A Novel Period Estimation Method for Periodic Signals Suitable for Vital Sensing

愛知県⽴⼤学情報科学部情報科学科神⾕幸宏研究室⼩久保律樹 2019/10/18 Rev. 5 [DRD-PUB-00002] Yukihiro
KAMIYA Laboratory Dep. of Information Science and Technology Ritsuki KOKUBO "JDIJ1SFGFDUVSBM6OJWFSTJUZ A Novel Period Estimation Method for Periodic Signals Suitable for Vital Sensing ICBSP2019

*OUSPEVDUJPO "/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH

*P5GPSNFEJDBMBQQMJDBUJPOT Patinet IoT device PC Respiration Heartbeat etc Vital
Sign FINE Staff Monitoring patients using their vital sign collected by the IoT device Improve care quality "/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH

$PNQPOFOUTPGUIFTZTUFN Patinet PC Vital Sign Sensing method 1 Period
estimation algorithm 2 "/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH IoT device

7JUBMTFOTJOHNFUIPE Conventional method: Electrodes Alternative method: Non-contact sensing Accuracy:
High Uncomfortable Accuracy: Low Comfortable 1 2 "/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH

0CTUBDMFPGOPODPOUBDUWJUBMTFOTJOH /PJTF Media + Noise SNR of signal degradation
due to noise is inevitable Robustness against noise Frexibility for installations where sensing device install ＝ 1 2 "/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH Non-contact vital sensing Important Robustness against noise

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH 7JUBMTJHO 3 [s] About 1 [s] About
1 2 [Hz] [s] Frequency Period ･･･ Sensor output Vital sign Respiration, Hearbeat etc… Respiration Heartbeat Period is very LONG

Resolution in Low frequency band is LOW "/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH 6TF%'5
DFT There is a problem for non-contact vital sensing Vital sign 1 2 PERIOD axis DFT spectrum

1SPCMFNPG%'5ʢʣ 1 2 Sensor output DFT 0 0.2 0.4
0.6 0.8 1 0.02 0.025 0.03 0.035 DFT spectrum FREQUENCY axis Estimated FREQUENCY DFT can estimate Frequncy = 1 Estimated PERIOD 50[sample] Conversion needed to estimate Period "/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH

1SPCMFNPG%'5ʢʣ FREQUENCY axis Regular intervals Low Frequency High Frequency
PERIOD axis NOT regular interval = 1 Resolution of long period signal is LOW Short Period Long Period "/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH 1 2 DFT spectrum

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH 1SPCMFNPG%'5ʢʣ Input signal DFT spectrum (FREQUENCY axis)
DFT spectrum (PERIOD axis) The resolution of low frequency band is low especially when the number of samples is few. 1 2 Non-contact vital sensing Important Resolution of low frequency band

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH 1SPQPTFENFUIPE

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH 'FBUVSFTPGQSPQPTFENFUIPE 1 2 “FVS” Proposed method Resolution
of low frequency band is high compared with DFT Robustness against noise is HIGH compared with ARS

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH #BTJDJEFB 1 5 10 15 20
25 -0.5 0 0.5 1 1.5 2 ･･･ <Period> 5 sample ･･･ Split sample sequence by 5 sample width same as the period Same waveforms Discrete time Sample value

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH #BTJDJEFB 1 5 10 15 20
25 -0.5 0 0.5 1 1.5 2 ･･････ <Period> 5 sample Split sample sequence by 6 sample width not same as the period NOT same waveforms Discrete time Sample value

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH #BTJDJEFB Same waveforms NOT same waveforms
･･･ = + + + ･･･ Accumulation result ▶ ･･･ Form of accumulation result is different

%JGGFSFODFPGEJWJEFETJHOBMBDDVNVMBUJPOCZXJEUIʢʣ Samples on window ( ) Same as
period NOT same as period Periodic signal Samples on window ( ) Accumulation of samples on window ( ) Accumulation of samples on window ( ) Discrete time Sample value "/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH

NOT same as period Same as period %JGGFSFODFPGEJWJEFETJHOBMBDDVNVMBUJPOCZXJEUIʢʣ
Split signals are same waveform <Accumulation result> NOT flat Variance: LARGE Split signals are NOT same waveform <Accumulation result> Flat Variance: SMALL Periodic signal Accumulation of samples on window ( ) Accumulation of samples on window ( ) Discrete time Sample value "/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH

4 7BSJBODFTPG BDDVNVMBUJPOPG TQMJUTJHOBMT 7BSJBODF ･･･ "/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH $POGJHVSBUJPOPGQSPQPTFENFUIPE
"DDVNVMBUJPO × "DDVNVMBUJPO × "DDVNVMBUJPO × 5 7BSJBODF 6 7BSJBODF ･･･ 41$ 41$ 41$ ･･････ Input signal Number of accumulation is different by the number of port of SPC (Width of window) Accumulation result is divided by the number of accumulation 6 6 6 5 5 5 4 4 4 4

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH 3FTVMUPGQSPQPTFENFUIPE 4 ･･････ 7BSJBODF "DDVNVMBUJPO ×
41$ ･･････ 4 ･･･ 3 4 5 6 7 Horizontal axis: Width of window Vertical axis: Variance of accumulation of split signals Maximum variance of accumulation Estimated Period: 5 [Samples]

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH $PNQVUFSTJNVMBUJPO

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH 4JNVMBUJPO 'VOEBNFOUBMXBWFGPSNT Fundemental waveform A Fundemental waveform
B 401 sample 402 sample Discrete time Sample value Discrete time Sample value

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH 4JNVMBUJPO *OQVUTJHOBM + + Noise DFT Proposed
method Period A: 401 B: 402 Number of samples of signal 32000 SNR -10 dB 0 200 400 600 800 1000 1200 -4 -2 0 2 4 401 sample/period A Sample value 0 200 400 600 800 1000 1200 -4 -2 0 2 4 402 sample/period B Discrete time Sample value ･･････

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH 4JNVMBUJPO 3FTVMU DFT Proposed method

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH $PODMVTJPO

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH $PODMVTJPO DFT Proposed method 7BSJBODF "DDVNVMBUJPO ×
41$ ･･････ 4 ･･･ 4 ･･････

END "/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH "QQFOEJY

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH 41$ʢ4FSJBMUP1BSBMMFM$POWFSUFSʣ Series on time axis Pararell on
time axis Hardware implementation of signal splitting ･･･ SPC Save some number of signal and output them at once. “Number of ports” : 3 samples

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH $PNQBSJTPOPGOPJTFSPCVTUBCJMJUZ 4JNVMBUJPOQSPDFEVSFʢʣ Repeat 1000 times 2Execute period
estimation Generate fundemental waveform (Random signal) Make periodic signal Add noise (AWGN) 1Generate signal

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH $PNQBSJTPOPGOPJTFSPCVTUBCJMJUZ 4JNVMBUJPOQSPDFEVSFʢʣ DFT ARS Proposed method Yes
Correct? +1 Yes Correct? +1 Yes Correct? +1 Generated signal 1Generate signal 2Execute period estimation Repeat 1000 times Count number of times when estimated period is correct Decide estimated period by peak of result a b a b

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH $PNQBSJTPOPGOPJTFSPCVTUBCJMJUZ 4JNVMBUJPODPOEJUJPO Number of period Noise Number
of simuration Number of signals Fundemental waveform Random signal 31 samples 30 periods -30 dB 〜 10 dB Peak acquisition range 20 samples 〜 40 samples 1 signal 1000 times

"/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH $PNQBSJTPOPGOPJTFSPCVTUBCJMJUZ 3FTVMU -30 -25 -20 -15 -10
-5 0 5 10 SNR [dB] 0 0.2 0.4 0.6 0.8 1 DFT ARS Strong noise Noise ability of proposed method is highest Weak noise Proposed Method Accuracy

END "/PWFM1FSJPE&TUJNBUJPO.FUIPEGPS1FSJPEJD4JHOBMT4VJUBCMFGPS7JUBM4FOTJOH

A Novel Period Estimation Method for Periodic S...

A Novel Period Estimation Method for Periodic Signals Suitable for Vital Sensing

More Decks by rpaka

Other Decks in Research

Featured

Transcript