Home

Design and Implementation of a Wellness Monitoring System via

1. Frequency Hz Fig 3 14 ECG filter frequency response In determining the frequency response of the ECG hardware the gain of the ECG circuitry was set to a moderate level to not be near the saturation voltage of the amplification circuits Maximal signal amplitudes at this gain level are 6 12 Vpp and 5 12 Vpp occurring at the frequencies 40 Hz and 80 Hz respectively By Figure 3 14 the removal of frequencies components near 60 Hz is im mediately apparent The center frequency of the notch filter was determined experimentally to be 61 0 Hz The lower and upper half power bandwidth frequencies of the notch filter were determined experimentally using the maximal amplitudes on opposite sides of the notch and were found to be 57 Hz and 66 Hz respectively The implementation of our notch filter matches nearly identically 40 Wellness Monitoring System via Body Area Networks 3 5 Tests and Results the desired specification for this filter In examining the frequency response of the high pass filter the cut off frequency was found experimentally to be 16 Hz Our desired cut off frequency for the high pass filter was 0 05 Hz This discrepancy is due to a trade off as described in the implemen tation section Looking lastly to the frequency response of the low pass filter the desired cut off frequency was specified to be 150 Hz In examining the data presented above the cut off frequency of the low pass filter was fo
2. if the difference between the local maximum and current measurement is larger than the prede termined threshold then the possible step satisfies the threshold requirement On the other hand when the difference is smaller than the threshold the possible step is invalid and ignored Next the step detection algorithm will check the time difference between the current possible step with the previous validated step The time interval comparison technique is used to get rid of frequent bouncing noise which produces small fluctuations in the signal that could be counted as valid steps Our algorithm will only count a step if the time difference between the current possible step and the previous validated step is large enough This time difference is required to be greater than 0 1 seconds which approximately represents the slow walking period for each detection of one peak Otherwise if the time difference is not larger than 0 1 seconds the possible step will be 62 Wellness Monitoring System via Body Area Networks 5 4 Implementation treated as bouncing noise and ignored A clear flow of the step detection algorithm will be shown by the flowchart in in Figure Another way to estimate the number of steps that a user has taken is to determine the rate at which the user takes steps and then multiply that by the time interval during which the user is walking To do this we utilized FFT for a set of samples and determined where th
3. Amplified and Notch Filter Filtered ECG Signal Output 1N4728A RFduino s GPIO6 Additive DC bias Fig B 6 Final stage amplifier with a gain of 25 V V and an adjustable DC bias 98 Wellness Monitoring System via Body Area Networks Appendix C Indoor Positioning Test Results These results are from the indoor positioning tests that compares the actual to the measured x and y coordinates which is discussed in Chapter 4 Table C 1 Indoor Positioning Module Experimental Data and Results A l Error m Error x Total error 0 72 2 050560899 0 24 0 024 4 193343296 0 24 1 46851762 12 024 2 88 4 32 12 2 249351907 6 2 4 Measured A E ul a No No OO SS 0 48 1 287975155 0 48 1 550927464 4 652782393 Hs 3 ll I ee re ae On OU UN SU Sh a As el Roa eee NO gt 00 KO GO CO N CO O 0 72 1 679314146 ae 2 2 2 64 4 281427799 A 3 84 4 688300332 2 2 Ha Pu ni NO O pa SN O pa a No 3 O 1 92 1 92 3 258348048 0 72 0 910735966 8 0 48 1 821471932 1 152 0 96 8 64 10 46752234 1 68 3 755060585 4 6 4 56 5 831555539 2 88 0 96 3 642943865 0 24 1 46851762 BE N On E Hs N 4 56 5 53979061 _ a O N 50 NO 00 A oo OO 00 O ES MEM 9 pl Md Ml Mar ESA rl eee Bl e El el See E il O E 5 a ES CO OU NT OU CO Sd DO BH C DO OI CO DO OU HY Od e ej ol CO
4. PLACE OF BIRTH YEAR OF BIRTH SECONDARY EDUCATION TERTIARY EDUCATION HONOUR AND AWARDS PLACE OF BIRTH YEAR OF BIRTH SECONDARY EDUCATION HONOUR AND AWARDS Yang Su Winnipeg Manitoba 1987 Fort Richmond Collegiate 2002 2005 University of Manitoba B Sc Microbiology Graduated with Distinction 2005 2010 Deans Honor List 2011 2012 amp 2014 Thomas H Gillman Electrical Engineering Award 2013 Esther R Steele Award 2012 Provincial Chess Competition 4th Place 2004 Haiyue Wang Sichuan People s Republic of China 1992 Deyang NO 1 High School 2007 2010 APEGM 3rd Year Scholarship 2013 2014 UMSU Scholarship 2013 2014 Financial Aid amp Awards Merit Scholarship 2013 2014 Dean s Honor List 2012 2014 International Undergraduate Student Scholarship 2012 2014 117 Wellness Monitoring System via Body Area Networks Curriculum Vitae Jeff Winkler PLACE OF BIRTH Winnipeg Manitoba YEAR OF BIRTH 1991 SECONDARY EDUCATION Murdoch MacKay Collegiate Institute 2005 2009 HONOUR AND AWARDS Dean s Honor List 2011 2013 University of Manitoba Merit Award 2010 Stella Wujek nee Shandruk Scholarship 2009 University of Manitoba Preliminary Entrance Scholarship 2009 Royal Canadian Legion Branch No 90 Scholarship 2009 118
5. s thigh by the exact same customized Velcro strap with a pocket shown in Chapter 2 Figure 2 7 The SensorTag was configured to have a sample frequency of 50 Hz to acquire the accelerometer measurements The data was transmitted by BLE to the developed mobile application for Android devices which processes the accelerometer measurements to determine the number of steps that a user takes We tested the step detection al gorithm based on the results of the step counter and the speed indicator shown on the mobile device As we specified in the design criteria of section the step detection algorithm should be able to detect steps from walking speeds ranging between 0 7 m s and 1 8 m s In order to test our algorithm with a well defined speed we used the PRECOR treadmill Model C966i which is an American model that has set speeds in mph in the Active Living Centre at the University of Manitoba We five individuals tested the accuracy of the algorithm for three different walking 65 Wellness Monitoring System via Body Area Networks 5 5 Tests and Results speeds which were 0 67 m s 1 5 mph 1 34 m s 3 mph and 1 79 m s 4 mph by counting out 50 steps for each speed The results of the experiment in terms of each individual are shown in Table below Table 5 2 Step Detection Experiment Results Test Test 1 Person 1 Haiyue Wang Walking Speed 0 67 m s 1 5 mph 1 34 m s 8 mph 1 79 m s 4 mph Average Steps Detected
6. 43 67 NN 42 67 NES 43 39 NES Average Accuracy 86 44 Person 2 Cassandra Aldaba Walking Speed 0 67 m s 1 5 mph 1 34 m s 3 mph 1 79 m s 4 mph Average Steps Detected 48 67 NS 44 33 os 43 Average Accuracy 90 67 0 67 m s 1 5 mph 1 34 m s 8 mph 1 79 m s 4 mph ESO as a anl 66 Wellness Monitoring System via Body Area Networks 5 5 Tests and Results 0 67 m s 1 5 mph 1 34 m s 3 mph 1 79 m s 4 mph Aena o unl m an 1 34 m s 3 mph 1 79 m s 4 mph Ansel ul The results showed that the accuracies for each walking speed varied between 85 97 while the average accuracies varied between 85 8 90 7 Based on all of the samples that we tested the average accuracy of the step detection algorithm was approximately 88 Compared to the accuracy we proposed in the proposal the accuracy of 88 which is much higher than the design specification of 70 For the speed detection we performed ten tests for three different speeds which were 0 67 m s 1 5 mph 1 34 m s 3 mph and 1 79 m s 4 mph This experiment used the same SensorTag worn just above the knee on the thigh of the user as shown in Figure The total average accuracy and the accuracy for each speed are listed in Table 5 3 below 67 Wellness Monitoring System via Body Area Networks 5 5 Tests and Results Table 5 3 Speed Detection Experiment Results 0 67 m s 1 5 mph 1 34 m s 3 mph 1 79 m s 4 mph Ac
7. Amplifier gt 200 The amount of voltage amplification that the Gain hardware should provide Leakage The amount of current feedback to the patient lt 10 yA l Current when using this device Notch Filter 60 Hz Frequency components at this value should Cut Off Frequency be removed 21 Wellness Monitoring System via Body Area Networks 3 4 Implementation 3 4 Implementation 3 4 1 Hardware Implementation Contained within the ECG wearable enclosure are two circuit boards two 9 V batteries one switch and one 3 5mm female stereo jack Figure 3 3 The larger circuit board Model BPS BR1 7 05 x 1 85 forms the detection and filtering circuit as it holds all of the electronic components required for amplifying and filtering the electrical signals measured from the electrodes The smaller circuit board Model BPS SB400 3 00 x 1 90 holds an Arduino based RFduino microcontroller as well as a Bluetooth Mate Silver wireless transmitter used in combination to process and then transmit the output of the detection and filtering circuit All aspects of the hardware were designed to be focused around user ease of use as well as product reliability Detailed schematics for each circuit can be found in Appendix B GROUP GO ECG MODULE PROTOTYPE UNIT JUuIZ UOQUES te 9U IZ7 3U0Q1e A6 He A al nsir 4 oluoseued i go tonez Fig 3 3 ECG module internal layout 22 Wellness Monitorin
8. Ol O a a aa Hs IS NO 99 Indoor Positioning Test Results Wellness Monitoring System via Body Area Networks E E a a El g Q a 9 0 910735966 0 910735966 2 894364179 1 550927464 Total error 2 656990779 1 44 1 44 0 24 on 12 0 48 1 92 1 92 1 68 2 64 fas a A 1 44 3 47991724 2 249351907 5 767 195506 0 72 0 72 0 24 0 3 3 47991724 0 96 4 193343296 5 738359347 2 683281573 2 42684981 Error y Error x 0 24 0 24 0 48 2 64 0 24 0 48 4 08 0 24 2 64 y 2 2 5 fase 34 ow AN el i pis par ow is pes ss a6 a 7 rfa ase EE fs resin re re 1 68 as 12 096 suos fas os aa rx c len lt hin Jeo rio Jar lo Ja JN Jon fi Jeo IR J00 Jar Jo 1a JN Jon fae o JR Joo Jaa Jo JG alo ho eo KR Joo Zo SUIS EN NES 69 60 E Sea GO Sea GE Gre rae Pe OR SS SS SS A ED eS A te o ESO 100 Wellness Monitoring System via Body Area Networks Indoor Positioning Test Results EN x y y Eror x Error y Total error 5 11 67 1396 20 3552 409613281 a o 105 162 18 14 2305127725 2 a a2 s5 26 18 3195216469 e a 4 7s a7 06 03 0670820393 sofe fsa a6 2er 416058057 Ce eja tens as 4503 asoma er 6 s 79 945 22 174 2868100417 es 4 ia 35 1225 06 09 1081665883 epops s os 12 15 pos 925 e9 03 225 229652185 n re 875 Ba
9. Olson Electrical Safety in Medical Instrumentation Application and Design 4th ed J Webster Ed New Jersey John Wiley amp Sons Inc 2010 p 659 116 A Bharadwaj and U Kamath Techniques for accurate ECG signal processing Feb 14 2011 Online Available http www eetimes com document asp doc_id 1278571 March 4 2015 17 R Nilsson and B Saltzstein Bluetooth Low Energy vs Classic Bluetooth Choose the Best Wireless Technology For Your Application June 8 2012 Online Available http www medicalelectronicsdesign com article bluetooth low energy vs classic bluetooth choose best wireless technology your application March 1 2015 18 Rafael Saraiva Campos and Lisandro Lovisolo RF FINGERPRINTING LOCATION TECH NIQUES Handbook of Position Location Theory Practice and Advances chapter 15 SEP 2011 119 Zahid Farid Rosdiadee Nordin andMahamod Ismail Recent Advances in Wireless Indoor Localization Techniques and System Journal of Computer Networks and Communications vol 2013 17 August 2013 20 M A Hussian and K S Kwak Positioning in Wireless Body Area Network using GSM International Journal of Digital Content Technology and its Applications vol 3 pp 3 4 2014 21 Jiugiang Xu Wei Liu Fenggao Lang Yuanyuan Zhang Chenglong Wang Distance Measure ment Model Based on RSSI in WSN Wireless Sensor Network 2010 2 606 611 22 Benkic K
10. turn on the Bluetooth service of your Android mobile device by tapping once the Turn On button found under the Bluetooth Settings heading of the module manager screen e You may turn off the Bluetooth service of your Android mobile device when you are finished monitoring data by tapping once the Turn Off button also found under the Bluetooth Settings heading of the module manager screen e Tap once on the Main Menu button to go back to the module selection screen as displayed in Figure C 4 Connecting to the Module s Sensors e Posture Recognition For posture recognition you should first power on the two SensorTag devices by depressing once their central on off switches Figure C 6 Once both devices are powered on tap once on the Posture button to enable the connection with the two SensorTag devices the status of posture will change from OFF to ON e Pedometer For step recognition you should first power on one SensorTag device by depressing once it s central on off switch Figure C 6 Once the SensorTag device is powered on tap once on the Pedometer button to enable the connection with the SensorTag device the status of pedometer will change from OFF to ON e ECG Power on the ECG wearable device by changing the position of its rocker switch from off to on Figure C 7 Once the ECG device is powered on tap once on the ECG button to enable the connection with the ECG device the status o
11. 21 Lrs 275806589 E p A r s a EEE 1 5 02 740 1323 2988 1476 3332674602 pra o owe 475 2102 33 3910702 fe s 0 7 48 12 aorrasisi ra ite 4 a 36 24 4306661531 a gt 4 96 24 288 37401081 3 8 536 10 14 2832 2568 3 822037001_ so 2 10 48 126 336 312 4585103562 Pe r 2 75 109 06 132 14965517 s 5 es o 09 o 2 asra 27 000 rom Ps 5 15 05 135 54 18 5602099788 rl ras s f ss 12 asoma Pes pijas 4 is 36 0 aj 90 sa s5 49 3 10 31881721 ro fife 5 10 48 48 6785225009 Por a is 49 1225 228 09 2451203786 ps fafa s ja joss 0 s 101 Wellness Monitoring System via Body Area Networks Indoor Positioning Test Results Actual i o Measured Tw pa SK QU o pia 1 12 238 09 2451203786 Jl os 12 130640786 8 OO J oil N SERES 00 alol e I ES CO CO on OS CO bo On pa E 99 SS Ot 102 Wellness Monitoring System via Body Area Networks Appendix D User Manual The wellNode Android application is a centralized wellness node that receives and processes the module s information and then displays the results within a graphical user interface In the following pages the application s user manual is shown that guides users throughout the application 103 Wellness Monitoring System via Body Area Networks User Manual Starting the Applica
12. ECG signal Bluetooth Mate Silver Fig 3 4 ECG data acquisition and transmission circuit Power To power all of the hardware components of the ECG module we decided to implement two 9 V batteries connected in a series configuration We selected this configuration since the input to the differential amplifier swings to a minimum voltage below zero volts specifically at the Q and S peaks of the ECG waveform To therefore adequately filter and amplify the ECG signal the op amps require a voltage supply range between symmetrical positive and negative voltages To accomplish this task we connected the negative terminal of the first battery to the positive terminal of the second battery Figure 3 5 This connection between the two batteries is used as ground throughout the electrical hardware components The positive terminal of the first battery is then utilized to provide the positive potential and the negative terminal of the second battery utilized to provide the negative potential 28 Wellness Monitoring System via Body Area Networks 3 4 Implementation GROUND 9 V 9V Fig 3 5 ECG battery implementation In determining what type of battery to implement key considerations included a compact size availability of replacements at local retailers and a voltage rating between 6 V 15 V For these reasons we decided to use two standard 9 V style batteries Both the RFduino microcontroller and Blu
13. ERED SRE HEEB EEE Gee xiii ENE Xiv ACTO Welw 6 aby oes eee eo ee AAA A ee A oe E xvi ee eaaa ea 1 naaa a e 1 anar E aaa 2 A 2 2 Posture Recognition 4 PE aaa rasa ara 4 Spee e tees hase eaeeeeunee ee eae e eos ee ee o 5 Cs AA 5 a As 8 24 1 o AITANA he 8 E OW 4 ge 6 Oe AA AAA 10 2 5 Tests and Results oros daa a ara 15 vl Wellness Monitoring System via Body Area Networks a a 18 a 18 a a ae a 19 dara aaa rara sae 21 aaa aaa al 22 3 4 1 Hardware Implementation 0 0 0 00 2 ee eee eee 22 AE 35 2 9 Tests and Results 2 4 2 oaoa aa ds as a Bs 31 4 Indoor Positioning 0 0 0 ee 42 E 42 Eo AAA 42 4 2 1 WiFi Triangulation Method e o 44 4 2 2 WiFi Fingerprinting Method a 46 4 3 Design Specifications s 002 oo eR Owe ww ee a 46 4 4 Implementation 2 2 a 47 4 4 1 Hardware Implementation 0 000 2 eee ee ee ee 47 AEREA III TEE 49 4 5 1 Otilitve Phasel e cca 4 40 2 a a sa 49 402 _ Online Pas y ss e le ess p aowa OS de e be a 51 4 Tests and Res ltsj s 4 s r essre sesers ROSE erada daneri 55 5 Step Detectioda a a 56 al Briet Itroquciion 4264462888048 Eeu creas sa 56 5 2 Design Specifications lt lt a a 92 A 57 5 3 Background Information ies a a 57 5 3 1 Mechanical Pedometer 0 e o 58 vii Wellness Monitoring System via Body Area Networks 5 3 2 Accelerome
14. Networks 5 5 Tests and Results and step size The test results show that the step detection module can determine the number of steps taken by the user with an average accuracy of 88 In addition the FFT algorithm of the speed detection can measure the average walking speed of the user with an average accuracy of 75 The FFT algorithm can determine the speed of slow walking users with extremely high accuracy However the results also indicate that the FFT algorithm should not be used for normal or fast walking speeds Thus our developed step detection algorithm is optimal for general walking speeds however our FFT algorithm has better accuracy with slower walking speeds 69 Wellness Monitoring System via Body Area Networks Chapter 6 System Integration All measurement devices worn by the user should have the ability to transmit data from the worn device to a server that can store the information for later analysis and examination by healthcare professionals Our wellness monitoring system achieves overall data synchronization by imple menting specific wireless data transmission techniques which transmit their measurements to the developed Android mobile application which then in turn uploads the information to an online server 6 1 Wireless Data Transmission We selected Bluetooth technology to accomplish wireless communications among our devices due to its widespread popularity and reliability However BLE is adequat
15. Posture Recognition Indoor Positioning Pedometer Module Electrocardiogram i Module E Module OE wate Module l i Hardware Hardware Android device l i Android device Io Android device Io SensorTag placed on thigh Hardware SensorTag placed on thigh Wireless Routers placed at i sampling at 50Hz ECG circuit Software Android Application Software Android Application transmission at 50Hz SensorTag pairing and l Data storage Software Android Application transmission at 10Hz Location display Bluetooth pairing and data Data storage Data storage transmission of 300 samples SensorTag pairing and i i Fingerprinting database i sec Algorithm Fast Fourier Transform to determine strides sec Low speed step detection Algorithm Algorithm Fuzzy logic sets to determine sitting standing bending or lying down postures Algorithm WiFi fingerprinting methods ECG data storage and dis play Fig 1 1 wellNode system modular overview The Android robot is reproduced or modified from work created and shared by Google and used according to terms described in the Creative Commons 3 0 Attribution License Wellness Monitoring System via Body Area Networks Chapter 2 Posture Recognition 2 1 Brief Introduction The purpose o
16. RFduino and the Android application The application would send a Ready signal to the RFduino to signal the microcontroller to send the next available ECG value However the times between each received value were not constant and ranged between 3 ms to 43 ms Next we had the RFduino send data in its maximum packet size of 20 bytes 10 2 bit ECG values but when we expected it to take 10 ms the time measured between received packets were approximately 20 ms The RFduino was tested to determine the maximum frequency at which it could transmit its data to the Android ap plication via handshaking It took on average 2929 4 ms to transmit 900 samples with a packet size of 10 2 bit integer data type ECG values These results showed that the RFduino s BLE module was able to meet our design specification however the inconsistent transmission time suggests that the ECG ADC value was not acquired at the desired sampling frequency Therefore the RFduino s BLE module was not used to transmit wirelessly the ECG data to our Android application 38 Wellness Monitoring System via Body Area Networks 3 5 Tests and Results The Bluetooth Mate Silvers wireless data transmission rate was tested by sending 900 consecutive analog read values At the same time a digital output was set to logic high 5V to signify the start of the data transmission then set to logic low 0 V to signify the end of data transmission As seen in the os
17. Serial Bus Universally Unique Identifier Wide Area Network Wireless Body Area Network Wireless Fidelity World Wide Web xvil Wellness Monitoring System via Body Area Networks Chapter 1 Introduction 1 1 Motivation In Canada there is a compelling need for a wellness monitoring system that can provide sufficient and easy access of wellness information to healthcare professionals There is a decreased level of physical activities and an increased in cases of obesity among the population ranging from young children to older adults In fact there is a financial burden on the Canadian economy due to the lack of physical activity costing 5 3 billion and the cost associated with obesity is 4 3 billion 1 The lack of physical activities in the individuals daily lives can lead to various chronic diseases such as cancer type II diabetes and heart diseases 2 In conclusion wellness monitoring can be utilized in the prevention of chronic diseases Currently there is an emphasis in our society for individuals to take responsibility of monitoring their own health to maintain a healthy lifestyle or efficiently recover within a wellness rehabilitation program Such sufficient wellness information includes information about their physical activities sleep patterns and cardiac activities that can describe their overall health aspects Many companies are providing personal wellness electronic devices for health conscious users how
18. and Malajner M and Planinsic P and Cucej Z Using RSSI value for distance estimation in wireless sensor networks based on ZigBee 15th International Conference on Sys tems Signals and Image Processing pp 303 306 June 2008 91 Wellness Monitoring System via Body Area Networks REFERENCES 23 V Honkavirta Location fingerprinting methods in wireless local area networks Master of Science Thesis vol 1 pp 22 Sep 2008 124 J E Graham G V Ostir S R Fisher and K J Ottenbacher Assessing walking speed in clinical research a systematic review Journal of Evaluation in Clinical Practice vol 14 issue 4 pp 552 562 2008 25 S Yang and M Gerla Energery efficient accelerometer data transfer for human body move ment studies Sensor Networks Ubiquitous and Trustworthy Computing SUTC 2010 IEEE International Conference pp 304 311 7 9 June 2010 126 N Zhao Full Featured Pedometer Design Realized with 3 Axis Digital Accelerometer Analog Dialogue 44 06 2010 27 G Thtier and T Verwimp Step detection algorithms for accelerometers E Lab M S the sis Dept Applied Engineering Artesis University College of Antwerp Antwerp Belgium 2008 2009 28 N Ichinoseki Sekine et al Improving the Accuracy of Pedometer Used by the Elderly with the FFT Algorithm Medicine and Science in Sports and Exercise vol 38 no 9 pp 1674 1681 2006 29 L H Sloot M M Va
19. debugging o o o o 0 Server Integration Server implementation and maintenance o Website implementation pe o ooo O O 111 Wellness Monitoring System via Body Area Networks ACKNOWLEDGEMENTS Acknowledgements We like to kindly acknowledge everyone that provided to us the mental and emotional support we required to reach the end of this long but enjoyable journey We would especially like to thank the following people who contributed to our overall success in the development and creation of this project Dr Behzad Kordi thank you for organizing the capstone final year project As well thank you for giving us the opportunity to showcase our engineering knowledge that we have acquired during our academic years in the Electrical Engineering Department of the Faculty of Engineering at the University of Manitoba Mr Sinisa Janjic thank you for supplying to us the various components from the ECE technical shop and as well for ordering the various components we required for this project Without your efforts we would not been able to reach our project s achievements and accomplishments We especially want to thank Dr Jun Cai as our absolutely wonderful supervisor of this project We cannot thank you enough for your continued advice and encouragement that drove the devel opment refinement and eventual completion of this project Your kindness and sincerity in your suggestions always made us f
20. dimensional axes with Euler angles where Ax Ay and A are the accelerometer measurements of the x axis y axis and z axis re Wellness Monitoring System via Body Area Networks 2 3 Background Information spectively The third Euler yaw angle y equation is not displayed because the accelerometer is insensitive to accelerations along the gravitational vector and cannot calculate the yaw angle 8 For example if a user is initially standing with their chest and thigh having a pitch angle of 0 then the user must be in a bending posture when his or her chest pitch angle is 30 while the thigh pitch angle remains at 0 The other method is based on fuzzy logic system that consist of combining imprecise inputs to fuzzy subsets to classify the users most likely posture Generally the fuzzy logic systems are defined in four stages 5 1 Fuzzier Maps numerical data into fuzzy sets in which the number of fuzzy sets describe the number of possibilities called membership functions 2 Rules Connects the previous membership function to the next membership function 3 Interference system Determines the final membership function commonly by an aggrega tion operator and finds the max value 4 Defuzzier Converts the fuzzy sets into a numerical value When comparing the accuracies of the two different methods each method depends on the relative environment user hardware and experiments executed Since the angular thresholding and f
21. is transmitted to an Android device with an embedded Bluetooth module The two SensorTags are placed on the user s chest and thigh encircled in red in Figure 2 4 By computing the Euler angles Equations and or implementing fuzzy logic systems the 3 axis accelerometer values can provide sufficient information to determine the desired postures from the orientation of a static Sensor Tag in 3 dimensional space In conclusion the orientation of the specific body part with an attached SensorTag can be determined through the accelerometer measurements that can be used further to determine the user s posture Wellness Monitoring System via Body Area Networks 2 4 Implementation Chest s SensorTag Android device with BLE v4 0 or greater CC2541 Microcontroller with BLE Fig 2 4 Posture Sensor Tags placement 2 4 2 Software The posture of the user can be determined by the orientation of specific body parts by many meth ods such as angular thresholding and fuzzy logic based systems The initial method we implemented 10 Wellness Monitoring System via Body Area Networks 2 4 Implementation was using an angular threshold which determines a users posture based on the Euler angles that describe the devices orientation in three dimensional space shown in Table 2 2 The pitch angles Equation 2 2 of both chest and thigh accelerometers within a particular range determines specific postures including standing ben
22. likelihood of the user being in a specific 11 Wellness Monitoring System via Body Area Networks 2 4 Implementation posture The wide range of each fuzzy logic function accounted for the different users that would execute each posture uniquely For each posture a score Equation 2 3 was calculated from the sum of the fuzzy subsets output Afterwards a user s posture was determined by the highest posture score O Posture L cPosturex Acz T FePosturey Acy T T cPoiturez Acz T 2 3 FiPosturez Atr T Ft Posturey Aty T FiPosturez Aiz where Sposture is the specific posture s score F ePosture is the specific posture fuzzy subset for the chest accelerometer for each x y and z axis FiPosture 18 the specific posture fuzzy subset for the thigh accelerometer for each x y and z axis and A and A are the axis accelerometer measure ments for the chest and thigh respectively Therefore our finalized design implemented a fuzzy logic method to determine each different posture hence eliminating the Euler angle complexity The posture recognition implementation on the Android application is described in Figure The application receives both SensorTag accelerometer measurements by requesting for the data at a rate of 10 Hz Therefore the algorithm will have semi synchronous data to determine a user s posture due to the latency of retrieving one SensorTag s accelerometer measurements after the other Howev
23. load the corresponding database e Tap once on the Start button to start the positioning process The location will update and display the determined user s coordinates e To record the location of the user tap once on the Record button The button will turn green in color indicating that the application has begun recording The location information with a time stamp will be continuously recorded into a csv file located in the path wellnodel Location By tapping on the Record button once again the button will turn back to red in color indicating the recording process has ended e View the position on a map by tapping once on the Map button The corresponding location will be shown in the map as indicated in Figure C 8 e Go back to the module selection screen as displayed in Figure C 4 by tapping once on the Return button g FE fh Location A ar Aet Map Location r Show results Press start pi g Current wif infolPress start sm read database and show EU C ET ex z O O lt O O Fig C 12 Location and map screens 111 Wellness Monitoring System via Body Area Networks User Manual Electrocardiogram Module Ensure the hardware of the electrocardiogram wearable device is properly connected and affixed to the user To begin displaying the electrocardiogram signal on the screen tap once on the Start button located on the bottom left of the electrocardiogram display
24. real time monitoring of four indications of wellness for patients users wearing the device posture recognition electrocardiography ECG measurements indoor positioning as well as step detection that estimates the users physical activity regime The non invasive devices send their data wirelessly to a local Android mobile device such that wellness monitoring can be achieved simply by virtue of being near to the user Healthcare professionals control and view the mea surements of individual sensors on our intuitive dedicated Android application Measurements are displayed within the application in which each module has a devoted display screen In addition to viewing the data of each sensor in real time measurements can be recorded and transferred to a PC environment for further analysis at a later time Alternatively long term or distant monitoring can be achieved by storing patient information in a database on a remote server that can be easily accessed by healthcare professionals The prototype devices were realized using a combination of hardware design infused with computer software programming to achieve a simple reliable and robust overall product Wellness Monitoring System via Body Area Networks CONTRIBUTIONS Contributions The four modules of our project posture recognition ECG signal acquisition indoor positioning and step detection were broken down and divided among smaller teams consisting of two members e
25. removed together from the enclosure at any given time Figure 3 8 A hole was drilled in the right most top side of the enclosure and the female jack affixed to the enclosure We connected the three pins of the female jack to the detection and filtering circuit board effectively connecting the female jack to the positive negative and reference pins of the INA118 differential amplifier 31 Wellness Monitoring System via Body Area Networks 3 4 Implementation Fig 3 8 3 5 mm stereo plug and jack electrode lead connection We modified the obtained ECG leads by shortening the leads to approximately 18 The shortened leads decreases the amount of induced current that could be created within a lead loop which could contribute additional noise to the ECG signal The electrode leads are grouped together by shrink tubing and a right angle 3 5 mm male plug Neutrik NTP3RC B is attached to the free ends of the three electrode leads Figure 3 9 In all the electrode leads that connect to the wearable enclosure resemble closely and function like a pair of in ear headphones that have become popular with today s mobile electronic devices 32 Wellness Monitoring System via Body Area Networks 3 4 Implementation Fig 3 9 ECG module electrode leads Enclosure The desire to maintain the shortest electrode leads necessitated that the ECG module be placed near to the user s heart We decided that the enclosure for the ECG module would
26. screen Figure C 13 To pause the waveform on the display tap once on the Stop button To record the electrocardiogram waveform measurements tap once on the Record button The button will change to green in color indicating that the recording process has begun To stop the recording process tap once on the green Record button once more The button will change to red in color indicating that the recording process has now been stopped Return to the module selection screen as displayed in Figure C 4 by tapping once on the Return button located at the bottom right of the electrocardiogram display screen JO OA A Eh Electrocardiogram Fig C 13 Electrocardiogram Display Screen 112 Wellness Monitoring System via Body Area Networks User Manual Navigating Through the Website Enter the wellNode website at www wellNode ca Clickthe Log in button to access your personal wellness information profile Figure C 14 Bo Bimpyiwsnotecas D C Bwemotecs wellNode A wellness monitoring application MACRO AAA Fig C 14 wellNode website If you have not registered a wellNode account yet click the Register button that will lead you to the registration page Figure C 15 o Fill in all the required fields that include your username password and personal information first name last name phone number date of birth sex weight and address o After click the Register butto
27. separate Shared Preferences Therefore the Shared Preferences is an integral part of our wellNode client server data flow As shown in Figure 6 11 the Android Shared Preference holds two different Preference profiles One of the preference profile is for user identification information which is displayed on the user profile activity while the other preference holds the wellness sensor information of the latest update The pathway for user identification starts at the log in activity screen that allows the user to either access the application or register a new account If the user attempts to log in the application will take the username and password and parse both pieces of data into a formatted command called JavaScript Object Notation JSON The parsed JSON data will then pass through a PHP script that takes the JSON data and manipulate it into a MySQL query command Once the MySQL query completes and returns the value of the user and password the Personal Home Page PHP script then confirms the validity of a user name and password by returning a JSON message back to the Android application In addition the rest of user identification information will also be returned to the Android device in the JSON format The returned JSON message will then be parsed by a JSON parser class and stored as key value pairs in the Preference Profile of the Shared Preference After a successful log in the user is led to his or her wellness profile containin
28. server to be viewed and analyzed by a health professional 18 Wellness Monitoring System via Body Area Networks 6 3 Server Integration 6 3 Server Integration 6 3 1 Server Integration Introduction The wellNode system was initially designed to operate locally only configured to operate with the WBAN of sensors transmitting their data to a central Android device that processes displays and internally stores the data However a WBAN on its own was not sufficient and feasible to provide adequate user wellness monitoring In order to perform a user s health analysis based on their wellness information the healthcare professional would have to manually transfer the user s wellness information to the healthcare professional s own computational device In addition the information stored only within the Android application would not be easily interpreted or analyzed by a healthcare professional Therefore we solved these two major issues by integrating our WBAN into a remote server located on the World Wide Web WWW Our server contains a centralized database managed by My Structured Query Language MySQL an easy to use and popular database management system The integration process is accomplished by adding a new background service to our wellNode Android application that wirelessly transmits the user s data through WiFi on a periodic basis to our integrated server The server integration allows the health care professional to exam
29. tag cannot be read when it does not align well with the reader therefore decreasing the ability to produce accurate positioning results 20 In contrast WiFi operates at 2 4 GHz and is ubiquitous in the modern society as almost every individual owns a smartphone that has a WiFi connection and almost every public facility has at least one WiFi router installed Since WiFi technologies are readily integrated with most public facilities the user s mobile device would receive the transmitted WiFi router signals that would be the basis infrastructure for a WiFi based indoor positioning system Therefore WiFi was the ideal solution for a cost effective indoor positioning module WiFi based indoor positioning system can utilize RSS in two different ways one being the trian 43 Wellness Monitoring System via Body Area Networks 4 2 Background Information gulation method and the other being the fingerprinting method 4 2 1 WiFi Triangulation Method The Triangulation algorithm is based on the lateration process 19 Assuming that a WiFi router is omnidirectional the RSS would be constant at a fixed distance away from the transmitter on the omnidirectional plane This can be pictured as a circle with the center being the transmitter and the radius being a function of the RSS Furthermore when we measure the RSS values from three different routers with the aforementioned relationship between the RSS and distance we can use the formulas below to
30. was first mounted to a flat ABS plastic backing plate which was then affixed to the base of the Ergo Case enclosure Figure 13 3 All electrical components were secured to the backing plate by the use of stainless steel socket cap bolts For each circuit board mounted to the backing plate a rubber spacer was placed in between the circuit board and the backing plate at each fastening position to alleviate mechanical stresses on the circuit board caused by jarring movements of the enclosure as a whole Figure 3 11 To keep the appearance of the backing plate neat all interconnect wires linking major components were routed underneath the backing plate For the interconnecting wires to 34 Wellness Monitoring System via Body Area Networks 3 4 Implementation pass through the enclosure holes were drilled in the backing plate at various locations A rubber grommet was placed within each hole in an effort to prevent the edges of the backing plate cutting into the insulation of the interconnecting wires Fig 3 11 ECG module internal backing plate structure 3 4 2 Software Implementation Our project description outlined that the detected ECG waveform should be transmitted wirelessly to a mobile device so as to be viewed within a dedicated Android application Typical ECG signals have a frequency range between 0 05 Hz and 150 Hz 14 therefore by Nyquist s theorem the ECG data should be acquired at a sampling rate more than 300
31. 34 The communication implementation has to read a single byte at a time We had the Bluetooth Mate Silver label the start of each new ECG data with the byte value Ox0A afterwards the Android application reads the incoming integer data type value in little endian format Finally the received ECG data can be placed in buffers for viewing the data on a graphical plot or saving the data in a csv file for later analyses 19 Wellness Monitoring System via Body Area Networks 6 2 Mobile Application Find and connect to Bluetooth Mate Silver BMS with pin Create a RF communication socket via SPP N Connected to BMS Return ee Are bytes available to read startint true Read byte 0x0A Y Read next incoming int type in little endian format Send int value to separate GUI plotting and saving buffers startint false Fig 6 4 Bluetooth Mate Silver flowchart 6 2 Mobile Application At the initial stage of project development the entire group agreed to display the wellness mod ules information on a mobile application because a large variety of people have these devices In the current market there are many operating systems that include Apple s IOS Windows Phone Blackberry and Android The Android platform surpasses all other mobile operating systems for a few reasons Firstly the platform does not require us to obtain a developer s license that could be a considerable expenditure As well th
32. 4 arar ada A 13 6 2 Sensorlag accelerometer data pollidg 02 eee 714 6 3 Asynchronous accelerometer characteristic TEad 714 6 4 Bluetooth Mate Silver flowchart e 76 ee ee ee ee ee ee 78 6 6 Server integration functionality o a a a a 80 6 7 User identification information table structure o 83 6 8 User identification information table e 83 a 84 nro a ea 84 a 86 aaa ao 95 ad ies 96 sees 97 B 4 ECG low pass filter ee 97 B 5 ECG notch filter uo we eae we Ree AAA 98 xi Wellness Monitoring System via Body Area Networks LIST OF FIGURES B 6 ECG final stage amplification xii Wellness Monitoring System via Body Area Networks LIST OF TABLES List of Tables 2 1 Posture Recognition Specifications 4 o o o o ee 5 2 2 Angular Threshold Determination of Postures 11 2 3 Posture Recognition Experimental Results o 16 2 4 Posture Recognition Specifications Comparison a e 17 3 1 ECG Proposed Specifications 4 o ee 21 3 2 ECG Specifications Comparison 2 1 e e e a 38 4 1 Indoor Positioning Specifications 4 o o oo ee 4T 4 2 Indoor Positioning s Database Format a 50 AS R Sat Position LLO se sase ny 29565 een reas S a es 53 AA RSS at Position 2 O looser rada roer 53 AE ee
33. 9 PO 0 E r r r E 1 0 5 0 0 5 1 Thigh Accelerometer Y Axis r r 0 r r E 7 1 0 5 0 0 5 1 1 0 5 0 0 5 1 Chest Accelerometer Z Axis Thigh Accelerometer Z Axis 1 1 0 5 a J 0 5 J 0 E r r r 0 E r r r E 1 0 5 0 0 5 1 1 0 5 0 0 5 1 a Standing Chest Accelerometer X Axis at Z 4 0 E r r 1 0 5 0 0 1 Chest Accelerometer Y Axis 1 E 0 55 0 r r r 1 0 5 0 0 5 Chest Accelerometer Z Axis Te 2 A 0 55 J Oo r A rae 1 0 5 0 0 5 1 Thigh Acceleromter X Axis 0 51 V4 1 0 E r r r 1 0 5 0 0 5 1 Thigh Accelerometer Y Axis 5 i l l J 1 0 5 0 0 5 1 Thigh Accelerometer Z Axis b Bending Chest Accelerometer X Axis off PON i 0 E r d E E 0 0 0 5 1 Chest Accelerometer Y Axis 1 g 0 5 J 0 r r r 1 0 5 0 0 5 1 Chest Accelerometer Z Axis 1 0 5 0 0 5 1 Thigh Accelerometer X Axis of 0 E r r r 1 0 5 0 0 5 1 Thigh Accelerometer Y Axis 15 z 0 5 J 1 0 5 0 0 5 1 Thigh Accelerometer Z Axis 1 0 5 0 0 5 1 c Sitting Chest Accelerometer X Axis of TOTO 1 0 E r r r at 1 0 0 0 5 1 Chest Accelerometer Y Axis iF 0 5 F j 0 E r r r a 1 0 5 0 0 5 1 Chest Accelerometer Z Axis q1 r r z 0 5 F J 0 ao E E 1 0 5 0 0 5 1 Thigh Accelerometer X Axis OR TA NO 0 E r E E a 1 0 0 0 5 1 Thigh
34. Accelereomter Y Axis iF z 0 5t 7 0 E r r p 1 0 5 0 0 5 1 Thigh Accelerometer Z Axis 1 z z z 0 5 a z z 0 1 0 5 0 0 5 1 d Lying down on back side Chest Accelerometer X Axis 0 r r r 1 0 5 0 0 5 Chest Accelerometer Y Axis 1 0 5 05 r r r E 1 0 5 0 0 5 1 Chest Accelerometer Z Axis of l l lt 1 0 5 0 0 5 1 Thigh Accelerometer X Axis 0 5f 0 E r r r 1 0 5 0 0 5 Thigh Accelerometer Y Axis 1 0 5 7 0 E r r r S 1 0 5 0 0 5 1 Thigh Accelerometer Z Axis oa 1 0 5 0 0 5 1 e Lying down on front side Chest Accelerometer X Axis os i E r r i 1 0 5 0 0 5 1 Chest Accelerometer Y Axis 1 opp NT 1 0 5 0 0 5 1 Chest Accelerometer Z Axis 1 0 5 y 0 r r r 1 0 5 0 0 5 1 Thigh Accelerometer X Axis 0p _ 4 1 0 5 0 0 5 1 Thigh Y Axis i opf 2 JN 3 1 0 5 0 0 5 1 Thigh Accelerometer Z Axis 1 0 5 p 0 r r r 1 0 5 0 0 5 1 f Lying down on left side Chest Accelerometer X Axis 1 0 5 0 0 5 1 Chest Accelerometer Y Axis 1 1 IAE y PNC 1 0 5 0 0 5 1 Chest Accelerometer Z Axis Thigh Accelerometer X Axis of E S s 1 0 5 0 0 5 Thigh Accelerometer Y Axis 1 z 0 51 IN 1 0 5 0 0 5 1 Thigh Accelerometer Z Axis 1 0 5 F m 0 r r r 1 0 5 0 0 5 1 g Lying do
35. Figure C 4 by tapping once the Return button e Enter the timeline to graphically display your past postures by tapping once the Timeline button The timeline display screen is shown in the Figure C 10b for reference e Tap once the Right arrow lt or Left arrow gt to navigate through previous to present posture recordings Return to the posture screen by tapping once the Return button at bottom right of the screen Current Posture Standing Past Pastunes DO Bending DO ts Ear Bhima Sitting Ore Laying dem Back ida OOhrs Fancy En Henney a Posture recognition main screen Y E Fo Gro pF if 17 Posture Timeline Legend 1 Sitting 2 Bending 3 Sanding 4 Lying down back 5 Lying down front 6 Lying down right 7 Lying down left z b Posture recognition timeline Fig C 10 Posture and timeline screens 109 Wellness Monitoring System via Body Area Networks User Manual Step Detection Module e Prior to commencing the step detection process tap once on the Option Menu located on the top right of the pedometer screen as shown in Figure C 11 e Tap once on the Settings button to enter the target setting screen as shown in Figure C 9 below e Enter values for the number of target steps and step size in metres then tap once on the Save Changes button to save the settings If you do not wish to overwrite previously i
36. Hz to have enough information to reproduce the original signal and avoid the effects of aliasing The RFduino was initially selected because it had a BLE module with a 16 MHz ARM MO Microcontroller and a 10 bit Analog to Digital Converter ADC We measured the average time it takes for the microcontroller to execute an analog read value to be 0 787 us per measurement Therefore it was concluded that the MO microcontroller was capable of acquiring the ECG data at a rate larger than 300 Hz as desired 30 Wellness Monitoring System via Body Area Networks 3 4 Implementation Unfortunately upon testing the BLE module of the RFduino it was discovered that BLE wireless communications would not be suitable for our implementation due to asynchronicity Specifically durations between the transmission of data packets by the BLE module were inconsistent Further information regarding this issue can be found in the following Tests and Results section There were several alternative options to achieving wireless communications between our ECG wearable device and our Android application namely standard Bluetooth ZigBee and Wireless Fi delity WiFi Further research indicated that standard Bluetooth transmission technology would be the most successful for continuous data transmission implementations whereas BLE technol ogy was more suited for infrequent data transmission applications such as relaying temperature humidity or pressure readings 17
37. Indoor Positioning 4 1 Brief Introduction The purpose of our indoor positioning module is to determine a user s position in an indoor facility where GPS produces inaccurate results or is simply unable to work Our indoor positioning module consists of the fingerprinting method which utilizes signal coverage from ubiquitous WiFi access points inside of a modern facility to determine the position of the user Typically the fingerprinting method results in an error within three meters 70 of the time 18 Our indoor positioning module is able to obtain slightly higher accuracy due to the consideration of the user s orientation within a facility Within this chapter the module specifications background information software algorithm development implementation and test results will be discussed in more detail 4 2 Background Information For many years accurate indoor positioning has been sought after by large public facilities such as wellness facilities hospitals and even airports Although the position of the user has no apparent relevance to their wellness wellness facility staff can greatly benefit from monitoring the user s position for two major reasons Firstly the user s position can be used to increase the efficiency of the facility and secondly knowing the user s position is vital for a quick response when harmful 42 Wellness Monitoring System via Body Area Networks 4 2 Background Information accidents occur wit
38. Overview The system we developed called wellNode is a wellness monitoring system that collects the user s wellness information via wireless body sensors and then transmits the data to a central node where the data is processed and stored locally or to a remote server that is accessible to health care professionals More specifically our wellness monitoring system is divided into four different modules Figure 1 1 The modular system includes posture recognition ECG acquisition indoor positioning and step detection The posture recognition module determines when a user is not Wellness Monitoring System via Body Area Networks 1 3 System Overview moving and if he or she is either standing bending sitting or lying down The ECG module displays and records the electrical signal of the heart that could describe the physiological wellness of a user The indoor positioning module determines a user s location within a wellness facility where the popular Global Positioning System GPS is unable to function Lastly the step detection module counts the user s steps at relatively low speed even shuffling can be detected All the modules transmit their data to our central wellNode Android application processing node on a compatible mobile device Wireless Body Area Network Android Device the location of interest and chest sampling at 10 Hz RFduino Software Android Application Bluet othtransmitter i
39. Posture Recognition Experimental Results Total 99 1 334 s The determination time was measured from the moment the subject settled in their specific static posture to when the application determined the exact posture The comparison between the pro posal and resulting specifications are shown in Table 2 4 The sampling frequency was programmed within the SensorTag and was obtained by BLE protocols without any latency hence the data was received at a synchronous rate The accuracy was 99 but was due to one subject not fully going into a bending position Also the average algorithm determination time was 1 334 s Therefore 16 Wellness Monitoring System via Body Area Networks 2 5 Tests and Results the experimental results showed that the posture recognition module was able to meet our proposed specifications By being able to adequately monitor a user s posture healthcare professionals can obtain sufficient wellness information An entire daily posture summary can be related to the amount of time a user spends doing physical activities in comparison to being sedentary Furthermore the rate that a user transitions from one posture to another posture can suggest that a user is doing some form of physical exercise Also a transition between a standing and lying down posture could signify that a user has fallen down and requires immediate assistance Therefore the recording of a user s posture provides compelling wellness informat
40. Server Integration Column pens cr id la d se string mame rrchar27 5 address Fig 6 7 User identification information table structure id creation date firstname lastname dob string sex phone weight address username password 34 2075 02 26 15 01 04 John sheppard 1980 10 10 Male 204123234 190 Earth newuser 123 Fig 6 8 User identification information table with a user s specific information The posture storage table Figure 6 9 also has simple structure which contains separate timetag posture and username columns The username column identifies the user that performed that specific action Similarly to the user identification information table the posture storage table s id column can account for a large number 101 1 of the current number of posture recordings The posture column shows the posture with data type varchar for easy user interpretation to 83 Wellness Monitoring System via Body Area Networks 6 3 Server Integration eliminate the ambiguity of a numerical value representation of the recorded postures Even though a numerical value representation allows the application of mathematical operators such as greater than equal to or less than each posture is not related to other posture in a tangible way therefore using math operators to sort these posture values show no immediate benefit An example of a user s posture information stored within our database is shown in Figu
41. Thus the Bluetooth Mate Silver by Sparkfun was selected to achieve transmission of the measured ECG data at a rate of at least of 300 Hz This Bluetooth module was easy to integrate with our RFduino via Universal Asynchronous Receiver Transmitter UART communications and functioned seamlessly with the Android application by using the Serial Port Profile SPP The SPP transmits data via the serial port just like a Universal Serial Bus USB connection Hence the Bluetooth Mate Silver is capable of transmitting data at rate of 2400 up to 115200 bits per second bps Additional information regarding the implementation of the SPP profile is discussed in Section 6 1 1 The RFduino s functional flowchart is shown in Figure Initially the RFduino waits until the Bluetooth Silver Mate is connected to the Android application and has received a command byte If the command byte is 0x00 this signifies the START command and the RFduino will start sending the measured ECG values In sending data the RFduino sends firstly an indicator signifying the start of a new measurement byte Ox0A then sends the measured value in its two byte little endian format After such the RFduino is paused for a period of 2 62 ms which corresponds to being less than a sampling period of 1 360 s 2 78 ms We decided to sample at a frequency 36 Wellness Monitoring System via Body Area Networks 3 5 Tests and Results of 360 Hz to eliminate aliasing issues T
42. University of Manitoba Department of Electrical amp Computer Engineering ECE 4600 Group Design Project Final Project Report Design and Implementation of a Wellness Monitoring System via Body Area Networks by Group 01 Cassandra Aldaba Tianqi Liang Yang Su Haiyue Wang Jeff Winkler Final report submitted in partial satisfaction of the requirements for the degree of Bachelor of Science in Electrical and Computer Engineering in the Faculty of Engineering of the University of Manitoba Academic Supervisor Dr Jun Cai Department of Electrical and Computer Engineering Industry Supervisor Mr Michael Zhang Wellness Institute of Seven Oaks General Hospital Date of Submission March 4 2015 Copyright 2015 Cassandra Aldaba Tianqi Liang Yang Su Haiyue Wang Jeff Winkler Wellness Monitoring System via Body Area Networks Abstract The ever increasing demands placed upon today s healthcare staff and systems lead to the re quirement for an efficient and effective means of connecting healthcare professionals to their many patients Current in hospital care methods force healthcare professionals to look after patients on an as needed basis and are often ineffective since physicians must make healthcare decisions based upon limited observations and short term physiological measurements he basis of this project was to create a set of wearable devices that when combined offer healthcare professionals con tinuous
43. a Networks 6 1 Wireless Data Transmission are negligible T herefore the data is retrieved at a nearly consistent sampling period Furthermore we selected BLE communications not simply for its compatibility with low transmission rates but also due to its ease of implementation BLE communication uses a generic attribute profile GATT that encapsulates requests commands and protocols 33 This structured framework has a collection of data within a service where each value within a service is called a characteristic These services and characteristics are labeled with a universally unique identifier UUID Related SensorTag accelerometer UUIDs are shown in Table The basic initialization of the BLE SensorTag is shown in Figure First the Android device discovers and connects to the SensorTag by its unique MAC address Table 6 2 Once the SensorTag is connected the Android device discovers the SensorTag s available services Afterwards the accelerometer of the SensorTag is enabled by retrieving the accelerometer service and configuration characteristic After the byte 0x01 is written to the configuration characteristic the accelerometer is ready to be polled for data In the posture recognition module the second SensorTag will be connected to after connecting to the first following which the second SensorTag will have its accelerometer enabled before polling of both accelerometers commences Table 6 1 SensorTag Specific Acceler
44. ach The use of this structure format allowed for members to collaborate together which facilitated the efficient and timely completion of the project As well this structure provided a layer of relief in that no one member would be accountable for an entire module on their own The division of labor among teams was assigned to specific members as is shown in the contributions table provided on the following page Legend Lead task O Contributed Wellness Monitoring System via Body Area Networks S 2 3 S E g T ES YN YN ES O Literature Review Final Report Indoor Positioning Module WiFi fingerprinting offline database implementation ele Yams ole ae Hone ve Jeff Winkler oo Tianqi Liang WiFi fingerprinting positioning algorithm WiFi fingerprinting indoor positioning testings Step Detection Module Algorithm development Algorithm tests Posture Recognition Module Angular threshold algorithm O Fuzzy logic subset function development e o Fuzzy logic algorithm e TT Fuzzy logic algorithm tests o Jo 9 o ECG Module Hardware circuit design O Data acquisition and wireles transmission e o Circuit construction and enclosure design e Hardwaetets e Wireless Transmission Protocols BLE protocol implementation O O Bluetooth protocol e Si Android Application Development Aesthetic design O O O O Modular integration a pa o Application testing and
45. also knew that with the specified circuit used for DC biasing it would have a cut off frequency described by the following equation 26 Wellness Monitoring System via Body Area Networks 3 4 Implementation E fe 27R C 3 12 We initially set fe 150 Hz to match the upper cut off frequency we desired and found C to be 100 uF When testing the entire circuit using C 100 uF the lower cut off frequency was as desired at approximately 0 05 Hz but the upper cut off frequency had been shifted up to 226 Hz Examining the output of the circuit just prior to the final amplification stage the filtering characteristics were exactly as desired so we knew that the addition of the DC biasing network was causing the discrep ancy We experimented with different values of C and found that when C 10 uF the upper cut off frequency was as desired at approximately 150 Hz At the same time however we realized the lower cut off frequency had now shifted up to approximately 16 Hz from the desired 0 05 Hz Nearing the end of the project not wanting to redesign the DC biasing network and risk damaging the circuit board or adjacent components we came to the conclusion that we had a trade off between meeting the specifications for the upper cut off frequency or meeting the specification for the lower cut off frequency Before making a decision we examined which case would be the most detrimental to the ECG waveform output from the filtering and d
46. ate to the server Thus the periodic update of sensor value is selected and preferred Once the server communication service is constructed the sensor information are parsed into JSON periodically and transmitted to another PHP script that stores the data to the database with the username as a label 87 Wellness Monitoring System via Body Area Networks Chapter 7 Conclusions We have successfully designed built and implemented our wellness monitoring system wellNode within the proposed specification as well as surpassing the specification in certain aspects More specifically the posture recognition module can detect three basic postures standing bending and sitting and four additional postures that describe the different lying down positions on his or her back front left or right side The ECG samples data at 360 Hz with a resolution of 10 bits and an adjustable gain of greater than 200 The indoor positioning module can detect the user s location within 2 3 m of error 80 of the time Lastly the step detection module utilizes two algorithms FFT and slow walk speed detection algorithm to determine the steps taken by the user to an accuracy of 90 with a slow walking speed of 1 67 m s The Android application can record process and display the data in a user friendly manner In addition to the proposed work we also integrated a server that allows the data to be collected by the Android mobile device and transmits the da
47. ated after every changed value This makes the data flow for the sensor information a little bit more difficult because the information for the sensor values can be updated and erased if the data is not stored onto the server beforehand There are a couple of solutions to resolve this issue The first solution 86 Wellness Monitoring System via Body Area Networks 6 3 Server Integration is to change the storage rate of the data onto the server while the second approach is to create a buffer to store the values and then the buffer is stored onto the server at a specific rate It is obvious that the first approach is unreliable when the WiFi connection is disrupted Therefore we created a sensor data buffer within the senor preference However we were were faced with the issue determining the update rate to store the buffer to the server at a timely interval or use a counter to flag when the buffer contains a specific number of samples and send the data altogether to the server The first choice is favoured because the update rate for the sensor values from each module are different hence attempting to synchronize the same amount of samples from each sensor would be disastrous On the other hand if we only store the sensor values when the buffer reaches a specific size then there is a possibility that particular sensor values does not change for a prolonged period of time while other sensor values are changing but not able to upd
48. ation of the user we can scan the area around the initial user s location By incorporating the initial location the number of calculations that need to be done is vastly reduced However the error will eventually accumulate and the algorithm might fail to find the right position as it could scan at a location that does not contain the actual location If we could utilize a RFID proximity sensor along with the WiFi fingerprinting method we would be given an accurate initial location via the proximity sensor and we can then run our algorithm based on that In the end we did not implement the idea of combining the fingerprinting method with RFID technology because we could not gain access to a RFID system However our database is structured in such a way that it will allow the eventual incorporation of RFID 4 5 2 Online Phase To determine the most likely location of the user we use the simple Euclidean distance method to calculate the difference between the sample and the fingerprinting reference locations within the database During the online phase the Euclidean distance between the sampled RSS value and each database value is computed and then sorted from highest to lowest The coordinate that yields the smallest Euclidean distance is selected as the optimal estimate of the user s position EuDi Y RSS RSS 4 5 0 1 ol Wellness Monitoring System via Body Area Networks 4 5 Software Implementation In thi
49. be worn on the user s lower chest area across the user s abdomen In an effort to ensure user comfort several spec ifications were decided upon for the enclosure design Specifically the enclosure should not have sharp corners or protrusions that would push into the user s chest As well the enclosure should follow the natural form of the user s chest not having the ability to shift or rock on the user s chest We chose to implement the Ergo Case style Acrylonitrile Butadiene Styrene ABS plastic container manufactured by OKW enclosures Figure 3 10 OKW enclosures offers their Ergo Case container line in several size options as well as colors and includes such features as a convex de sign for comfortable wearing internal screw pillars for mounting circuit boards integrated loops for waist belt strap attachment all of which were desirable and would meet our design specifications 33 Wellness Monitoring System via Body Area Networks 3 4 Implementation Fig 3 10 ECG module enclosure with lid affixed The enclosure lid is affixed to the base with the use of four Philips head screws which can be accessed on the backside of the base of the enclosure No components were affixed to the lid of the enclosure In this way the front lid may be removed easily at any time without the worry of damaging components when pulling the halves of the enclosure apart For the internal layout of the enclosure all of the electronic hardware
50. by the rod inside the mechanical pedometer will wear down and the swing angle of the rod may change These changes in the hardware will affect the detection process of one step leading to inaccurate step detection and difficulty detecting steps from slow walking speeds and shuffles Therefore the calibration of the mechanical pedometer will be hard to implement since we cannot replace and adjust the pendulum rod that is inside of the box Despite the low power consumption the mechanical threshold based pedometer was not selected due to the deterioration of the rod that would lead to an unreliable and inaccurate step detection module 5 3 2 Accelerometer based Pedometer The accelerometer based pedometer is generally accurate regardless of where the user wears the device 26 The user can wear the pedometer on the waist or on any other location on the body to detect a single step Since the accelerometer measurements provide an indication of how the part of the body is accelerating we can analyze these measurements to detect and count a user s 58 Wellness Monitoring System via Body Area Networks 5 4 Implementation step based on peak detection pattern matching and Fast Fourier Transform FFT methods The algorithm that we developed utilizes a combination of threshold and pattern matching algorithms With the proper step detection algorithm of the accelerometer sensor we do not need to do the calibration of threshold for each i
51. calculate the intersection of the three circles which would indicate the user s position La 21 Ya Y h 4 1 La 22 Ya Ya da 4 2 Za 23 Ya ya 4 3 1 2 and x3 represent the known WiFi access point AP locations while d d2 and d3 are the distances between the AP locations Then Za Ya is the unknown coordinate that will be computed for a 2 D situation For example in Figure 4 1 the three access points are located at 2 2 1 2 and 2 1 With the d d2 and dg distances known the unknown point 1 1 will be calculated as the user s possible position 44 Wellness Monitoring System via Body Area Networks 4 2 Background Information Triangulation Method Fig 4 1 Position estimation for triangulation method The distance of each APs to the unknown location can be estimated by the signal propagation equation P d P do 10nlog 4 4 In this equation n is the path loss rate relative to a specific distance P d is the signal strength value measured at a reference position P dg and d is the unknown distance to be calculated In practice it is not easy to accurately convert the signal strength to distance since varying factors such as antenna gains interference from objects within the signal s path signal reflections and the dielectric properties of various entities need to be taken into consideration On the other hand varying ra
52. cilloscope screen capture Figure 3 13 the entire data transmission took 2 5 seconds thus this meant that the data was sent at a rate of 360 Hz In conclusion the testing results showed that the Bluetooth Mate Silver with the RFduino as its microcontroller is capable of sending the ECG data to the Android application at a sufficient rate larger than 300 Hz A Sims 4 Mar 15 15 35 Fig 3 13 Oscilloscope capture of ECG Bluetooth transmission test Signal amplification as well as filtering characteristics were analyzed by connecting the ECG module to a signal generator and examining the output characteristics by visualization on an oscilloscope Specifically an Agilent 33220A signal generator was connected to the electrode leads of the ECG device and the final output of the ECG circuitry was monitored on a Tektronix TBS 1052B digitial oscilloscope An input waveform of amplitude 24 8 mVpp and varying frequency was used to test the frequency response of the ECG circuitry To ensure the filtering hardware was removing the desired frequencies we measured the peak to peak amplitude of the output signal at frequencies 39 Wellness Monitoring System via Body Area Networks 3 5 Tests and Results ranging from 0 05 Hz to 165 0 Hz The measured amplitudes were then plotted as a function of frequency the result of which is shown in the graph below Figure 3 14 Frequency Response O 10 Signal Amplitude dB 13 20
53. curacy 98 67 58 27 69 83 Average Accuracy 75 59 According to the experimental results shown in the Table the accuracy of speed detection for the slow walking speed of 1 5 mph was very accurate However the accuracy decreases for faster walking speeds Overall the average accuracy of the speed detection algorithm by using the FFT method was approximately 75 which is acceptable as we are only interested in capturing the speed of users with slow walking speeds 28 The reason why the speed detection did not perform very well for the faster speeds is that the FFT method mainly focuses on slow walking speeds as opposed to fast walking speeds Also the manner in which a person walks on a treadmill differs from the way in which they walk otherwise 29 Table 5 4 Step Detection Specifications Comparison Detectable Walking Speed 0 7m s 1 8m s 0 67 m s 1 8m s Speed Detection Accuracy Not Defined In conclusion the step detection module meets all of the design specifications that we suggested in our proposal The comparison of our final results and the proposed specifications are listed above in Table The step detection module was able to count the steps of users with walking speeds ranging from 0 67 m s 1 79 m s with the SensorTag sampling rate of 50 Hz Based on these results the accuracy of the pedometer module varies with the user s walking speed walking style 68 Wellness Monitoring System via Body Area
54. d disadvantages of the server integration and assumed that our wellNode system is only for demonstration purposes to show the feasibility and benefits of wellness monitoring Thus we assume that there are no cost concerns and that the users have no privacy concerns and willingly allow their wellness information to be stored and analyzed 6 3 3 Server Implementation The server has three main functions to transfer the user s wellness information to the server The server authenticates user access to the database by validating the user s log information Once the user has acesss to the server the server gathers and stores the users identification information from the user s Android device to the MySQL database The users identification information includes their user name password and personal information such as weight sex date of birth phone number address Then health care professionals are able to run queries on the database to obtain specific wellness information that they desire The server needs to be accessible anytime to all the users which means that the server could either reside on the wellness facility s Local Area Netowrk LAN or on a Wide Area Network WAN such as the WWW In our project the server is hosted on the WWW to demonstrate the large accessible range of our system irregardless of location and the ease of accessing the information for healthcare professionals as long as they are connected to the in
55. ding sitting and lying down However when we incorporated the fuzzy logic system to determine different lying down positions on one s back front left or right side the algorithm was not able to determine some postures because specific postures had the same determinant conditions Therefore we required to increase the complexity of the angular threshold by including all of the three possible Euler angles to determine the postures with a high level of accuracy Since the accelerometers cannot measure the yaw angle the accelerometer is insensitive to any changes about the gravitational vector we were required to incorporate a gy roscope into our implementation which is another motion sensor that measures angular velocity 8 Table 2 2 Angular Threshold Determination of Postures gt 30 Lying down on back side The finalized method was based on the fuzzy logic system It was implemented by first creating fuzzy logic functions for each posture to create the fuzzy subsets then we created the classifier algorithm based on the previously determined functions to classify the most likely posture Given an accelerometer measurement the most likely posture was determined by the largest sum of the fuzzy subsets output By analyzing the measured range of each accelerometer axis we created fuzzy subsets for each posture and for each SensorTag thigh and chest as shown in Figure The defined fuzzy subsets are then used to describe the
56. dio conditions at a site caused by environmental factors such as changes in time and humidity may also alter the effectiveness of the signal propagation model In fact it is shown in literature that RSS is a poor indicator of distance 22 45 Wellness Monitoring System via Body Area Networks 4 3 Design Specifications In addition the triangulation algorithm due to poor distance estimation could lead to nondeter ministic results produced from non intersecting circle solutions We tried to resolve and eliminate the possibility of nondeterministic results by transforming the systems of equations into an op timization problem that allows the x and y coordinates to produce the smallest error However optimizing the system of equations resulted in poor location determination with errors in the range of 10 m 20 m As a result it would not be feasible to implement the triangulation method in our indoor positioning module 4 2 2 WiFi Fingerprinting Method The alternative approach is to use the WiFi fingerprinting method which records the RSS values of different predefined reference areas within the indoor area of interest The RSS values are used as unique area identifiers as it is assumed that different areas would have different RSS characteristics Also when we utilize this method we do not need to consider the antenna gains and interference from objects This approach typically consists of an offline phase and an onli
57. e application development process could be a long process Wellness Monitoring System via Body Area Networks 6 2 Mobile Application for there are many aspects with the IOS firmware does not permit Also the Android platform is heavily supported by the open source community Since our group had limited mobile programming development experience we felt the Android platform to best the best match to our project Once the Android platform was selected the user interface had to be designed to fit within the context of our wellness monitor The main objective of our application is to behave as a central node to collect the user s health information and transmit the information to our online server The application overall hierarchy is shown in Figure 6 5 The development of the entire application allows the user to easily navigate through each individual module explained further in Appendix D For the positioning system the user s position is monitored and displayed on a visual map The ECG module allows a user to view his or her ECG signal that can be recorded for later analyses The posture recognition module summarizes the user s daily posture activities in a visual pie chart and time line plot Lastly within the step detection module users can set a step counter goal and track the number of steps they take Overall the application encompasses all the user s wellness information in a visual graphical form 7 Wellness Monitoring Sy
58. e components For example the RFduino GPIO pin 6 which accepts input from the filtering and amplifying circuit has a upper voltage limit of 3 6 V The filtering and amplifying circuit output can reach values above 3 6 V so we implemented a Zener diode which conducts at a breakdown voltage of 3 3 V and placed it between the GPIO 6 pin of the RFduino and ground Similar diode circuits were implemented for the positive and negative inputs to the differential amplifier The reference pin of the differential amplifier did not require a protection circuit as it maintains in contact with the floating ground of the system 30 Wellness Monitoring System via Body Area Networks 3 4 Implementation Patient connection Three electrode attachment cables leads were provided to us by the ECE technical staff The electrode leads were essentially insulated wires with button style snaps molded onto one end of the cable and an exposed conductor at the opposite free end The button style snaps are designed to work with the stick on style 3M Red Dot Ag AgCl single use electrodes Figure 3 7 Fig 3 7 3M Red Dot Ag AgCl single use electrode The product s reliability decreases during user motion as tension and flexion of the electrode lead cables cause stress on the connections to the circuit board Instead we decided to use a stereo style 3 5 mm headphone plug and accompanying jack to allow the three separate electrode leads to be easily
59. e for the slow transmission rates such as used with the step detection and posture recognition modules while standard Bluetooth protocols are required to achieve the fast transmission rate needed for the ECG module 6 1 1 Bluetooth Bluetooth technology is a viable wireless data transmission method for our application due to its popularity and reliability in regards to data transmissions Most of today s modern mobile devices TO Wellness Monitoring System via Body Area Networks 6 1 Wireless Data Transmission have a BLE v4 0 or greater integrated module within their hardware architecture These mobile devices include popular Android devices such as the Samsung Galaxy series and Google Nexus series cellular phones 30 Therefore we selected to utilize Bluetooth technology as a wireless data transmission protocol to make our device accessible to the many individuals who have an Android mobile device Not only is Bluetooth technology readily accessible to many individuals this protocol has error correction and error detection as well as frequency hopping to ensure the reliability of transmitted data Error correction and detection is important in poor RF environments Bluetooth provides solutions throughout its generic data transport layers via the baseband layer that does forward error correction by the receiver and detects any error after corrections Also the logical link control and adaption protocol L2CAP layer check
60. e locations will be described as how far they are away from the red cross or the origin However the look up table is not feasible as it is not intuitive and each determined location does not contain information about its spatial relationship which could provide more ben eficial user information Next we looked into another system that utilizes a 2 D index system similar to the index of a matrix The origin is still defined at the same location but instead of having the origin at the 0 0 point the origin is located at 1 1 The location 1 unit north of the origin would therefore be 1 2 Since we have chosen to map the Atrium in a grid format with uniform grid length and width it is easily seen that an index location can be translated to the relative location without a look up ta ble In fact we only need to multiply the index by the length of the grid to find the relative location 50 Wellness Monitoring System via Body Area Networks 4 5 Software Implementation In addition to the benefit of the direct mapping of the index to the relative location of the user there is a huge benefit to using 2 D indexing We were concerned that our fingerprinting database could become excessively large as we expand the WiFi fingerprinting area Thus the simple look up table s performance time would increase because we have to scan all locations to get a valid position result However by using 2 D indexing within a definitive initial loc
61. e overall goal of the step detection module is to record the number of steps taken by a user with an impaired walking ability The pedometer module specifically caters to users that walk at a slower than average speed which is considered to be within the range of 0 7 m s to 1 8 m s 24 As listed in Table provided below it is necessary for the accelerometer data to be collected at a sampling frequency of no less than 50 Hz 25 26 At this benchmark for the sampling frequency the pedometer should have a step detection accuracy of at least 70 Table 5 1 Step Detection Specifications Value or Range Measurement Explanation The sampling rate of Ey ee the Sensor Tag The walking speed of the user that Detectable Walking Speed 0 7 m s 1 8 m s the step detection algorithm can detect steps with at least 70 accuracy The accuracy of the step detection A gt l i algorithm in the pedometer module 5 3 Background Information The pedometer is a step detection module which is a device that can count the number of steps that a user takes by sensing the impact of the user s foot with the ground Currently the pedometer uses one of two different types of technologies namely mechanical threshold based or electrical motion sensor based The mechanical threshold pedometer has a pendulum or a swing arm system with moving parts inside the device When a user takes a step the pendulum arm in the mechanical threshold based ped
62. e peak occurred The frequency at which the highest peak occurs is the rate at which steps are taken During step detection we can determine the cadence which is the walking speed of the user by using the FFT algorithm In order to implement the FFT algorithm we need to choose a larger data set for better evaluation of the FFT data points For the data array the number of samples is 256 and the sampling frequency is 50 Hz We chose a sample number of 256 because the number of data points for the FFT array can only be a power of 2 and we want to evaluate and update the speed about every 5 seconds We determine the FFT of the data array and find the maximum peak within the 256 samples by using the following equation N 1 gt y a NN mEZ 5 1 n 0 In this equation N represents the sample size which was chosen to be 256 in our algorithm while m is the frequency index that corresponds to the sample number which ranges from 0 to N 1 We determine the power spectrum of the signal and find the frequency index at which maximum power occurs excluding the spectra at a frequency of 0 which is the DC component The corresponding frequency index defines the frequency of one step that has been taken inside the data array of Equation Tepo Mmazx fs 63 Wellness Monitoring System via Body Area Networks 5 4 Implementation step Detection Slope detection of two data points Case 1 Delta Current time Previous step time X a
63. ea Networks 3 3 Specifications 3 3 Specifications A summary of the required specifications for our ECG module is provided in Table 3 1 below For appropriate viewing of the ECG waveform the detection hardware should amplify the measured analog signal by a factor of 200 or greater A hardware filter should eliminate noise and stray signals in the analog signal by filtering out frequencies outside of the range of 0 05 Hz 150 Hz 14 As well the filter should try to eliminate power line interference in the analog measured signal by removing the 60 Hz component of the measured ECG signal In converting the analog measured signal to a digital signal the device must sample the analog ECG signal at a rate in excess of 300 Hz and quantize the signal amplitude with a resolution of 10 bits Lastly to ensure adequate patient safety and for our device to meet the IEC 60601 1 2005 medical device standard the maximum leakage current from the device either by electrode leads or enclosure to the human body must not exceed 10 yA 15 Table 3 1 ECG Proposed Specifications Value or Range Measurement Explanations Pon gt 300 Hz How often to sample the analog ECG signal 10 bits The quantization steps of the digitized signal High Pass Filter 0 05 Hz Frequency components below this value should Cut Off Frequency be removed Low Pass Filter 150 Hz Frequency components above this value should Cut Off Frequency be removed
64. eases with the number of cameras needed Also the user s posture can only be determined within the field of view of the cameras Another alternative uses multiple 3 axis accelerometers placed on specific body parts of the user The accelerometers are electronic devices that can mea sure accelerations relative to their orientation These can be placed on both the chest and thigh or on the head as well Differentiating between various users can easily be determined by monitoring Wellness Monitoring System via Body Area Networks 2 3 Background Information the user s specific accelerometer measurements Contrarily accelerometers are more susceptible to vibration and motion noise that can affect the overall performance of the posture recognition algo rithm Since our module should be able to monitor a single user regardless of his or her location the accelerometer device was the best option for our implementation The accelerometer data can be processed in various ways to determine a user s posture Such methods include angular thresholding 6 and fuzzy logic systems 7 Within the angular threshold method the accelerometers are placed on the user s chest and thigh and individual accelerometer orientations can be calculated based on axial measurements If the accelerometer s 3 dimensional axes are defined as in Figure 2 1 the two Euler angles roll and 0 pitch can be calculated as follows 6 8 Fig 2 1 Defined 3
65. eel proud of our work we truly could not have done it without you IV Wellness Monitoring System via Body Area Networks Mr Michael Zhang thank you for providing to us a unique and captivating capstone project as well as providing to us industry insights in regards to our project Thank you Mr Daniel Card for providing to us your technical insights when we were faced with difficult decisions and issues regarding our design We like to thank Mr Travis Rogozinski of OKW Enclosures Inc for timely providing us with multiple wearable enclosures suitable for our ECG module Thank you Ms Aidan Topping for providing to us the technical writing advice that allowed us to successfully communicate our project s achievements and accomplishments Thank you Ms Clara Lee for guiding us through the basics of database structures and implemen tation Thank you Ms Jennifer Winkler for helping us in the revision of this document Lastly we like to thank all our loved ones family members friends and fellow colleagues Even though we started to go a little crazy after all these years you still stuck beside us and continued to provide your support We could not have survived without you Wellness Monitoring System via Body Area Networks TABLE OF CONTENTS Table of Contents o A EEE Ew ED ESS i Contributions 2 ee ee il Acknowledgements 2 6 1 a a iv MistOr Piles covers IA x Dist Or Tables o cu e 24 eG Ge SREEESRESe
66. er a BLE module typically transmits data faster than our desired 10 Hz 11 We determined the average latency time between retrieving both SensorTags measurements by recording the times that the data is received from a specific SensorTag The received measurements timestamps were logged and displayed on the Android development environment Eclipse The latency times were calculated by the difference between the two SensorTag data timestamps and all were less than 0 03 s Therefore the latency time between receiving data from both devices was less than our sampling period of 0 1 s and we can consider the received data as being semi synchronous Since a user can not transition into a new posture within the 0 1 sampling period the semi synchronization of data 12 Wellness Monitoring System via Body Area Networks 2 4 Implementation will not have a great affect on the algorithm s performance Also in order to ensure that the user was not transitioning to a new posture the algorithm removed any acceleration magnitude vector greater than 1 6 m s or less than 0 5 m s which would indicate that a user was jumping running or falling by the following equation A 4 Az Ay Az 2 4 13 Wellness Monitoring System via Body Area Networks Chest Accelerometer X Axis 0 66 D 0 E r E 1 0 5 0 0 5 1 Chest Accelerometer Y Axis 1 z n oE N 1 Thigh Accelerometer X Axis 0
67. esired cut off frequency of approximately 0 05 Hz We chose a quality factor of 0 5 for the filter so as to balance the roll off and flatness characteristics of the filter 23 Wellness Monitoring System via Body Area Networks 3 4 Implementation With Q 0 5 we were able to choose resistor and capacitor values so as to set the cut off frequency of the high pass filter to be 0 05 Hz Specifically we used the equation provided below to determine values of R and C Je 27 RC We thus chose R 16kQ and C 2004F Low Pass Filtering Stage The low pass filter stage utilized a second order Sallen Key topology design Appendix to re move frequency components of the measured ECG signal above the desired cut off frequency of approximately 150 Hz with the same quality factor of 0 5 With Q 0 5 we were able to choose resistor and capacitor values so as to set the cut off frequency of the low pass filter to be 150 Hz Specifically the cut off frequency is the same as Equation Hence we chose R 7kQ and C 0 15 pF Band Reject Filtering Stage It is known that 60 Hz power line interference overlaps with the frequency range containing useful information for ECG signal acquisition and analysis As such we decided early on that we wanted a notch filter with a very high quality factor to precisely reject the 60 Hz frequency component We required a high quality factor to reject a narrow frequency band To achieve a hig
68. ess Control MAC addresses multiple Aps and two different operating fre quencies 2 4 GHz and 5 GHz Therefore the offline database acquisition and data analysis would be much more complicated As a result we purchased six TP LINK routers model TLWR740N Figure 4 2 which all operated at a frequency of 2 4 GHz and had a total cost of 120 These routers are ideal for the indoor positioning module because they operate using a 5 dBi omnidirec tional antenna which allows for better signal coverage to acquire a sufficient database The WiFi routers are set up and used in two different applications one of which is open area testing in order to determine the position of the user in the Atrium s open space which is shown in Figure 4 3 The other experiment will be carried out in a small closed space on the first floor of the EITC building in order to identify the specific room that the user is in as shown in Figure 4 4 If we consider the total cost of implementation for our project 120 for an open area of approximately 400 square meters we have an average cost of 0 3 per square meter Alternatively if we choose to identify the room in which the user is in the hardware implementation cost averages to 20 per room 4T Wellness Monitoring System via Body Area Networks 4 4 Implementation Fig 4 2 TL WR740N WiFi router used in the indoor positioning module E2 205cor ee o e 5 ee
69. ess Monitoring System via Body Area Networks Table A 1 Project Budget Budget System Item Part Number Supplier Quantity Unit Cost Subtotal All Cr2032 Coin Cell Battery P189 ND Digikey 30 0 35 10 50 Enclosure Ergo Case OKW Enclosures Inc 3 20 00 Domain Name NA www namescheap com 2 35 00 Customized Velcro Straps NA China NA 9 00 ECG RTL 12578 A E Ge Real RTL 12578 Abra Electronics 1 48 32 48 32 30 454 9 V E A Hess Dan en e 30 454 Abra Electronics 6 2 24 13 44 os FH 3 Abra Electronics 1 0 58 0 58 Female Header Receptacle Br 2325 2HAN 6V Coin Cell Battery P143 ND Digikey 10 3 94 39 58 Male Headers amp Female Socket ECE Tech Shop ECG Cables ECE Tech Shop ECG Electrodes ECE Tech Shop Miscellaneous 7 i i i i electronic components activie and passive ie SB BE 19 05 7 5556 1173 INA118 Amplifier 595 INA118P 3 15 20 45 60 5648 12 06 Posture Recognition Texas Instruments SensorTag 296 35645 ND Digikey 2 31 66 Texas sa Instruments SensorTag Developer Kit Bee eta ey ENSS j Patient Positioning Wiresless Wifi Router TL WR740N FutureShop 6 18 78 127 35 Step Detection Texas Instruments SensorTag 296 35645 ND Digikey 1 31 66 31 66 Shipping Costs 19 09 Grand Total 412 23 Note These items were personal purchases 94 Wellness Mon
70. etection circuit We wanted to remove frequencies between 0 Hz 0 05 Hz because frequencies in this range cause baseline wander of the ECG signal Frequencies in the range of 0 05 Hz 16Hz would contain useful ECG information but for heart rates that were very slow Frequencies in the range of 150 Hz 226 Hz would most be comprised completely of high frequency noise For this reason we chose to lose some of the ECG information from 0 05 Hz 16 Hz but remove all high frequency noise components above 150 Hz Data Processing and Transmission Circuit The analog output signal of the ECG filtering and detection circuit is acquired by the RFduino s General Purpose Input Output GPIO 6 and the measurement is transmitted wirelessly by the Bluetooth Mate Silver module Figure 3 4 The RFduino was selected because of its Arduino 27 Wellness Monitoring System via Body Area Networks 3 4 Implementation based Advanced RISC Machine ARM MO Microcontroller 16 MHz clock frequency with a 10 bit analog to digital converter that sufficiently quantizes the signal within an adequate range of values Afterwards the microcontroller sends the data to the Bluetooth Mate Silver module via UART communication to be transmitted by Bluetooth protocols to our Android application discussed further in Section 6 1 1 ECG Data Acquisition and Transmission Hardware 9 V battery source poe ae a a View and save ECG with Bluetooth signal
71. etermined the offset was removed to align the user to the 0 90 180 and 270 direction which is shown in Figure Then the corresponding orientation database was loaded to determine the user s position A flow chart showing the overall online phase algorithm is given in Figure 4 5 Initialize the coordinate x 1 y 1 Scan and get qument ASS and user s orientation Set database wifilata9o 7 a TB 5 lt Current Get database degree lt 225 wifiDatal8o Sum values of x and y for the five nearest locations Set database wifiData270 Average x and y locations Get database witilat 0 Calculate Euclidean distance between sample and all reference locations Sort and ist nearest five distance Fig 4 5 Online phase flowchart 94 Wellness Monitoring System via Body Area Networks 4 6 Tests and Results 4 6 Tests and Results We conducted our open area testing in the University of Manitoba s EITC Atrium To begin 100 different locations were randomly chosen by Matlabs randi function The tester walked to the randomly generated locations and then used the wellNode Android application to determine their current location T he results were compared with the actual location via the error equation shown below All of the experimental data is shown in Appendix C error 1 2 za 1 2 Ya yl 4 8 where a Ya were the actual randomly selected posi
72. etooth transmitter require a power supply no greater than 3 6 V We incorporated a voltage regulator MC33269t 33 that converts our 9 V battery power supply to a steady 3 3 V power supply The voltage regulator is placed on the data process ing and transmission circuit board and powers both the RFduino microcontrolller and Bluetooth transmitter simultaneously A dual pole single throw rocker switch was mounted on the top left of the ECG enclosure and is used to power off on all electrical circuits housed within the enclosure Figure 3 6 Referring to Figure one pole of the switch was used to break establish the connection to 9 V whilst the second pole was used to break establish the connection to 9 V 29 Wellness Monitoring System via Body Area Networks 3 4 Implementation Fig 3 6 ECG main power switch Protection Circuitry The ECG device circuitry was designed to include two types of protection circuits specifically input over voltage protection for the electrical components as well as current limiting circuitry to protect users from harmful levels of feedback current from the hardware components To limit the current provided to the user through the electrodes in contact with their skin we used 10 kQ resistors in line with each electrode lead For input over voltage protection of the circuit components we utilized simple diode circuits that will provide a path to ground when input voltages are large enough to damage sensitiv
73. ever few devices Wellness Monitoring System via Body Area Networks 1 2 Project Scope provide the connection between these individual users and health professionals The health pro fessionals can analyze the individual s health information and provide the necessary health advice Therefore the designed project provides the means to connect individuals with health profession als to achieve optimal health advice to increase physical activity and prevent obesity and chronic diseases 1 2 Project Scope The primary focus of this project is to design and implement a wellness monitoring system by creating a Wireless Body Area Network WBAN of sensors and mobile devices In our project we display three important wellness information which include posture recognition ECG signals and step detection that aid in the monitoring of a user s wellness and early signs of possible chronic diseases Also the monitoring of a user s location within a wellness facility is another important aspect in order to find a missing user or find an injured user that requires further assistant All the sensors information is collected in one central node by an Android application called wellNode on a compatible mobile device The sensor information is then processed and the results are displayed on the application or the data can be saved for later analyses within the device s internal memory or within our constructed server 1 3 System
74. f ECG will change from OFF to ON 106 Wellness Monitoring System via Body Area Networks User Manual o e ana MP Module Manager Bluetooth Settings Main Menu Modules Posture A a Pedometer Fig C 5 Module Manager Screen rae LN te a Fig C 6 SensorTag ON OFF switch encircled in red Fig C 7 ECG ON OFF Switch 107 Wellness Monitoring System via Body Area Networks User Manual Note If it is your first time connecting to the ECG module e Enter your Android mobile device s settings e Enter the Bluetooth settings to see all the paired and unpaired devices Figure C 8 e Select GO1ECG under Available devices and enter the pin 1234 Figure C 9 AA E Bluetooth n Palred dewines O YANGLAPTOP O Sonilgble devices 4 a Nexus 5 EA A E Bluctooth pairing request GOIECG 1234 iy O ne 1744 U PIN contains detlers or symbols You may also need to type this PIN on the other eine Fig C 9 ECG module pairing request pin 108 Wellness Monitoring System via Body Area Networks User Manual Posture Recognition Module e With the two SensorTags of the posture recognition connected the posture screen starts to update the pie chart and display past postures as shown in Figure C 10 e Tap once the Clear button to clear the previous pie chart data and posture summary information e Return to the module selection screen as shown in
75. f our posture recognition module is to monitor a user s posture over a set duration of time T his information is useful for healthcare professionals in that they can use such information to infer the wellness conditions of the user Specifically data that indicates extended periods of siting or lying down may indicate an increased risk of developing a chronic disease 3 Hence we selected a subset of specific postures to recognize which were standing sitting bending as well as various lying down positions The hardware we selected to determine a users posture was the Texas Instru ments Sensor Tag due to its light weight small size and overall versatile nature The SensorTag is worn by the user and collects and simultaneously transmits its accelerometers measurements to our Android application via Bluetooth Low Energy BLE The posture recognition algorithm utilizes fuzzy logic systems to provide continuous determination of the user s posture and timely transmits the user information to an online database Within this chapter the posture recognition mod ules specifications background information hardware selection software algorithm development implementation as well as tests and accompanying results are discussed in more detail Wellness Monitoring System via Body Area Networks 2 2 Design Specifications 2 2 Design Specifications The posture recognition module s proposed specifications are summarized in Table 2 1 be
76. g their identification information from the server In contrast when the user is registering a new account on the application the data is parsed into JSON and then sent to the PHP script The PHP script first verifies is there are no pre existing username of the desired registrant If there are no pre existing usernames the script will store the new user identification information into the 85 Wellness Monitoring System via Body Area Networks 6 3 Server Integration MySQL database Sensor Services User Profile i Activity User Information Sensor Information pedometer Preference Preference posture position Login Activity Sensor Buffer Data Legend Activity Container Requests Translated to JSON Service Server Communication Service runs periodically Buffer PHP SCRIPT FOR LOGIN Database REGISTRATION Requests Translated to JSON HTML Scripts Direction of dataflow PHP SCRIPT FOR INSERT DATA INTO DATABASE USER IDENTIFICATION INFORMATION Table Fig 6 11 wellNode Server Integration Overview The second pathway of data flow is the transferring of sensor information from the user s mobile device to the MySQL database The sensor data flow follows the same basic pattern as the user identification information flow However the user identification preference will only be updated once after logging in while the sensor values within the sensor preference are upd
77. g System via Body Area Networks 3 4 Implementation Filter and Detection Circuit The larger of the two circuit boards housed within the wearable ECG enclosure contains all of the components required to both amplify and filter the electrical signals measured at the surface of the patients skin The electrodes apply the measured signal firstly to a differential amplifier stage then to sequential stages as depicted in the figure below From the differential amplifier the signal is passed through a high pass filter a low pass filter a 60 Hz centered notch filter and finally through a final stage of amplification with a DC bias added to the measured signal Differential Amplification Stage We chose to use the INA118 differential amplifier for its remarkably high input impedance 100 MQ to match a user s high skin impedance for better signal acquisition 16 To set the gain of the differential amplifier the following equation is known 50 kQ g G 1 3 1 We decided to use a 10 k 2 potentiometer for Rg With this design we were able to adjust the gain of the ECG waveform at any time if specific patients require a little more or little less gain such that the ECG waveform remains within the RFduino s analog range of 0 V 3 3 V High Pass Filtering Stage The high pass filter stage utilized a second order Sallen Key topology design Appendix B to re move frequency components of the measured ECG signal below the d
78. h quality factor we decided to use the active twin t style notch filter as shown in the Appendix B The active twin t notch filter also has the added benefit that the quality factor of the filter may be adjusted by configuring resistors R4 and R5 which adjust the voltage seen at the non inverting input of the lower operational amplifier U3 in Appendix B We chose Q 5 so as to balance a very narrow 24 Wellness Monitoring System via Body Area Networks 3 4 Implementation width but also a considerable depth of the notch Since the cut off frequency is 1 Je 27 RC 3 3 where R and C correspond to Ry Ra and C1 Ca respectively in Appendix B By convention of the active twin t notch filter C3 201 2 0 3 4 Ri Ra R 5 3 5 To determine R4 and Rs we first have to determine K given that weve already chosen a value for Q gt 3 6 in 5 3 7 K 1 3 1 7 0 3 8 Then Ra 1 K R 3 9 R5 KR 3 10 Final Amplification Stage with Added DC Bias To achieve our target amplification factor specification of 200 V V or greater we implemented a final amplification stage using a 741 operational amplifier set up in an inverting configuration Appendix B It is known that for the inverting configuration the gain of the amplifier is set by 20 Wellness Monitoring System via Body Area Networks 3 4 Implementation the combination of resistors R13 and R14 as seen in Appendix B Specifically the gain is
79. hardware devices the iBeacon and SensorTag that contain all our desired parameters The iBeacon requires an Apple developer license to integrate the device with our project therefore we selected the SensorTag since Android development does not require a license The hardware that we selected was the small Texas Instruments SensorTag shown in Figure which contains a 3 axis accelerometer and a BLE module The KXTJ9 accelerometer communicates at a maximum rate of 3 4 MHz via I C with the SensorTag microcontroller 9 In addition the BLE module can transmit at maximum of 236 7 kbps 10 This implies that 1f we take 10 samples per second of each axis with each sample taking up 3 bytes the total transmission would be 90 bps 10 samples s x 3 axes x 3 bytes sample plus the overhead bytes in BLE packet transmission Therefore the transmission rate is more than sufficient to meet our desired sampling rate of 10 Wellness Monitoring System via Body Area Networks 2 4 Implementation Hz Lastly the SensorTag can be easily implemented on most Android mobile devices with a BLE v4 0 or greater which will be discussed in further detail within Section under Bluetooth Low Energy Overall the SensorTag allowed for quick software development without the need to design our own customized BLE accelerometer device Fig 2 2 Texas Instruments SensorTag The overall system hardware diagram is shown in Figure where the accelerometer data
80. heart s conduction system can be visualized as a function of time In specific the difference in potential measured between electrodes one and two are used in conjunction with a third reference electrode to produce an ECG waveform The third electrode is required to be placed on the body such that 1t is not between electrodes 1 and 2 so as to act as a reference for measuring potential difference 19 Wellness Monitoring System via Body Area Networks 3 2 Background Information FRONT Fig 3 1 Electrode placement Each valley and peak in an ECG waveform represents a critical event taking place within the heart As is the case an ECG waveform should follow an extremely specific form The ideal form for a single period of an ECG waveform is displayed in Figure 3 2 provided below PR Rs Complex ST aa AA Segment FR Interval g OT Interval Fig 3 2 Ideal ECG waveform The above waveform depicting the sequence of events taking place within the heart is repeated once for every heartbeat Any deviations to the above shown waveform in regards to peak amplitude interval times between events and a host of many other distortions can be linked to specific heart conditions Therefore a healthcare professional may be able to diagnose patients with a possible heart condition by studying the measured ECG waveforms produced by a patient s heart 20 Wellness Monitoring System via Body Ar
81. hin the wellness facility It is also possible to use the signal propagation time to accurately determine the distance between transmitters and receivers before applying the triangu lation method to determine the user s position which is similar to how GPS works However the propagation time for indoor positioning is very short thus the transmitter and the receiver must be highly synchronized in order to accurately determine the distance between the transmitters and receivers The cost associated with synchronization of this system makes the implementation of this methodology unfeasible for our project 19 Other researchers have investigated economical and relatively accurate indoor positioning systems via visual determination floor sensors Radio Frequency Identification RFID tags proximity sensors and motion sensors However the method that provides the best results involves the use of the Relative Signal Strength RSS of electromag netic waves from transmitters operating in the radio frequencies RF i RFID and WiFi are two important technologies that operate within the RF range and can be used for an indoor positioning module RFID operates just below 1 GHz within a frequency band that has less interference from other devices and can effectively propagate through walls Unfortunately RFID is a costly expense as one RFID reader costs well above the 500 budget allocated for this project Also an RFID
82. his delay allows for a sufficient overhead such that the RFduino may check for any received command bytes If the RFduino receives the byte OxFF this signifies a STOP command and the RFduino should stop sending ECG measurements to the Android application Ane bytes available toread Read command and save Command 0x00 ACQuIre ECG value via ADC pin Send 2 byte ECG value via little endian Delay by 2 62 me Fig 3 12 RFduino data acquisition and transmission flowchart 3 5 Tests and Results Following the completion of the ECG module hardware the circuitry was analyzed for compliance utilizing several testing methods Below is a summary table of the desired specifications and test 37 Wellness Monitoring System via Body Area Networks 3 5 Tests and Results results of several key features of the ECG wearable device Table 3 2 ECG Specifications Comparison Proposed Specification Sampling Rate gt 300 Hz 300 Hz High Pass Filter Notch Filter 60 Hz 61 Hz Center Frequency Low Pass Filter l l Adjustable gt Leakage Current lt 10 yA Initial tests were done with the BLE and the Android application to determine the wireless data transmission rate The Android application had problems synchronizing with the RFduino BLE consequently the application would miss bytes from a continuous stream of data being transmitted from the RFduino The solution was to create a handshaking protocol between the
83. ine the user s wellness information regardless of their time and location 6 3 2 Server Background Information The main objective of the server would result in the ease of access of the information on the patient In addition there are also some other benefits that is associated with storing the user s information on a database within the server For example as described from Figure below the healthcare professional can run queries on either an individual user or a group to identify any wellness patterns After anaylzing the wellness patterns the healthcare professional can provide any necessary beneficial wellness advice for the user When they see that a user is mostly inactive they can advise the user to become more active by setting a goal for the user to walk more via a 19 Wellness Monitoring System via Body Area Networks 6 3 Server Integration milestone set on the pedometer Then the user will receive the advice and have their pedometer target step number set to that specific goal number to achieve All of the events can be documented through the usage of database Ultimately the healthcare professional can gain great insight to the well being of the patient through the server integration as well as providing adequate wellness advice Te ee ee l l User viewing User viewing their recommendation Use r own activity from hea Ith care professional l l l wel
84. ion Table 2 4 Posture Recognition Specifications Comparison Proposed Specifications Sapling vats Average update delay L344 7 sitting standing bending Number of static postures gt 3 standing sitting and lying down and 4 lying down positions on back front left or right side 17 Wellness Monitoring System via Body Area Networks Chapter 3 Electrocardiogram 3 1 Brief Introduction The overall purpose of the ECG module is to measure and display the electrical signals controlling an individual s heart In a clinical environment viewing the waveform produced by the heart s conduction system can be utilized to examine patterns among heartbeats where abnormalities in the rhythm can indicate various heart conditions 12 For many years the electrical hardware required to measure and display ECG waveforms has been both large and immovable essentially confining users to a chair or bed next to the equipment The goal of our ECG module was to design a portable enclosure to be worn by the user housing a set of hardware that can extract the ECG signal from user worn electrodes The hardware would then transmit the ECG signal to a local mobile device through wireless communications Due to the very specific nature of our requirements we decided to construct our own ECG module by utilizing simple and reliable off the shelf components Within this chapter the ECG module s background information specifications hardware selec
85. ion Table 4 3 RSS at Position 1 10 61 5 V 5 58 5 6 5 Ca a n a a e Table 4 4 RSS at Position 2 5 a a om e a e gt s o u e o w a From each column of both tables the RSS values for the same AP clearly vary for different orien 3 1 ws e tations As an example for AP5 at position 2 5 the RSS was 44 dBm at 180 and 62 dBm at 270 hence the RSS difference of 20 dBm corresponded to a power attenuation factor of 100 This variation in the RSS values is due to a user s own body shielding some of the APs signals and the orientations having different reflected signals Therefore when we tested the fingerprinting method in a different orientation than the orientation of the recorded database the determined position error increased These results indicate that a user s orientation within the indoor positioning area is important to incorporate within the indoor positioning module in order to obtain a minimal error The different orientations were incorporated into our original database by collecting and saving the RSS values for four different directions each with their own database The orientation of the user was determined by using the magnetometer sensor built into the mobile device that runs our 59 Wellness Monitoring System via Body Area Networks 4 5 Software Implementation Android application Once the orientation was d
86. itoring System via Body Area Networks Appendix B ECG Circuit Schematics The following figures are the ECG signal amplification and filtration that occurs in a specific stage order which are as follows differential amplifier high pass filter low pass filter notch filter and final amplifier The electrode placement described in Figure corresponds to the inputs and reference nodes in Figure B 2 electrode 1 connected to Vin electrode 2 is connected to Vin and electrode 3 is connected to Ref Fig B 1 Electrode Placement 95 Wellness Monitoring System via Body Area Networks ECG Circuit Schematics Patient Protection Vin Differential Output Vin INA118P Ref c5 0 1uF Circuit Protection Fig B 2 Differential amplifier with a gain of an adjustable gain as well as patient and circuit protection 96 Wellness Monitoring System via Body Area Networks ECG Circuit Schematics C6 gt High Pass Filter Fig B 3 Sallen Key high pass filter with cut off frequency of 16 Hz High Pass Filter Output Low Pass Filter Output Fig B 4 Sallen Key low pass filter with cut off frequency of 149 Hz 97 Wellness Monitoring System via Body Area Networks ECG Circuit Schematics C1 C2 v2 9V 10nF 10nF R1 R2 Low Pass Filte NAA AMA Output 265KQ 265kQ Notch Filter Output 133kQ 20nF Tf 741 R5 v3 95kQ 9 v Fig B 5 Notch filter with center frequency of 61 Hz
87. ized that the scalability criteria is satisfied by the second method but not the first because the amount of tables allowed within the database is limited while the size of the table itself is virtually unlimited Additionally the second method would allow healthcare professionals to study the complex relationship that might exist between groups of users and their wellness For example a healthcare professional can look at either the user s activity at a particular time of the day or the statistical information of a group in regards to their overall activity Therefore we adapted the second method and created separate tables for user identification information and for the wellness posture information of the user The table for the user identification information Figure includes separate columns for the creation date first name last name date of birth sex phone number weight address username and password The ID column has an data type of int that allows for a sufficient number 101 1 of registered user to store their specific identification and wellness information The indexes we used for the user identification table has the type Binary Tree BTREE because BTREE is an ordered key value map that can retrieve data in a range or in particular order For example we can easily retrieve all the users that has the same birthday or retrieve the users with similar last names 82 Wellness Monitoring System via Body Area Networks 6 3
88. k Dither DiE Month Day Year all all ail Wien Fig C 3 Registration screen Enter your username and password into the corresponding fields and click the Login button to continue to the module selection screen as shown in Figure C 4 oh wellNode gt e NAME GENDER D O B g WEIGHT 0 Posture Summary Steps Taken 0000 Location Pedometer Location Fig C 4 Module Selection Screen 105 Wellness Monitoring System via Body Area Networks User Manual Navigating Through the Application e Tap once on the Posture button to go directly into the posture monitoring screen as shown in Figure C 6 e Tap once on the Pedometer button to go directly into the step monitoring screen as shown in Figure C 7 e Tap once on the Location button to go directly into the location monitoring screen as shown in Figure C 8 e Tap once on the Electrocardiogram button to go directly into the electrocardiogram monitoring screen as shown in Figure C 9 e To enable the Bluetooth connectivity between your mobile device and a specific module sensor select the Module Manager option by tapping once on the three dots located in the top right of the module selection screen The module manager screen is shown in Figure C 5 Managing Bluetooth Settings e Before connecting to the Bluetooth transmitter of the ECG wearable device or the BLE transmitter of a Texas Instruments SensorTag
89. lNode MySQL Database Table USER SENSOR VALUES l l A A o e e e E EE E AAN SN ARA RE ASA AAA A ae A fre aay ee ee ee we oe ee ee er Seg ey o gre eg E me ety meg ene me eg ey Pee ew eee So O ey ee ee N N Health Care Professional a e 7 es imple Query ubgroup of Users nvestigate pattern as Individual User by sex age etc for each sub group Investigate what is every user doing at particular time of l the day Investigate the Investigate events indestipsiatimeor amount of times occurred at j i certain event articular time of tien a parecia nvestigate what p event haseccured event occurred occurred the day l most often for all l users Significant Pattern Changes Changes to be l made l WA A A i A A A A A A ii ia Fig 6 6 Server integration functionality However the ease of accessing the user s wellness information has some disadvantages First of all 80 Wellness Monitoring System via Body Area Networks 6 3 Server Integration there is a possibility for a breach of privacy that becomes the biggest concern with any sorts of people monitoring and data storage systems Secondly the incorporation of a server would increase the initial cost during the implementation due to maintenance protection and associated database expenses We analyzed the advantages an
90. lerometer data characteristic Request a characteristic read Has sampling period passed Hi Fig 6 2 SensorTag accelerometer data polling Accelerometer data characteristic read Store data in array Return Fig 6 3 Asynchronous accelerometer characteristic read Standard Bluetooth Standard Bluetooth is necessary for high data rates such as our ECG module that requires the signal to be acquired at a frequency above 300 Hz As explained in Chapter 3 the RFduino s BLE could not handle the high data rate because of the issue of synchronizing the data transfer between the RFduino and Android application and inconsistent data transmission times Thus we selected the Bluetooth Mate Silver with its standard protocols to handle the ECG data transmissions from the RFduino to the Android application for viewing and file saving This standard protocol allows 14 Wellness Monitoring System via Body Area Networks 6 1 Wireless Data Transmission the two devices to communicate through the Serial Port Profile SPP with a UUID of 00001101 0000 1000 8000 00805f9b34fb The SPP creates a virtual connection that allows the data transfers between devices be transmitted and received instantaneously The SPP communication Android implementation is shown in Figure In contrast to the BLE protocols the standard Bluetooth is connected by matching the devices name G0O1ECG and a entering a pin combination which is 12
91. low The posture recognition module has to be able to recognize at least three different postures which include standing sitting and lying down with an accuracy of 80 or greater The accelerometer sampling rate should be no less than 10 Hz to accurately measure the user s posture and adequately determine a new users posture in at most 2 seconds Table 2 1 Posture Recognition Specifications Feature Value or Range Measurement Explanations Sampling rate 10 Hz The sampling rate of the SensorTag The average time the system Average update delay lt 28 uses to determine a users posture once they become static The accuracy of the posture recognition algorithm The number and types of postures the system should recognize Accuracy gt 80 Number of static postures gt 3 sitting standing and lying down 2 3 Background Information There are many different ways of determining a user s posture utilizing various hardware or software techniques Within the hardware selection some studies used vision dedicated cameras or multiple accelerometers placed across the user s body In the vision dedicated camera method various cam eras are placed throughout an area focusing on a user within its vicinity then the user s posture is determined by applying image signal process techniques 5 However using multiple cameras can be problematic when distinguishing between different users and the cost of implementation incr
92. measure a larger magnitude of accelerations than the Sensor Tag placed above the knee Therefore the ankle accelerometer measurements will have more noise that will affect the overall step detec 99 Wellness Monitoring System via Body Area Networks 5 4 Implementation tion performance Also individual users have unique stride lengths that increase the variability of the magnitude of the ankle accelerometers measurements furthermore increasing the variability of accuracy among individual users We therefore selected to wear the SensorTag just above the knee on the thigh to have optimal accelerometer measurements with minimal noise for the step detection s algorithm Fig 5 1 Pedometer Sensor Tag placement 5 4 2 Software Implementation The step detection algorithm is based on a combination of three conditions namely slope detection static threshold and time interval comparisons Slope detection can determine the slope between two data points as being either positive or negative The static threshold is a constant value that acts as a minimum limitation of the peak value in the step signals The time interval comparison method will determine the time difference between the previous step and the current possible step in comparison to a minimum time required for slow walkers to take one step Overall all three conditions should be satisfied in order to validate and detect a single step We defined a one step cycle a
93. n der Krogt and J Harlaar Self paced versus fixed speed treadmill walking Gait amp Posture vol 39 no 1 pp 478 484 2014 30 Bluetooth Developer Portal Bluetooth Smart Ready products Online Available http www bluetooth com Pages Bluetooth Smart Devices List aspx March 1 2015 31 Bluetooth Developer Portal Data Transport Architecture Online Available https developer bluetooth org TechnologyOverview Pages DataTransport aspx Feb 25 2015 32 Bluetooth Developer Portal Architecture Overview of Operations Online Available https developer bluetooth org TechnologyOverview Pages O 33 Bluetooth Developer Portal Generic Attribute Profile GATT Online Available https developer bluetooth org TechnologyOverview Pages GATT aspx March 3 2015 34 Texas Instruments May 27 2014 SensorTag UserGuide Online Available http processors wiki ti com index php SensorTag_User_GuideA Accelerometer_2 Feb 24 2015 92 Wellness Monitoring System via Body Area Networks Appendix A Budget A complete list of components that were purchased and received are listed in Table The project s budget provided by the Electrical and Computer Engineering Department was 500 00 The total project cost was 412 23 and below the provided budget Also throughout the entire development of the project we did not use the estimated machine time of 1 5 hours for enclosure modifications 93 Welln
94. n that should lead you to cer a card 2 rod Register Lianan Fig C 15 wellNode registration page on the website 113 Wellness Monitoring System via Body Area Networks User Manual e If you have an account enter in your username and password and click the Log in Button you should arrive to your personal wellness information profile Figure C 16 Fig C 16 wellNode s personal wellness information profile 114 Wellness Monitoring System via Body Area Networks Appendix E Software Repository All the Android and Arduino software files can be found at https github com ECE4600G1 115 Wellness Monitoring System via Body Area Networks Appendix F Curriculum Vitae PLACE OF BIRTH YEAR OF BIRTH SECONDARY EDUCATION HONOUR AND AWARDS Cassandra Aldaba Winnipeg Manitoba 1992 Garden City Collegiate 2006 2010 NSERC Undergraduate Student Researcher Award 2014 amp 2015 UMSU Scholarship 2014 Financial Aid amp Awards Merit Scholarship 2014 Dean s Honor List 2011 2014 University of Manitoba s Undergraduate Research Award 2013 UM Queen Elizabeth Scholarship Engineering 2011 University of Manitoba Entrance Award 2010 Tianqi Liang PLACE OF BIRTH HeBei People s Republic of China YEAR OF BIRTH 1991 SECONDARY EDUCATION BaoDing No 1 High School 2009 2012 HONOUR AND AWARDS Dean s Honor List 2013 2014 116 Wellness Monitoring System via Body Area Networks Curriculum Vitae
95. ndividual 27 Moreover the algorithm of the accelerometer can remove some false steps caused by noise or unnecessary vibration of the pedometer itself 26 For our project we used the accelerometer based pedometer because the accelerometer can be placed anywhere on the body and measures the way the body accelerates in time Furthermore it can capture the motion of a step regardless of how tiny the step is By implementing a well designed algorithm the step can be detected whereas in the mechanical system the threshold algorithm only provides you with a value of 1 or 0 regardless of what we do We cannot exploit the data to get more information 5 4 Implementation 5 4 1 Hardware Implementation The hardware that we selected was the Texas Instruments SensorTag which contains a 3 axis ac celerometer and a BLE module for reasons similar to those explained in Chapter 3 For the step detections specific sampling frequency of 50 Hz the SensorTag can be customized by manually changing the initial configuration of the hardware device Theoretically the accelerometer based pedometer can be placed arbitrarily on the users body Af ter a few experiments we found that placing the SensorTag just above the knee on a user s thigh provides a better result than a SensorTag placed on a user s ankle Figure 5 1 If the users place the accelerometer on the ankle while they walk the sensor will undergo a larger range of motion and
96. ne phase Firstly in the offline phase which is also called the calibration phase a fingerprinting database is created containing all of the RSS values associated with all of the predefined reference areas Afterwards the online phase is implemented on an Android application which compares a sample of the RSS values at the user s position and the RSS values stored within the fingerprinting database to determine the user s most likely position 4 3 Design Specifications The indoor positioning module specifications proposed in our project namely requiring the module to be able to determine the users indoor position within eight meters 70 of the time are summa rized in Table 46 Wellness Monitoring System via Body Area Networks 4 4 Implementation Table 4 1 Indoor Positioning Specifications Proposed Specification 8 meters 70 of all the measurements 4 4 Implementation 4 4 1 Hardware Implementation The main advantage of using WiFi technology is the existence of WiFi infrastructure in most facili ties However during our initial testing in the University of Manitobas EITC Atrium we observed that the universitys WiFi router would decrease the accuracy and increase the complexity of the fingerprinting method The fingerprinting accuracy would decrease because the signals vary de pending on the time of the day In addition the complexity would increase because each router had multiple Media Acc
97. nput values tap once on the Cancel button to go back to pedometer screen Note If any of the fields in the target settings screen are left blank a message will prompt the user to enter the values again e Tap once on the Start button to start the step detection process of the pedometer The steps taken and walking speed of the user will be updated on the pedometer screen e Tap once on the Stop button to stop the step detection process of the pedometer e Tap once on the Reset button to clear the progress and records of the pedometer e Tap once on Return button to return to the module selection screen as shown in Figure C 4 24 ZP rma 1 Pedometer th TargetSetting Target Steps Target Steps 100 Step Ta Step Size Meter Walking Speed 0 Fig C 11 Pedometer and target setting screens 110 Wellness Monitoring System via Body Area Networks User Manual Indoor Positioning Module e Choose an environment to start determining the user s position in the location screen as displayed in Figure C 12 e To determine the location within a specific room tap once on the Room button the button will change its status to show Room True e To monitor the location in an open space i e The Engineering Atrium tap once more on the Room button the button will now change its status to show Room False e Tap once on the Load button to detect the user s orientation and
98. oe a ee eee 55 5 1 Step Detection Specifications A ooo 57 ee E E ae ea ee ee 66 ee ee aa aaa 68 eee Sete eRe hes GEOR eee 2 68 6 1 SensorTag Specific Accelerometer UUIDs B4 0 ee 72 as 73 a eee 94 aaa 99 x111 Wellness Monitoring System via Body Area Networks NOMENCLATURE Nomenclature 6 S posture Atz F Posture FPosture n EuDi RSS RSS x Xj Description X axis accelerometer measurement g Y axis accelerometer measurement g Z axis accelerometer measurement g Euler roll angle rotation about z axis Euler pitch angle rotation about x axis Euler yaw angle rotation about y axis Fuzzy logic posture score Magnitude of 3 axis accelerometer g Chest SensorTag x axis accelerometer measurement g ee ee Chest SensorTag y axis accelerometer measurement g Chest SensorTag z axis accelerometer measurement g e g Thigh SensorTag z axis accelerometer measurement g Lol Thigh SensorTag x axis accelerometer measurement Thigh SensorTag y axis accelerometer measurement Specific posture fuzzy subset function for the chest SensorTag Specific posture fuzzy subset function for the thigh Sensor Tag Number of access points Euclidean distance m RSS database values dBm RSS value measurement dBm Most probable x coordinate position Closest x coordinate reference location to sample point from KNN xiv Wellness Monitoring System via Bod
99. oe o e e To E2 249 e E221 T r e a E2 200st eoe ooo o o lee oeoo EO OO ooo eeeeeoeoeeeeeeee eeeeeoe lt e35usoeweieee3svee ee eeeeeeec3 ex3eeenee E2 247 ee 0 0 bex2 9 e ae eeeeeeeteete3usvse3e30e O E eeeeeeeeeeev3ee3se ee ro o eje eoe o o o RkO Cd lt Oo q DSDS gt SEASO CECOCOOoO co E2 295 gt eeeeeeeecsdv3eceseeeeeeeeteeeee eeeeeee e eeeeeeeeeeeeee Origin x 1 y 1 O o e e e eoe o e e e o e Go 0 ES eee es SERES REA E2 201 Na i 1 I il Legend EITC E2 a e gt 75 CHANCELLORS CIRCLE Finger Printing igi Router Location Origin 1 1 Reference Locations Fig 4 3 Open area WiFi fingerprinting map 48 Wellness Monitoring System via Body Area Networks 4 5 Software Implementation TUNN a ES hs 6 b i E2 1 lt EJE Ra 2 110 Se m m r a S i E2 171V y y E2 150 E E2 160 m e de othe a 2 164 E2 158 m L eos l E2 154 mlin Lh L E2 166cor IX AA N N E Na N PA a E2 145cor e e e a citer ES 994 Fig 4 4 Indoor positioning module room testing setup 4 5 Software Implementation 4 5 1 Offline Phase In the offline phase a mobile device was used to measure the RSS in dBm of six APs at the fingerprinting reference locations We initially mapped 27 equally dispersed locations in the Atrium however the position errors were above eight meters mos
100. ometer UUIDs FOOOAA10 0451 4000 B000 O00000000000 FOO00A A11 0451 4000 B000 000000000000 FOO00A A12 0451 4000 B000 000000000000 12 Wellness Monitoring System via Body Area Networks 6 1 Wireless Data Transmission Start of BLE sensorTag initialization Discover and connect to sensorTag Discover sensorTag services Get accelerometer service amp configuration characteristic Write byte 0x01 to characteristic Poll sensorTag accelerometer data Fig 6 1 BLE initialization Enable accelerometers Table 6 2 SensorTags MAC Addresses Sensor Tag MAC Address 90 59 AF 0B 82 F4 Thigh 90 59 AF 0B 82 D9 OSA ZABOLOR The SensorTag accelerometer values are polled as shown in Figure The polling loop will read any stored data in an array and process the data with its corresponding module algorithm Next the Android application will retrieve the accelerometer service and data characteristics and then request to read the data Afterwards the application will wait until the sampling period time has passed An asynchronous function Figure 6 3 is called once the accelerometer data is retrieved this function simply stores the accelerometer data in an array so that the polling loop can process 13 Wellness Monitoring System via Body Area Networks 6 1 Wireless Data Transmission the data further Poll sensorTag devices loop Read stored data in array if available and process Get accelerometer service and acce
101. ometer will swing in attempt to reach a certain predefined threshold Once the threshold is reached either a mechanical or electrical counter is incremented An alternative method to the mechanical threshold pedometer is the electrical motion sensor pedometer which utilizes motion sensing accelerometers to determine the number of steps taken by the user by ov Wellness Monitoring System via Body Area Networks 5 3 Background Information pattern matching or peak detection methods 5 3 1 Mechanical Pedometer The mechanical threshold based pedometer is power efficient and accurate when the user wears the step detection device on the waist or when it is securely placed in the users pocket However when the device is placed on a different location on the body or simply put inside a bag the result will be inaccurate due to the inability to reach the threshold Since the mechanical pedometer contains a rod inside the rod can swing within a specific angular range that detects one step once it connects to the circuit inside the pedometer and creates induced current and voltage in the circuit For arbitrary placement of the pedometer the rod may not be able to connect to the circuit inside and register that one step has been taken Also it is difficult to design the algorithm with threshold detection and step patterns because the mechanical pedometer implements the hardware circuit with a moving rod inside the box As time goes
102. re posture a wo o To pme pateo A chnged on Jimeno No CURRENT TESTAN es a NN E EME y Indexes e Cardinality Collation Keyna procar erere xe xo fa pomo Ja wo Fig 6 9 User posture information table structure id posture timetag changed _on username h Standing 2015 02 25 18 04 47 2015 02 25 19 20 00 yang 2 Laying down back side 2015 02 25 18 04 51 2015 02 25 19 20 01 yang Ww Standing 2015 02 25 18 05 10 2015 02 25 19 20 01 yang 4 Laying down back side 2015 02 25 18 05 17 2015 02 25 19 20 01 yang mn Standing 2015 02 25 18 05 28 2015 02 25 19 20 02 yang Fig 6 10 User posture information table 6 3 5 Client Server Data Flow The data flow to and from the server can be classified as two different pathways even though the mechanism of data transfer is the same One pathway is for user identification information re 84 Wellness Monitoring System via Body Area Networks 6 3 Server Integration trieval storage and authentication The second pathway is for wellness sensor value storage Both of the pathways start on the client side wellNode Android application running on the user s mobile device In the wellNode Android application we utilize an Application Program Interface API called Shared Preferences which allows simple storage and retrieval of key value pairs within the mobile device s internal memory The user information and sensor information are both stored inside
103. re SensorTags placement 2 aa a a 10 oe eee ee eae eee eee ee ee ea ee 14 2 6 Posture recognition flowchart 2 0 e 15 2 7 SensorTag customized Velcro strap ooo a a a a a 16 3 1 Electrode placement oo aoaaa es Sew eee eee Pw ew ada 20 ee E ee ee ee ee ee 20 3 3 ECG module internal layout ee 22 3 4 ECG data acquisition and transmission circuit 0 28 aras as sa 29 3 6 ECG main power switch a a a 30 3 7 3M Red Dot Ag AgCl single use electrode 2 020002008 31 3 8 3 5 mm stereo plug and jack electrode lead connection 32 3 9 ECG module electrode leads 2 e 33 3 10 ECG module enclosure with lid affixed 0000000 34 cerrara 35 3 12 RFduino data acquisition and transmission flowchart Of Wellness Monitoring System via Body Area Networks LIST OF FIGURES 3 13 Oscilloscope capture of ECG Bluetooth transmission test 39 aaa asa a 40 4 1 Position estimation for triangulation method 04 45 4 2 Indoor positioning module open area testing a ee 48 4 3 Open area WiFi fingerprinting man 48 4 4 Indoor positioning module room testing setup a 49 4 5 Online phase flowchart 2 0 a a a 54 5 1 Pedometer SensorTag placement e 60 AEREA 62 5 3 Step detection owchartl a 6
104. rent measurement is smaller than the value of the previ ous data point One possible step is defined when the slope of the measurement becomes positive and then changes to negative In other words we define one possible step with three piecewise components with a positive slope and two piecewise components with a negative slope as shown in Figure 5 2 The red dots are the accelerometer measurements and one possible step is represented by the total five measurements having a positive then negative slope The five points were selected because we assume that a possible single step would take approximately 0 1 s After a possible step is detected the algorithm will validate the step by using a predetermined threshold 61 Wellness Monitoring System via Body Area Networks 5 4 Implementation Local Maximum Steps peak S bS Signal from Accelerometer Predetermined Threshold e Data Sample Time difference Fig 5 2 Step detection algorithm Once a possible step detection peak is detected the local maximum and the current accelerom eter measurement will be compared to a predetermined threshold The finalized predetermined threshold was 0 2 g which was found by analyzing the recorded accelerometer measurements of a past walk The threshold requirement is satisfied by comparing the difference between the local maximum and the current measurement value to the predetermined threshold As seen in Figure
105. s a user planting one foot on the ground lifting the same foot to move forward and then planting the foot on the ground again in the step detection module Before the 60 Wellness Monitoring System via Body Area Networks 5 4 Implementation step detection starts the algorithm will determine the orientation of the SensorTag for accurate data acquisition The algorithm will compare the accelerometer values of each x y and z axis every 10 seconds every 500 samples This calibration period was selected because we assumed that the orientation of the SensorTag will not change frequently and that 10 seconds will be fast enough for the algorithm to make the corresponding modifications The accelerometer measure ment that has the largest magnitude signifies that its axis will align with the gravitational vector when it is parallel to this vector hence the axis determines the Sensor Tag s orientation Once the Sensor Tag s spatial orientation is determined the step detection algorithm uses the axis that could align with the gravitational vector as the accelerometer measurement input to the step detection algorithm The first stage of the step detection algorithm is the slope detection to determine a possible step The slope is determined by comparing the value of two consecutive measurements We have a positive slope when the current measurement has a larger value than the previous data point and a negative slope is when the value of the cur
106. s equation n represents the number of APs RSS is the RSS value from the database and RS S is the RSS value measured during the online phase However when using the Euclidean distance method it is possible to produce inaccurate results as the WiFi signal varies with many factors such as attenuation and reflections Therefore we applied an enhancement algorithm to produce a more accurate result regardless of the varying WiFi signals The K nearest neighbour KNN algorithm selects K nearest results found using the Euclidean method and then averages them to get an enhanced estimation of the users position 23 i L Na 4 6 y 7 gt y 4 7 Here K is the number of nearest reference locations chosen Through experimentation we decided to use five as it gave us more accurate results During our next tests we found that the orientation of the user within the Atrium makes a sig nificant contribution to the accuracy of determining that user s position The two sets of RSS values shown in Table 4 3Jand Table 4 4 were randomly selected from the database for two different positions 1 10 and 2 5 with four different predefined directions According to the map 4 3 0 is south of the point at the red cross 90 is north of the point at the red cross 180 is west of the point at the red cross and 270 is east of the point at the red cross 52 Wellness Monitoring System via Body Area Networks 4 5 Software Implementat
107. s for undetected errors by the baseband layer 31 As developers we did not need to implement our own customized error detection and correction protocols and could readily process and analyze the data received from any Bluetooth device Lastly the ability of Bluetooth to perform frequency hopping is a key feature in the situation when there are many other devices that are wirelessly transmitting data within the same frequency band The Bluetooth frequency hopping ability of the transceiver acts to prevent interference and fading of the wireless communication between devices 32 In conclusion Bluetooth was selected in our project due to its popularity and data transmission reliability but BLE and standard Bluetooth protocols perform differently It is thus important that one choose the appropriate protocol either standard Bluetooth or BLE depending upon their desired intentions Bluetooth Low Energy BLE BLE is used for the SensorTag in our step detection and posture recognition modules due to its compatibility with low data transmission rates and ease of implementation BLE technology is ideal for these modules as its low sampling frequency of 10 Hz for the posture recognition and 40 Hz for the step detection are adequate to meet our desired specifications When the accelerometer data is polled at a set sampling period any delays due to data correction detection and retransmissions 71 Wellness Monitoring System via Body Are
108. se the scalability and accessibility 89 Wellness Monitoring System via Body Area Networks REFERENCES References 1 SS P T Katzmarzyk and I Janssen The economic costs associated with physical inactivity and obesity in Canada An Update Canadian Journal of Applied Physiology vol 29 1 pp 90 115 2004 Public Health Agency of Canada Benefits of Physical Activity Internet http www phac aspc gc ca hp ps hl mvs pa ap 02paap eng php Jan 20 2011 Feb 21 2015 K I Proper A S Singh W van Mechelen and M J M Chinapaw Sedentary behaviors and health outcomes among adults A systematic Review of Prospective studies American Journal of Preventive Medicine vol 40 2 pp 174 182 2011 C Aldaba T Liang Y Su H Wang and J Winkler Design and Implementation of a Pervasive Health Monitoring System via Body Area Networks September 2014 project Proposal for ECE4600 D Brulin Y Benezeth and E Courtial Posture recognition based on fuzzy logic for monitoring of the elderly IEEE Transactions on Information Technology in Biomedicine vol 16 no 5 2012 Q L John et al Accurate fall fast detection using gyroscopes and accelerometer derived pos ture information in 2009 Sixth International Workshop on Wearable and Implantable Body Sensor Networks pp 138 143 2009 M Aloqlah R R Lahiji K A Loparo and M Mehregany A headband for classifying human po
109. set by the following equation Ce 3 11 As we had already established a gain of by the initial differential amplifier we chose R13 and R14 such that the amplifier would supply a gain of 25 Specifically we chose Ry3 25 kQ and Ri4 1 kQ In addition to amplifying the ECG signal prior to the input to the RFduino microcontroller the ECG signal also required the addition of a DC bias voltage due to the fact that the RFduino mi crocontroller cannot accept input voltages below OV on any of its input pins To bring the baseline of the ECG signal above 0 V we used a 25 kQ in combination with a 10 kQ potentiometer to create a voltage divider and applied the bias voltage to the non inverting input of the 741 operational amplifier By using the 10 kQ potentiometer we were able to adjust the DC bias applied to the ECG waveform at any time if specific patients require a little more or little less bias to have the ECG waveform centered within the RFduino s analog range of 0 V 3 3V We required the use of two capacitors Cio and C41 as seen in Appendix B for the purposes of AC coupling of this stage of the circuit Essentially the capacitors act to configure this stage of the circuit so that it only amplifies signals that change with time We encountered difficulties in choosing the value of Cio and C11 We knew that the combination of R and C components would have an effect on the filtering characteristics of our circuit overall We
110. stem via Body Area Networks 6 2 Mobile Application Welcome Screen Login Screen User Registration Screen Wellness Modules Management and Connection Screen User s Main Personal Wellness Information Screen Step Posture Positioning ECG Screen i o Detection Screen Recognition Screen System Fig 6 5 Android application hierarchy flow Some of the module s procedures occur within the current screen or within a background service Any updates to the current visual screen that a user looks at may only occur when a user is on that particular screen This ensures these high prioritized modules programmed functions occur before any other lower prioritized modules However today s mobile microprocessors allow applications to perform many background functions which are called services in Android development They can execute without interfering with the user s interaction with the application If there are no services the application functions will execute in a linear path that only one function can be executed at a time Therefore we decided to execute all the Bluetooth protocols posture recognition and step detection algorithms within their own individual background services The services allow the user to continuously interact with our application without the application having any noticeable delayed reactions Furthermore without the users knowledge their wellness information can be transmitted to an online
111. stures in 32nd Annual Internation Conference of the IEEE EMBS online Buenos Aires Argentina 2010 vol 2010 pp 382 385 M Pedley Tilt sensing using a three axis accelerometer Freescale Semiconductor Inc Docu ment Number AN3461 2013 Kionix 2g 49 89 Tri axis Digital Accelerometer Specifications Rev 4 Ithaca NY 2012 10 C Gomez J Oller and J Paradells Overview and Evaluation of Bluetooth Low Energy An Emerging Low Power Wireless Technology Sensors vol 12 pp 11734 11753 2012 90 Wellness Monitoring System via Body Area Networks REFERENCES 111 Digi Key Corporation Aug 15 2014 Moving forward with Bluetooth Low Energy Online Available http www digikey com en articles techzone 2014 aug moving forward with bluetooth low energy Feb 24 2015 112 Mayo Clinic Staff Oct 24 2012 Tests and Procedures Electrocardio gram ECG or EKG Online Available http www mayoclinic org tests procedures electrocardiogram basics definition pre 20014152 March 1 2015 113 P E Paulev and G Zubieta Calleja New Human Physiology Paulev Zubieta 2nd Edition Chapter 11 Cardiac Action Potencials and Arrhythmias Online Available http www zuniv net physiology book chapter11 html March 1 2015 114 K Soundarapandian and M Bereaducci Analog Front End Design for ECG Systems Using Delta Sigma ADCs Texas Instruments Dallas TX Rev Apr 2010 15 W
112. style than they have been in the past In rehabilitation centers and medical fitness facilities patients are required to participate in certain physical activities to reduce their recovery time or to reach their fitness goals T he most common way to achieve these objectives is to increase the amount of time spent doing exercises especially walking and running T herefore the implementation of a pedometer which is a step detection counter is useful in monitoring the user s time spent doing a physical activity The pedometer is typically worn as an external piece of hardware on a belt however modern mobile devices have evolved to include motion sensors that can combine with a step detection algorithm The commercial pedometers that are currently available on the market are designed to monitor the steps for daily fitness Most of the pedometers work quite well for users that walk with a normal walking speed However users that walk slowly or shuffle make it difficult for pedometers to detect their steps Thus we attempt to resolve this problem by focusing on developing a pedometer specifically for users with slow walking speeds or a shuffle walking style In this chapter the step detection s specifications background information hardware selection software development and performance tests will be discussed in detail 56 Wellness Monitoring System via Body Area Networks 5 2 Design Specifications 5 2 Design Specifications Th
113. t of the time Therefore we increased the positioning module s accuracy by creating more reference locations within the Atrium The second attempted WiFi fingerprinting database was constructed using a total of 270 reference locations that were separated by 1 2 m as illustrated by the blue dots in Figure 4 3 49 Wellness Monitoring System via Body Area Networks 4 5 Software Implementation Database Construction The database was constructed by acquiring the RSS measurements and then storing the measure ments in a concise format At each reference location represented by an x and y coordinate we measured the RSS values and stored the measurements in a dbs format specified for database storage The finalized constructed database was implemented as a table using the MySQL database in our Android application The database is built in the form shown in Table which contains the coordinates and the RSS values for each of the APs Table 4 2 Indoor Positioning s Database Format a am amare Jas Jas The database s construction is integral for our algorithm We initially constructed a database that sorted using simple numerical tags such as location 1 through location 27 Then we had a look up table to translate the 1 to 27 locations into a relative location system using a red cross the origin At the origin and north of that point is the positive y axis while east of the red cross is the positive x axis Thus all of th
114. ta to a remote location T he stored data is a accessible to healthcare professionals that have the right qualifications 88 Wellness Monitoring System via Body Area Networks 7 1 Future Work 7 1 Future Work The wellNode demonstrates the feasibility of a wellness monitoring system at a low cost However the system can improve in its functionality and desirability Specifically for the posture recognition module we like to implement a calibration and machine learning algorithm to better assess the user s static posture and dynamic exercises In addition we would like to implement a fall detection algorithm The positioning system also requires work in the future to better implement a system to collect the finger printing map to reduce the resource cost for generating the fingering printing map initially For pedometer we would like to incorporate machine learning algorithm so the step detection is tailored towards the specific user The ECG module needs algorithms implemented to detect heart beat and signal irregularity monitoring T he system as a whole can be improved if we can smartly integrate different sensor information together For example if the user is walking we could use location system to point out where they are and use pedometer to show the speed of which they are walking and the direction they are heading towards Lastly the server integration requires database management and perhaps specialized database structures to increa
115. ter based Pedometer 2 0 0 ee 58 aa aaa A 59 5 4 1 Hardware Implementation 0 e 59 AAA 60 5 5 Tests and Results copiosa srta AAA RA 65 6 System Integration 0 0 0 0 0 2 ce 70 POOR eH Paeet ee eho asa ss 70 Ol Buctoothi s 4 4262 6S eR REDE REECE RE AAA 70 6 2 Mobile Application 2 444448 ee ee Beate howe oe PB eRe ey ew 76 ee eee ee ee ee T9 6 3 1 Server Integration Introduction o e a a a 19 6 3 2 Server Background Information 0 0 00 0800 19 6 3 3 Server Implementation a a a 81 634 Database Structural 82 6 3 5 Client Server Data Flow 84 MAA 88 Tok aa senaeacsaxvtutbeeasetoeeebbueerenh eee ee bheieed chee e 89 iaa sra a A A 92 Appendix A Budget 0 0 00 0 2 93 Appendix B ECG Circuit Schematics a 95 Appendix C Indoor Positioning Test Results 99 Appendix D User Manual e 103 Vill Wellness Monitoring System via Body Area Networks TABLE OF CONTENTS Appendix E Software Repository 0 0 a 2 ee 115 Appendix F Curriculum Vitae 0 0 0 0 00 2 ee ee 116 1X Wellness Monitoring System via Body Area Networks LIST OF FIGURES List of Figures 1 1 wellNode system modular overview e 3 2 1 Defined 3 dimensional axes with Euler angles 6 2 2 Texas Instruments Sensor Tag ee 9 AENA 10 2 4 Postu
116. ternet The server is currently located at www wellnode ca and contains a simple webpage a discussion forum as well as a centralized MySQL database for user identification information as well as the user s wellness information posture step detection and position Due to the importance of the high ECG sampling rate normal wireless server protocols are not recommended and the topic on its own is beyond the scope of our project 81 Wellness Monitoring System via Body Area Networks 6 3 Server Integration 6 3 4 Database Structure The server database stores all the information of every registered user The database management system we picked for our project is MySQL due to wide adaptation and easy implementation lt is not the best database for health related information storage because its performance suffers when the database has a high concurrency level which means the database slows down when data are stored or read very frequently The structure of the database should be simple and efficient and scalable First the database should be simple to allow an ease of data storage and retrieval by implementing separate tables for user identification information and the user s wellness informa tion However there are two ways to store the wellness information One way is to create separate tables for each individual user with their wellness information The second possibility is to create tables for each type of sensor measurements We real
117. tion e Navigate to the application menu on the Android mobile device and locate the application titled wellNode Start the Android application by tapping once on the wellNode icon depicted as shown below The welcome screen should immediately appear and should resemble Figure C 1 as provided below e wellNode lt O LJ Fig C 1 Welcome screen e Tap once on the Start button at the bottom of the welcome screen to continue into the user login screen Your screen should now resemble Figure C 2 and have blank fields for entering your username and password 20 2 dh Login ee 3 Fig C 2 Login Screen The Android robot is reproduced or modified from work created and shared by Google and used according to terms described in the Creative Commons 3 0 Attribution License 104 Wellness Monitoring System via Body Area Networks e Ifyou are a returning user please skip to the next step User Manual e Ifyou are a first time user of the wellNode application tap once on the Register button to lead you into the registration screen Figure C 3 To register a new account input your personal information into the required fields Once you have completed all of the text entry fields tap once on the Register button located at the bottom of the registration page you will now be taken back to the login screen Figure C 2 er Mura Parma Firgi nares LASA rm Aches Para E Er Female Mi
118. tion software development implementation as well as performance tests and accompanying results are discussed in more detail 18 Wellness Monitoring System via Body Area Networks 3 2 Background Information 3 2 Background Information In the most basic sense the human heart is an electrochemical generator that is suspended in a conductive medium 13 The potentials generated by the heart s conduction system are used to open and close valves within the heart which permit the flow of blood into successive cham bers The heart s conduction system as any insulated electrical conductor is not perfect and it is known that electrical potentials generated within the heart leak into local tissues surrounding the heart Since the internal tissues of the body being comprised mainly of ion rich fluids act as a good conductor we can hence measure the sum of electrical activity at the surface of the user s skin An ECG is a waveform displaying the sum of variations in electrical potential generated by the heart as a function of time Electrical potentials are generated at the sinoatrial SA node located inside the upper right chamber of the heart and pass down the heart s conduction system until they reach the apex bottom left region of the heart a finite time later By placing two electrodes on the surface of the skin in parallel with the path of the heart s conduction system see Figure 3 1 the electrical potentials passing down the
119. tion coordinates and x y were the most likely determined position coordinates For the room testing we conducted our experiment on the first floor of the EITC building as shown in Figure The tester randomly picked a room before walking to each green starred location The Android application was used to test the users room location at those starred positions The results were definitively accurate with absolutely no error in room detection To summarize Table 4 5 shows the final indoor positioning module specifications compared to the proposed features Table 4 5 Indoor Positioning Specifications Comparison Feature Proposed Specification Results Average error within the Atrium is 3 02 m Room accuracy 100 Update Delay Not defined lt 3 seconds Accuracy 8 m 70 of all the measurements In summary the results show that the average error in the open area test is 3 02 m while the room testing results were 100 accurate By comparing these results with our proposed specifications we achieved and superseded all requirements In addition the cost of the hardware implementation was very low only 0 3 m is required for a desired coverage In conclusion our final indoor positioning algorithm shows great feasibility in patient monitoring 59 Wellness Monitoring System via Body Area Networks Chapter 5 Step Detection 5 1 Brief Introduction People today are more concerned about maintaining a healthy life
120. und experimentally to be 149 0 Hz The implementation of our low pass filter matches nearly identically the desired specification for this filter Amplification of the ECG circuitry was tested in the pass bands of the combined three filters namely at 40 Hz and 80 Hz With the gain of the circuitry set to a maximum just below a level at which clipping of the output signal was observed the gain was found to be 543 1 V V and 520 2 V V at 40 Hz and 80 Hz respectively Our implementation exceeds our specification by a factor greater than two fold and in addition can be adjusted and fine tuned to be work with a wide range of individuals In researching further about the techniques for measuring device leakage current we made a critical realization in that our device would not pose any risk in this regard Leakage currents are only applicable to medical devices that are physically connected to ground either by a 120 V 60 Hz wall outlet or other means If the physical path to ground were to be broken the current that would normally flow to ground the leakage current would pose a risk for the user as the current must now pass through the human body to reach ground In the case of our ECG wearable device all components are supplied power by batteries which employ a floating point ground design As is the case the idea of harmful leakage current does not apply to our device Al Wellness Monitoring System via Body Area Networks Chapter 4
121. uzzy logic system methods are not too difficult to implement we decided to investigate both methods We began by selecting a suitable 3 axis accelerometer device and then implemented an angular thresholding method and later a fuzzy logic method to classify different defined postures We discovered that combining the two methods resulted in difficulties determining certain postures Therefore the finalized algorithm is based on fuzzy logic systems only Wellness Monitoring System via Body Area Networks 2 4 Implementation 2 4 Implementation 2 4 1 Hardware Overall the general hardware selection requires a microcontroller BLE module and a 3 axis ac celerometer We initially decided to use various Arduino microcontrollers such as Arduino Minis or Arduino Unos to eliminate the additional prototyping time if we were to design our own microcon troller component However the BLE modules are typically a surface mount device that is difficult to prototype and wire together In addition if we used a readily made Arduino microcontroller board the size of the overall device with an accelerometer and BLE transmitter would be large and ultimately cause the user some discomfort Thus we decided to look into hardware alternatives that are currently in the market and contain the following desirable parameters e Small size e Easy to program and implement e Contains a 3 axis accelerometer and a BLE module Currently there are two
122. wn on right side 2 4 Implementation Fig 2 5 Posture fuzzy logic subset functions where the x axes are the accelerometer measurements and the y axes are the likelihood values 14 Wellness Monitoring System via Body Area Networks 2 5 Tests and Results Has 0 15 passed E Poll sensorTags for accelerometer measurements Is sensorTag data available Acceleration magnitude lt 1 pr Moving Average Filter T Li ad a Calculate posture scores 3 E Sit E L Stand TG Find highest posture score om Bend Lying Down Positions Fig 2 6 Posture recognition flowchart 2 5 Tests and Results The SensorTags were adhered to the user s thigh and chest by a customized Velcro belt with a pocket that holds the device Figure 2 7 This belt can be lengthened by adding more Velcro belt segments hence the belt ensures that the SensorTag is tightly strapped against various body sizes The algorithm was tested to determine the accuracy and average determination time by using the 15 Wellness Monitoring System via Body Area Networks 2 5 Tests and Results project s Android application to determine and display the user s posture All five group members performed a series of 35 uniformly distributed random postures and the determination time was manually measured by a stopwatch The results of the tests are summarized in Table Fig 2 7 SensorTag customized Velcro strap Table 2 3
123. xis orientation Orientation iteration 500 2 ase 2 Y axis orientation ositive slop a Case gt Z axis orientation Y Positive slope counter increment Determine maximum value lesative slope J Negative slope counter increment Peak Maximum value current value Mime difference Megatiwe slope counter gt 17 Step counter increment iteration increment Clear positive slope counter Clear negative slope counter Gear the maximum peak value Record the current values of data points Fig 5 3 Step detection flowchart 64 Wellness Monitoring System via Body Area Networks 5 5 Tests and Results In this equation frrr is the step frequency determined from fs that represents the sampling frequency which is 50 Hz N is the sample size and Mmax is the frequency index at which the maximum peak occurs other than DC component We use the step size entered by the user and the step frequency obtained from the FFT computation to figure out the speed during a period of time in terms of miles per hour The walking speed of the user will be determined after every 5 seconds by the following equation v stepsize frrr 5 3 Since we conducted our tests using the walking speed displays on the treadmill we must convert the speeds determined by the FFT algorithm into miles per hour instead of kilometers per hour 5 5 Tests and Results The SensorTag was placed just above the knee on the user
124. y Area Networks NOMENCLATURE Symbol y y Description Most probable y coordinate position Closest y coordinate reference location to sample point from KNN Number of nearest reference positions Discrete Fourier Transform of accelerometer measurements Sample size Frequency index Frequency index at which the maximum peak occurs for the step frequency Step frequency Hz Step detection accelerometer sampling frequency Hz Walking speed m s XV Wellness Monitoring System via Body Area Networks ACRONYMS Acronyms Symbol ABS ADC AP API ARM BLE BTREE ECG FFT GATT GPS GUI JSON KNN LAN MAC MySQL PHP RF RFID Description Acrylonitrile Butadiene Styrene Analog to Digital Converter Access Point Application Program Interface Advanced RISC Machine Bluetooth Low Energy Binary Tree Electrocardiogram Fast Fourier Transform Generic Attribute Profile Global Positioning System Graphical User Interface JavaScript Object Notation K Nearest Neighbor Local Area Network Media Access Control My Structured Query Language Hypertext Preprocessor Radio Frequency Radio Frequency Identification xvi Wellness Monitoring System via Body Area Networks ACRONYMS Symbol RSS RSSI SPP ToA UART USB UUID WAN WBAN WiFi WWW Description Received Signal Strength Received Signal Strength Indicator Serial Port Profile Time of Arrival Universal Asynchronous Receiver Transmitter Universal

Design and Implementation of a Wellness Monitoring System via

Contents

Download Pdf Manuals

Related Search

Related Contents