Doppler Radar for Biomedical Measurements - 2004
Contents
1. Name Derek2 22 04 3ft 2 4t Date Of Collection 4 19 2004 Start Time 17 18 08 du UU E UU End Time 17 20 12 n 17 18 22 17 18 24 17 18 26 17 18 28 r Apnea Episodes Heart Rate Scan Rate 1 200 in 17 18 20 17 18 22 17 18 24 17 18 26 17 18 28 Jump to Time Index i Respiration Rate 17 18 25 1073820 17 18 22 17 18 24 17 18 26 17 18 28 Movement 05 Figure 15 A Screenshot of the GUI 36 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto When using the Jump To Time Index box the user can enter time indices in five manners where d stands for digit 0 9 dd dd dd d dd dd dd dd d dd or d Dd dd dd or d dd dd This is translated by the GUI to be absolute time For example if a dataset begins at 21 00 and the user enters 22 35 00 it would move to 1 35 00 into the data Dd dd or d dd This is translated by the GUI to be absolute time without seconds for data sets longer than one hour or as only minutes and seconds for data sets within a single hour For example if the user enters 15 15 in a data set within an hour the data moves to h 15 15 where h is the hour in which the data set falls However if the data set were longer than an hour then the GUI would interpret that as 15 15 00 and move there D This is translated by the GUI to mean seconds from the start of the data If the data started at 1 00 00 and the user entered 350 the GUI would displ2. Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 2 4GHz units were built to test an array configuration The 2 4GHz units performed well and were used throughout the project A block diagram of the 2 4GHz radar units is available below Lower Ant Gain SdBi SdBm lidBm 1dBm 15dBm i 10dBm i i i 10dBm 4 i ANT i 2dBm i 22dBm i ATS 11dBm 4dBm 12dBm 295dBm High Ant Gain 20dBi Figure 3 Block Diagram for 2 4GHz Radar 7 1 1 Transmitted and Received Signals RS The radar will transmit a single frequency signal T t cos 2aft t where f is the oscillating frequency and o t is the phase drift of the oscillator The received signal can be found as 2af 1 20 2a 1 20 R t cos 2af S EE t po e c c where c is the velocity of propagation and d t is the distance to the target The signal will be reflected by the target which is at a distance d The total distance to the target will 11 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto also have a time varying displacement component x t which will consist of the vital signs and any other movement Thus the distance between the transmitter and receiver is d t d x t After substituting this equality for d t the received signal becomes d t d t e r 1 2 29 R t cos 2aft c 1 E dj t a E c c where A
3. i O 120303 TC Thesis J 042004 J Thesis Presentation Preparation 04 20 04 04 30 04 Thesis Presentation 04 30 04 Doppler Front End Consult w Dr Scott on Existing Designs Training w Alan Macy and Dr Scott Purchase 2 5 GHz Prototype Supplies 10 0903 amp 2 Determine Conceptual Antenna Design 15403 Determine Final Array Configuration 120303 Colect Small Apnea Data Ses omma 2 Collect Abnormal Data Sets 02 27 04 10 Collect Movement Data Sets 02 2704 1 2 ColedlongApneaDaaSets onoo 8 Collect Long Sleeping Data Sets 0 19 04 47 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto Manpower Hours DSPEnim O o 1 Determine needed DSP Engine Specs mozo 60 Determine Hardware or Software Implementation 10203 Determine Preliminary Design mozo 1208608 60 Develop Respiration Algorthm 30104 40 UI System B cdi c o4 y nvestigate Existing Techniques and Methods 10 10 03 11 02 03 Evaluate Existing GUI 10 10 03 11 02 03 Determine needed GUI Specs 11 0208 Determine Redesign or Modification 11 02 03 G Preliminary Front End Done 12 03 03 03 22 04 Final Front End Done 03 22 04 04 20 04 Back End Development 03 22 04 04 20 04 Back End Complete 042004 S ystem
4. used IC devices Table 1 indicates the reliability data for a single radar antenna Component Failure Analysis per year million ne UroS300P vco 1887438 08288 VNA 25 Amplifier A 1 876000 0 1142 LATA2 Attenuator 8 60000000 001 ALY44MH Mixer 2937784 0 3404 VarabeCapado oos O O 04007 Variable Resistor OJ 019 o 218037 Surface Mount Capacitor 000181 0 2066 ema T 6609 incline Terminator Resistor 900877 0 6587 BNCioSMACabes o ooa Power Supply S S o 8800 Total Failure Rate per million hours MTTF Hours 30964 0412 MTTF Years 3 5347 Table 1 Component by Component Failure Data The mean time to failure for a single radar unit is 3 53 years However since we are using two radar devices that can work independently the system MTTF is 7 07 years 17 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto The system is redundant because should one device malfunction the other would still collect valid data The MTTF for these designs is lacking and largely cut short by the failure rate of the variable resistor There is also a design flaw on the existing prototypes related to this resistor The potentiometer will short out the power supply when adjusted to zero ohms since there is no additional resistance in series with the potentiome
5. A DSP for Doppler Radar Sensing of Vital Signs IEEE 2002 Polk Charles and Elliot Postow Handbook of Biological Effects and Electromagnetic Fields Florida CRC Press 1986 43 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 18 19 20 2 22 23 24 25 26 Proakis John G and Dimitris G Monolakis Digital Signal Processing Principles Algorithms and Applications New Jersey Prentice Hall 1996 Ramachandra K V Kalman Filtering Techniques for Radar Tracking New York Marcel Dekker 2000 Stearns Samuel and Ruth A David Signal Processing Algorithms in Matlab New Jersey Prentice Hall 1996 Yuen C K and D Fraser Digital Spectral Analysis California CSIRO 1979 Proakis John G Dimitris G Manolakis Digital Signal Processing Principles Algorithms and Applications Third Edition Prentice Hall New Jersey 1996 DeMaw Doug Practical RF Design Manual Prentice Hall New Jersey 1982 Prat Timothy et al Satellite Communications Wiley amp Sons 2003 Droitcour Amy et al Range Correlation and I Q Performance Benefits in Single Chip Silicon Doppler Radars for Noncontact Cardiopulmonary Monitoring IEEE 2004 www radiolab com 44 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 9 2 Appendix B Budget AB The following items were purchased with Biopac funds during our first
6. 5 1 Vital Sign Monitoring and Apnea 5 1 1 Description of Apnea DL Sleep Apnea also referred to as Sleep Disordered Breathing SDB 2 affects approximately 6 to 7 of the American population about 18 million An episode of apnea regardless of nomenclature is defined as the cessation of breathing for ten or more seconds Depending on the severity of apnea in a given patient there may be anywhere from five episodes of apnea in an hour to hundreds of episodes in a night The number of apnea episodes in an hour is referred to as the Apnea Hypopnea Index AHI 3 5 1 2 Apnea Studies AB The most expensive and reliable method of diagnosing the severity of apnea is found at sleep clinics and is known as polysomnography It is considered the gold standard in determining the seriousness of sleep disorders Polysomnography uses physiologic sensor leads that monitor brain electrical activity eye and jaw muscle movement leg Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto movement airflow respiratory effort chest movement heart rate and oxygen saturation 5 Video cameras are also used to monitor the patient s body movement throughout the sleep study period This study however requires a large number of sensors directly on the patient to monitor the values of interest To study the brain s electroencephalogram six electrodes are needed at certain places on the head Contact with a multitude o
7. 60 72 j8 76 377 431 72 j8 16 0 72 j0 0344 _ 138422 72770 6 73834 42 76 737 138345 76605 191675 5 191598 8 76 737 Therefore the calculation below shows around 72 of energy will be reflected back t 0 72 j0 0344 50 Ball Lamppa Marrero Selina Sugimoto Doppler Radar for Biomedical Measurements 9 5 Appendix E Permittivity and Conductivity of Biological Tissue as a Function of Frequency DVt t 60 0 t0 0 el I I 80 10 L00 co 0 Ol 1L 0 89 0 gs 0 89 0 390 L0 0 c0 0 090 wa poo q HOUA y r 910 soo tLS0 0 LETOO LI0 0 rrl00 10 0 6710 0 9c10 0 auog oc LV 17 680 M S800 LO 690 t9 0 St 0 8t 0 seo 410 danew oad uig SI C781 c8 0 18 0 080 6 0 68 0 1 0 8t 0 sto 8t 0 9t 0 IE 0 67 0 ot 0 870 170 610 710 Sr o ctUo apeu mya ueg 9 00 Ez 01 L6 0 6 0 T00 L O L 0 99 0 So t6 0 Ls 0 0 0 89 0 r9 0 6 0 1t 0 ScOo rco amp aupry ut S Ait nonpuo Lc 00 SZ 0c t r11 601 c00 9L 0 t L0 Sot 8 0 6 0 t 9 0 T90 uaads a 8t v OT 0 1 6 0 t L0 rai 860 t1L 0 09 0 ts 0 0L 0 tto 9r 0 cFr 0 9r 0 iro oro LTO 910 s10 Ilo si o 960 0 10 760 0 10 6800 tio dun JINI 87 400 LC SI J 8t c8 0 L 0 800 6 0 660 660
8. permeability and conductivity as well as the wave s frequency Refer to the CRC Handbook of Biological Effects of Electromagnetic Fields 17 for a discussion and the governing equations for transmission attenuation and skin depth calculation Here is a table of calculated skin depths for various tissues at various frequencies Frequencies of interest are the 915 MHz and the 2450MHz rows Follows is the equation that governs skin depth for all materials 55 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto DEPTH OF PENETRATION OF AN ELECTROMAGNETIC WAVE IN BIOLOGICAL TISSUES AS A FUNCTION OF FREQUENCY Tissue Frequency Muscle Fat MHz Saline Blood skin Lung bone Depth of Penetration cm 433 2 8 3 7 3 0 4 7 16 3 915 2 5 3 0 2 5 4 5 12 8 2 450 1 3 1 9 1 7 2 3 7 9 5 800 0 7 0 7 0 8 0 7 4 7 10 000 0 2 0 3 0 3 0 3 2 Figure 17 Skin Depth as a Function of Frequency The human body of course does not consist of a homogenous material single tissue with a single boundary at the skin Each tissue has its own permittivity and conductivity characteristics The biology of this is interesting but only worth a cursory mention factors such as the average water content of a tissue cell size and shape intra and extra cellular ion concentration and plasma membrane structure play a part in defining the cell s characteristics A table of the conductivities and permittivities of the various tis
9. Integration TE Waters We coed ncorporate GUI DSP Engine and Radar 02 01 04 02 20 04 Prototype Finished 04 01 04 04 20 04 Deliver Prototype 05 01 04 04 30 04 Determine Preliminary Design 11 02 03 12 03 03 48 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 9 4 Appendix D Reflection and Absorption Supporting Calculations KS The complex amplitude of the reflected wave Intrinsic wave impedance of free space 7 7 A x 4z x1 pal code SEN E T E Oo 107 0 107 eo J o 36r 2xQ 5x10 367 Intrinsic impedance of skin E To 10 ej 33x10 1 2 92x10 j1 59x10 z u Ar x107 4 An x107 J 9 367 2z 2 5x10 E 4z x1077 2 92 x10 j1 59x10 8 67x10 5 j1 998x10 2 92x10 j1 59x10 2 92x10 7 j1 59x 10 8 526x10 7 2 53x10 7 jJ J 43 321x10 1 8x10 1 Lg 73 l 2e Nile a jb r 2 41 1x107 3 24x10 41 42x107 23 77 x10 r 4 324x10 0 tan tan 2 945x10 16 41 1 1x107 Polar Form wl jzC1641 1 f 3 77x10 2e 61 4678295 49 Ball Lamppa Marrero Selina Sugimoto Doppler Radar for Biomedical Measurements Trigonometric Form f 61 4e7 61 4 cos 8 205 jsin 8 205 61 4 0 989 0 143 60 72 j8 76 _ 60 72 j8 76 377 _ 316 28 j8 76 _ 316 28 j8 76 437 72 j8 76 437 72 j8 76 437 72 j8 76
10. SL 0 69 0 96 0 c6 0 L8 0 98 0 90 88 0 8 0 t8 O t 0 8tC 0 6 0 9 0 0tr 0 90 800 0 800 cs 0 9L0 0 cso 9 0 0 Ts o sejnoipuadsad Ipsa pAs a pojuauiouou Jayfeaed 3psaur IENS v og ZHD 4 67 87 ZHD 14 97 HW 0 Lc sc t ZHN of c Ic ls 6l 8I ZHN ZH 001 7H 01 amp xuanbo44 51 Ball Lamppa Marrero Selina Sugimoto Doppler Radar for Biomedical Measurements 0t t L t SL S P Ol X 7 Ol x Ol x I T 0 t 0 Sr c 0 8 96 8 6 SL 9 L9 t9 9 8 8 tL CL L9 c ooz LE OrO c L8 OrL c 000 t O87 0187 0t9 006 c 000 I 008 sl 6 L1 0 Is Lp St 08 9 06 06 68c Lt C 08 cet osc Sc 99 LS 161 061 007 607 91 LC8 tv 008 00t 096 Ol 8 Lt 0t 1 S lp 1 8 t9 9 S6 68 T0C 061 66r I Eb 069 C 06 c 00 71 006 01 8t c 9 0 C Is os 18 9L 1L 8 Olp TSE Ice 0st l 097 301 01 OI Ol AVAL oAne os tist 001 9 LE 8E tE ttr ES cr 6t Lt se 9t 89 9 St 6L LL S9T IST 00 8t 0L61 0L6 l Ol X PI 09L 6 XST Ol x os X 8 Ox el X op Ol x 8 xst OMXS 0 0l LORS S Lt t 0t L0 7 8r 8s 6S LS 0L t9 t 89 tL L9 181 791 TOC L8I 061 0Z1 0 17 006 1 0te c o0
11. Safety of Radiation AB Radiation safety issues need to be considered when using radar on a human subject Radiation standards are governed by OSHA and standards have been set by IEEE and ANSI The primary health effect caused by non ionizing radiation is heating when the subject absorbs the transmitted energy Prolonged radiation exposure above the OSHA levels can produce adverse health effects At the operational frequency of 2 4GHz radiation power density at the subject cannot exceed 2mW cn based on the ANSI and IEEE standard Figure 4 is the IEEE standards of transmission 15 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto Power Density mw cm2 Figure 4 IEEE Standards for Frequency of Transmission The power density p at the subject is as follows p refx 977 Inserting values for the above we obtain p lt W m 2204 8 11 6 02 dB 29 02 dB m 0 00125 w m Zr The IEEE ANSI standard is F 2m w cm 20 w m The power density at the target is well below the maximum allowable levels of radiation Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 7 1 5 Radar Reliability RS The mean time to failure MTTF for our two radar antenna array is 7 07 years This is based on data obtained from an online component reliability database http www sercoassurance com the srda and from Minicircuits the manufacturer of the
12. ZEE H B5 H S5 B Figure 9 Control Pulse in Time left and Frequency Magnitude Response right One expects the pulse signal to appear periodic and it does What was unexpected was the harmonic nature of the magnitude response The first and second harmonics occurring roughly two and three times above the characteristic peak frequency They appear as a result of the density variances in the blood that the sensor picks up To extract a similarly appearing pulse signal from the radar these signals must be 28 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto approximated from the radar signal However the radar does not pick up these harmonics and so they will have to be artificially created from the ambient noise occurring at these frequencies First though a general discussion of extracting the pulse signal from the radar follows Because the pulse signal is a few orders of magnitude smaller than the respiration signal the respiration frequency data must be removed before pulse analysis can begin The lower frequency higher magnitude respiration signal is cut off with a lowpass filter with a stop band up to 1 Hz The resulting signal then has a lower average amplitude and is fitted for analysis Next the positive and negative peaks of the 1 to 3Hz range relative maxima and minima respectively are found and tallied The positive peaks are sorted and the peak ratio the ratio of one peak s amp
13. much more complex than what is necessary for the application A smaller and simpler data acquisition system designed for this application would make the system more economical and practical to setup An ideal feature would be onboard memory to store the night s data for later analysis In addition it would be appropriate to develop a stand that can be adjusted in width and height to setup the radar antennas in a person s home for a temporary installation Hardware Design Flaw The potentiometer on the radar boards will short out the power supply when adjusted to zero ohms since there is no additional resistance in series with the potentiometer to prevent current flow This causes the monolithic amplifier in the circuit to fail Two fixed resistors arranged as a voltage divider could provide the target control voltage without the use of the potentiometer in future prototype models Not only would this eliminate the flaw but it would also replace the least reliable component present in the prototype and increase the MTTF per unit to 10 88 years increasing the system MTTF to 21 76 years 40 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto Poor Heart Rate Recognition The heart rate signal is very difficult to locate in the radar data Noise from surrounding electronics and from drift with the VCO make the characteristic frequency peak nearly impossible to distinguish from noise in the signal A more accu
14. the more complex oscillations that characterize power transmission in the near field 17 When the near planar wave of energy hits the body tissue the system acts as a typical wave hitting a boundary with different characteristics than the current medium In this case the wave traveling in free space air contacts the skin of the patient As with any such interaction part the wave is reflected back to the source while the remnant is transmitted into the skin The reflection coefficient that defines what is reflected back is dependent upon the frequency of the wave the permeability conductivity and permittivity of both sides of the boundary Its value is determined by using a ratio of the 54 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto pre and post boundary impedances defined by the aforementioned characteristics A discussion of the coefficient can be found in the CRC Handbook of Biological Effects of Electromagnetic Fields 17 pages 14 15 The governing equation is found below pak 1 mt Where T is the reflection coefficient and 77 is the wave impedance of medium 1 and 2 Continuing the discussion of the air to skin transmission the transmitted wave inside the boundary will continue to penetrate the body but it will attenuate at an exponential rate dependent on the skin depth of the material The skin depth 8 is also a function dependent on the medium s permittivity
15. the peak finding algorithm This frequency becomes the center of the bandpass filter whose transition bands are then placed on the nearest relative minima on either side of the peak This band of data can then be put back in the time domain via the Inverse Fast Fourier Transform IFFT and displayed for each window The phase shift is still present in the extracted respiration from the radar but only visible when compared directly to the control respiration signal By averaging the respiration frequency the main peak for the time windows in a given time spread we can easily determine the respiration rate for a whole dataset or any subset therein Figure 8 shows 50 seconds of original radar data in blue while the purple waveform is the isolated respiration data The main frequency component of the respiration signal has been extracted and can be clearly displayed 26 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 0 25 0 2 0 15 5 1 15 20 25 3 35 40 45 50 Figure 8 Original Radar Blue with Extracted Respiration Purple This algorithm proves to be very robust in finding the respiration from the radar data under all conditions and positions of the body beneath the radar antenna system When apnea episodes occur in the same time window the frequency response is not drastically affected the apnea effects which appear as noise because respiration stops leaving only pulse data and sy
16. 9r c 008 1 00t tl 00 LC 008 tC Ol X 01 Ol xe 901 Ol x eI sot Ol x Ct 901 Al 88 UL US x ol XTT QI t ZH9 0I E IE oE 7H5 fe 8c ra 7HD fs St vC ZHW w s 44 Ic 7HW Zr 61 8I ZHN fi 91 t 7H ofn l el 7p of Il Ol 6 ZH L 9 7H w t 7H afz I t 7HD O14 TE it 52 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 9 6 Appendix F Interactions of RF Energy and Biological Tissues DL This section will give a brief introduction into the theory of boundary conditions of a traveling RF wave hitting the various tissue types of the body Also explored are the permeability and conductivity of said tissues When thinking about interactions between electromagnetic waves and biological tissue the concepts are no different than considering the interactions between a wave of any frequency and a boundary and material of any conductivity permeability and permittivity In all types of body tissue magnetic permeability is very closely equal to Ho or the permeability of free space This fact makes most calculations involving body tissues simpler The more important concepts to consider at the frequencies the prototype will operate at 900MHz and 2 4GHz are the reflectivity coefficients and skin depth of penetration because from these energy absorption and the distribution of energy can be derived While the specifics of energy
17. B ISM Band Industrial scientific and medical radio frequency bands Apnea The cessation of breathing for more than 10 seconds during sleep Hypopnea Slow light breathing during sleep Pulse Plethysmograph Sensor that detects changes in blood density to determine pulse rate AHI Apnea Hypopnea Index A measure of how many apnea episodes occur in one hour An AHI of five five episodes per hour is defined as minor sleep apnea while cases with anywhere from 15 20 episodes an hour are defined as significant and pathological this is the cutoff line above which treatment is recommended Polysomnography Most reliable method of diagnosing apnea It considers brain activity chest movement and heart rate eye and jaw muscle movement leg movement airflow and oxygen saturation Dataviewer Team generated tool for visual manipulation of datasets GUI Graphical User Interface DSP Digital Signal Processing VCO Voltage Controlled Oscillator Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 5 Introduction RS The purpose of this project is to use a RF device to monitor a subject s heart rate and respiration without making physical contact This will be accomplished using continuous wave Doppler radar antennas and data post processing in Matlab to extract the vital signs The subject s movement and apnea episodes are also monitored and displayed on the Matlab GUI
18. B eene en nnne enn enne en nnne enne nns 45 9 3 Appendix C Project Timeline DL eese nennen ener enne 47 9 4 Appendix D Reflection and Absorption Supporting Calculations KS suus 49 9 5 Appendix E Permittivity and Conductivity of Biological Tissue as a Function of Frequency 51 9 6 Appendix F Interactions of RF Energy and Biological Tissues DL sss 53 9 7 Appendix G Breakdown of the Major Radar Front End Components KS 58 9 8 Appendix H Radar Evaluation and Design KS 60 9 9 Digital Appendices eec eiie tete ete e CH e eL Eee dene et a eie de ds 66 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 2 LIST OF TABLES Table 1 Component by Component Failure Data sss enne 17 3 LIST OF FIGURES Figure 1 Sleep Study Patient in Full Gear esee entree nennen nete ennt 7 Figure 2 Test Setupis s sedet Lr b tti eene dta eret emet nie eds 10 Figure 3 Block Diagram for 2 4GHz Radar eese eene nennen enne rennen enne 11 Figure 4 IEEE Standards for Frequency of Transmission c cccssesseesseesseeseeseeeeeceseessecnsecnaeenseenaeeneeaes 16 Figure 5 Sample Screenshot of the Dataviewer essere eene nennen 21 Figure 6 Control Respiration Black vs Radar Signal Red sene 24 Figure 7 Sample
19. Doppler Radar for Biomedical Measurements April 20 2004 A THESIS Submitted to the faculty of the Electrical Engineering Department Of New Mexico Tech in partial fulfillment of the requirements for the course EE 482 482L Senior Design Project BY Senior Design Team 4 Aghavni Ball Derek Lamppa Nicolas Marrero Robert Selina Katsuya Sugimoto Electrical Engineering New Masco Institute of Mining and Technology Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto ABSTRACT The Doppler Radar for Biomedical Measurements Project is sponsored by BIOPAC Systems a medical research equipment manufacturer They requested that the Electrical Engineering design team modify a preexisting radar prototype to extract a subject s respiration and heart rate from a distance without making contact with the patient After discussion with the customer Senior Design Team 4 decided to focus on developing the prototype towards apnea studies The system consists of a two antenna continuous wave radar array that receives raw data and transfers it to a signal processing system that consists of adaptive bandpass filters that extract the heart rate and respiration signals from the reflected data signal Movement in the signal is also identified in addition to identifying the patient s respiration rate heart rate and apnea specific data including apnea episodes per hour and per session This is done with a Matlab gene
20. G G ESI Wa ug O Ax R Signal to Noise Ratio S N The signal to noise ratio needs to be considered at this stage The S N ratio shows can be used to determine how difficult it will be to extract the desired signal from the noise present in the system The noise is defined as N kTB where k is the Boltzmann s constant 228 6dBW T is temperature and B is the noise bandwidth Combining two equations gives us the S N ratio expressed as follows S BG G N 42 R kTB A signal to noise ratio of greater than 10dB is the typical way to set up these system s efficiency Power calculation at the transmitter receiver and target The power density at the target is obtained from following equation p PG G W Any R Pt 20dB 10dBm Gr Gt 8dB 8dB1 The parameters are divided to the numerator and denominator After converted into dB the parameters which are listed in the numerator are simply added and the parameters on the denominator are subtracted Knowing this figure one can adjust the power 62 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto transmitted and received to determine if sufficient signal strength is available at the RF input of the mixer This is discussed in greater detail in Appendix G Phase Noise In order to obtain the highest functionality of the system typical signal distortion such as phase noise or jitter needs to be considered Let say that a per
21. Radar Magnitude Response for a 10 second Window eee 25 Figure 8 Original Radar Blue with Extracted Respiration Purple eee 27 Figure 9 Control Pulse in Time left and Frequency Magnitude Response right sss 28 Figure 10 Resulting Pulse signal extracted from radar Black Against Control Pulse Sensor Red 30 Figure 11 Data Collection Interrupted by Subject Movement essere 31 Figure 12 Apnea Episode from Control Sensor and Radar Signal esee 33 Figure 13 Standard Windows Menu Interface in the GUL essere 35 Figure 14 Patient Information Dialog Box sees ener nrenneneen enne ntn 36 Figure 15 A Screenshot of the GUI eee ee epe t eicere beet ge i tertie edo eds 36 Figure 16 An Example of User Oversight essent rennen ennt enne 38 Figure 17 Skin Depth as a Function of Frequency eese enne nee nnne 56 Figure 18 LO Circuit Equivalence eerte te eerta b tete Dp ete e de eda entes 58 Figure 19 Fundamental Oscillator Circuit Representation eese enne 59 Fig re20 VCO Representation uen eee hee egre deerit e etes e bands 59 Figure 21 Phase Noise Diagramm ne egere IR tert E dre cT Bice tete ce re es 64 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 4 LIST OF ABBREVIATIONS AND DEFINITIONS A
22. adar Evaluation and Design H Digital Appendices I 42 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 9 1 Appendix A References l 2 3 10 11 12 13 14 15 16 17 http www sleepnet com sleepapnea html http www lungusa org diseases sleepapnea html http www sleepapnea org slpaprsk pdf http www sleepclinic org apnea html http classes kumc edu cahe respcared cybercas sleepapnea trenpoly html http www talkaboutsleep com sleepbasics viewasleepstudy html http hyperphysics phy astr gsu edu hbase sound wavplt html c2 Balakrishnan A V Kalman Filtering Theory New York Optimization Software 1987 Baranski S and P Czerski Biological Effects of Microwaves Pennsylvania Dowden Hutchinson amp Ross 1976 Battocletti Joseph H Electromagnetism Man and the Environment Boulder Westview Press 1976 Chen Kun Mu et al An X Band Mircrowave Life Detection System IEEE 1986 Droitcour Amy et al A Microwave Radio for Doppler Radar Sensing of Vital Signs IEEE 2001 Edmonds D T Electricity and Magnetism in Biological Systems New York Oxford 2001 Embree Paul M and Bruce Kimble C Language Algorithms for Digital Signal Processing New Jersey Prentice Hall 1991 Johnk Carl T A Engineering Electromagnetic Fields and Waves New York John Wiley amp Sons 1988 Lohman B et al
23. and the black is the reference signal data 23 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 2 100 102 104 106 108 110 112 114 115 118 120 ul Figure 6 Control Respiration Black vs Radar Signal Red The radar data was normalized for a better visual comparison Looking at Figure 6 it is clear that the radar carries the respiration rate data In order to extract the pure waveform an adaptive bandpass filter was created which locates the characteristic frequency peak of the respiration in the frequency domain and follows it through time windows that progress through the dataset The adaptive bandpass filter beings with the data being filtered using a low pass filter with a stop band beginning at 3Hz to remove all noise in the radar greater than the limits of the biological signals respiration can occur at a rate up to 1 2 Hz while heart rate is assumed to stay below a 3Hz 180 bpm rate The data is viewed in a ten second window that slides over the data in five second increments The time domain data in the window is then run through a process called the 24 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto Fast Fourier Transform FFT which yields a set of complex numbers containing the frequency magnitude and phase responses of the data in the time window The magnitude response is of interest because it is here that the characteristic freque
24. applications Pulsar recommends as a minimum the use of a 17 dBm level mixer 58 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto Voltage Control Oscillator A fundamental oscillator circuit is shown in Figure 19 Figure 19 Fundamental Oscillator Circuit Representation In order to oscillate the total phase shift of this closed loop has to be 360 degrees and the A l nour project we used a 2 4GHz oscillator V 1 BA gain must be one Tuning Voltage Crystal resonatar Output Frequenc Amplifier i y Figure 20 VCO Representation Figure 20 is the common configuration of the crystal oscillator Most oscillators operate P dx at parallel resonance where the reactance vs frequency slope df 1s inversely proportional to C1 the motional capacitance of the crystal unit The principal mechanism of the voltage controlled oscillator is that the circuit is only marginally stable Thus the output of the system oscillates at the constant amplitude and frequency 59 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 9 8 Appendix H Radar Evaluation and Design KS In this section we will introduce some concepts that would have been necessary in the full design of a radar system The design flow and power are discussed in the flow chart below 1 Determination of adequate signal and frequency whic
25. ay data at 1 05 50 If any of these times fall outside the range of the data then the GUI will move as far towards these times as the data allows When displaying data the GUI automatically adjusts the y axis of the viewing windows to show the data as large as possible for that window Because of this the y axis of the viewing windows changes every time the data being displayed changes If the user wishes to stop this behavior the Freeze Axes menu under the Edit menu holds indicators for all four axes In addition the user can Freeze Unfreeze all four axes at once When an axis is 37 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto frozen its minimum value stays at the minimum value of the entire dataset likewise for maximum Three additional features incorporated into the GUI are e The ability to modify patient information under the Edit menu if a mistake is made initially e The ability to modify the sampling rate if it is other than the default 200 Hz This is done in the Scan Rate input box e The ability to create Debug logs that are helpful in troubleshooting or modifying the functionality of the GUI The Debug Mode of the GUI produces an output file GUIDebug txt which will show the reader what callbacks and subroutines the GUI is in along with any pertinent data about variables in those routines With access to the debug file and the source code for the GUI debugging any er
26. c f Since the period of oscillation for the vital signs is much larger than d c and x t d we can approximate the received signal as R t cos at E MD di 24 c The received signal is similar to the transmitted signal with two differences There is a delay due to the distance d between the transmitter and the target and there is also a phase change due to the periodic motion x t The mixer multiplies the received signal with the local oscillator signal extracting the changing phase component The resulting signal can be characterized as B t cos 0 n do where 12 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto Ag t dt d E 2d C is the constant phase shift related to d and Ard 10 0 J A is the changing phase shift Thus one can see that the respiration and heart rate signals will be visible on the output of the mixer regardless of the fact that they do not produce a Doppler shift 25 7 1 2 Power Density Transmitted KS Knowing that the vital signs should be present in the received signal we need to confirm that their magnitudes will be measurable We determined the power density transmitted by the radar and found that it is the maximum permissible under current ANSI IEEE regulations The power density of the radar is described as follows p EE S Re Ext I7 This can be expanded to 13 Doppler Radar for Biomedical Measureme
27. ce menus Figure 13 the user can load a saved data set save a processed data set close the current data set print the current data set exit the program modify patient information modify the data display axes obtain help on using the GUI obtain version information about the GUI or put the GUI in a special operating mode called Debug Mode Fie Edt Help Open Patient Information x cS Save Freeze Axes b Radar Axes na ad a Close Heart Rate Axes n UH Debug Made Print Respiration Axes i Movement Axes Exit All Axes Figure 13 Standard Windows Menu Interface in the GUI When a user loads a data set the GUI asks for the patient s name date of collection and collection start time as shown in the dialog in Figure 14 It then processes the data by sending it out to the Respiration Heart Rate Movement and Apnea algorithms and stores the returned data presenting it to the user The user can then scroll through the data using the mouse or using the Jump To Time Index box as shown in Figure 15 35 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto Bl x1 Please Enter Patient Information Patient Name Derek2 22 04 1 5ft 2 4G Date of Collection 4 19 2004 Start Time hh mm ss 10 00 00 Cancel Figure 14 Patient Information Dialog Box Doppler Radar For Biomedical Measurements os ni x File Edit Help E Radar Data Patient Info
28. distribution inside a tissue is rather complex and beyond the scope of our research we are still interested in net energy and thus power absorption to determine if our power output will be within OSHA defined limits It is worthwhile to note that for the antennas to be used and at the frequencies transmitted all occurrences will happen in the far field region of the antenna The far field region is defined as the region of space a minimum distance from the emitting antenna to infinite that the transmitted energy waves can be seen as a predominantly plane wave character electric vector E is perpendicular to the H field vector 17 The 53 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto region of operation for our device 0 5 m at the closest to 2m nearly assures operation in the far field region of the antennas which is defined by the equation 17 Where D is the larger dimension of the antenna and with wavelength A as defined by 15 Wavelengths for the frequencies of our devices are 33 cm for 900MHz and 12 5 cm for 2 4GHz These yield far field regions beginning at 27 cm D 20 32 cm and 16 cm D 10 8 cm respectively The fact that the prototypes will be operating in the far field region assures simpler power calculations and considerations the power density travels as one over the square of the distance traveled in the far region as opposed to having to calculate
29. enna gain can be obtained by the following equation 47A xD ie er Antenna gain is proportionally related to the area A of the antenna and inversely proportional to the wavelength For this project one reason why the 2 4GHz band was selected was the smaller antenna size From the above equation above the antenna efficiency 77 also can be determined as follows GA sa EU Ta MA 65 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 99 Digital Appendices The following files are available on the CD ROM Executable File Source Code Sample Data Sets User Manual 66
30. f sensors can cause discomfort in the patient and make the results more inaccurate due to changes caused by higher stress levels The respiration monitor is also a source of uneasiness as seen below in Figure 1 It is a sensor that requires a tight secure fastening around the chest which is understandably not comfortable Figure 1 Sleep Study Patient in Full Gear The specific values of interest that polysomnography collects for sleep apnea research are brain waves respiration heart rate and body movement While brain waves cannot be monitored without electrodes it is ideal to find a way to monitor the other three without making physical contact with the patient in order to reduce stress and collect more natural data The development of such a product is the focus of this project Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 6 Background Information on Doppler Radar Applied to Apnea Studies 6 1 Doppler Radar 6 1 1 Radar Equation RS The Doppler effect is defined as a shift in the frequency of a wave caused by the relative motion of the transmitting source the reflecting object or the receiving system The Doppler effect will influence the data received from the radar in the following ways The Doppler Effect or Doppler shift can be described mathematically as follows where the v is the relative velocity of the targets respect to the radar f is the tran
31. fect VCO would produce an ideal sine wave V t Asin at However there is inherently some noise introduced to the signal This can be represented by fluctuations in the amplitude of the signal and by fluctuations in the signal phase We can represent the noisy oscillator signal as V t A a t sin 2af t t where a t represents the amplitude noise and P t represents the phase noise Amplitude noise can be removed to some degree by using automatic level control ALC systems However phase noise is another matter It is very difficult to remove The equation can be rewritten ignoring amplitude noise V t Asin 22f t t The picture below shows the effect of e t 63 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto SineWave with Phase Modulation ideal sinewave phase modulated Voltage o 3 0 00E 00 5 00E 07 1 00E 06 1 50E 06 2 00E 06 time s Figure 21 Phase Noise Diagram Phase noise is a serious source of interference such as timing error in the RF design Especially in our system which obtains information from the phase shift measurement Minute amounts of phase noise on a transmitter signal can result in the transmitter causing significant interference to other services whereas minute amounts of phase noise on a receiver local oscillator can severely reduce the receiver selectivity or cause other undes
32. h can provide sufficient information of the target Measurement T Doppler Shift Phase Shift Detection Signal to Noise Ratio S N 2 Power calculation at the transmitter receiver and target Received Power Density at the Target Power Absorbed by the Target ES Reflectivity Radar Cross Section Safety of Radiation ANSI OSHI FCC Phase Noise Impedance Matching 60 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 3 According to the information above choose antenna types and specifications Antenna types Antenna power and gain Antenna dimension Based on the desired power density the antenna specifications such as antenna type gain dimensions and power can be determined 4 Fundamental RF circuit design and analysis Component types Ens Semiconductors Three primary components VCO Mixer Filter Determination of adequate signal and frequency Higher frequency operation shorter wavelength can make the antenna size smaller It can also increase the range of the unit and make any phase changes more distinguishable When designing a Doppler radar we need to determine if the target moves fast enough to give us a Doppler shift If not a phase detector type application is the alternative Actual received power at the antenna is obtained by the following equation 61 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto P
33. ime domain the x axis is always in terms of seconds referenced from the start of the data collection and the y axis is the voltage ofthe signal For the frequency magnitude domain the x axis is expressed in Hz for ease of recognition of the characteristic signal peaks The six checkboxes on the left below the window labeled Options are functions that allow a user to further explore the datasets e Hold Enabled Should a user wish to overlay two different channels for a more direction comparison he should use the Hold function This holds whatever datasets are currently displayed is in the window and plots the next selected dataset over it This can be repeated indefinitely 20 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto e Zero Padding Adds zero padding to the frequency domain output to produce smoother curves Team4 Data Viewer ie E3 M LI E M ri i eeg Figure 5 Sample Screenshot of the Dataviewer 21 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto e DC Offset Removal This subtracts from the dataset or time window the mean of the set This removes the DC component which appears as a large peak at 0 Hz in the Frequency magnitude domain Removing this large peak makes displaying the rest of the frequency components easier e Time Windows This function allows a user to divide the dataset into windows of a user defined leng
34. irable effects 26 Minimizing phase noise is one of most important factors in the selection of a VCO and is frequently a selling point in the VCO spec sheet Impedance Matching Impedance matching is another important aspect of RF circuit design Matching the source impedance and the load impedance produces maximum output power and avoids creating standing waves A mismatch caused by these can result in interference when they add or subtract signals on the transmission line This is avoided by careful design 64 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto during the board etching process Our boards were professionally designed and etched by MiniCircuits Antenna Types There are many different types of antennas a few examples are helix yagi and patch antennas Patch antennas are employed in our project because of their small size but yagi antennas were very appealing due to their increased directionality Antenna Dimension It is appropriate to compare the antenna s dimensions with the size of the wavelength instead of just evaluating its physical size For instance a 30m high AM antenna is small in terms of its wavelength which is 300m One advantage of using microwave frequency bands is that the wavelengths and thus the antennas are rather small In our project the antenna has a physical size of 9 9cm x 10 7cm This is efficient when we consider the wavelength which is 12 5cm Ant
35. isting of 200 samples a second 200Hz for accurate analysis in the processing system of the Matlab program 7 2 2 Matlab and Algorithms NM The data collected using AcqKnowledge is post processed in Matlab The raw feeds from both radars are received and the following are extracted e Heart Rate Signal e Respiration Rate Signal e Movement Signal The numeric heart rate respiration rate the magnitude and duration of motion and occurrences of apnea The Graphical User Interface GUI is a final version of a test program written for the project called the Dataviewer 19 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 7 2 3 Dataviewer NM For visual manipulation of the data a user controllable program called Dataviewer was created to test the algorithms written for the project It has the capability to display any portion of any channel in the time domain frequency magnitude domain and the phase domain It also has numerous other functions that will be individually described below Figure 5 on page 21 shows a screenshot of the Dataviewer Datasets are loaded in via the Load Data From File button see Figure 5 The main window of the dataviewer is the graphical display presenting the data from the selected channel over the window The window is fully scalable by using the mouse controlled zoom function inherent in Matlab plots This allows a user to investigate data in detail For the t
36. litude to the next is considered If this peak ratio is greater than 0 5 then the peaks are not far enough apart The pulse characteristic frequency peak stands above the surrounding noise by about 4 times larger so that when the highest ratio differential is seen the initial peak is treated as our characteristic frequency peak of the pulse data Once this peak is located the two surrounding relative minima become the cutoff bands of the adaptive bandpass filter that follows the main component of the pulse signal Once again this signal is IFFTed and displayed in the GUI This only reproduces a single sine wave that contains the main frequency component of the pulse similar to Figure 8 the respiration signal To generate a signal that looks more like the reference pulse the harmonics have to be artificially induced The radar does not pick up the harmonics from the interaction with the movement of the heart so a technique must be created to simulate their presence This is done by 29 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto multiplying the characteristic frequency bandwidth range by two and three and amplifying the signal noise held therein This produces a reasonable approximation of the natural harmonics of the pulse signal and yields the results as seen in Figure 10 25 Ts 0 5 0 5 Al 0 200 400 600 800 1000 1200 1400 1600 1800 2000 Figure 10 Resulting Pulse signal ex
37. lter As a result only the phase component remains V cos 6 2 Technical Contributions to Prototype AB Biopac contributed their two functioning radar prototypes and the design of the 2 4GHz radar to the project They also supplied a copy of their AcqKnowledge data acquisition software and the necessary MP100 acquisition hardware The 2 4GHz radar design and the AcqKnowledge MP100 system are used in the final prototype Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 7 Apnea and Vital Signs Monitoring Subsystems 7 1 Antenna Array AB Our antenna array consists of two 2 4GHz radars mounted perpendicular to the subject and sixty centimeters apart They are positioned one meter from the subject This distance can be increased to two meters though signal quality is degraded A photo displaying a test setup is shown below in Figure 2 Figure 2 Test Setup The purpose of two antennas is to be able to have a signal reflected off the back of the subject 1f they are lying on their side Data quality is diminished when the subject is lying on their side but the respiration signal is of a higher magnitude when reflected off the subject s back Two antennas allow us to collect data in this manner in a wider variety of sleeping positions The 2 4GHz radar designed by Dr Chip Scott for Biopac was never constructed After testing the prototype 1 85GHz and 900MHz units a second 900MHz unit and two 10
38. m peaks for each second increment are also found Similar to the movement algorithm the average value and the maximum and minimum peaks are compared but for the case of apnea the peaks must be within a certain bandwidth threshold That is if a given second has both its maximum and minimum peaks within 0 1 of the average value for the second it is safe to conclude that there is no respiration component present and that it may be part of an apnea episode thus raising a flag for the second Using the tallying system of the movement algorithm if ten or more consecutive seconds occur with raised flags then the time set is considered to be an episode of apnea It is considered as such until two consecutive seconds have lowered flags then the apnea episode has a beginning time index length and an ending time index Once a dataset has been analyzed in this manner the total number of episodes are counted up and divided by the total length of the data set in hours This number is then displayed in the GUI as the Apnea Hypopnea Index AHI This is the tell all signifier that indicates whether or not a patient has a diagnosable level of sleep apnea 34 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 7 3 GUI Overview Data Display and Manipulation NM The GUI serves as an interface between the user and the algorithms developed to extract signals from the radar data From the standard Windows interfa
39. ncy peak is found with the peak finding algorithm Figure 7 shows a sample magnitude response in the frequency domain of an average time window 150 100 50 Figure 7 Sample Radar Magnitude Response for a 10 second Window The DC component has been removed using the DC offset algorithm The DC bias in the radar data comes from the bleed through from the LO to the RF input of the mixer on the radar device hardware This resolution is typical of a ten second window for visual purposes the signal can be zero padded for higher resolution but it is clear that the respiration characteristic peaks are visible here In this time window these peaks correspond to a respiration rate of approximately 0 2Hz one breath every five seconds which is perfectly reasonable for a person at rest 25 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto Because the time window is a rectangular function being applied to a specific portion of the dataset one may expect to see sidelobe leakage in the frequency domain 22 This is typically something to be concerned about and using a Blackman or Hanning window instead of a rectangular window can diminish this effect However for the purposes of this project it is not necessary There is not significant energy leakage to make these peaks indistinguishable they are simply too significant to be lost in the leakage This characteristic frequency peak is easily found with
40. nd Phase Shift RS sss 8 6 2 Technical Contributions to Prototype AB nennen nenne 9 Apnea and Vital Signs Monitoring Subsystems sse 10 7 1 Antenna Array AB 5 iere coser dtes Seed te tee Er RE CA EE eae eS 10 7 1 1 Transmitted and Received Signals RS sse 11 7 1 2 Power Density Transmitted KS esses 13 7 1 3 Retlected Enetey IS ati nd wiih t eoe E e te dues 14 7 1 4 Safety of Radiation AB 14 o salvis e deeem tete et eie ate pe deg 15 7 1 5 Radar Reliability RS edet petet beet a tabe aee d cee 17 7 2 Radar Data Acquisition and Signal Processing NM sssssssssseseeeeeee 18 7 2 1 The MP100 and AcqKnowledge NM sss eene 18 7 2 2 Matlab and Algorithms NM eese 19 7 2 3 Dataviewet NM ine cette ie ehe EE R E t e ea t dud ees 20 7 2 4 Respiration A D E AON EE tetas eisetat iae ende aeree s 23 7 2 5 Heart Rate DD xa cuia etes tiet e a tes 27 7 2 6 Movement Algorithm DD a3 optet be ede ie e ies pite nies 30 7 2 7 Apnea Episodes RS sse eene enne nnne ener innen enne nenne 33 7 3 GUI Overview Data Display and Manipulation NM sese 35 Recommendations for Future Work RS sss ener nnns 40 APPENDIGCES 4 E A A tite ee t CP ER ER EOD er E evo Pap onam bits vee teat Gs 42 9 1 Appendix A References ee a eee tu tuere t aimed 43 9 2 Appendix B Budget A
41. nts Ball Lamppa Marrero Selina Sugimoto aa leis da zo sz x10 Bt 2 5 Em 0 72 G7 6815x107 W m i 2 1 E 1 70 72 0 0345 la mi jj Dz ml ml Where the intrinsic impedance of free space is H 7 E The transmitted wave from our antenna is left as notation in this calculation ml because Et will vary depending on the antenna gain 7 1 3 Reflected Energy KS It is very important for the radar system to receive enough reflected energy so that it is able to compare the differences between the RF and LO signals For this reason the ratio of absorbed and reflected energy must be calculated The ratio of the energy that is reflected back from hitting a target is expressed in the following equation a ml E 7E 12 Where 7 is the intrinsic impedance of the objects 14 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto In this case the boundary between free space and skin is used For our research the parameters such as relative o conductivity relative permittivity and relative u permeability of human skin are 1 0 33 and 4nx10 respectively We obtained the ratio of 1 0 72 0 034 j This result shows that around 70 percent of the energy is expected to be reflected back to our radar Supporting calculations are available in Appendix D 7 1 4
42. r as movement they are all output as one six second episode of movement The output is sent to the GUI which isolates these ranges from the radar data and displays them in the movement channel of the GUI The GUI backend also makes note of these ranges for use by the other algorithms which disregard them in their processes 32 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 7 2 7 Apnea Episodes RS In the radar signal data set an apnea episode is characterized by the disappearance of the respiration signal for a ten second interval As a result the amplitude variations of the radar drop significantly and the signal is reduced to the heart rate and the system noise Figure 12 shows the control respiration sensor picking up an apnea event red while the black curve is the raw radar data feed The apnea episode occurs between 30 and 40 seconds Figure 12 Apnea Episode from Control Sensor and Radar Signal 33 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto Since the large signal of the respiration has disappeared the radar signal only varies by fractions of what it once did As a result a modification of the movement algorithm can be used to determine the starting time and duration of apnea episodes The beginning of the algorithm is still the same the signal is partitioned into one second increments and analyzed The average value maximum and minimu
43. rate VCO and a less noisy environment could make this data more visible Nulls in Range When the local oscillator and the received signal are either 0 or 180 degrees out of phase null points occur Thus nulls are found with a target distance of 4 4 from the radar With our transmitted frequency of 2 4GHz these nulls occur every 3cm which makes them nearly impossible to avoid when monitoring a patient A quadrature radar transceiver can eliminate these null spots and would be a valuable improvement on the existing radar hardware Information on a quadrature receiver designed for vital signs monitoring can be found in the IEEE paper Range Correlation and I Q Performance Benefits in Single Chip Silicon Doppler Radars for Noncontact Cardiopulmonary Monitoring by Droitcour et al Post Processing Post processing was determined to be acceptable for this application Should the system be modified to run in real time however it could be made substantially more versatile and could be used in applications such as infant monitoring to prevent sudden infant death syndrome 41 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 9 APPENDICES APPENDIX TITLE SECTION References A Budget B Timeline C Reflection and Absorption Supporting Calculations D Permittivities and Conductivities of Biological Tissue E Interactions of RF Energy and Biological Tissues F Breakdown of the Radar Front End Components G R
44. rated GUI Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto ACKNOWLEDGMENTS The authors would like to sincerely thank the following for their assistance and support throughout the project Dr Robert Bond Dr Ali El Osery Alan Macy Chris Patscheck Chris Pauli Betty Scott Dr Chip Scott Dr Scott Teare Carol Teel Andrew Tubesing In addition we would like to thank MiniCircuits www minicircuits com and Mouser Electronics www mouser com for providing hardware samples and reliability information Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 1 TABLE OF CONTENTS Und U Lb No TABLE OF CONTENTS oneee esee tene eite he Le eae eoe e e eee ehe eee ende ere redeo rre ede Rene 3 LISTOF TABLES riari 4 LIST OF HGURES ieee 4 LIST OF ABBREVIATIONS AND DEFINITIONS ABJ essere ener 5 Introduction RS o mec enge E t Rd tr E RIA 6 5 1 Vital Sign Monitoring and Aptriea citet ER DRY TR SER R ded 6 5 1 1 Desctiption of Apnea DL teer e REESE EUST eer ERE VERE ese 6 5 1 2 Apnea Studies AB 3 sce tee eR RISE A ines I RET NER e Ste ieriees 6 Background Information on Doppler Radar Applied to Apnea Studies sess 8 6 1 Doppler Radar e EI as is ees fe eee os TE he eee eee eee eet 8 6 1 1 Radar Equ tiot RS ete ette e e I RE TENTE E Seat aa 8 6 1 2 Doppler Frequency a
45. rds While the power output for the range of our prototypes have not yet been measured the designer of the circuitry estimates that the power output of the antennas are well below this limit This is beneficial because if the resulting data is not discernable we will be able to increase our output power to receive signals back with higher amplitude 57 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 9 7 Appendix G Breakdown of the Major Radar Front End Components KS Mixer There are two signals that are compared by the mixer One is from the local oscillator the VCO and is referred to as LO The second is the reflected signal that comes from the RF antenna and is appropriately referred to as the RF If we represent LO as cos of and RF as coso t we can evaluate their relationship in the mixer as follows cos t cos t A circuit diagram demonstrating how this relationship is achieved is shown below Figure 18 LO Circuit Equivalence As you can see from the circuit a certain level of power is required on the LO input to properly driver the mixer diodes LO power level has to be sufficient in order to turn on the diode in the saturation mode The LO power level must also be much larger than the RF power level in order to prevent load distortion The LO drive level plays a critical role in determining the IP 1 dB Compression Point and Dynamic Range of a mixer For higher performance
46. rors that arise should take much less guesswork than standard debugging Are You Sure You Want To Quit Yes No Figure 16 An Example of User Oversight 38 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto In addition the GUI contains some limited user oversight program redundancies to prevent a user from losing data For example if the user closes a file before saving it the GUI asks the user if they wish to save the current data set In Figure 16 the user is asked to confirm if they want to exit the program If the user enters values incorrectly that the system needs to understand a request in the Jump To Time Index or Patient Information Start Time boxes for example then the GUI will display an error message In some cases the GUI will simply not comply In these cases the Debug Mode output file will contain more specific error messages Code for the GUI is available in the Digital Appendix CD ROM 39 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 8 Recommendations for Future Work RS The prototype fulfills its intended function but is lacking in the following regards e Bulky Hardware and Inconvenient Setup e Hardware Design Flaw e Poor Heart Rate Recognition e Nulls in Range e Post Processing Bulky Hardware The necessary acquisition hardware and PC make this setup impractical In addition the AcqKnowledge data acquisition system is
47. rworth low pass filter and special filters like a manual version of the respiration rate finding algorithm are found in here and can be applied to a given dataset The Load Script and Run Script functions allow a user to employ scripts that automatically use all the previously described functions to display a complex product one such script included with the Dataviewer CheckRespiration is an automated version of the respiration rate finding algorithm using the Dataviewer functions Most of these functions are implemented by the back end of the GUI automatically These algorithms are discussed in more detail in the following sections 7 2 4 Respiration DL The radar waves pick up a phase shift as they encounter the moving chest cavity this phase shift is dependent on the relative velocity of the chest when the transmitted wave encounters the boundary of the chest This phase shift is responsible for the appearance of the respiration rate in the sampled waveforms To understand what sort of periodic signal to extract from the sampled radar data it is helpful to have a reference signal that is the exact respiration signal To this end Biopac provided a pressure sensitive sensor that wraps around the chest and creates an output voltage based on the chest cavity movement Figure 6 shows the use of the Dataviewer to overlay the radar signal over the control respiration signal for a given time window The red curve is the radar data
48. semester VCO 2 4 GHz Amplifier 5 2 5 GHz Mixer 900 MHz VCO 900 MHz Attenuator 12dB Attenuator 15dB Mixer 2 4 GHz Low Pass Filter DC 2 95 GHz Low Pass Filter DC 1 GHz 2 4 GHz Antenna SMA plug SMA to BNC plug SMA jack Variable Capacitor 3 1 2 pF Variable Capacitor 8 8 pF 45 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto In addition to these items we also purchased the following using NMT funds ROS 960PV VCO VNA 25 Amplifier RMS 5H Mixer BLP 150 LP Filter Parts Bin Tin Foil Freight Charges We also received 5 additional LAT 12 attenuators from Minicircuits and three variable capacitors from Mouser Electronics free of charge 46 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto 9 3 Appendix C Project Timeline DL Hours Lp smt End Documentation 0 Statementof Work 100103 60 Preparation for Conceptual Design Review 10 01 03 100703 60 Conceptual Design Review 1007038 o Preparation for Preliminary Design Review 10 07 03 10 2103 60 Preliminary Design Review 10243 Preparation for Critical Design Review 10 21 03 11 0403 60 Critical Design Review 19590403 y Formal Report
49. smitted frequency Ais wavelength and C is the velocity of radiation propagation 6 1 2 Doppler Frequency and Phase Shift RS In this project any change in frequency will be unreadable due to its small size relative to the carrier frequency Respiration produces a contraction or expansion of the chest with a velocity of less than 3cm sec The heart contracts at a typical rate of 6cm sec Inputting this data into the radar equation yields m 10V _ pondo _ 0 1236 A 0 125 Sa 0 9888 Hz 0 125 S Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto This Doppler shift will not be identifiable with a conventional VCO which will vary from its center frequency Any observable information will be produced by a change in the phase shift between the received and transmitted signals How this phase shift is readable is shown below Vo V Foe cosa V cos ot cosy sin ot sin cos wt cos t d cos ot cos o sin at sind cos at 9 1 I I Foro 1 cos 2 t cos o t5 sin ot wt 3 sin ot ot sin 1 1 1 Foe coso 5 cos 2at cos Q A sin 2 t sin 1 5 o Les o cos 2 t cos 9 sin 2oit sin d ZVV eos Q9 cos 2 2z f t cos o sin 2 2z f tsin Where V a COS t is the transmitted signal and V cose et 9 is the received signal cos 2 27f tcos and sin 2 27 tsin are high frequency terms and are removed by the low pass fi
50. stem noise This is effectively zero padding having minor influence on the magnitude of the very low frequency components 7 2 5 Heart Rate DL The pulse rate is picked up by the reflection of the RF energy off the periodically varying position of the heart as it beats Because the radiated energy has numerous boundaries to interact on the approach and return path from the heart skin ribs cartilage blood and 27 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto fat and it also is hitting a moving target with smaller total displacement the returned pulse signal is understandably much smaller than the respiration signal The respiration amplitude is on average one hundred times larger than that of the pulse amplitude This makes isolating the heart rate more difficult and a different approach is required The amplitude of the noise from the system also makes it more difficult to discern the pulse rate from the signal since the ambient noise is of approximately the same amplitude Once again it is helpful to have a reference signal against which to compare the time varying reflected waveforms Biopac provided a sensor for this purpose one that detects changing densities in blood in the tip of a finger Figure 9 shows a reference signal response and its characteristic magnitude response in the frequency domain ENS 8 HB BN
51. sues can be found in Appendix E These properties are noticeably different from tissue to tissue and warrant calculations for each transition 56 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto Because of this heterogeneity numerous different boundaries must be considered for example the path between the skin and the heart has layers of fat muscle bone and cartilage all in varying amounts depending on the patient The calculations to do this are too elaborate to examine here Chapter 6 in Engineering Electromagnetic Fields and Waves contains the procedure for considering multiple heterogeneous regions At the frequencies that the prototypes will be operating at the penetration and thus power dispersal the majority of the transmitted energy is absorbed as the attenuating wave passes through the body After this point most of the power will have either been reflected back or absorbed by the tissues through which the wave has propagated with the minimal remaining amounts of energy continuing on Thus the tissues of concern are the focus of safety considerations The OSHA standard limit for continuous exposure for the frequencies of our operation is 10 W cm Other organizations also release comparable recommendations for legal limits ANSI American National Standards Institute and IEEE release their own limits with the former being stricter than the latter whose values parallel the OSHA standa
52. ter to prevent current flow This causes the monolithic amplifier in the circuit to fail 7 2 Radar Data Acquisition and Signal Processing NM Our Data Acquisition and Signal Processing System consists of two steps The data is collected using Biopac s provided MP100 system in real time and then post processed using a team generated Matlab program GUI 7 2 1 The MP100 and AcqKnowledge NM AcqKnowledge and the associated MP100 hardware are used to digitize the data and save it as a text file for future manipulation AcqKnowledge is Biopac s proprietary data acquisition system and it has more functionality than is required for this application The Biopac resources are for proof of concept any future marketable version of this prototype will include a much simpler data acquisition system that lacks almost all of the functions of the AcqKnowledge program 18 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto Regardless of the final program used to sample data our created processing system requires a text file input consisting of two to four tab delineated columns each corresponding to one set of sampled In the final phase of our product the input to the Matlab program only consists of the two radar channels in two columns For data validity verification purposes heart and respiration control rates can be received as well in the other two channels columns of data The input data needs to be cons
53. th and select any portion of it for display This is helpful to simulate the window by window data processing algorithms and to look at the characteristics of a smaller portion of the data e Normalize Data When comparing two sets of data using the hold function the data can be of different magnitudes Using this function one can define the normalization factor to which the dataset is scaled to allowing overlapping comparison between datasets e Peak Detection This is the peak finding algorithm used in the respiration and pulse rate finding algorithms It analyzes the windowed data and finds the relative peaks throughout placing a mark on each This is best used with the hold function on so that the peaks can be overlaid on the original graph Below Options is the Domain section which allows a user to select the domain that the plot displays The sampling frequency below is used to calculate the time and frequency index used in the plots obviously data sampled at a frequency of 100 Hz could not be displayed accurately on the same time index as that of a 200 Hz dataset To the right of that is the channel select in which a user can choose the channel to be displayed To the right of the channel select is the filter selection in which standalone 22 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto filter programs can be dropped in and applied to the data Standard filters like the Butte
54. to go through the radar datasets and mark sections of the data that are movement To accomplish this the function is passed anywhere from a small time window to the whole data set which parses the passed subset into an X by 200 matrix where X is the number of seconds in the data The maximum and minimum values for 31 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto each second are found and stored Then the mean of each second is stored in an array For any given second if the maximum or minimum peak is above or below the average by a given threshold the second is marked as containing movement The function then tallies the number of seconds in a row that are marked and outputs the duration of the movement as well as the start of movement the second it started based within the data timestamp reference The function has a few additions in it to assure robust operation First since there may be a case where a second of data is characterized by the downward drop of a periodic signal which may be outside the governing thresholds any cases of movement that only last one second are disregarded Also if in the middle of a string of seconds of movement should one second inside the string not meet the criteria to be considered movement it is assumed to be inside a longer string of movement For example should 2 seconds of movement and 3 seconds of movement be separated by one second that does not registe
55. tracted from radar Black Against Control Pulse Sensor Red With these two basic algorithms complete the project was ready to move into the case specific design of creating a program for sleep monitoring To do this algorithms to detect movement and apnea episodes are required 7 2 6 Movement Algorithm DL Unfortunately one aspect of conducting a sleep study on a fully mobile patient is that he or she is just that fully mobile Because the patient is not movement limited by sensor gear or restraints he is able to exhibit normal movement patterns during sleep This 30 Doppler Radar for Biomedical Measurements Ball Lamppa Marrero Selina Sugimoto movement effectively nullifies the radar data during the movement and for a period afterwards Figure 11 shows a sample radar waveform that has movement data When the subject moves he or she causes abrupt amplitude jumps and uncharacteristic frequency response Note that after the movement episode the data does not immediately settle back down into the predictable periodic signal that it used to be On average it takes two to three seconds to fully reacquire the respiration and pulse rates 2 5 Ws 0 5 0 5 180 185 190 195 200 205 210 Figure 11 Data Collection Interrupted by Subject Movement To effectively tally these regions in a collected set of data so that the other algorithms won t consider them a movement locating algorithm had to be created This algorithm is
Download Pdf Manuals
Related Search