 4 2 4 4 46 48 50 5 2 5 4 5 6 5 8 6 0 time msec Mean 1 91 msec Standard Deviation 1 28 msec 134 Whitening Filter Coefficients 2048 Priority Normal Moving Histogram 70 60 I 1 1 1 0 15 2 0 2 5 3 0 3 5 4 0 4 5 5 0 55 6 0 65 7 0 time msec Mean 2 87 Standard Deviation 1 51 Priority High Moving Histogram 80 70 o 2 60 2 5 50 amp 40 5 30 V 220 2 10 o i 1 1 0 LS 2 0 29 3 0 ays 4 0 45 5 0 5S 6 0 6S AO time msec Mean 2 69 msec Standard Deviation
33. of freedom 11 Moreover researchers at the University of New Brunswick have done a tremendous amount of work in the area of prosthetic limb control which can be seen in 4 5 and 16 Hence EMG signals can be used in the development of advanced prosthetic devices that have various degrees of freedom 2 2 2 Gait Rehabilitation Another area in which EMG signals are utilized is in gait rehabilitation for brain injury and stroke patients A typical neurorehabilitation program involves manual assisted treadmill therapy While clinical research has shown that this type of therapy is beneficial it has several limitations Most notably manual assisted treadmill therapy places a significant burden on physical therapists because the rehabilitation procedure 15 requires several physical therapists to physically move the legs of the patient through each step 32 An alternative to manual assisted treadmill therapy is to use robotic devices to automate gait therapy Automating gait therapy has several advantages For instance only one physical therapist is needed for each rehabilitation session Also automated therapy sessions result in longer sessions that are not only safer for patients but are easily reproducible While there are several advantages to automated gait therapy there is a glaring disadvantage The human gait cycle is complicated and the robotic system must limit the degrees of freedom through which the subject is able to wal
34. section of the report In particular the fact that Windows XP utilizes a non preemptive scheduling policy was presented Upon further review of the operating system it is apparent that Windows based systems use clocks with a resolution of one millisecond In particular the time quantum or amount of time that a given thread is allowed to execute before a decision is made to either perform a context switch or continue execution is approximately nine milliseconds Hence if a context switch occurs while LabVIEW is executing a latency of at least nine milliseconds will occur This large latency periodically introduced by the operating system is thus the cause of the low MTBF 13 6 5 Chapter Summary In this chapter we discussed our method for testing the performance of LabVIEW as areal time signal processing system Our results indicate that data samples are missed on the order of every second with no computational load for frequencies greater then 500 95 Hz For frequencies below 500 Hz the MTBF was rather high when there was no computational load However when a computational load in the form of a simple point by point FIR was introduced at 200 Hz the MTBF decreased dramatically These low MTBF are a result of the high latency outliers that frequently occur as a result of Windows XP performing context switches Hence a PC based system utilizing LabVIEW would be well suited for non real time systems or soft real time systems that
35. software application or even in a text based language LabVIEW allows you to write the function in another language and use it in LabVIEW In particular our workstation will use LabVIEW to interact with both C code and MATLAB The following sections will give an overview of how LabVIEW can be linked with C and MATLAB 2 6 3 Linking MATLAB with LabVIEW To use MATLAB in LabVIEW one must create a MATLAB script node inside the block diagram of a VI This node which can be found in the standard LabVIEW functions library first allows the user to define input and output variables The user is then able to connect and wire other functions controls and indicators to these inputs and outputs like any other standard function control or indicator found in LabVIEW After inputs and outputs have been defined the user can write MATLAB code inside the script node using the inputs and outputs that were previously defined in LabVIEW When the VI executes the MATLAB routine entered in the script node will be run and executed in MATLAB 31 36 Figure 2 10 displays an example of how LabVIEW and MATLAB can be linked together In this particular configuration the user is able to enter two numbers on the front panel MATLAB will then compute the product of these two numbers and LabVIEW will display the corresponding result The example also shows that MATLAB can manipulate vectors and matrices that it receives from LabVIEW In this example a spreadsheet fil
36. that our system can acquire process and analyze data in hard real time However since the board must obtain a minimum of 1024 data samples before issuing a data transfer request this will cause a latency to incur before any processing is done Assuming that a sampling rate of 4096 Hz is used with all 32 analog input channels 128 Hz per channel with the minimum count of 1024 one can find that the latency incurred before doing any processing is 1024 4096 1 32 7 8 msec While an initial latency of approximately 8 msec does not seem significant for a soft real time system this latency is quite large for a system dependant on real time control Therefore the minimum count requirements imposed by the Measurement Computing hardware makes the DAQ board seem insufficient to handle hard real time processing To the contrary NI s PCI 6229 is able to transfer a single piece of datum every sampling period using single point hardware timed As a result the National Instruments board does not contain this initial latency before processing 4 3 2 Simultaneous Input and Output As previously discussed it is desired for the workstation that is being created to be utilized in experiments involving control applications As a result it is desired that the DAQ board will be able to perform simultaneous input and output That is while data is being collected it should be processed and then sent to analog output 61 The Measurement Comp
37. that prompts the user to configure the virtual instrument For instance when a Filter Express VI is placed on the block diagram a configuration window opens asking the user to specify whether an FIR or IIR is desired along with other filter configuration parameters The chief benefit of using Express VIs over standard array based functions is that Express VIs are simple to configure and provide numerous different function capabilities in a single module For instance all of the standard array based FIR and IIR functions are available in a single Filter Express VI However despite these benefits Express VIs do have some glaring limitations In particular Express VIs do not allow for the level of parameter control that standard array based functions allow For instance using a standard FIR array based filter the user is able to model a filter by providing an array of filter coefficients However when using Express VIs the user is not able to specify filter coefficients and is limited to selecting cutoff frequency values Similarly Express VIs do not allow for as complex calculations as regular array based VIs For example when choosing the taps for an FIR filter for a standard array based filter the user is able to select any integer number for the size of the taps However with an Express VI the user 40 is only able to select a taps value up to 511 Consequently while Express VIs offer a simpler and more convenient user interface than st
38. the amplitude of the EMG si to give the model of the noise free EMG signal ri The additive measurement noise is represented as the zero mean random process vi and is lastly summed with rj to Si i complete the model of the surface EMG 8 Wi Figure 2 5 Model of EMG Used for Adaptive Whitening Filter 8 Figure 2 6 shows the three stage of the adaptive whitening filter proposed by Clancy and Farry The first stage of this process whitens the noiseless EMG amplitude s along with a filtered version of the additive noise v The second stage optimally estimates the noiseless whitened signal by adaptively removing the noise via a Weiner 11 filter The final stage of this process applies an adaptive gain This gain is determined by the transformations of the EMG signal in the previous steps Adaptive Whitening Filter H e s Hj e Hy e s d s Fixed Whitening Filter H e Scat Adaptive Wiener i Filter H e s Fe 7 Sp le m rR v VSun e s5cu nle re Figure 2 6 Clancy and Farry s Adaptive Whitening Stage of EMG Amplitude Estimation 8 For the purpose of our project this entire process can be implemented as a time varying FIR filter Consequently numerous different FIR filters will be pre compiled and stored in a matrix Each sample is filtered by one of these many FIR filters stored in the matrix The filter cho
39. the results from multiple loop executions is presented along with the mean and standard deviation of these multiple iterations In our testing we varied the sampling rate of the DAQ board 200 500 1000 and 2500 Hz and the Priority Level LabVIEW was running at Normal High and Real time and produced histograms of times between failures for an input signal to write to analog output without any computation We then used these data to calculate MTBF and standard deviation of these data sets Next to determine if the number of channels utilized affected performance we performed the MTBF test described above using one two and four analog input channels Lastly we executed this VI with a sampling frequency of 200 Hz Postulating that the MTBF would be small for this sampling frequency we then added a point by point FIR filter since previous results indicated point by point filters induced the lowest latency with varying lengths of filter 87 coefficients 500 5000 and 50000 to determine the effect on the MTBF The results calculated from these experiments can be viewed in the following section 6 2 Results The following section displays the results that were garnered by conducting the tests described in the previous sections Tables 6 1 6 2 and 6 3 respectively display the results that were garnered when the MTBF test was conducted when using 1 2 and 4 analog input channels Again these tests involved calculating the MTFB when
40. the user is able to choose from 18 The next step in generating a VI is formulating the block diagram of the instrument The block diagram is where the VIs are actually defined The block diagram method of coding is not the typical sequential programming that most other programming languages utilize Instead it is a type of object oriented programming known as dataflow programming In this type of programming the user places blocks into their program which represent controls and indicators on the front panel as variables or functions that will work on the data supplied by these blocks The user then uses virtual wires to connect these blocks together This type of programming is known as dataflow programming because the wires are used to represent the flow of data in the program Program execution consequently occurs as data are available on the wires when a block has new data available at its inputs it will process these data and produce an output that can be passed to another function object or stored in a front panel variable 18 34 Similar to the palette of front panel objects there is a palette that contains functions that can be used in LabVIEW block diagrams These functions range from standard mathematical processing such as adding and subtracting to array manipulation to signal processing LabVIEW even provides some advanced functions for generating reports and saving data to a given location on a disk Moreover this palett
41. the user to 70 specify how many times this test is performed Once all of the trials are complete the results are displayed in a histogram and the mean and standard deviation of the results are also presented 71 Place the functions or VIs you want to time here Array of Time ms er A 132 Initial Tick E K Values b QUN A D Ca naasa vy lt disable indexing EEX feta pes esa Figure 5 1 VI to Test Latency of LabVIEW Functions 5 2 2 Analysis Method Comparison Procedure With a method created to measure the time it takes to execute a LabVIEW algorithm an analysis of computational latencies induced by LabVIEW can be conducted First a 100 Hz square produced by a function generator was acquired from two analog input channels at a sampling frequency of 500 Hz for approximately five minutes and stored in a buffer Next using the timed loop VI described earlier in the chapter each EMG amplitude estimation VI was tested by utilizing these stored data During testing we kept the amount of filter coefficients consistent between the smoothing filter high pass filter and whitening filter Moreover the MVC value was set to a constant value rather than calculating it from acquired data This was to ensure that external calls to C and MATLAB code did not affect the elapsed times and that these time delays were strictly due to the DAQ board hardware operating system and LabVIEW processing Also in order to conduct
42. 1 39 msec 135 Priority Real time Moving Histogram 70 Ny woO Ao DH O O O O O I I I i I Number of Occurrence _ I 0 iy ji i i I i I I 20 22 24 26 20 3032 34 J6 30 40 42 4 446 40 5 0 52 54 56 56 60 62 6 4 66 6 070 time msec Mean 2 88 msec Standard Deviation 1 41 msec Sampling Frequency 500 Hz Number of Samples 10 Whitening Filter Coefficients 512 Priority Normal Moving Histogram 70 gt Number of Occurrence N Q gt wo an eae GD E o I I 1 I 1 _ I 1 i 1 1 1 0 0 a5 1 0 1 5 2 0 2 5 3 0 3 5 4 0 4 5 5 0 5 5 6 0 6 5 7 0 time msec Mean 0 94 msec Standard Deviation 1 17 msec 136 Priority High Moving Histogram 60 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 I 1 1 1 0 0 0 2 0 4 06 0 8 10 12 14 16 18 20 22 24 26 26 30 32 3 4 36 38 4 0 42 4 4 46 48 5 0 time msec Mean 0 81 msec Standard Deviation 0 95 msec Priority Real Time Moving Histogram 60 i i I 1 I j I I 00 02 04 06 0808 10 12 74 16 186 Z0 22 24 26 2 0 30 32 J4 36 30 40 4 2 4 4 tG AD S0 time msec Mean 0 80 msec Standard Deviation 0 92 msec 137 Whitening Filter Coefficients 1024 Priority Normal Moving Histogram 80 0 Number of Occurrence 1 1 1 i 34 36 386 40 42 44 46 48 5 0 i i GU Ose Giese reels Se arate aa aca le Sha time
43. 229 also has additional functionality besides acquiring analog signals This DAQ board contains four 16 bit analog output channels When only one of these analog output channels is used the maximum sampling rate is 833 kHz When all four analog output channels are used the maximum update rate is 625 kHz per channel Again there are two methods of transferring analog output between the PC and DAQ board The method of storing the data into a FIFO buffer on the board until a certain buffer count value is obtained is the first option For analog output there is 8191 sample FIFO buffer that is shared with all of the output channels The second option for acquiring data is the hardware timed single point configuration that was described previously Although digital input and output signals will not be used during the course of this project it should be noted that the PCI 6229 has 48 digital input and output lines two 32 bit general purpose timers and a phase locked loop PLL In order to control all of the various functions of the DAQ board this NI device uses six DMA controllers Consequently there is an individual DMA controller dedicated to analog input analog output digital input digital output the general purpose timers and the PLL Also in order to gain access to all of the pins of the DAQ board a SCB 68 noise rejecting shielded I O connector block must be utilized This I O connector block is connected to the DAQ board via the SHC68 68 E
44. 23 Another type of operating system is the real time system These systems are distinguished by having time as a key consideration For instance in control applications real time computers have hard deadlines that must be met For example in an assembly line at a car manufacturing plant if a welding robot welds too late or too early then the car will be ruined Hence if an action absolutely must occur at a certain moment or within a certain time interval the system is deemed to be a hard real time system Another kind of real time system is a soft real time system In these types of system missing an occasional deadline is acceptable For instance an example of a soft real time system would be a live audio video system In this type of system violation of timing constraints can result in degradation of audio and video quality but the system will still continue to operate 14 2 4 3 Scheduling When a computer is multiprogrammed it frequently has multiple processes that are competing for the CPU at the same time If only one CPU is available in a system a choice has to be made regarding which process is to be run next The part of the operating system that makes the choice about which process is to be run next is called the scheduler Furthermore the algorithm that the scheduler uses is called the scheduling algorithm Information regarding general scheduling procedures and scheduling procedures in two specific types of operating systems
45. 6 drray of Coefficients peg Misioa RIE yw a Instance y Figure EJ 7 Express VI EMG Amplitude Estimation Algorithm 177 Point by Point Implementation The final version of the EMG amplitude estimation VI utilizes point by point analysis A block diagram of the stages of this routine can be seen in figure EJ 8 Again this routine is similar to the array based version depicted in Figure EJ 1 Absolute Multiple Channel Average Filter Whitening VI Value Absolute Multiple Channel Point by Point Divide by Value SCRE MVC Lowpass FIR Figure EJ 8 Block Diagram of Point by Point VI In Point by Point Highpass FIR Point by Point Adaptive Point by Point Lowpass FIR Filter Out Filter However the main difference in this algorithm is that a one dimensional array is taken as input Each piece of datum in this array comes from a unique analog input channel The algorithm is then the standard algorithm described in section 5 2 however point by point FIR filters are utilized instead of array based or Express VI filters Moreover the adaptive whitening filter is also modified to become compatible with point by point analysis While the method to calculate the matrix of possible whitening filter coefficients is the same as described in the beginning of this Appendix the adaptive whitening function is slightly modified This slightly modified adaptive whitening module can be seen in Figu
46. AS6402 16 The first DAQ board that was analyzed to determine if the choice of DAQ board affects the ability of a PC based system to operate in hard real time was the PCI DAS6402 16 developed by Measurement Computing In section 4 1 1 a review of the hardware of the board will be given Then in section 4 1 2 the LabVIEW Universal Library function set which is used to configure Measurement Computing DAQ boards will be described 48 4 1 1 Hardware Overview The PCI DAS6402 16 has a total of 32 differential analog input channels with a 16 bit A D resolution and a 200 kHz maximum sampling rate However while there are multiple analog input channels the PCI DAS6402 16 only contains one A D converter Consequently when more then one channel is being utilized to acquire data the sampling rate for each channel is defined as Sampling Frequency Number of Channels Used Hence when all 32 differential analog input channels are used the maximum sampling rate for each channel is 6 25 kHz Acquired data are stored in a 2048 sample First In First Out FIFO buffer located on the DAQ board Data are then transferred using the PCI bus of the computer via direct memory access DMA in either demand or non demand mode Data are transferred from the FIFO to the computer once the amount of samples stored in the FIFO reaches a specified value which is known as the count value The minimum count value for this particular Measurement Computing board is
47. As Figure 6 5 shows when computation was added the MTBF went to not having any failures after seventeen hours to having an average failure in the magnitude of seconds With a simple point by point filter which was shown to induce the lowest computational latency causing such dramatic MTBF changes it is readily apparent that LabVIEW is not optimal for control experiments requiring hard real time control However if an application is capable of missing data points on the order of every second or less then a LabVIEW based system certainly not be used Mean Time Until Failure with a Sampling Frequency of 200 Hz and a Processing Filter 16 14 12 10 8 MDA DD Mean Time Until Failure Seconds oO 500 5000 50000 Number of Taps Figure 6 5 MTBF w Computation Load at 200 Hz Sampling Frequency 94 6 4 Limitation of Windows Operating System As shown in the discussion of our results the MTBF for our system is entirely too small While it may be acceptable for some of the experiments that would be conducted with our workstation for data points to be missed on the order of once every few hours losing data points on the order of every second is unacceptable The reason for the high frequency of data points being missed can directly be attributed to the limitations of the Windows operating second The basic scheduling procedures of Windows XP were discussed in the background
48. Block Diagram of Array Based VI In Express VI FIR Filter Out This routine again utilizes a high pass filter and smoothing filter However instead of using array based filters Express Filter VIs are utilized The parameters of these filters are configured by clicking on the filter blocks in the block diagram and entering the desired parameters in the graphical palette Another notable difference in this implementation is that this VI takes a time stamped waveform containing the analog input from various channels as input A for loop can not be utilized to obtain time stamped data from each channel Consequently in order to separate these data from each channel so that they can be combined the LabVIEW waveform parser function must be utilized In the version of the amplitude estimation function depicted in Figure EJ 7 a parser that only separates the waveform 175 into two channels is utilized However by simply clicking on the parser on the block diagram the user is able to separate more channels if desired Also the same adaptive whitening filter that was detailed in the previous section is utilized in this example However this previously detailed filter takes an array of data in the form of a double data type as input Hence in order to use this module with time stamped data a LabVIEW function that converts a time stamped waveform to a standard multidimensional array of double values is utilized 17
49. Filter Coefficients High Pass Filter Coefficients Whitening Coefficients Sampling Frequency 500 Hz Number of Samples 1 Whitening Filter Coefficients 512 Priority Normal Moving Histogram g 1 I g 1 1 I g i 1 g 1 i 1 I 1 g g 1 0 0 0 2 0 4 06 0 6 10 12 1 4 16 16 2 0 22 24 26 28 30 3 2 34 36 3 6 4 0 42 44 46 48 5 0 Time msec Mean 70 msec Standard Deviation 78 msec 130 Priority High Moving Histogram a 8 8 8 Number of Occurrenc 5 9 1 1 g 1 g 1 i 0 0 0 2 04 06 0 6 10 1 2 1 4 16 16 20 22 24 2 6 28 30 3 2 34 36 3 6 40 42 4 4 46 48 5 0 Tene meee Mean 61 msec Standard Deviation 79 msec Priority Real Time Moving Histogram 5 flee 5 30 vJ 520 fio 2 o 0 0 0 2 0 4 0 6 0 8 10 L2 14 16 L8 2 0 22 24 2 6 28 3 0 3 2 34 3 6 3 6 40 4 2 44 46 4 8 5 0 Teme meee Mean 67 msec Standard Deviation 92 msec 131 Whitening Filter Coefficients 1024 Priority Normal an Histogram 60 40 20 Number of Occurence ah 0 2 04 0 6 0 8 1 0 12 14 L6 L8 2 0 22 24 26 28 3 0 3 2 3 4 3 6 3 8 40 4 2 4 4 4 6 4 8 5 0 Tene mee Mean 1 18 msec Standard Deviation 77 msec Priority High Moving Histogram 100 80 60 ade 20 o i Number of Occurence 0 0 0 2 0 4 0 6 0 8 1 0 12 14 16 18 20 22 24 2 6 2 8 3 0 3 2 3 4 36 3 8 40 42 44 46 48 5 0 Tame ec Mean 1 10
50. Gain Maximum design gain at any frequency scalor B Each row contains the moving average coefficients of the complete adaptive whitener corresponding to the EMG amplitude found in the corresponding element of sCalVec sCalVec Equally spaced vector of EMG amplitudes specifying the amplitudes for which the filters in B are calibrated EMG Amplitude Estimation Toolbox Ted Clancy WPI April 9 1999 Updated 23 January 2004 o Q o99909909099090090 SETEC EEGECETIETIsgIs Process Command Arguments 6 ssssssssSSSSSSSSSS Check number of input output arguments ArgsIn nargin ArgsVar length varargin if ArgsiIn lt 3 ArgsIn gt 9 114 error Bogus number of arguments int2str ArgsIn expected 3 RADI end o Look for option pairs NNfft 256 Default FFT length power of 2 MMaxGain 100 Default maximum fixed whitening gain FlagNfft 0 0 denotes option not yet used 1 gt used FlagMax 0 0 denotes option not yet used 1 gt used NotDone 1 while ArgsVar gt 2 if isstr varargin ArgsVar 1 switch varargin ArgsVar 1 case Nfft if FlagNfft gt 0 error Repeated Nfft selection end FlagNfft 1 if isempty varargin ArgsVar 0 NNfft varargin ArgsVar ol ol ole oe end if length NNfft gt 1 error NNfft not a scalor end if 2 nextpow2 NNfft NNfft error NNfft not a power of 2 3 end case Ma
51. PM shielded cable Further information 56 regarding the hardware specifications of the PCI DAS6402 16 and these two other peripheral components can be found in 24 4 2 2 NI DAQmX LabVIEW Functions The NI PCI 6229 is completely plug and play Consequently there are no switches or jumpers to set on the board and like the Measurement Computing device the hardware is configured using software Since LabVIEW is a National Instruments product this software package contains a library of functions that is used to configure NI DAQ products This library of functions can be divided into two sets The first set of functions can be utilized with older NI E series devices while the second set of functions can be utilized with newer NI M series devices Since the PCI 6229 is an M series device the function set applicable to only M series devices will be discussed This particular function set is referred to as the NI DAQmx function library There are two main ways to configure the PCI 6229 device using the NI DAQmx function library The first choice is to use the DAQ Assistant Express VI The DAQ Assistant Express VI can be seen in Figure 4 3 The NI DAQ Assistant provides a graphical interface for configuring the DAQ board task out error out error in if la timeout sec DAQ Assistant data ER DAQ Assistant Figure 4 3 NI DAQ Assistant 57 By placing the DAQ Assistant on the block diagram and right clicking it with t
52. Project Number ECE EXC 0501 Data Display Acquisition and Feedback System for Biomedical Experiments A Major Qualifying Project Report submitted to the Faculty of the WORCESTER POLYTECHNIC INSTITUTE in partial fulfillment of the requirements for the Degree of Bachelor of Science by Patrick J Bonneau Anthony Fortunato Jesse King Date February 15 2006 Approved Professor Edward A Clancy Major Advisor Abstract Biomedical signals have various research applications in prosthetic limb development and other control applications Consequently a workstation that can be used to conduct biomedical experiments using EMG and other similar signals can be beneficial to the continuation of research in this growing field We have investigated the possibility of creating a PC based workstation to conduct these experiments using National Instrument s LabVIEW Our work suggests that such a system can not be used with experiments that require hard real time control ii Acknowledgements The authors would like to thank the advisor of this project Ted Clancy for his guidance continual support and own personal work towards the advancement and completion of this project The authors would also like to thank Garret Earnshaw for his help in explaining the functionality of the PCI 6229 and Jason Thurlow for his prompt response to all of our questions related to the PCI DAS6402 16 iii Authorship
53. The next two functions that are necessary are the DAQmx Read and DAQmx Write functions The Read function reads samples from the specified acquisition task while the write function writes samples to the specified task The final function that is necessary is the Clear Task function This function will clear a specified task and consequently release all system resources 4 3 DAQ Board Comparison The previous two sections have given an overview of the functionality of the two DAQ boards that were chosen for comparison In this section architectural differences of the two boards will be examined In particular the manner in which each board signals for data to be transferred between the PC and the board as well as how each board handles simultaneous input and output will be discussed 4 3 1 Data Transfer As discussed earlier in this chapter in order for data that is stored in the FIFO of the Measurement Computing PCI DAS6402 16 board to be transmitted to the PC for analysis the board must issue a DMA transfer request The PCI DAS6402 16 issues this request when the onboard FIFO is filled with half of a packet of data The minimum packet size for the PCI DAS6402 16 is 2048 Hence a minimum of 1024 data samples must be collected before the board can generate an interrupt and send data to the PC 30 60 A possible application for our system is for it to be used during experiments that involve real time control As a result it is desired
54. VI routines the point by point routine is shown to execute at a lower latency Moreover this graph displays that Express VIs while convenient incur substantially larger latencies then the other analysis methods Average Latency for 512 Taps EMG Amplitude VI with Buffer Size of One 2 5 2 a E s 1 5 B Normal Priority 5 B High Priority g 1 D Real Time Prioirty D gt lt 05 0 Express VI Pt By Pt Array Figure 5 5 Average Latency for Buffer Size of One Moreover Figure 5 5 indicates that the priority that LabVIEW is executed at does not have a great effect on the average latency of a LabVIEW routine Furthermore running LabVIEW at a higher priority does not guarantee that a given function will execute at a lower latency The idea that running LabVIEW at higher priorities does not guarantee the smallest latency can be seen in Figure 5 6 which displays the average latency of the point by point algorithm running at varying priorities As the graph indicates while the latency incurred while executing the LabVIEW application in real time priority induced a minimum latency when utilizing only 512 taps executing 80 LabVIEW in Windows in real time priority garnered the highest latency when 1024 taps were used Average Latency for EMG Algorithm Using Point by Point Analysis g 0 25 D 0 2 a 0 15 e Normal 5 High J 0 1 z Real Time
55. able to visibly notice slowdown in data being displayed on a monitor when the screen refresh rate approaches 30 Hz As a result convention holds that 60 Hz is an acceptable screen refresh rate for real time systems while a 30 Hz refresh rate is deemed highly unacceptable 28 Consequently the goal of our system is to determine if LabVIEW can be configured to conduct routines that operate with a maximum worst case latency of 5 milliseconds Operation at or below this worst case latency will allow for the screen to be updated at the conventional 60 Hz and thus allow for our system to maintain real time operation 46 3 3 Objectives and Tasks The previous two sections identify the major goals of our project In order to achieve this goal numerous other objectives must be met These objectives can be seen below 1 Determine the effect using different data acquisition cards has on system performance 2 Develop a method that indicates how fast LabVIEW functions operate 3 Develop a method that determines the latency associated with acquiring data via the DAQ board utilizing LabVIEW tasks and indicate if processing can keep up 4 Develop EMG amplitude estimation VIs 5 Determine if Windows XP Priority setting affects system performance The following chapter in the report will describe how we met these objectives and the results that we obtained 3 4 Summary This chapter has described broadly the goal of our project as deter
56. ach design option offers a tradeoff between functionality and maintainability 103 9 Conclusion In order to help continue research related to the study of EMG signals it is helpful to researchers to have a system that can conduct biomedical experiments While there are a myriad of design options available to create a workstation to conduct biomedical experiments our task was to investigate to what extent a PC based system utilizing the Windows XP operating system and the LabVIEW software package can be used to conduct biomedical experiments A possible desired application for our PC based system would be to conduct control experiments that involve simultaneous input and output our work has suggested that using standard LabVIEW in conjunction with Windows XP is not currently suitable for such an application For instance although our PC based workstation was successfully able to input and then output a function sampled at 200 Hz for 17 hours without conducting any processing once computation in the form of a FIR filter was added it was determined that every 14 seconds a datum point was missed Consequently losing points of data so frequently is not acceptable for a hard real time system and therefore it is apparent that using LabVIEW in conjunction with Windows XP does not provide desired results for control experiments Although a PC based system utilizing LabVIEW does not seem appropriate to conduct control experiments this PC base
57. action for any length of time the motor units must be repeatedly activated This repeated activation gives rise to a series of MUAPs that can be modeled as a pulse train in classical signal processing terms This series of MUAPs that is produced is referred to as a motor unit action potential train MUAPT When measured using a surface electrode the electromyogram can be defined as the superposition of numerous MUAPTs firing asynchronously Figure 2 3 reinforces the notion that the superposition of motor unit action potentials gives rise to surface EMG 1 Dirac Delta impulse yeta na Trains Prvetogical a Neuron on i firing Potentials Pyrs Detection Site nit Observed Electrode stem EMG ang Ise Signal Recording Equipment Figure 2 3 Superposition of Motor Unit Action Potential Give Rise to Surface EMG 12 The surface electromyogram signal typically does not exceed 5 10 mV in amplitude with the majority of signal information being contained between the frequencies of 15 and 400 Hz The signal is an interference pattern which grows with muscle effort As a result the amplitude of the EMG contains a great deal of the signal information which can be modeled as a Gaussian random process The EMG amplitude can thus be defined as the time varying standard deviation of the EMG signal and is a measure of the activity level of the muscle under observation 8 The next section contains further details regarding techni
58. al experiments are formulated Now that we know what the system must do the question we ask in this project is Can a hard real time biomedical signal acquisition display and feedback system be implemented on a commodity Windows based PC using National Instruments LabVIEW software If so how functional can we make this system for the end user In particular a PC with 1 68 GHz Intel Pentium 4 CPU and 512 MB of RAM running Windows XP Profession Version 2002 Service Pack 2 is the computer that is being utilized to answer the previously stated question In the following section specifics regarding the hard real time requirements of our system will be discussed 45 3 2 Hard Real Time Requirements As stated previously the goal of our project is to determine if we can develop a PC based system to conduct biomedical experiments in hard real time The experiments that will be conducted with this workstation involve having graphical data displayed on a computer monitor As a result careful attention must be made to the screen refresh rate of the monitor The screen refresh rate refers to how fast the images displayed on a computer monitor are updated Most computer monitors operate with a screen refresh rate of 60 Hz However if the time it takes to output data onto the monitor exceeds approximately 17 milliseconds the screen refresh rate for the monitor will scale down to 30 Hz or one update every 30 33 milliseconds 28 Most humans are
59. al outliers Hence a second test was performed to determine if a simultaneous input and output task executing the previously tested point by point EMG amplitude estimation technique XIV could keep up without losing datum points any datum points do to these occasional outliers Our tests showed that for sampling frequencies above approximately 500 Hz datum points were lost on the order of every fraction of a second without any computation being performed When a sampling frequency of 250 Hz was utilized it was viewed that over seventeen hours passed without any datum point being lost without any computation being conducted However once a simple FIR point by point filter with 500 filter coefficients was introduced as computation a datum point was lost on the order of once every fourteen seconds This final test indicated that our PC based system failed at conducting hard real time applications due to the frequency of datum points being lost with only minimum computation and not even a full EMG amplitude estimation routine However while hard real time experiment can not be conducted with this workstation it is possible to create a data logging application That is data can be streamed to disk and then be operated on offline XV 1 Introduction The electrical activity of human muscle as well as the various mechanical parameters related to human motion can be used in a wide array of research studies For instance electromyogram
60. allow for several data samples to be missed and not the hard real time tasks that we desire to perform 96 7 Using LabVIEW in a Data Logging Application Although the past few chapters have suggested that Lab VIEW is not suitable for the types of hard real time control application we were aiming to implement namely for real time signal processing of EMG signals a PC based system using LabVIEW does have some other potential uses In particular such a system can be used for offline processing In offline processing as the name suggests data are not processed in real time Rather an amount of data are collected and then processed at a later time with delay not being a concern With the NI PCI 6229 a FIFO of 4095 samples is available Consequently approximately one second of buffering can be obtained when a sampling frequency of 4096 Hz With this kind of delay between data coming in some processing in between new buffers of data being available for processing can be conducted The reason that some processing can occur between new packets of data being available is that most processing while not wholly deterministic takes on the order of milliseconds rather than seconds to complete in Windows 15 Another option available using an offline system is to stream recorded data to RAM While data can be saved to hard disk performing such an operation can cause for additional latencies to occur In particular when disk seek operations must b
61. an not use more then 512 Taps when using Express VIs 146 Appendix G Testing to Evaluate LabVIEW Real Time Performance In Chapter 6 of the report information regarding how each EMG amplitude estimation algorithm each utilizing one the three different LabVIEW analysis methods was tested to determine which algorithm had the least latency While summary plots were presented this Appendix contains all of the collected raw data Each test was conducted at the Normal High and Real time priorities Details of the parameters set are shown with the histograms below Furthermore it should be noted that the simultaneous input and output was testing with the PCI 6229 using the single point hardware timed transfer method 147 Raw Data Iteration 100 Intervals 25 Channels 1 Computation None Sampling Frequency 200 Hz Priority Normal 17 hours without a failure Sampling Frequency 500 Hz Priority Normal Moving Histogram 40 30 I pi y o k 5 ot Yu o cs 5 ed f o 5 o a 1 1 2 0 3 0 4 0 5 0 6 0 7 0 8 0 3 0 10 0 11 0 12 0 Time sec pag 5 a o Mean 4 09 sec Standard Deviation 4 11 sec 148 Priority High Moving Histogram 124 i Number of Occurrence oci i 1 j i 1 I 1 i i i if j I j lel they 10 15 20 ioe ha hoe oc nt eon HEL SS GU GSC el ee Wee DS S0 SE UTES LAOIS Time sec Mean 5 21 sec Standa
62. and redisplaying recently captured data 7 Hence this software system allows for the user to create customizable EMG signal processors Moreover the basic architecture of this system can be used to analyze other types of biomedical signals as well In systems where algorithms are executed via hardware adapting a system is very difficult if not impossible In the currently available system a user must merely create new software modules in the same basic form as the other modules and connect them with other modules until the desired processing 31 sequence is formed Thus this PC based system is highly convenient for the user because all customization can be achieved via software 7 When Desired Go to Edit Mode Yes Figure 2 8 Software Flowchart for EMG Workstation 7 32 2 5 3 Discussion The previously described system s design allows for the user to implement EMG signal analysis algorithms via customizable software instead of customizable hardware Although the current system functions it has some limitations Most notably the software is a DOS based program Although DOS can guarantee real time because it can only run one task at a time support for DOS is now very difficult to maintain For instance DOS lacks modern software and features available in other operating systems and drivers for new hardware components are not readily available in DOS 17 As a result in order for this workstati
63. andard array based VIs Express VIs have limited functionality The second type of analysis that is available in LabVIEW is point by point analysis Point by point analysis is a scalar oriented process that analyzes each datum point as soon as it becomes available The buffering associated with array based analysis can potentially cause large delays thus making array based analysis the non ideal method for low latency applications However point by point analysis is designed specifically for low latency applications 3 Table 2 1 outlines further the differences between point by point and array based analysis 41 Data Acquisition and Analysis Characteristic Array Based Analysis with Point By Point VIs Compatibility Limited compatibility with Compatible with real time systems real time systems backward compatible with array based systems Data typing Array oriented Scalar oriented Interruptions Interruptions critical Interruptions tolerated Operation You observe offline You control online Performance and Compensate for startup data Startup data loss does not occur programming loss 4 5 seconds with initialize the data acquisition system complex state machines once and run continuously Point of view Reflection of a process like a Direct natural flow of a process mirror Output a report and an event in real time Run time behavior Delayed processing Table 2 1 Difference Between Array Based and Point By Point Analy
64. ate that MATLAB script nodes produce large latencies and as a result are not desirable in low latency applications Figure EA 1 depicts a test that was done to indicate the large latencies associated with MATLAB script nodes In this test two real numbers were added together The amount of time taking to execute the addition operation was then recorded one hundred times at three different Windows priority levels Normal High and Real time The results of these tests which are displayed in this Appendix show that in the best case operation under real time priority the latency induced by MATLAB was an average of 8 53 msec IN Place the Functions or VIs you want to time here MATLAB Script ount Real pei DBI b lapsed Time sec Dara Coun 4 7 lt disable indexing o 7 ED E st EL Btandard deviation E Moving Histogram 2 raz Histooram bing istogram Intervals 32 Figure EA 1 VI to Test Latency of MATLAB Script Node Adding Two Real Numbers 119 Raw Data Collected The following histograms display the raw data collected from our experiment Each test was executed 100 times for three Windows XP priority levels Normal High and Real Time Histograms displaying the latency recorded for each experiment can be seen below Priority Normal Moving Histogram 60 Number of Occurrence N w on o o 1 i I i I p Ae ies S I DAN pie A
65. ation algorithm with an adaptive smoothing window length Each of these modules and modules to be made in 30 the future are designed so that it contains input and output buffers Consequently any desired sequence of modules can be made by connecting the output of one processing module to the input of another processing module This customized processing configuration is developed by entering specific command line arguments or by having the program read a configuration file when the workstation program is initialized 7 A flowchart of the major activities that the software performs is depicted in Figure 2 8 Once the program is initiated it enters display mode Display mode functions as a continuous loop that can only be interrupted if the user presses a particular button on the keyboard or if the capturing of a section of data has been completed In display mode the computer first acquires a scan of A D samples Then if desired the acquired data can be down sampled Next the data processing specified during the initial program configuration is performed and the display screen is continuously updated showing the most recent data Then outputs are written to any appropriate devices such as D A converters and data can even be saved to the RAM disk Once this loop is ended the program then enters edit mode In edit mode the user is presented with a menu of command options These commands include numerous options such as exiting the program
66. cally necessary in order to avoid saturation 2 1 2 2 Whitening A whitening filter is used to improve EMG amplitude estimation by removing EMG signal correlation and measurement noise A fixed whitening filter is a finite impulse response FIR filter that has the characteristic shape similar to a high pass filter Ideally the optimal shape of the complete whitening and noise rejection filter is a function of measurement noise and muscle activation level However for contraction below 10 of the MVC it is hard to distinguish between signal and noise Adaptive whitening filters that whiten the signal for strong contractions but progressively turn off whitening for weak contractions have been developed These adaptive whitening 10 techniques have resulted in an amplitude estimation improvement of approximately 10 when compared to unwhitened amplitude estimates 8 In order to alleviate as much additive noise as possible Clancy and Farry recently modeled the noise levels of the EMG and modified the adaptive whitening process Figure 2 5 shows the functional model of the EMG proposed by Clancy and Farry The signal wj is a zero mean random Gaussian process that is used as a start for modeling the EMG The signal wi is passed through a shaping filter Hiime to create the signal ni This shaping filter serves to create the low pass characteristic shape of an EMG signal while still maintaining unit standard deviation Next ni is multiplied by
67. cessing while the DAQ is acquiring data In background mode the computer is still able to function while data are being collected The user also must select whether the board will operate in continuous mode or in default mode In default mode the DAQ board will stop acquiring data after one scan In continuous mode the board will continue to acquire data in an endless loop until the acquisition program is halted Lastly it must be selected in software if the board will operate in burst mode When burst mode is enabled the sample data is acquired on adjacent channels at the maximum sample rate This mode is used to minimize time between samples or channel skew on different channels 30 Measurement Computing has provided a library of software functions in C that can be used to configure the previously described settings of the board This function library is referred to as the Universal Library The Universal Library can in turn be 50 utilized in LabVIEW by using the CIN Figure 4 1 depicts a LabVIEW VI that was created that will acquire data from the data acquisition board in background continuous mode The primary function used is AlnScBg This VI scans a range of A D channels in the background and then stores the samples into an array In order to manipulate data as it is being collected another VI needs to be used called GetStatus This virtual instrument takes as input a cluster data structure referred to as context from AlnScBg GetSta
68. csseceesceseeccecesseeceseeecsseeeesaes 109 Appendix B C Code for Choosing An Array of Filter Coefficients 111 Appendix C MATLAB Code to Simulate Adaptive Whitening Filter Parameters 13 Appendix D MATLAB Code that Creates Adaptive Whitening Filter Coefficients EE Pe clei ccoaad earcyatea E E A waded Sates bua said 114 Appendix E MATLAB Script Node Latency Test cccccecceceeseeceseeeeeseeeeees 119 Appendix F Latency Test Results for EMG Algorithm 0 ccccceecceceseeeeees 122 Appendix G Testing to Evaluate LabVIEW Real Time Performance 147 Appendix H Detailed Description of EMG Amplitude Estimation VIs 169 viii Table of Figures Figure 2 1 Motor Neuron Induced Electrical Fields in Muscle Fibers 25 csseeeeseeeeeeeed Figure 2 2 Motor Unit Action Potential 14 EN aE 0 Figure 2 3 Superposition of Motor Unit Action Potential Give Rise to g Surface EMG 12 iaraa 7 Figure 2 4 Standard Six Stage Filtering Process for EMG Amplitude Estimation 10 9 Figure 2 5 Model of EMG Used for Adaptive Whitening Filter 8 0 0 cece eee ee ee eee 11 Figure 2 6 Clancy and Farry s Adaptive Whitening Stage of EMG Amplitude Estimation 8 ccc ccc cece eee n eee e eee e eee nee etna cease eae eee e nae eeeeed 12 Figure 2 7 Block Diagram of EMG Workstation Hardware 7 2 eee ee cece eeeee eee
69. ctrum of JUST signal o Htime min 1 sqrt Snn MMaxGain Assign zero phase o Iteratively compute Wiener filter gain and FIR filters Svv_tilda Svv Htime Htime PSD of noise after fixed whitener for i 1 length sCalVec Hwiener sCalVec i sCalVec i sCalVec i sCalVec i Svv_tilda 116 Htw Htime Hwiener Cascade Htime Hwiener both real B i fir2 Order f Htw First pass no gain scaling AchievedH abs fregz B i 1 NNfft 2 1 Achieved filter magnitude Sinput sCalVec i sCalVec i Snn Svv Input PSD at this EMG amplitude d sqrt mean Sinput mean Sinput AchievedH AchievedH Bia Bigs d Proper gain scaling end return CESECECEEEEEEEEEEEEEEEEEEEEEEEE SESE SEES EEE ES SESEESECEEEEEEEEEEEEEEEEEEES SESESE SS EEEE SESS EEE ESSE SESESESESESSSESS ClearEPS SSSSSSSSS5SSSSSSESEESESSEESES EEEECCCEEEEEEEEEEEEEEEEEEEEEEEE SEES SESE SESE SESESECEEEEEEEEEEEEEEEEESEES The minimum value in input Snn is found in all locations where an eps has been inserted This function replaces this value with a smooth interpolation so that the ensuing inverse does not blow up Originally writen by Punit Prakash 2004 function Snn ClearEPS Snn epsl find Snn min Snn find locations where minimum value occurs in noise minval min Snn store minimum value if eps1 1 1 first find Snn gt minval
70. currently available usually only have one degree of freedom As a result these devices provide nowhere near the amount of control as the original limb which they are intending to replace Through clinical research it has been shown that amputees and partially paralyzed individuals typically have intact muscles that they can exercise varying degrees of control over As a result research is being conducted in regards to utilizing the signals from these intact muscles to control robotic devices with multiple degrees of freedom 11 14 The EMG has been used in two manners in the area of prosthetic limb development The first approach is for the subject to exert a force with a particular muscle This force results in a steady state EMG signal amplitude estimate A degree of freedom of a robotic limb is then moved in proportion to the EMG amplitude This described approach is used in the control of a standard prosthetic gripper that has one degree of freedom 11 The second manner that EMG signals are used involves discrete actions When a discrete action is performed such as the quick movement of the hand or arm the surface EMG is obtained from various muscle cites The temporal structure of the transient EMG activity is then analyzed Upon analyzing the transient EMG activity various movements can be classified For instance a research group at the University of Washington was able to use EMG signals to control a robotic device with four degrees
71. d system can be utilized in data logging applications That is due to the large FIFO buffer on the PCI 6229 LabVIEW can be utilized to acquire and save data on a PC Then after these data have been recorded analysis can be performed on these data using LabVIEW in a non real time application 104 Therefore while LabVIEW appears inappropriate for such applications as controlling a prosthetic limb in hard real time our PC based system can be utilized in other off line experiments However if it is desired to use LabVIEW in a PC based hard real time application NI FPGA and other real time boards can potentially be utilized These boards contain on board processors which are solely dedicated to performing desired real time tasks and thus can potentially eliminate the deficiencies caused by utilizing a computer running a non real time operating system 105 Works Cited 1 2 3 4 5 6 7 8 9 10 11 Basmajian J V and De Luca C J Muscles Alive Their Functions Revealed by Electromyography Williams amp Wilkins Company 1985 Bida Oljeta Influence of Electromyogram EMG Amplitude Processing in EMG Torque Estimation WPI MS Thesis 2005 Campellone Joseph M D MedLine Plus Medical Encyclopedia Electromyography 2004 http www nlm nih gov medlineplus ency article 003929 htm Chan Adrian and Englehart Kevin Continuous Myoelectric Control for Powered Prosth
72. dix A Compiling CIN Source Code Tony Appendix B C Code for Choosing an Array of Filter Tony Coefficients Appendix C MATLAB Code to Simulate Adaptive Whitening Jesse Filter Parameters Appendix D MATLAB Code that Creates Adaptive Whitening Filter Coefficients Appendix E MATLAB Script Node Latency Test Pat Appendix F_ Latency Test Results for EMG Algorithm Pat Appendix G Testing to Evaluate LabVIEW Real Time Pat Performance Appendix H_ Detailed Description of EMG Amplitude VIs Jesse e Final Corrections were performed by Pat and Jesse vi Table of Contents Fabl of Contents 1 02 dsstise cise Grins he ai ea ada i A vii Tableof Tables ireen a E E A EAE EETA x Executive Summary lt 5 Ge dea hd eA eS xi UPS Try ta Ce 8 C0 0 a era ne at Ten ne eS aa R oe Re en pee Ae 1 Ze Background S a ited emit shade teases e ahaa teats 4 2 1 Introduction to EMG iy sicscsecctsceiitvsniaas sa tcaaiaaesssntasassaasnanydcansrisnea catasoiaaa etebameaandcatinae 4 2 1 1 Sources and Characteristics of EMG 00 ccccccceceseceeeceeeeeeeeeenseeneeeeeeeenaees 4 2 1 2 Methods for Amplitude Estimation 0 0 0 0000 0 cc ccccceccceecceeeeeeneeeeeceteeeeeeenaees 8 2 12 11 NOISE RE ECUOM es sciiicsutarrii elastic totais vache das dele ut atthe abhi seek ae 9 21 2 2 WGN seis e aea Pola oma cheasadss Stier aiei andains 10 2 1 2 3 Multiple Channel Combination and Gain Normalization
73. e When a resource is time multiplexed different programs take turns using it Determining how the resource is time multiplexed is the job of the operating system The other kind of multiplexing that the operating system is responsible for is space multiplexing In some instances devices get parts of numerous resources at a time For instance main memory is normally divided up among several running programs or processes Similarly allocating disk space and keeping track of what process uses each disk block is also the job of the operating system 29 2 4 2 Types of Operating Systems In the previous sections the general functions of an operating system were presented However there are a variety of different types of operating systems that are available each with their own strengths and characteristics For instance there are operating systems that are appropriate for use with mainframes servers and multiprocessor computers The most widely known operating system is the Personal Computer Operating System The main job of this operating system is to provide a good interface to a single user Microsoft Windows is the most widely known of this type of general purpose operating system Moreover as previously discussed there are operating systems for embedded systems These operating systems do not provide the simple interface that PC operating systems provide but instead are concerned with deterministic control over multiple routines 29
74. e is read by LabVIEW and stored as a matrix The first row of this matrix is connected to the vecin terminal MATLAB will then compute the sine of this vector and the answer will be displayed in LabVIEW Similarly the entire matrix is manipulated as well Each element inside the matrix in this example is multiplied by a factor of two and then displayed to the front panel Consequently LabVIEW allows the user to utilize all of the standard functions in MATLAB along with custom M files in a VI MATLAB Script Hnum frut mult numi num2 vec sin 2 pi vecin matrix2 matrix1 2 Number 2 osinon Fiza tril bost ine of Input Vector pratrix2 feu D Ea Figure 2 10 Example of Linking LabVIEW with MATLAB 37 2 6 4 Linking C code with LabVIEW Another very useful feature of LabVIEW is its ability to link with C code Although it is convenient to program in the graphical format of Lab VIEW or higher level text formats of MATLAB there are some programming tasks which are simply left better to C One example is hardware interfacing By allowing the use of C code any piece of hardware that can be interfaced with in C can also be interfaced with in LabVIEW Linking LabVIEW with C code is similar to linking with MATLAB First the user must place a Code Interface Node CIN on the block diagram and set up its input and output terminals Next the user must wire the inputs and outputs of the CIN as if i
75. e of block diagram contains execution control and branch constructs that can be used in a given application For instance the user is able to use while and for loops as well as typical case structures that can be found in other text based programming languages Figure 2 9 displays both the front panel and block diagram of an example VI This example VI known as AddandSubtract allows the user to enter two numbers in the front panel The sum and difference of these two numbers are then displayed as output on the front panel The block diagram contains the functions that actually implement the addition and subtraction of the two numbers entered on the front panel i gt AddandSubtract vi Front Panel File Edit Operate Tools Browse Window Help gt m 13pt Application Font ordi i ett s907 j t 4 EE tamini Hp pt Dy gt AddandSubtract vi Block Diagram IES Ble Edit Operate Tools Browse Window Hif Resme E Figure 2 9 Example Front Panel and Corresponding Block Diagram 21 35 2 6 2 Calling Code from Other Languages One of the distinctive features of LabVIEW is that LabVIEW has the ability to link with programs written in text languages and in other software packages As a result functions that have been created on other development platforms can be used in LabVIEW Similarly if it is easier to write a function that is needed in a LabVIEW VI in another
76. e performed latencies can occur that are on the order of tens of milliseconds By streaming data to RAM and then transferring to disk in non real time these latency issues can be averted 30 With data saved in memory later processing by LabVIEW MATLAB or any other sort of processing program can be conduct Hence while it is apparent that LabVIEW is 97 not suitable to conduct hard real time control experiment LabVIEW can be utilized as a data logger for offline applications 98 8 Future Recommendations From our work described in the previous chapters it is apparent that the PC based system that we designed is not optimal for conducting experiments that involve hard real time control In this chapter we will investigate further analysis and platform options that can be chosen to further advance the creation of a work station to conduct biomedical experiments 8 1 Other LabVIEW Design Options While the current system is not capable of meeting our hard real time needs as a result of the Windows operating system it may be possible to still utilize the standard LabVIEW software package to implement our design In particular LabVIEW is now available for the LINUX operating system By using the LINUX version of LabVIEW one can make a determination if a change in operating systems will allow for LabVIEW to be a more affective and viable option While using LabVIEW on another operating system is an option there are other Natio
77. e sampling frequency increases the MTBF drops to levels on the order of milliseconds Mean Time Until Failure Using 1 Analog Input Channel 9 2 g 3 L 6 e Normal c 59 r High 3 Real Time c 2 o 0 500 1000 2000 2500 Frequency Hz Figure 6 2 MTBF 1 Analog Input Output Channel Moreover as Figures 6 3 and 6 4 indicate these low MTBF results are almost identical for utilizing 2 and 4 analog input and output channels These similar results are 91 appropriate due to the architecture of the PCI 6229 As previously discussed there is only a single DMA controller to control the analog input task and a single DMA controller to control the analog output task and only one A D and D A converter for input and output routines As a result for multiple channels the sampling frequency per channel is equal to the total sampling frequency divided by the number of channels These garnered results reinforce this notion and indicates that the DAQ has the same processing delay no matter the amount of analog input and output channels used Mean Time Until Failure Using 2 Analog Input Channels gt w e Normal a High Real Time k ad Mean Time Until Failure sec N ouomH Onawon foo 500 1000 2000 2500 Frequency Hz Figure 6 3 MTBF 2 Analog
78. eaaes ue abode eae 47 JA S MMA ps Sepsis A ratte nto slesteUhe rs Sill oot i deuon ac Sal ed tabetha att 47 4 Data Acquisition Board Performance Analysis 0 cccccccsceceteceeeceeeeeeeeeneeens 48 4 1 Measurement Computing PCI DAS6402 16 0 000 000 cceeceeceeeeeeteeeeeeeteeeees 48 4 1 1 Hardware OVer Vie W sieiissisiistessicaesscssac cits sotaaasetcantaeisd aia sbasia ned siete n iS 49 4 1 2 LabVIEW Universal Library Functions 0 00ccccccccccceesceceseceesseeeeees 50 4 2 National Instrument s PCT 6229 000 000 cccccccceseecssececsseeeceseeecsseeecseeeeeeeeess 55 4 2 1 Hardware OVvervieW iin seccvnsessstsasssivassavrassspheaneeaunteanisbsstensereaneeigeaetamtaan 55 4 2 2 NI DAQmX LabVIEW Functions 0 cccccccccccccsesscceceesseeeceesssceeeensaaes 57 4 3 DAQ Board Comparison iiss isiasinishcctisiiaa vieviectaseavte eas hs ies Hs oe 60 BS Date Transfer encaminan a a aaa E E E O ETS 60 4 3 2 Simultaneous Input and Output ccc ccesccesceceseeecsseeecsseeeesseeeesaes 61 44 DISCUSSIONS in a AR R ooineraassvaesete nega dada R E NERT 65 4 5 Chapter Summary is3 5 esiesyi cd sotediand io oe is does ees 66 5 LabVIEW Analysis Methods Comparison cccccceeceeesseceeeceeeeeeeeeeaeeeaeens 67 5 1 EMG Amplitude Estimation VIS 0 0 0 cccccceccceecceecceenceceaeceeeeeeeeeeseecsaeeneeeaes 67 5 2 Latency Test Procedure sis c4 scsistcnrihoe oadctesesasetis cinaade st ianaeaaes d
79. ective PC based system Similar large prices are apparent for the other RT devices as well 8 2 Other Alternatives While possible alternatives for implementing a PC based system were shown that use additional LabVIEW modules many other alternatives exist each solution possessing unique advantages and disadvantages For instance proprietary systems such as VXWORKS QNX and REAL IX are potential options because they are designed specifically for applications similar to the ones that we wish to conduct While these systems do provide the functionality that is desired their major drawback is that they are expensive and not widely available Another viable option is to utilize a real time Linux based system known as RT Linux RT Linux is a free open source operating system that provides the deterministic results that are the chief characteristic of real time operating systems RT Linux has 101 already been adopted by researchers who are conducting biomedical applications that are similar to the tasks we have attempted to create For instance a research group at the University of Texas developed a RT Linux system to detect and minimize heart arrhythmias by using electrical stimuli on the heart muscle 6 While RT Linux does prove to be a reasonable alternative to LabVIEW the Linux operating system is not as widely used as Windows As a result people would need to become familiar with this new operating system and programming in it if RT L
80. ed to implement the smoothing filter 172 Next the smoothed data are normalized by the MVC and placed in the adaptive whitening VI which is labeled with a 1 on the block diagram Figure EJ 4 depicts the block diagram of this adaptive whitening filter module The function takes as input first the array of data to be whitened It then takes a multidimensional array of filter coefficients and a scaled amplitude value Based on the scaled amplitude value entered one of the vectors of filter coefficients is selected A C code interface node is used to select the vector of filter coefficients The code used to implement this CIN can be viewed in Appendix B ilter Coefficients Scaled Amplitude DBL Figure EJ 4 Array Based Adaptive Whitening VI Next with the appropriate array of filter coefficients selected the LabVIEW FIR filter is utilized This function implements the FIR filter specified by the vector of coefficients entered Consequently with the filter implemented this function can be utilized in the EMG amplitude estimation VI The array of filter coefficients is implemented by a module that utilizes a MATLAB script Figure EJ 5 depicts the block diagram of the module used to construct the array of whitening coefficients In this 173 module two MATLAB functions are utilized cal_sim and e_cal_wh to calculate this matrix of filter coefficients The function cal_sim is a function created by the authors to create a vect
81. eeded 70 S 2t imed Loop VU rise esterase ae OO A Aa 70 5 2 2 Analysis Method Comparison Procedure c cccccesseeseeeeteeeeeeeees 73 5 3 Timed Loop Test Results 3 250505 ac 2c AR ese Bs a eos BOSS 74 5 4 DESCUSS HOM 5535 0s sies easiness sacks n a eet a eed ceda E Wusdad E V Suma Meda 71 5 5 Chapter Summary Rien er ARE a n a en ee G a ia 83 6 Hard Real Time LabVIEW Performance 0cccccccceescecesscecesececeseeecsseeeesseees 85 6 1 Final Testing Procedure MT BB o 0 cece cecccecseccecseeeeceeneeceeeeeceeeeenneeeeeeeeees 85 6 2 RESUS rynna e sens tava Mad a aN eiai 88 6 3 Discussion of Results ccccccccccssccsesseseesceceesceceeeeecsseeeceeeeceeeeceseeecseeeeseeeess 91 6 4 Limitation of Windows Operating System cccccccesecesseeceseceesseeeesseeeees 95 6 5 Chapter Summary Seen eee ne Are Se mera ed ne a ee Sn nea enn 95 7 Using LabVIEW in a Data Logging Application 0 0 cccccccceseceseceeeseeeeseees 97 8 Future Recommendations ccccccccsccecesssececessnececeseeecececeesaeeecseaeeeeesseeaeesesenaes 99 8 1 Other LabVIEW Design Options cccccccccceccecesececeseeecsseeecsseeecseeeesseeeees 99 8 2 Other Alternatives occ ccisciatigevessi ats ian n A R E RENA 101 8 3 Chapter Summarys 36210 canis oss oe 102 9 Concl sion oen eR A E R 104 SVOPKS ted Boo eee e ea eee 106 Appendix A Compiling CIN Source Code ccccccccc
82. ement these three processes vary depending on whether RMS or MAV is being used to provide an amplitude estimate 2 When an RMS amplitude estimate is conducted detection consists of squaring all the points in the signal window Once detection is complete the information is then placed through a smoothing filter A smoothing filter can be implemented as a moving average filter or any other FIR lowpass filter The estimate is then relinearized by taking the square root of the smoothing function output 13 When an MAV amplitude estimate is conducted detection consists of taking the absolute value of each point in the signal window Again once detection is complete the information is smoothed by utilizing either a moving average filter or another FIR lowpass filter When a MAV amplitude estimate is conducted there is no relinearization step 2 2 Uses of EMG With a basic overview of EMG signals presented we will now give an outline of some of the various manners that EMG signals are used in research activities These are by no means all of the applications of EMG signals Rather this section serves to give an idea of the type of experiments that our workstation will help to conduct 2 2 1 Prosthetic Limb Development As previously discussed EMG signals provide a non invasive measure of ongoing muscle activity Therefore EMG signals can be potentially used for controlling robotic prosthetic devices Most prosthetic devices that are
83. emoved In the future 69 this step can be implemented in LabVIEW to obtain a more complete EMG amplitude estimation routine 5 2 Latency Test Procedure In the previous section three modules developed to perform an EMG amplitude estimation routine utilizing three different LabVIEW analysis methods were presented These algorithms represent the type of computation and analysis that will be conducted with our PC based workstation As a result it is important to grasp the latency associated with running a typical processing application and the affect that utilizing different LabVIEW analysis techniques and Windows XP priority levels has on these levels Consequently in this section a testing procedure created to determine the latency associated with running each of these versions of the EMG amplitude estimation algorithm is presented 5 2 1 Timed Loop VI In order to make a comparison between the latency induced using three different LabVIEW analysis methods we have created a VI to measure the time that it takes for a LabVIEW function to execute Figure 5 1 depicts the VI that was formulated to conduct this testing LabVIEW has a module that allows access to a 1 kHz system clock Hence our VI allows a routine to be placed inside of a loop and measures using this system clock the amount of time that it takes for the routine to execute a single iteration Moreover in order to gain statistical results over multiple trials the VI allows
84. en array based analysis is utilized a buffer is first defined Data are only analyzed when a buffer unit is available Choosing which analysis method to utilize in a given application simply involves choosing which function library to utilize in LabVIEW If point by point analysis is desired one must merely choose functions listed in the point by point subsection of the LabVIEW function palette In order to use array based analysis the user has two options First the standard functions available use array based analysis Second the user can use Express VIs Express VIs provide numerous different function capabilities in a single module and can be figured with a simple graphical user interface In order to determine which analysis method induced the smallest latency a module was created to record the time it takes for a LabVIEW routine to operate Three EMG amplitude estimation routines were created using each of the previously identified LabVIEW function routine and were then timed with this module Our results indicated that on average our point by point and standard array based algorithm using a buffer size of one garnered average latencies that were below our maximum latency standards Thus initial results indicated that it seemed possible to create a hard real time control experiment using LabVIEW However while the average latencies induced by these EMG algorithms were below our maximum acceptable latency our statistics showed occasion
85. en each implementation of the EMG amplitude algorithm was that point by point Express VI and standard array based functions were respectively utilized in each implementation of the algorithm Consequently now knowing that the same algorithm was utilized for each routine a discussion of how each stage of the algorithm was implemented can be discussed First the highpass filter stage in all three routines was implemented by utilizing one of the three versions of the FIR highpass filter that is available in LabVIEW Second as discussed previously in the background section of the report a smoothing filter can be implemented as a moving average filter or another FIR lowpass filter In all three modules this smoothing stage was chosen to be implemented by utilizing a standard FIR lowpass filter Third the absolute value stage was simply constructed by taking the absolute value of a sample Fourth the combining stage consists of taking an average of all of the signals that are used in the routine Fifth the normalization step was 68 executed by merely dividing each sample by the maximum voluntary contraction MVC Last the adaptive whitening filter was represented as a matrix of pre processed FIR filters The filter chosen to analyze a given datum point is determined by the current normalized MVC value In all three versions of the EMG amplitude estimation routine the user is able to enter values for the cutoff frequencies for the highpass filt
86. er and the smoothing filter The user also is allowed to enter the value for the MVC and the amount of filter coefficients that will be used with the adaptive whitening filter Also our VI not only outputs the graph of the final signal but it plots the output after each of the nine major steps of the algorithm For instance the user can look at the graph of the waveform after it has passed through the highpass filter or even after it has been whitened Appendix H provides a more detailed view of how each routine was implemented in LabVIEW While three EMG amplitude estimation routines have been created all three routines have some limitations First the gain normalization routine which was discussed in the background section of the report is not included in these functions This step was originally developed using a MATLAB script node and the two MATLAB functions e_uncorr and e_cal_sp which were previously designed by Ted Clancy who advised this project However MATLAB was not designed to be utilized in real time applications In order to support the inability to use MATLAB in real time applications a simple test was conducted to measure the time it takes to execute a MATLAB script node computing the simple addition of two real numbers The results of this test can be viewed in Appendix E Therefore in order to obtain a more realistic comparison of the latencies utilizing each of the different analysis methods this gain normalization step was r
87. esigned to plug directly into a personal computer s bus with all the power required for the A D converter and associated interface components being directly obtained from the bus Moreover it should be noted that a DAQ board is more than a simple A D function on a board A data acquisition may include discrete digital bi directional I O lines counter timers and D A converters for outputting analog signals for control applications The purpose of a DAQ board is to convert analog data to digital data that a computer is able to manipulate There are three primary methods available to transfer digitized data between the DAQ board and computer memory These three methods for data transfer are direct memory access DMA interrupt and programmed I O transfers For programmed I O transfers data are transferred between the CPU and the PC whenever the CPU receives a software code to either acquire or generate a single data point Interrupt data transfers occur when the DAQ board sends an interrupt to the CPU This interrupt causes the CPU to either read acquired data from the DAQ board or write data to the DAQ board DMA transfers use a DMA controller instead of the CPU to move acquired data between the board and computer memory Even though high speed data transfers can occur with interrupt and programmed I O transfers they require the use of the CPU to transfer data Consequently DMA transfers are able to acquire data at high speeds and keep the CPU f
88. esis using Hidden Markov Models IEEE Transactions on Biomedical Engineering Vol 52 No 1 January 2005 Chan Francis Fuzzy EMG Classification for Prosthesis Control IEEE Transactions of Rehabilitation Engineering Vol 8 No 3 Sept 2000 Christini David J Kenneth M Stein Steven M Markowitz and Bruce B Lerman Practical Real Time Computing System for Biomedical Experiment Interface 1998 Clancy Edward A PC Based Workstation for Real Time Acquisition Processing and Display of Electromyogram Signals Biomedical Instrumentation amp Technology 1998 Clancy Edward A and Farry Kristin A Adaptive Whitening of the Electromyogram to Improve Amplitude Estimation IEEE Transactions on Biomedical Engineering Vol 47 No 6 June 2000 Clancy E A Bouchard S and Rancourt D Estimation and Application of EMG Amplitude During Dynamic Contractions IEEE Engineering in Medicine and Biology Vol 20 No 6 Nov Dec 2001 Clancy E A Morin E L and Merletti R Sampling Noise Reduction and Amplitude Estimation Issues in Surface Electromyography Journal of Electromyography and Kinesiology 12 2002 Crawford Beau Real Time Classification of Electromyographic Signals for Robotic Control Technical Report No 2005 03 05 Department of Computer Science University of Washington March 2005 106 12 13 14 15 16 17 18 19 20 21
89. experiments conducted can be seen in Tables 5 1 5 2 and 5 3 As these data indicate the standard deviation for all of the point by point experiments is quite small on the order of fractions of milliseconds However the standard deviation from the mean for the other experiments is quite large giving evidence that there should be a concern for outliers Outliers are defined as datum points that are far away from the rest of a set of data Figure 5 7 and 5 8 displays the raw data collected when 2048 taps were utilized for the array based algorithm with a buffer size of one and for the point by point routine when the priority of LabVIEW was set to real time While the average latency for this point by point experiment was computed to be 82 22 milliseconds the raw data indicates that outliers of one millisecond are possible For our hard real time requirements this is not a concern However when examining the array based routine using a buffer size of one the concern of outliers can be seen Although the average latency for this routine was 2 88 milliseconds which was below our defined maximum latency outliers can be viewed that are as great as 7 milliseconds As this example shows while the average latency is an important characteristic of our system the worst case latency is even more critical Although our algorithms may execute quicker then our average latency most of the time a large latency that exceeds our maximum latency indicates sy
90. experiments under the same conditions LabVIEW was the only foreground application running on the PC while testing was conducted For each EMG amplitude estimation routine the recorded square waves were used as input Next using 512 filter coefficients all three filter values the time loop VI was executed 100 times The default priority for a process in Windows XP is Normal Consequently in order to determine the effects of using different priorities has on the latency induced by the EMG algorithm these testing procedures were performed using high and Real Time priorities as well Moreover in order to get results using increasing amounts of filter coefficients this procedure was repeated using filter coefficient values of 1024 1536 and 2048 Again each of these various tests were conducted at all three 73 priority levels Furthermore while the routine implementing point by point analysis operates on one piece of datum every interval the array based and Express VI routines operate on packets of arrays Consequently for these two routines two different array sizes were utilized array sizes of one and ten While the processing routines were configured to process data as quickly as possible these graphs were configured to update after every ten datum points were analyzed 5 3 Timed Loop Test Results The results of these various tests can be seen in Tables 5 1 5 2 and 5 3 In particular Table 5 1 indicates all experimental results
91. fferential input channels was used Consequently the hardware used in this existing EMG workstation does not require the formation of customized circuits that perform algorithms in hardware Rather the system is PC based and all of the computing hardware is readily available for purchase As a result all of the customizable algorithms can be developed in software Now that the hardware components of the EMG workstation have been described this customized software can be discussed 29 Subject Reference Signal Gain Up to Sixteen k i 1 25 EMG Variable for Electrode each channel Amplifiers Figure 2 7 Block Diagram of EMG Workstation Hardware 7 2 5 2 Software Overview The design of the current EMG workstation allows the user to design custom EMG algorithms via software The software developed for the workstation was programmed in Borland C version 4 5 and all software is run under DOS It should also be noted that all interaction with the A D board was programmed in C using a software function library provided by the board manufacturer The EMG workstation software program works as a configurable state machine Generic signal processing and EMG analysis modules can be connected in any desired sequence to create a customizable signal processor For instance there is a module called SumDif that performs simple addition and subtraction and another module named MAVadp that implements an advanced EMG amplitude estim
92. ficult Also there are several EMG amplitude estimator systems that rely on the development of custom hardware to implement their algorithms Such customized hardware systems have several drawbacks First it is not only fairly expensive but often difficult to duplicate such systems Systems based on customized hardware are hard to modify for different applications For instance while the basic signal acquisition systems are very similar an individual using hardware to develop an EMG workstation would have to develop an entirely new system to analyze ECG signals Similarly when new EMG amplitude estimation algorithms are developed the entire hardware configuration of the system will have to be modified to accommodate the new algorithm 7 2 3 2 PC Based System Another option available for creating a system to acquire and analyze biomedical signals is to develop a system that is based around a standard PC A PC based system is comprised of a standard desktop computer that contains signal analysis software However in order for the computer to acquire the biomedical signals that are analyzed via software a commercially available data acquisition DAQ board must be installed in the computer 19 2 3 2 1 Data Acquisition Board Fundamentals A DAQ board is a basic A D converter that is coupled with an interface that allows a PC to control the actions of the A D and capture the digital output information from the converter A DAQ board is d
93. from the point by point EMG function Table 5 2 indicates all experimental results from the Express VI function and Table 5 3 indicates all experimental results from the standard array based VI It should be noted that for the Express VI version of the algorithm test results are only displayed for utilizing 512 taps for all three filters The reason for this limitation is that 512 is the maximum amount of filter coefficients that are allowed to be used by this function The raw results from all of the experiments described in the previous section can be seen in their entirety in Appendix F 74 Taps for all Windows Mean Standard Three Filters Priority Setting ms Deviation ms Normal 0 12 0 33 512 High 0 11 0 31 Real time 0 10 0 30 Normal 0 13 0 34 1024 High 0 14 0 35 Real time 0 16 0 37 Normal 0 19 0 39 1536 High 0 16 0 37 Real time 0 18 0 38 Normal 0 23 0 44 2048 High 0 18 0 38 Real time 0 22 0 41 Table 5 1 Experimental Results of Point by Point EMG Function Array Windows Mean ms Standard Size Priority Deviation Setting ms Normal 2 36 3 49 1 High 2 18 3 62 Real time 2 13 3 33 Normal 2 39 3 60 10 High 2 71 3 72 Real time 2 40 3 60 Table 5 2 Experimental Results of Express VI EMG Function where 512 Taps were used For All Three Filters 75
94. gain when another full buffer unit of data becomes available 20 An obvious problem with this method of analysis is that long delays can occur during simple processing routines For example suppose that 100 samples of a signal sampled at 1000 Hz are captured and then passed through two cascading array based FIR filters The delay induced from signal passing alone for this example would be 3 100 samples 1000 Hz 0 3 seconds Consequently for time critical tasks such a large delay without even performing any analysis can be unacceptable However it should be noted that if a smaller buffer size is utilized such as a size of one this large signal passing latency can be decreased However while there are only two analysis methods available in LabVIEW array based analysis can be performed two different ways The first implementation of 39 array based analysis uses standard LabVIEW functions that take in a standard array of data input The second implementation of array based analysis uses VIs that are referred to as Express VIs Express VIs take time stamped waveforms as input parameters instead of arrays of numeric data types A time stamped waveform is an array of data that has each piece of datum stamped with an associated time value from the clock of the computer Moreover an Express VI unlike standard LabVIEW functions is configured by a graphical interface Once an Express VI is placed on the block diagram a window opens
95. h Hebe lO Cire Cet Cece Css yl ieee safle S0 60 time msec Mean 2 04 msec Standard Deviation 1 34 msec 140 Whitening Filter Coefficients 2048 Priority Normal Moving Histogram 70 60 1 1 1 1 1 i I 1 I 1 1 10 15 20 25 30 35 40 45 50 55 60 65 time msec 1 1 1 1 1 1 I 1 1 7 0 75 680 65 90 95 10 0 10 5 11 0 Mean 2 96 msec Standard Deviation 1 76 msec Priority High Moving Histogram 80 a I Number of Occurrence w EN o i Li 1 1 i 1 4 0 4 5 5 0 5 5 6 0 6 5 7 0 7 5 8 0 time msec o l 5 ee in N o w in w o w in Mean 2 62 msec Standard Deviation 1 49 msec 141 Priority Real Time Moving Histogram 100 80 60 40 20 Number of Occurrence 0 1 1 fi 1 0 L5 2 0 2 5 3 0 3 5 4 0 i 4 5 time msec I 5 0 5 5 6 0 6 5 7 0 7 5 8 0 Mean 2 59 msec Standard Deviation 1 42 msec 142 Iteration 100 Intervals 25 Channels 2 Express Method Smoothing Filter Coefficients High Pass Filter Coefficients Whitening Coefficients Sampling Frequency 500 Hz Number of Samples 1 Whitening Filter Coefficients 512 Priority Normal Moving Histogram 100 80 60 40 20 Number of Occurrence 0 1 I i I 1 j I J 1 1 I 1 1 I 1 I 1 CO 20 40 60 00 100 120 14 0 160 16 0 00 220 240 26 0 20 00 Z0 A0 60 H0 yalue Mean 2 36 msec Standard Deviation
96. half of the size of the FIFO which is 1024 data samples 30 The PCI DAS6402 16 also has additional functionality besides acquiring analog signals This DAQ board contains two 16 bit analog output channels Each of these analog output channels is able to simultaneously output data at a maximum rate of 100 kHz per channel Analog output is also stored in the 2048 sample FIFO and like analog input data are transferred out of the FIFO buffer once a specified count value is reached Again this count value must be at least half of the size of the FIFO Although digital input and output signals will not be used during the course of this project it should be noted that the PCI DAS6402 16 four digital input and output channels This board 49 utilizes a single DMA controller to control all of the various functions of this board Further information regarding the hardware of the PCI DAS6402 16 can be found in 30 4 1 2 LabVIEW Universal Library Functions The PCI DAS6402 16 is completely plug and play Consequently there are no switches or jumpers to set on the board and instead the hardware is configured using software Hence parameters such as the number of analog input and output channels used and the sampling frequency and count value are set in software Other parameters that must be set in software include whether the board will be operated in background or foreground mode In foreground mode the PC is not allowed to do any form of pro
97. hannels 89 Table 6 3 MTBF Results Obtained Using 4 Analog Input Output Channels 89 Table 6 4 MTBF Results for 200 Hz Sampling Frequency using Varying Computation Loads at Normal Priority with One Analog Input Channel cee ceeee cece eee neee ence eens 90 Executive Summary The electrical activity of human muscle as well as the various mechanical parameters related to human motion can be used in a wide array of research studies For instance electromyogram EMG signals are currently being utilized in the development of artificial limbs and in the gait rehabilitation of stroke and brain injury patients 4 5 16 23 Hence a workstation that can be used to acquire analyze and display EMG and other similar signals can be beneficial to the continuation of research related to biomedical signal analysis A workstation that will be used to acquire and analyze the electrical activity of human muscle may need to operate in hard real time for certain control applications Operation in hard real time refers to the notion that the system must not only perform logically correct computations but also operate under critical time constraints as well 14 The reason that a hard real time system is required is that most of the previously described applications for this system involve feedback in a timely manner For instance signals analyzed by the workstation might be used to control the move
98. he mouse the configuration palette that can be seen in Figure 4 4 will be launched Using this graphical palette the user is able to select which analog input channels and output channels will be used the desired sampling frequency and method for data transfer can be selected gt DAQ Assistant in won w wen w ree 289P Fato Resstexe Votege Ei Vex Source Wen vae Stren Corfa on rier 25 fd Oredge I Lead Reastance 0 o Task Tng A Task Inggerng Acqare serow Acqure N Samples Acqure Cortirscutly 1000 00 Rate Mz 10D Samples To Read Advanced Clock Settings Cock Type Interna v Figure 4 4 NI DAQ Assistant Configuration Palette While the DAQ assistant provides a simple graphical user s interface using this method does not allow the user to configure certain aspects of the DAQ board in regards 58 to analog output In particular when the option is chosen to utilize the FIFO when performing an analog output task the DAQ assistant configures the board to operate in regeneration mode In regeneration mode data that is in the DAQ FIFO is continuously reused For instance a single period of a sine wave can be written to the FIFO buffer Regeneration allows for this period to be continually generated thus giving the appearance of a continuous waveform While such a regeneration feature might be appropriate in some applications if it is desired to have the output waveform change while the board i
99. imum Yalue 2pe Physical Channels 2i dip ati ji et a vl Waive Sample Clock Ratel ost 4 OK message warnings oes Hardware Timed Single Point_ Physical Channels 0m 0 ee i es eA Pe y 40 voltage Sample Clock gt gt a Ls Ce Finished Late i 1 F E E Ee ol gt me Wat frear wc POBL Etandard deviation ppr istogram Intervals 32 an peu Soz Figure 6 1 MTBF Test VI 86 In the VI that is depicted in Figure 6 1 an analog input and an analog output task are first created The output routine is located inside a loop whose timing is controlled by the hardware timer on the DAQ board unlike output routines utilizing the FIFO which are dependent on the timing of the operating system When a loop is configured with a hardware time there is a signal available in the loop labeled finished late that is asserted when the loop has not completed by the time the hardware clock triggers it to run again In our VI this signal is used to stop the internal hardware timed loop When the loop is stopped the number of iterations that was completed before failure is passed to an outer loop These data are then passed to a function which converts the number of iterations completed to a length of time that the loop ran before error Like the timing VI created for measuring function latency this routine can be configured to run multiple times As a result a histogram exhibiting
100. inux was chosen Yet another option is to utilize an embedded system such as a digital signal processing DSP board to implement this system For instance in 1999 students at the University of Alabama built a holter system using a Texas Instruments data acquisition board 19 However as previously discussed embedded systems are often more difficult and cumbersome to configure then PC based systems In particular individuals programming embedded systems must be more familiar with all of the hardware components of the board and in most instances write their own personal drivers for other input and output devices Hence an embedded system although it provides the capability to execute hard real time routines is difficult to maintain 8 3 Chapter Summary Our results and analysis indicate that a PC based system using LabVIEW can be used in data logging applications However standard LabVIEW on a Windows machine provides nondeterministic results and therefore is not suitable for hard real time control experiments The proceeding section outlines future recommendations to individuals who will continue working on creating a workstation to conduct biomedical experiments 102 These design options include using different LabVIEW modules such as LabVIEW FPGA and LabVIEW RT or even utilizing a different operating system such as RT LINUX While all of these options are feasible alternatives to a PC based system using standard LabVIEW e
101. ion of the muscle fiber tissue and a following depolarization waveform This waveform travels the length of the muscle fiber and is known as the action potential AP Figure 2 1 depicts the generation of electric fields in muscle fibers NEURON an p Ww ELECTRICAL FEY 2a a Figure 2 1 Motor Neuron Induced Electrical Field in Muscle Fiber 25 Muscle fibers are excited by nerve branches from one motoneuron in groups known as motor units These motor units are defined as the fundamental unit of contraction and can range from a few muscle fibers for small muscles such as those in the hand and fingers to thousands of muscle fibers in large muscles Because each motor unit contains a number of muscle fibers that are attached to the motor neuron at various points the electrical signal of a motor unit is the summation of the action potential of each muscle fiber which may be phase shifted from the other muscle fibers in that unit This notion is reinforced in Figure 2 2 The electrical potential due to contraction of all fibers in a motor unit during a single activation is referred to as the motor unit action potential MUAP This MUAP can be recorded by using electrodes placed on the surface of the skin above the muscle 1 motoneuron muscle fiber h t Eiry Figure 2 2 Motor Unit Action Potential 14 Also a muscle is not typically excited by only one action potential In order to hold a contr
102. ion set which is used to configure NI M series DAQ boards will be described 4 2 1 Hardware Overview The PCI 6229 has 16 differential analog input channels with a resolution of 16 bits and a maximum sampling frequency of 250 KHz Again like the Measurement Computing device the PCI 6229 only has one A D converter Consequently when more then one channel is being utilized to acquire data the sampling rate for each channel is again defined as Sampling Frequency Number of Channels Used There are two methods available for transferring analog input data from the PCI 6229 and the computer First like the Measurement Computing board input data can be stored in a 4095 sample First In First Out FIFO buffer located on the DAQ board Data is then transferred using the PCI bus of the computer via direct memory access DMA Data is transferred from the FIFO to the computer once the amount of samples stored in the FIFO reaches a specified value which is referred to as the buffer size The minimum buffer size for this NI board is two samples The other method that is available for transferring analog input data from the PCI 6229 to the PC is referred to as hardware timed single point In this transferring mode configuration a single sample of datum is 55 transferred using the PCI via DMA every sampling period In this configuration the FIFO is completely bypassed by utilizing the hardware clock present on the DAQ board 24 The PCI 6
103. ize a different LabVIEW function type was presented In this section a detailed description displaying how each routine was developed in LabVIEW will be presented Array Based Implementation The first implementation of the EMG amplitude estimation algorithm is the standard array based algorithm The block diagram of this VI can be seen in Figure EJ 1 while the LabVIEW block diagram can be seen in Figure EJ 2 Adaptive Absolute Multiple Channel Whitening VI Value Average In Array based Highpass FIR Filter Array based Lowpass FIR Filter Out Absolute Multiple Channel Array based Divide by Value Average Lowpass FIR MVC Filter Figure EJ 1 Block Diagram of Array Based VI The input of this function is a multidimensional array of data DAQm x functions which will be used to acquire data for use with this VI hold data from multiple channels in a multidimensional array Each row in the multidimensional array contains data from each individual channel As figures EJ 2 depict the input signals are first placed through an array based highpass FIR filter inside of a for loop The user is able to enter the 169 parameters for this filter on the front panel The parameters that the user is able to configure include the number of taps used and the type of window if any that the filter should implement It should be noted that the for loop is utilized here and throughout this VT to allow each row in the
104. k due to safety and control complexity As a result robotic devices are unable to take into account all of the key gait determinants 32 Consequently testing needs to be done to determine if muscle activation patterns exhibited during treadmill walking are altered when a robotic aid is used This analysis of muscle patterns is performed by analyzing EMG signals Such tests have been performed at Catholic University in Washington DC where devices are being developed for gait rehabilitation One of these devices is a robotic gait orthosis known as the Lokomat In order to test the effectiveness of this device EMG signals generated in normal subjects while walking on a treadmill were compared to those generated by individuals using the Lokomat Therefore EMG signals provide a mechanism for testing the effectiveness of gait therapy 32 16 2 2 3 Other Control Systems While EMG signals have been used in the areas of prosthetic limb development and gait rehabilitation EMG signals have also been used in a variety of other control applications for the paraplegic and partially paralyzed One application has been developed that enables disabled individuals to utilize computers One of the chief components of a computer is the mouse In order to operate many computer applications the user is expected to move the mouse cursor on the screen and click on various graphical buttons Individuals with severe muscle disabilities are unable to move their a
105. le Channels cceeeee eee enone 172 Figure EJ 4 Array Based Adaptive Whitening VI cece ce eee eee ee ence nee ee ene enenneeaens 173 Figure EJ 5 MATLAB Script Used to Construct Multidimensional Array of Whitening Filter Coefficients nner e eree e A S el ead E Aa NiE 174 Figure EJ 6 Block Diagram of Array Based VI 0 cee cce nce ee ence eee eee eee en ene ences 175 Figure EJ 7 Express VI EMG Amplitude Estimation Algorithm ccceeeee seen ee ee es 177 Figure EJ 8 Block Diagram of Point by Point VI 0 eee ce cece cee eeeneee eee enaenaeaee 178 Figure EJ 9 Point by Point Adaptive Whitening Filter cece eee e eee ee eee e eae 179 Figure EJ 10 Point by Point EMG Amplitude Estimation Algorithm e ee eee 180 1X Table of Tables Table 2 1 Difference Between Array Based and Point By Point Analysis 3 00 42 Table 5 1 Experimental Results of Point by Point EMG Function cece seen eee eeeee 75 Table 5 2 Experimental Results of Express VI EMG Function where 512 Taps were used For All Three Filters 0 aore ea ria an E E AAA EAPO EENET EEE a T 75 Table 5 3 Experimental Results of Array Based EMG Function sce ece eee ee eee e eee neee 76 Table 6 1 MTBF Results Obtained Using 1 Analog Input Output Channel 0 88 Table 6 2 MTBF Results Obtained Using 2 Analog Input Output C
106. lized when performing a routine that requires simultaneous input and output 15 63 Hardware Timed Single Point Y Rate Hz Peri Dev1 ai0 1 v DAQ mx tm DALM Dev1 ao0 1 7 An i __ amp Error out 40 Voltage Sample Clock z ai SampleClock v Figure 4 5 Simultaneous Input and Output Using PCI 6229 64 4 4 Discussion As discussed in the previous two sections the PCI DAS6402 16 has two limitations that hinder the development of our workstation First the board places a limitation on the minimum of samples that must be acquired before an interrupt can be generated This limitation causes a latency that will not allow for our workstation to be used with hard real time control experiments Second the PCI DAS6402 16 does not support simultaneous input and output Since the goal of this project is to determine if LabVIEW can be used to implement a workstation to conduct biomedical systems some of which involve hard real time control it is apparent that the PCI DAS6402 16 is not a suitable DAQ board to create a workstation to conduct such biomedical experiments On the contrary the PCI 6229 manufactured by National Instruments is designed to perform simultaneous input and output Moreover the NI device does not place restrictions on the minimum number of points that must be acquired before data is transferred between the board and the PC Our comparison between these two devices thus shows that all DAQ
107. ly be used with LabVIEW s array based analysis As a result a slight modification needs to be made to the previously described model in order to make the data collected compatible with point by point VIs In order to extract one piece of datum at a time the multidimensional array of voltage values is placed through two nested for loops By placing the multidimensional array through two for loops a single point of datum from each of the analog input channels will be extracted during each loop iteration Figure 4 2 displays this modified module that can be used to perform point by point analysis Using the analog output channels requires the same basic procedure as depicted in Figure 4 1 However the functions AOutScBg and FromEng are used instead of AlInScBg and ToEng FromEng takes voltage values and converts them to A D count values Then the function AOutScBg outputs these created A D count values to specified output channels at a specified sampling frequency 53 Figure 4 2 VI That Acquires Data in Background Continuous Mode for Use with Point By Point Functions 54 4 2 National Instrument s PCI 6229 The other commercial DAQ board that was analyzed to determine if the choice of DAQ board would affect system performance was the PCI 6229 M series device manufactured by National Instrument s In section 4 2 1 a review of the hardware of the board will be given Then in section 4 2 2 National Instruments DAQmx LabVIEW funct
108. m is to develop a PC based system that runs a particular software package One widely used software package is National Instrument s LabVIEW An overview of LabVIEW and some of its advanced features were presented In the following chapter our specific project goals and objectives will be presented 43 3 Problem Statement While there are a myriad of ways to implement a signal acquisition display and feedback system to be used in biomedical experiments the goal of our project was to determine if a PC based system running LabVIEW can meet the hard real time constraints that are necessary for various EMG control applications This chapter presents the problem statement of our project along with the overall goal of our project team Further objectives that needed be met to accomplish the overall goal are presented 3 1 Problem Statement and Project Goals Biomedical signals such as the EMG can be used in various applications Consequently our system will be used to acquire and analyze EMG signals Moreover multiple EMG signals are typically recorded at a time during experiments Hence our system will be configured to acquire and analyze as many EMG signals as possible at a time As discussed in the background section of the report the bulk of EMG signal information is at frequencies below 100 Hz As a result a sampling frequency of 256 Hz for each analog channel utilized on the DAQ board was chosen to be utilized in experiments co
109. me of the benefits of using an FPGA based system over a microprocessor based system can be seen in Figure 8 1 Another design option is to utilize the LabVIEW Real Time RT module LabVIEW RT delivers increased reliability in time control by utilizing an embedded real time operating system on a National Instruments hardware device This embedded hardware device is used to execute a LabVIEW program that is present on a PC The benefit of using LabVIEW RT is that deterministic results can be garnered In particular unlike standard LabVIEW a system of internal LabVIEW priority controls can be utilized As a result the speed of which certain aspects of LabVIEW code executes can 100 be configured Hence a processing routine can be configured to run at a frequency of 1 KHz while a graphical output routine can be configured to run at a frequency of 60 Hz 7 Clearly such a system would be ideal for our hard real time processing needs Hence LabVIEW RT and FPGA offer two alternatives for utilizing LabVIEW to conduct our hard real time experiments However while these options appear to offer a viable solution for creating our workstation the cost of these devices is rather high In particular a LabVIEW FPGA board with 8 analog input and output channels can cost approximately 4000 without including the price of any of the necessary FPGA software drivers Hence spending this much money on a component eliminates the benefit of using a cost eff
110. ment of a prosthesis In order to accurately work like a human limb the prosthesis must move within a certain window of time when it receives the appropriate stimulation and consequently not react after a long delay There are at least two fundamental ways that a system that can be used to conduct various biomedical experiments can be designed The first option is to design an embedded system that can only perform the acquisition task While this is a feasible approach there are several drawbacks to such an approach A standalone embedded xi system is often difficult to program and maintain over long periods of time Moreover an embedded system might require the development of customized hardware to implement various algorithms The drawback to making customized hardware is the notion that duplicating the hardware can be difficult and the hardware might need to be frequently updated when new analysis algorithms are implemented A second approach is to design a PC based system This type of system utilizes a standard desktop computer with a data acquisition DAQ card and commercially available components to amplify and isolate signals The PC would then have a software package that can be executed to perform desired biomedical signal acquisition experiments A PC based workstation is cost effective as it bypasses the need for custom instrumentation by using a currently available PC and inexpensive acquisition equipment However the compone
111. mining if a PC based system executing LabVIEW can be used to conduct biomedical experiments that require hard real time execution In particular our goal will be to determine if the system can operate with a maximum latency of 10 milliseconds The objectives and tasks necessary in achieving this goal are also described 47 4 Data Acquisition Board Performance Analysis In section 2 3 2 the concept of using a PC based system to conduct biomedical experiments was introduced In particular the notion that a DAQ board needs to be installed in a PC in order to allow for analog signals to be acquired and then analyzed in software by the PC was discussed In chapter three the main goal of our project and the various objectives that we met to achieve these goal was presented One of the questions that was answered in order to accurately determine if a hard real time biomedical signal acquisition display and feedback system can be implemented on a commodity Windows based PC using National Instrument s LabVIEW software was if the DAQ board used with the PC affected system performance In order to meet this objective a comparison of two commercial DAQ boards was performed The two boards that were analyzed were Measurement Computing s PCI DAS6402 16 and National Instruments PCI 6229 In the following sections a description of these two boards and a comparison of the performance of both boards will be presented 4 1 Measurement Computing PCI D
112. multidimensional input array and thus all of the data from each channel to be processed 170 array of i ima ow cutoff Freq Fl 2 DB1 igh cutoff freq Fhi D658 Lo T pir tic E gt a E Sampling Frequency Frequency Fi 132 low cutoff low cutoff freq fl2 fl2 meee nd Gates 2 high cutoff Freq Fh high cutoff Freq Fh Freq fh Figure EJ 2 EMG Array Based Amplitude Estimation VI 171 After the output of the highpass filter is absolute valued data is passed to a module labeled with the number 2 This module is used to average the data from multiple channels and the implementation of this module can be seen in Figure EJ 3 This average module takes a two dimensional waveform sums each element in a particular column together and then divides the entire array by the divisor to create a one dimensional array result It should be noted that the LabVIEW summation function sums each element in a row As a result in order to have each element in a column summed the input array must first be transposed Consequently this module is used to average these data after it has been absolute valued Figure EJ 3 Module that Averages Data from Multiple Channels These averaged data are next placed through a smoothing filter This smoothing filter was chosen to be implemented as an FIR windowed filter The user is able to select on the front panel the type of filter number of taps and window utiliz
113. n the two basic forms of biomedical signal acquisition systems will be discussed 2 3 Biomedical Signal Acquisition Systems From the previous sections of this chapter it is apparent that EMG signals have a wide array of applications in various research areas While only EMG signals were discussed in the previous section other biomedical signals are equally important in research In order for these biomedical signals to be used in their various applications a system must be available that can acquire analyze and display these various signals Although there are many design options available for creating such a workstation two of the most common approaches are to either develop an embedded or PC Based System 2 3 1 Embedded System Biomedical signal acquisition systems are conventionally implemented as an embedded system An embedded system is some combination of computer hardware and software that is specifically designed for a particular application For instance using 18 digital signal processing DSP boards individuals are able to design and implement customized signal processing algorithms Although embedded systems typically have a real time operating system individuals working with embedded systems have to be aware of the low level details of how to program and interact with all of the hardware components of the system Consequently programming embedded systems to perform signal analysis routines is often cumbersome and dif
114. nal Instruments products that are available for use with standard LabVIEW These products can be utilized on a Windows PC that can potentially allow for our real time specification to be met Two of these products that can be utilized to help achieve hard real time performance are the LabVIEW FPGA Field Programmable Gate Array and LabVIEW RT Real Time modules 99 Microprocessor Based System FPGA Based System Performance limited to 1 kHz Closed loop performance beyond 1 MHz Serial execution single rate control Parallel execution multi rate control Performance slows as app grows No slow down as application grows Operating system runs control logic Control logic in dedicated hardware iO modules have fixed functionality Custom circuitry requires board layout iO functionality is reconfigurable Software defined gate array LIII Separate motion control system Motion integrated with other control logic Figure 8 1 Impact of FPGA based system 18 The LabVIEW FPGA module offers additional command over NI hardware by giving the user control of FPGA chips on select NI RIO Reconfigurable I O hardware This module allows the user to program the FPGA chip with user created hardware execute multiple tasks simultaneously and customize timing with a resolution of up to 25 nanoseconds Implementation of an FPGA system would require purchase of a NI R series DAQ board which contains the RIO hardware used by LabVIEW FPGA An outline of so
115. nducted with this workstation Moreover our system will be configured to be able to conduct control applications such as those involved with prosthetic limb development and gait therapy In order to accurately control prosthetic limbs the limb should respond within a specified interval to the stimulus Similarly if feedback is needed regarding the muscle activation of an individual participating in rehabilitation gait therapy by a certain amount of time 44 our system should be able to meet this deadline for providing feedback Therefore our workstation must process signals in hard real time By definition a hard real time system is a system whose correctness of computations not only depends on the logical correctness of the computation but also upon the time in which the results are produced Hence if the specified timing constrains are not met system failure is said to occur 14 As a result of the applications in which our workstation will be used for it is desired that our system function as a hard real time system Besides acquiring and synthesizing signals our system should be able to display results give appropriate feedback to the user and when prompted save selected signals to the hard drive Also since our system will be used for a variety of biomedical experiments it needs to be easy to use and operate Similarly the system must also be easy to modify since it will be continually modified and updated when different biomedic
116. nts chosen to operate with the PC must be selected carefully in order to optimize system performance In particular DAQ boards do not follow a universal convention of operation This notion can be seen by comparing two commercial DAQ boards National Instrument s PCI 6229 and Measurement Computing s PCI DAS6402 16 While both boards have a FIFO buffer and transfer data between the PC and computer via DMA data transfer is handled differently with each board The PCI DAS6402 must collect a minimum of 1024 data sample before the board can generate an interrupt and send data to the PC To the contrary the PCI 6229 does not have this minimum collection requirement and can even transfer a single datum point per sample by using a clock that is present on the device and bypassing its FIFO Moreover in order to conduct a control application the DAQ board must be able to simultaneously input and output data While both of the previously described boards contain analog input xii and output channels only the PCI 6229 has the architecture to support simultaneous input and output Consequently we have created a PC based workstation which will be utilized to conduct these various biomedical experiments with the PCI 6229 DAQ board Moreover the software that we utilized is National Instruments LabVIEW LabVIEW is a graphical programming language for the Windows operating system that is used for various signal processing applications Since LabVIEW
117. o two function types standard array based and Express VI functions In order to investigate the feasibility of creating a hard real time workstation it is important to investigate the latency associated with using each type of LabVIEW VI Consequently EMG amplitude estimation algorithms developed using all three of the analysis methods will be presented These three different EMG amplitude estimation algorithms were tested to determine the latency induced by each method Once the results of these tests for each method is displayed observations regarding if any of the analysis methods operate under our maximum latency of five milliseconds can be made 5 1 EMG Amplitude Estimation VIs In the background section of the report the theory behind EMG amplitude estimation techniques was presented When implementing the previously discussed algorithm we can represent the amplitude estimation process utilizing adaptive whitening with nine stages A block diagram of this entire nine stage process can be seen in Figure 5 1 67 In Highpass Adaptive Absolute Combine Filter Whiten Value Absolute Combine Smooth Normalize Value Figure 5 1 EMG Amplitude Estimation Block Diagram Utilizing Adaptive Whitening Out While three versions of the above algorithm were created using the three different LabVIEW function types all three routines were constructed in the same basic manner Hence the only significant difference betwe
118. oftware Pgs 142 146 2005 lt http portal acm org citation cfm id 1086228 1086256 gt 108 Appendix A Compiling CIN Source Code Complete the following steps to build CINs using the Visual C integrated development environment IDE on WINDOWS 1 Launch Microsoft Visual Studio 2 Select File New Project to create a new DLL project 3 Select Win32 from the Visual C menu as the project type You can name your project whatever you want 4 Select An empty DLL project when prompted to choose the type of DLL that you want to create and click Finish 5 Select Project Add To Project Files and navigate to your c file Select your c file and click the Open button to add the c file to the project 6 Select Project Add To Project Files to add CIN objects and libraries to the project Select cin obj labview 1lib lvsb 1lib and lvsbmain def from the Cintools subdirectory You need these files to build a CIN 7 Select Project Settings and change Settings For to All Configurations Click the C C tab and set the category to Preprocessor Add the path to your Cintools directory in the Additional include directories field Be sure to place the path to the directory in double quotations 8 Select Project Settings and change Settings For to All Configurations Click the C C tab and set the category to Code Generation Select the Struct member alignment tab and select 1 byte 109 9 Choose a run time library Select Pr
119. oject Settings and change Settings for to All Configurations Select the C C tab and set the category to Code Generation Select Multithreaded DLL in the Use run time library control 10 Make a custom build command to run 1lvsbutil Select Project Settings and change Settings for to All Configurations Select the Custom Build tab and change the Commands field as follows with the code all on a single line lt your path to cintools gt lvsbutil TargetName d ProjectDir O0OutDir Note that the path to cintools must be placed in double quotations Change the Output fields to OutDir TargetName 1sb 11 Add lt CINTOOLSDIR gt lvsbmain def to the Linker Input Module Definition File field 12 Click the File View tab in the Work Space window 13 Open your c file andreplace Insert code here with your code 14 Select Build Build projectName dll where projectName is the name of your project 110 Appendix B C Code for Choosing An Array of Filter Coefficients Below is the C code that when given a matrix of filter coefficients will output one of the coefficient vectors depending on the scaled amplitude that is given as input CIN source file include extcode h define typeCode Ox0A define numDims 2 define paramNum 2 Typedefs typedef struct int32 dimSizes 2 Float64 elt 1 joe TDI TD1 TD1Hd1 Fh typede f struct int32 dimSize Floa
120. on to be functional well into the future new operating systems and software packages need to be investigated that can potentially be used to upgrade this system In the following section one of these options will be discussed 2 6 Programming in LabVIEW One programming language that is readily used in soft real time PC based signal processing systems is National Instrument s LabVIEW LabVIEW is a graphical programming language utilized on the Windows operating system which can perform many of the same features of other programming languages while allowing for an extensive library of data acquisition board drivers In this chapter some of the basic and advanced features of LabVIEW that will be utilized in our workstation will be discussed 33 2 6 1 Basics Programs in LabVIEW are designed to appear and function like physical instruments such as oscilloscopes and multimeters As a result LabVIEW programs are called Virtual Instruments VIs Building a VI in LabVIEW is a two step process First the programmer must create a user interface known as the front panel This user interface is comprised of virtual controls and indicators Controls such as pushbuttons and dials provide inputs that can be used in the program Indicators such as LEDs and graphical displays serve as the output devices of the system While the user is free to create their own controls and indicators LabVIEW comes with a palette of these front panel objects that
121. on to conduct biomedical experiments 2 Background This chapter presents information on the background of the data display acquisition and feedback system for biomedical experiments that we have created First an introduction to EMG signals is presented along with the various applications that can be conducted by utilizing these signals Second the general structure of biomedical signal acquisition systems is discussed Third the properties of operating systems that are used on PC based acquisition systems are presented Fourth an existing acquisition system is described and its deficiencies are explored Last a common software package that is used in signal processing applications LabVIEW is introduced 2 1 Introduction to EMG There are various biomedical signals that will be analyzed with the workstation that we have created In this chapter an introduction is given to one of these many biomedical signals the electromyogram EMG signal After giving basic information about the characteristic of EMG signals examples depicting the type of applications that EMG signals are used in are presented 2 1 1 Sources and Characteristics of EMG The nervous system controls the voluntary movement of various body parts in humans by contracting and relaxing various skeletal muscles To instantiate a contraction a neuron generates a small electrical potential on the surface of the muscle fiber This electrical potential causes depolarizat
122. or containing a calibration sample of the EMG signal acquired during a constant force constant posture nonfatiguing contraction and a vector containing a calibration sample of the noise signal The function e_cal_wh which actually designs the adaptive whitening filter coefficients was previously created by the advisor of this project These MATLAB functions can be seen in Appendix D MATLAB Script mvcO my cSO cal_sim Inth rt B sCal ec e_cal_wh myc50 0 5 mycO Filter Coeffecients Array of Coefficients Figure EJ 5 MATLAB Script Used to Construct Multidimensional Array of Whitening Filter Coefficients The remainder of the EMG algorithm follows the block diagram located in the beginning of the chapter First the whitened signal is absolute valued Next the data is averaged again using the previously described module Finally the averaged signal is placed through the previously defined smoothing filter The output of this final smoothing filter gives the EMG amplitude estimate 174 Express VI Implementation Next a block diagram version of the same EMG amplitude estimation algorithm described in section 5 2 implemented with Express VIs can be seen in Figure EJ 6 Figure EJ 7 displays the LabVIEW block diagram of the routine Express VI Adaptive Absolute Multiple Channel FIR Filter Whitening VI Value Average Absolute Multiple Channel Express VI Divide by Value Average FIR Filter MVC Figure EJ 6
123. or data acquisition boards are readily available for use with LabVIEW As a result by using LabVIEW on a Windows PC the need for writing customized hardware drivers was avoided While we used LabVIEW for a hard real time application LabVIEW was originally intended for soft real time applications Moreover while Windows XP allows for the availability of data acquisition board drivers it is not a real time operating system Therefore the purpose of this project was to determine the extent to which LabVIEW can be used for a hard real time application In particular our goal was to determine if biomedical experiments can be conducted with LabVIEW with latencies no greater then five milliseconds This five millisecond worst case latency was chosen to ensure that a computer monitor used to display graphical data will have a real time screen refresh rate Although tests conducted indicated that the average latencies generated by LabVIEW VIs could be on the order of fractions of milliseconds outliers that well exceeded our latency requirements were frequently observed These large outliers are a result of the limitations of the Windows XP operating system Hence our results indicate that a PC based hard real time system with our given latency requirements can not be implemented using standard LabVIEW running on the Windows XP operating system Consequently future recommendations are provided that offer alternative methods to implementing a workstati
124. predictable but instead attempts to allow for each available process to have a fair share of CPU time The latest version of Microsoft s operating system is Windows XP Windows XP uses a preemptive thread based priority scheduling algorithm While a process is traditionally perceived to only have one thread of execution Windows allows for 26 multiple threads to exist in the same process Consequently Windows XP has 32 priority levels with O being the lowest and 31 being the highest The highest 16 levels 15 31 are characterized as real time levels Although threads at these levels are characterized as real time the name is misleading Threads running on these priority levels are not guaranteed to receive process time On the contrary threads running on the highest priority may receive no processors cycles because the processor may be busy handling interrupts Moreover system level threads are not running on these levels Consequently setting an application to a real time level may block a system task and cause for system instability 32 During execution the difference between real time levels and lower levels called dynamic levels is that the scheduler will never change the priority of a thread running on a real time level Processes running on lower levels will occasionally have their priorities boosted by the operating system to avoid process starvation Therefore before a thread is run on Windows XP it must first wait for hardwa
125. ques used to perform EMG amplitude estimates 2 1 2 Methods for Amplitude Estimation Currently the most popular EMG amplitude estimation techniques are the statistical moving average mean absolute value MAV and moving average root mean square RMS Below are the equations for these two estimation techniques MAY gt x 1 y j r N 1 In the above equations N is the number of samples in each smoothing window t is the time at which the interval starts and x is the signal being smoothed in the time domain Research has shown that amplitude estimates using MAV and RMS calculations exhibit similar results 8 Other signal processing techniques are often using in conjunction with MAV and RMS calculations to improve EMG amplitude estimates In particular research has indicated that whitening an EMG signal prior to MAV and RMS calculations results in enhanced estimates Figure 2 4 displays a six stage process that may be used to compute EMG amplitude estimates with whitening prior to detection of the signal This six stage process is comprised of noise rejection whitening gain normalization detection smoothing and relinearization m t Noise Filter Spatial Uncorrelate and Gain Normalize m t Noise Relinearize 1 d Reject Filter Figure 2 4 Standard Six Stage Filtering Process for EMG Amplitude Estimation 10 2 1 2 1 Noise Rejection As stated previously the majori
126. rd Deviation 4 26 sec Priority Real Time a Histogram lat at N ay _ i i TE Number of Occurrence Wet ful aia Hoe P ae Cdr A a Geer A o S Tara T ase e o e aee Aa Ube G a Aa S a O ob Na Aa Time sec Mean 7 74 sec Standard Deviation 3 62 sec 149 Sampling Frequency 1000 Hz Priority Normal Moving Histogram 40 30 Number of Occurrence N o I i EN i 0 1 0 2 0 3 0 4 0 5 0 6 Time sec o oa Mean 36 sec Standard Deviation 0 18 sec Priority High Moving Histogram feni n N o n 1 fi 1 Number of Occurrence on I o al j 0 1 0 2 0 3 0 4 os 0 6 Time sec o Mean 0 35 sec Standard Deviation 0 19 sec 150 Priority Real Time Moving Histogram 100 80 60 40 Number of Occurrence 20 0 1 0 1 0 2 0 3 0 4 0 5 0 6 Time sec Mean 0 54 sec Standard Deviation 0 04 sec Sampling Frequency 2000 Hz Priority Normal Moving Histogram Number of Occurrence 0 0 0 1 0 2 0 3 0 4 0 5 0 6 Time sec Mean 0 20 sec Standard Deviation 0 15 sec 151 Priority High Moving Histogram 2 N S t 1 1 1 Number of Occurrence n N O1 g2 0 3 0 4 HS 0 6 Time sec o Mean 0 30 sec Standard Deviation 0 20 sec Priority Real Time Moving Histogram 80 0 Number of Occurrence i
127. re EJ 9 This module uses the same CIN as previously described to select the appropriate array of filter coefficients based on the scaled amplitude entered However this function uses the point by point FIR filter to implement the desired filter specified by the array of coefficients instead of the array 178 based function Figure EJ 10 depicts the entire EMG amplitude estimation routine developed using point by point analysis ilter Coefficients Scaled Amplitude O51 gt Figure EJ 9 Point by Point Adaptive Whitening Filter 179 Filter Coefficients one Filter type Ko B E gt gt Pp RD ow cutoff Freq Fl ow cutoff freq fl 2 eeptbe gt e igh cutoff freq Fh ps1 DBLP o E T 7 as2 taps 2 Sampling Frequency a JBL Filtered x gt 5 gt low cutoff freq fl2 l r high cutoff freq Fh window 2 F AEN Figure EJ 10 Point by Point EMG Amplitude Estimation Algorithm 180
128. re and software interrupts to finish After the interrupts finish the waiting thread then must wait for other threads with higher priorities to execute Then when a thread is finally allowed to execute whenever an interrupt arrives or a higher priority thread becomes ready the running process will become preempted and have to wait for the other interrupts or threads to execute Consequently the current kernel architecture for Windows XP is not ideal for high precision real time applications 27 2 5 Existing System From the previous chapter it is apparent that a PC based system which incorporates some form of operating system can potentially be used to acquire and analyze biomedical signals Currently there is a PC based system that is used by the Center for Sensory and Physiologic Signal Processing C SP that can acquire and analyze EMG signals This system can acquire EMG signals and process them in real time Furthermore this workstation can display selected signals on the computer screen provide feedback information to a user and store signals to the hard drive A description of the existing workstation s hardware and software components follows 2 5 1 Hardware Overview Figure 2 7 provides a block diagram of the hardware used in the EMG workstation EMG signals can be observed if a differential electrode pair is applied to the skin surface above a muscle In order to analyze and manipulate EMG signals more effectively the
129. ree for performing other tasks at the same time 20 2 3 2 2 Benefits of a PC Based System While using standalone embedded systems is a viable method of acquiring and analyzing biomedical signals a PC based system is a cost effective alternative Individuals are able to share the cost of buying the PC by utilizing it for other general purpose applications such as word processing or web browsing Moreover the other external components that are utilized in the system can be purchased commercially Hence the development of customizable hardware can be avoided or reduced However if a PC based system is implemented another piece of software must be taken into account that is not utilized in a system designed to execute only one task That piece of software is known as an operating system and it will be discussed in the following section 2 4 Operating System Information In the previous section the notion that a biomedical signal acquisition system can be either embedded or PC based was discussed In both implementations a piece of software is capable of being utilized known as the operating system An operating system allows for numerous applications to be executed on a single workstation instead of just one customized application In the following sections some of the basic functions of an operating system will be discussed 21 2 4 1 Operating System Functions As discussed in the previous section the key software component that
130. rent that both the array based and point by point EMG amplitude estimation algorithms have latencies that are below our defined five millisecond maximum 78 Array Based vs Point by Point High Priorty Latency Results 3 00 a 2 50 77 E 2 00 a e Array Based Buffer Size 1 g 1 50 s Array Based Buffer Size 10 z Point by Point 1 00 0 50 a pE E 0 00 512 1024 1536 2048 Taps Figure 5 3 Latency of EMG Array Based and Point by Point Algorithm High Priority Array Based vs Point by Point Real Time Prioirty Latency Analysis 3 50 T 3 00 2 50 2 00 e Array Based Buffer Size 1 1 50 a Array Based Buffer Size 10 1 00 0 50 0 00 Point by Point Average Latency msec 512 1024 1536 2048 Filter Taps Figure 5 4 Latency of EMG Array Based and Point by Point Algorithm Real Time Priority The previous three graphs did not depict results obtained when using the Express VI implementation of the EMG routine since Express VIs as discussed previously are limited to utilize only a maximum of 512 taps for filters Consequently Figure 5 5 depicts results for executing all three versions of the EMG algorithm with a constant 79 value of 512 taps for each filter utilized in the routine Although a buffer size of one is utilized for both the array based and Express
131. rms or feet and therefore are not able to use the traditional point and click features of a mouse 22 While these individuals are unable to move their arms or legs many of these patients still have control over their facial muscles As a result a system has been developed that allows the user to move the cursor on a screen by the movement of their cranial muscles In particular the current system detects the EMG signals associated with the temporalis muscles and muscles associated with the raising and lowering of the eyebrow through three electrodes These signals are then processed through a digital signal processing board and result in the movement of the cursor on the screen 22 2 2 4 Discussion One of the chief biomedical signals that will be analyzed on our workstation is the electromyogram EMG signal As shown throughout this chapter EMG signals are used for a variety of activities ranging from the development of prosthetic limbs to the 17 analysis of gait rehabilitation therapy By showing some of the research activities that EMG signals are used in the function of our workstation becomes more apparent Our workstation is to acquire biomedical signals and perform analysis to allow for the various experiments associated with the different applications of EMG signals to be conducted Furthermore our workstation should also be able to analyze other similar biomedical signals in other biomedical experiments In the next sectio
132. rt Slnth length of sample in samples srt sampling rate in Hz randn state sum 100 clock Sstates random number generator mvcO 0 03 randn 1 i1nth b a butter 2 200 2 rt mvc50 0 5 randn 1 linth rt mvc50 filter b a mvc50 mvc50 mvc50 end lnth l end 0 03 randn 1 1nth return 113 Appendix D MATLAB Code that Creates Adaptive Whitening Filter Coefficients Below is the MATLAB M file created previously by Ted Clancy that creates a matrix of filter coefficients for adaptive whitening function B sCalVec e_cal_wh SandN sCal Noise varargin SE_CAL_WH Design calibrate adaptive and fixed whitening filters B sCalVec e_cal_wh SandN sCal Noise Order sSpec Design a set of FIR adaptive whitening filters based on the three stage cascade technique of Clancy and Farry In Preparation SandN Calibration sample of signal plus noise vector sCal MVE level corresponding to SandN scalor Typically 0 5 for 50 MVC Noise Calibration sample of noise vector gt Options Order Order of the combined filter coefficients 1 scalor Default 60 sSpec EMG amplitude level specifier same scale as sCal vector Default specifies 100 filters from just above 0 to 2 sCal Option Pairs Nfft NNfft FFT length power of 2 for spectral estimates scalor MaxGain MMax
133. s outputting then this mode is highly undesirable Consequently in order to gain greater control and flexibility there are individual DAQnmx functions that are present By linking these various functions together such as with the Universal Library function set the board can be configured completely Although using these various functions adds a certain amount of complexity compared to using the single DAQ assistant function by using these various functions the user is able to have full control of their particular application In order to configure an analog input or output task there are five basic functions that must be utilized The first function is the Create Virtual Channel function A virtual channel maps in software to a physical channel Consequently this function specifies what particular analog input or analog output channels are being utilized in the application The second basic function is the Sample Clock Timing function This function is used to specify what type of data transfer method is going to be used in the application For instance one can either specify that a continuous acquisition utilizing the FIFO is going to be used and the size of the buffer or that a single point hardware timed acquisition will be performed 59 Next the Start Task function is the third function that must be utilized to configure an analog input or output task The Start Task function takes a task that has been defined and actually executes it
134. sen to operate on a given sample is selected depending on the current value of the amplitude estimate 2 1 2 3 Multiple Channel Combination and Gain Normalization This next step involves the combination of EMG recordings obtained from several electrodes placed adjacent to each other on the skin surface above the same muscle The combination of multiple channels is performed because the signal to noise ratio SNR 12 improves with an increase in the volume of muscles recorded Since the gain and the distance from the muscle differ for each electrode used the gain normalization process occurs after the channels are combined Research has indicated that using several electrodes for measuring the EMG of a muscle greatly increases EMG amplitude estimation accuracy In particular SNR performance improvements of up to 91 have been observed when using multiple channels compared to using a single channel 8 While there are benefits of using multiple channels there are also some disadvantages In particular the chance of defects that may arise due to noise shorted electrodes and other abnormalities increases when the amount of channels used is increased 2 1 2 4 Detection Smoothing and Relinearization Once the signal is whitened and combined if multiple channels are used the actual amplitude estimation process begins The remaining three parts of amplitude estimation are detection smoothing and relinearization The methods used to impl
135. sis 20 Choosing which type of analysis method to use in an application simply involves choosing which function library on the functions pallet to use The standard functions available use array based analysis If array based analysis is desired using Express VIs one must pick the Express VI which maps to the standard array based function on the Express VI function palette Similarly if point by point analysis is desired one must merely chose from the functions listed in the point by point subsection of the functions palette Most of the functions available to perform array based analysis have an analogous function available to perform point by point analysis Figure 2 12 displays three types of filter functions that utilize each of the three analysis techniques While the 42 array based and point by point filter displayed only implement a Butterworth filter the Filter Express VI implements various types of FIR and UR filters F d Instance 2 b ae Filtered Signal ip eae ee oe oe Figure 2 12 Array based Express VI and Point by Point Butterworth Filters 2 7 Chapter Summary This chapter has detailed background information that is associated with our project In particular information is presented detailing the type of signals that our workstation will be analyzing along with the types of applications that our workstation will be involved in One design option for creating a biomedical signal acquisition syste
136. stem failure However the frequency with which these outliers occur is an important measurement to acquire While these large outliers cause for system failure if they occur infrequently such as on the order of hours LabVIEW may still prove capable to meet our hard real time needs As a result another test must be conducted to examine the frequency with which these outliers occur 5 5 Chapter Summary In this chapter we investigated ways to determine the latency delay caused by LabVIEW functions We developed a module that measures this latency and then analyzed an EMG amplitude function utilizing three different methods of processing We found that functions that utilize point by point analysis exhibited the lowest latency and functions that make use of Express VIs exuded the highest latency However while the average latency is acceptable for the point by point and array based routine using a buffer size of one these algorithms occasional may exhibit outliers that are above our maximum latency threshold As a result further tests must be performed to determine 83 how frequently these large outlier values occur If these outliers occur infrequently such as on the order of hours then it is possible that this LabVIEW system can still be utilized to meet our hard real time requirements 84 6 Hard Real Time LabVIEW Performance In the previous chapter latency tests were conducted that compared system performance when different f
137. t were a normal function Figure 2 11 displays an example of a CIN that is connected to two 32 bit signed integers After everything has been connected in LabVIEW the C code c file can now be created By right clicking on the CIN a c file template will be opened This template provides the structure in which the C code must be written to compile correctly After the c file is created the user must compile the CIN source code and load the CIN object code 31 Figure 2 11 Example of CIN Configuration 31 Appendix A offers detailed instructions on how to compile and load the source and object code using Microsoft Visual Studio NET Once the C code has been successfully loaded 38 into the CIN the CIN will act as a standard function in LabVIEW that will perform the specified function outlined in the c file 2 6 5 LabVIEW Analysis Methods Another unique feature of Lab VIEW is that there are two distinct types of analysis methods available The standard analysis method used in LabVIEW is known as array based analysis In array based analysis all computations are array oriented Consequently the first step in using this type of analysis is to define a buffer size Once a buffer unit of data is available this buffer will be able to be analyzed by LabVIEW functions and routines After a buffer unit of data has been analyzed LabVIEW VIs then wait for another buffer unit of data to become available Analysis is only performed a
138. t64 Numeric 1 TD2 TD2 TD2Hd1 typede Fh typede Macros MgErr CINRun TD1Hdl InputArray float64 NormalizedValue TD2Hd1 Array2 MgErr CINRun TD1Hdl InputArray float64 NormalizedValue TD2Hd1 Array2 int32 index Index of the filter array selects which row to use ine32 1 MgErr err nokErr float64 result Insert code here int32 numOfFilters InputArray gt dimSizes 0 Number of Rows in the array the number of filters defined int32 numOfTaps InputArray gt dimSizes 1 Number of Columns in the array number of filter taps if err SetCINArraySize UHandle Array2 paramNum numOfTaps return err 111 Array2 gt dimSize numOfTaps index int NormalizedValue index int32 NormalizedValue numOfFilters if index lt 0 index 0 if index gt numOfFilters 1 index numOfFilters 1 result Array2 gt Numeric result 0 for i 0 i lt numOfTaps i result InputArray gt elt index numOfTaps i resultt return noErr 112 Appendix C MATLAB Code to Simulate Adaptive Whitening Filter Parameters Below is a MATLAB M file that was created to simulate the parameters that need to be entered into Ted Clancy s M file to create a matrix of filter coefficients for adaptive whitening function mvc0O mvc50 cal_sim lnth
139. tandalone embedded system is often difficult to program and maintain over long periods of time Moreover an embedded system might require the development of customized hardware to implement various algorithms The drawback to making customized hardware is the notion that duplicating the hardware can be difficult and the hardware might need to be frequently updated when new analysis algorithms are implemented The second approach is to design a PC based system This type of system utilizes a standard desktop computer with a data acquisition card and commercially available components to amplify and isolate signals The PC would then have a software package that can be executed to perform desired biomedical signal acquisition experiments A PC based workstation is cost effective as it bypasses the need for custom instrumentation by using a currently available PC and inexpensive acquisition equipment Consequently we have created a PC based workstation that will be utilized to conduct these various biomedical experiments Moreover the software that we utilized is National Instruments LabVIEW LabVIEW is a graphical programming language for the Windows operating system that is used for various signal processing applications Since it is a graphical programming language it is fairly easy to learn and operate Similarly since the workstation will be run on a Windows PC it can readily be used and adopted by a wide audience Furthermore drivers f
140. tency This notion is enforced in Figures 5 2 5 3 and 5 4 These three figures display the average latency in milliseconds versus the number of filter coefficients used for the highpass smoothing and adaptive whitening filters in the EMG amplitude estimation routine These three graphs respectively depict results for when LabVIEW is run with normal high and real time priority 11 Array Based vs Point by Point Normal Priority Latency Analysis Q 3 50 3 00 J 2 50 2 200 e Array Based Buffer Size 1 2 s Array Based Buffer Size 10 S 1 50 Point by Point 1 00 S 0 50 les aan lt x 0 00 512 1024 1536 2048 Filter Taps Figure 5 2 Latency of EMG Array Based and Point by Point Algorithm Normal Priority As all three graphs indicate the point by point routine always executes with the lowest latency In fact the average latency for the point by point routine remains relatively close in value while the latency incurred while running the array based algorithm grows steadily when an increased number of filter coefficients are utilized Moreover even when a buffer size of one is utilized for the array based routine the point by point routine executes at a substantially lower latency This latency difference can be attributed to the extra overhead that is associated with executing an array based function in LabVIEW However even with this additional overhead it is appa
141. the beginning of this chapter it was noted that there was another data transfer method available with the PCI 6229 This other data transfer method allowed for the user to specify a buffer size Once this buffer size is reached data are then transferred between the PC and DAQ board However using this data transfer method is not optimal 62 for simultaneous input and output routines The buffers that are used to hold analog output before it is transferred to the DAQ board are formulated in software in LabVIEW As a result while the analog input is controlled by the clock located on the DAQ device the analog output is controlled by the operating system and the PC As a result the analog input and output routines although they might have the same sampling frequency are not guaranteed to operate and update at the same time since Windows XP is not a real time operating system and a thread context switch can result in system delay The hardware timed single sample transfer method is thus desired for simultaneous input and output routines because the same clock located on the DAQ device is utilized As a result hardware timed single sample transfer is the data transfer that should be utilized if simultaneous input and output is desired This notion was further cemented by Garrett Earnshaw an applications engineer for National Instruments who reiterated that single point hardware timed data transfers are the only type of transfer that should be uti
142. tus will check the status of a background operation that is currently running and will output a Boolean called running which indicates the status of the background operation GetStatus also outputs the data array containing the inputted data that were acquired via the AlnScBg VI While GetStatus outputs the array containing the acquired data this data is represented as an array of A D count values In some instances in order for more meaningful results that can be analyzed these A D count values should be converted to equivalent voltage values The Universal Library VI ToEng accomplishes this conversion by taking the data array outputted from GetStatus and outputting a multidimensional array containing the appropriate voltage values Each row in this multidimensional array contains data collected from each analog input channel Lastly in order to stop any background operation that is in progress the VI StopBg must be used This VI simply takes the context data structure as input that the GetStatus VI outputs The StopBg function will then output an error code that is passed as input to the ErrMsg VI This ErrMsg VI will then translate the error message into a readable string 51 DataArray buts ect Figure 4 1 VI That Acquires Data in Background Continuous Mode 52 The previously described module successfully acquires data from the DAQ board However this method acquires data as an array and thus these acquired data can on
143. ty of EMG signal information is contained in the frequencies between 15 and 400 Hz However when EMG signals are measured noise is also present with the valid signal information There are various sources for this noise present in EMG signals One source for this EMG noise is a result of electrode motion artifact This electrode motion artifact can be caused by mechanical disturbance of the electrode charge layer and deformation of the skin under the electrodes Artifact due to electrode motion is present below 20 Hz Also cable motion artifact is another possible source for EMG noise The cables that connect the electrodes to the electrode amplifiers have an intrinsic capacitance Consequently if these wires are subjected to an electric or magnetic field current is generated which in turn will induce a voltage The cable motion artifact typically has a frequency range of 1 to 50 MHz 10 In order to suppress the noise present in EMG signal a high pass filter is commonly used This high pass filter can be incorporated into the hardware instrument that is acquiring the EMG or it may be implemented in software Implementing the filter in software generally makes it easier to implement higher order filters Consequently these filters can have cutoff frequencies as close to 15 Hz as possible and thus the loss of useful signal information is minimal However even if a highpass filter is modeled in software a hardware highpass filter is still typi
144. unction types were utilized to conduct an EMG amplitude estimation algorithm Results from these tests indicated that the average latencies of the point by point and array based routine using a buffer size of one were below our maximum latency threshold However while the average latency was viewed as acceptable by looking at the raw data it is apparent that outliers that exceeded our maximum latency threshold were present Thus another latency test was conducted that served as a measurement of how frequently these outliers occurred If the frequency of these outliers is small then it is still possible for this PC based system to conduct our previously described hard real time experiments This chapter presents our testing procedure and the results that were garnered when this test was executed 6 1 Final Testing Procedure MTBF For our final latency test procedure we wanted to determine how long our system could run at various sampling rate before losing a data point due to a large outlier In many hard real time systems a dropped point can be tolerated without issue as long as too many are not dropped in a specific time period In order to find out how often our system was dropping points we came up with a system to measure the Mean Time Between Failures MTBF of our system This created procedure can be seen in Figure 6 1 85 Hardware Timed Single Point Y ai SampleClock Sample Clock v Minimum value 26 m jax
145. uting PCI DAS6402 16 board uses a single FIFO and DMA controller for both analog input and analog output 30 Thus when one background operation is using the FIFO for either analog input or output another background operation cannot be initiated While this would not be a problem if one wanted to simply input data and process it at a later time the presence of a single FIFO and DMA controller for both analog input and output tasks is a large problem when it comes to creating a hard real time system with analog output and input As a result the lack of simultaneous input and output limits the board s uses in areas such as prosthetics and control While the PCI DAS6402 16 is incapable of doing simultaneous input and output the PCI 6229 contains a separate DMA controller for analog input and analog output Consequently the PCI 6229 is architecturally designed to conduct simultaneous input and output routines However if simultaneous analog input and output is desired one will need to perform single point hardware timed analog input and output Figure 4 5 depicts how simultaneous input and output is conducted using the hardware timed single point transfer method The process data module located in this function block diagram is a generic process that is meant to represent any data processing that can take place to the data acquired from the analog input channels before the data is passed as output to one of the analog output channels However in
146. voltages acquired by the electrodes must be amplified This amplification occurs by connecting electrode amplifiers to each electrode In the currently available system up to sixteen EMG electrode amplifiers can be used While these electrode amplifiers can be commercially bought electrode amplifiers designed by WPI students have also proven to be effective These amplifiers are made from an Analog Devices AD620BR Instrumentation Amplifier 26 These electrode amplifiers are connected to a signal conditioning box where each EMG signal is first placed through an eighth order Sallen Key high pass filter with an 28 embedded gain of 10 and a cutoff frequency of 15 Hz This filter is used to eliminate motion artifacts and offset potentials and allows for more accurate signal measurements After being high pass filtered the signals are then amplified with separate gain settings available for each channel In the current system there are selectable gains of 1 2 4 8 16 32 64 and 128 Each signal is then electrically isolated for patient safety Electrical isolation is achieved by using an Agilent HCNR201 Photo coupler in conjunction with operational amplifiers Next each signal is low pass filtered with a fourth order Sallen Key low pass filter with a cutoff frequency of 1800 Hz The signals are then available to be acquired by an A D converter on a PC In this system a 16 bit Measurement Computing PCI DAS6402 16 A D converter card with 32 di
147. xGain if FlagMax gt 0 error Repeated MaxGain selection end FlagMax 1 if isempty varargin ArgsVar 0 MMaxGain varargin ArgsVar end if length MMaxGain gt 1 error MMaxGain not a scalor end if MMaxGain lt 0 0 error MMaxGain must be non negative end otherwise error Bogus option varargin ArgsVar 1 end else break end ArgsIn ArgsIn 2 ArgsVar ArgsVar 2 end Required arguments and regular optional arguments if ArgsiIn lt 3 ArgsIn gt 5 error Bogus number of non option pair arguments int2str ArgsIn expected 3 5 end if isstr SandN error First argument SandN not numeric end SizeRes size SandN if length SizeRes gt 2 SizeRes 1 1 amp SizeRes 2 1 error First argument SandN not a vector end if length SandN lt 2 NNfft error First argument length too short end if isstr sCal error Second argument sCal not numeric end if length sCal gt 1 error Second argument sCal not a scalor end 115 if sCal lt 0 0 error Second argument sCal must be non negative end if isstr Noise error Third argument Noise not numeric end SizeRes size Noise if length SizeRes gt 2 SizeRes 1 1 amp SizeRes 2 1 error Third argument Noise not a vector end if length Noise lt 2 NNfft error Third argument length too short end Order 60 Default if ArgsVar gt 0

Data Display, Acquisition and Feedback System for Biomedical

Contents

Download Pdf Manuals

Related Search

Related Contents