An Efficient Algorithm for Remote Detection of Simulated Heart Rate Using Ultra-Wide Band Signals

Transcription

1 American Journal of Biomedical Engineering 213, 3(6): DOI: /.abe An Efficien Algorihm for Remoe Deecion of Simulaed Amad Hashemi 1, Alireza Ahmadian 1,*, Mehran Baboli 2 1 Deparmen of Biomedical Engineering and Medical Physics & Research Cenre for Biomedical Technology and Roboics, RCBTR, Tehran Universiy of Medical Sciences, Tehran, Iran 2 Deparmen of Elecrical Engineering, Universiy of Hawaii a Manoa, Honolulu Absrac Ulra-wideband (UWB) signals have become aracive because of heir paricular advanage of having a narrow pulse widh, which makes hem suiable for non-invasive remoe sensing of vial signals. In his paper, we presen an efficien algorihm based on power specral mehod applied on UWB signals for non-invasive monioring and measuremen of simulaed hear raes. The obecive of his sudy is o evaluae pracical algorihms for wireless deecion of he human hear rae using UWB signals in a noisy environmen. To perform his, he hear rae is firs simulaed by our designed moving phanom wih variable speed and range of moion. Then, hese daa were enered in he proposed simulaion framework including a new mulilayer UWB channe l o resemble he human body o deec accurae hear rae using an UWB ransceiver wih 3.2GHz of bandwidh in a noisy environmen. We evaluae and compare he moion rae deecion echniques based on variance, Fourier ransform, wavele ransform, and power specrum densiy, PSD. Experimenal resuls show ha our approach based on PSD is well adaped o deec he simulaions by he moion phanom. The resuls prove an accuracy of 98% achieved by PSD mehod for deecion of various periodic movemens of he moion phanom. The resul shows ha his algorihm is capable of being used in real ime applicaions and is promising for daily clinical use. Keywords Ulra-wideband, Wireless Deecion, Virual Moion Phanom, Power Specrum Densiy, Fourier Transform, Wavele Transform 1. Inroducion Monioring of paien s hear and respiraion raes is frequenly performed in clinical reamen programs. Mos of he commonly used monioring ools in clinical applicaions need o be in direc ouch wih he paien s body ha obviaes paien compliance. Recenly, new echnologies have been developed o remoely monior hese vial signals wih wireless ransmission by employing an ulra-wideband (UWB) echnique as a promising candidae[1-3]. In February 22, Federal Communicaions Commission (FCC) legalized he use of UWB by releasing a se of specral masks[4, 5]ince hen, UWB signals have found widespread use in medical applicaions, e.g., in he early sage deecion of breas cancer[6-8], and in wireless racking sysems[9, 1]. Because of high spaial resoluion in comparison wih he hear and respiraion frequency, as well as he high daa rae and resisance o amming, UWB can be used for deecion of hear and ches caviy movemens. However, hese signals have low power and hey are considerably aenuaed in he propagaion environmen. * Corresponding auhor: Ahmadian@sina.ums.ac.ir (Alireza Ahmadian) Published online a hp://ournal.sapub.org/abe Copyrigh 213 Scienific & Academic Publishing. All Righs Reserved Therefore, hey need robus and sophisicaed algorihms for deecion[3, 11, 12]. The deecion of hear and respiraory raes employing UWB signals has been an acive research area[13, 14]. A UWB-simulaed framework based on a layered-based channel model was proposed o evaluae he effecs of simulaion and hear moion parameers such as he hickness of channel layers on he accuracy of hear rae measuremens. Based on simulaed sysem proposed by he auhors of his paper[15], a MATLAB framework presened which simulaes an environmen whereby he UWB signals can be observed and evaluaed a differen locaions of he simulaed channel for wireless measuremen of hear rae. Main pars of he simulaed sysem include UWB signal generaor, ransmier and receiver, as well as a new channel which models differen layers of human body. The proposed sysem enables us o evaluae he effec of differen parameers, such as hickness of channel layers and hear moion parameers on he accuracy of hear rae measuremen. One of he main feaures of ha work is employing a five layers channel model wih arbirary aenuaion parameers defined by user for simulaion of human body. This muli-layer channel model was used o simulae and evaluae he effecs of UWB wave propagaion hrough he human body[15, 16]. The proposed mahemaical framework of he UWB channel model has

2 2 Amad Hashemi e al.: An Efficien Algorihm for Remoe Deecion of Simulaed been adoped in his paper. A deecion algorihm followed by real experimens was proposed o remove respiraion from he daa and o calculae he hear rae[17]. The main disadvanage of his algorihm was is sensiiviy o he noise. Respiraion was deeced in he real daa using an algorihm based on coninuous wavele ransform[14]elecion of he moher wavele in his algorihm was based on he shape of he ransmied pulse; hence, i has limiaions in applicaion o he daa acquired by anoher UWB ransceiver. Anoher wavele based algorihm using full wavele packes wih slow calculaions was previously repored and esed on he real daa[16]. A saisical algorihm o deec respiraory and hear raes has been repored previously[18, 19]. The main conribuion of hese sudies was providing an analyical framework for he developmen of signal processing algorihms o esimae respiraion and hear raes and corroboraion of he echniques wih measuremens. In his paper we have exended our previous works o become more precise and accurae in deecing he arge s hear rae in an environmen conaining oher obecs wihou any wave absorber. Furhermore, a simulaed sysem is proposed o compare he simulaion resuls and hose of real experimens. 2. Maerials and Mehods 2.1imulaion Framework In his secion, a simulaion framework o simulae he UWB sysem and o deec he hear rae is presened. In his simulaed sysem he ransmier sends a Gaussian monocycle pulse wih a bandwidh of 3.2GHz (Fig. 1). Anennas are exacly locaed in fron of he arge s hear o achieve minimum wave aenuaion[16]. mulipah propagaion are no considered. This model is valid only for shor ranges of wave propagaion. Aenuaion depends on he frequency of he signal which is assumed o be 3GHz[16]. Air (A) Fa (F) Muscle Carilage (M) (C) Figure 2. Channel model Lung (L) Hear (H) Based on he condiions proposed by Baboli[16], ransmied and received average power densiies can be wrien as: Transmied power: S av = η 1 2η 2 2. e 2α 2z 2 i η 2 η 2 +η av (1) 1 Received power: r S av = η 2 η 1 2. e 2α 2z 2 i η 2 +η av (2) 1 where z is he layer hickness, and α and η are he aenuaion consan and inrinsic impedance, respecively, as calculaed in[16], Figure 3chemaic of he muli-layer J +1 Figure 3 shows he schemaic of a muli-layer model o ake ino accoun he aenuaion effecs of differen issues and propagaion media on UWB signal. According o such a model when differen propagaion layers are consecuive, he wave ha passes hrough he firs inerface (S av,1 ) is he inciden wave for he second inerface (S av,2 ). Hence he ransmied power for each layer can be wrien as follows[2]: S av,1 = η 1 2η 2 2. e 2α 2z 2 i η 2 η 2 +η av,1 (3) 1 S av,2 = η 2 2η 3 2. e 2α 3z 3 η 3 η 3 +η av,1 (4) 2 Figure 1. Gaussian monocycle pulse, Time domain and Frequency domain The propagaion channel only aenuaes he signals and obviaed o consider he effecs of scaering and mulipah propagaion. The model of UWB channel is shown in Fig. 2. This UWB channel consiss of five layers and he hickness of each layer can be changed. Transmission and he reflecion coefficiens for differen layers can be calculaed and o be used for he daa processing and hear rae deecion. Here i is assumed ha waves ravel only perpendicular o he planar inerface beween wo differen layers. Therefore, effecs of mulipah fading as a resul of scaering and S av, S av,3 = η 3 2η 4 2. e 2α 4z 4 η 4 η 4 +η av,2 (5) 3... = η 2η e 2α +1z +1 η +1 η +1 +η av, 1 (6) Consequenly, he oal aenuaion of he signal afer ransmiing hrough layer is calculaed using he nex equaion: a 1 = η k 2η k e 2α k+1z k +1 k=1 (7) η k +1 η k +1 +η k Le us assume ha he wave is refleced from he (+1)h inerface. We have:

3 American Journal of Biomedical Engineering 213, 3(6): a 2 = η +1 η η +1 +η (8) Then, he wave reurns from he same pah o he receiver. Aenuaion in he reurning pah is as follows: a 3 = η k+1 2. e 2α k z k k=1 (9) η k η k +1 +η k Finally, he oal aenuaion is calculaed from (5) as follows: a = a 1 a 2 a 3 (1) For simpliciy, we can consider all of hose calculaions in decibels: 2η k a 1 db = η k 2η k+1 2. e 2α k +1z k +1 k=1 η k+1 η k +1 +η db (11) k a 3 db = η k e 2α k z k k=1 η k+1 +η db (12) k η k 2η k a db = a 1 db + a 2 db + a 3 db (13) 2.2. Experimens Experimens are designed o verify he deecion algorihm and he simulaion resulseveral experimens are performed, and he goal of each sep was o develop and complee he simulaion framework and improve he deecion algorihm for deecing he hear beas. Anennas are locaed near he ches caviy and boh he human arge and radars are covered wih wave absorbing maerials o resric unwaned inerferences. Under such a siuaion, he propagaion channel only conains he human body. Physical geomery of he propagaion environmen (e.g., disance beween radars and he ches caviy), ransmi pulses and relaed parameers (e.g., ransmission frequency), as well as he expeced hear rae are similarly chosen in he simulaion (Fig. 4). The experimens are performed using he Time Domain PulsON P22 Evaluaion Ki. Is specificaion and he parameers of he ransceiver are shown in Table 1. Table 1pecificaion and he parameers of ransceiver PRF(Pulse Repeiion Frequency) Cener frequency Band widh Power consumpion Raw daa rae 9.6 MHz 4.7 GHz 3.2GHz 5.7 W 9.6 Mbps For full conrol of arge frequency, a virual horax moion phanom wih he capabiliy of seing he frequency and movemen range was buil and used in he experimen. The frequency and range of moion were conrolled using an ATMEGA32 microconroller. For seing he iniial posiion of he movemen, he sysem uses an infra-red sensor. The phanom is shown in Fig. 5. Figure 5. A virual horax moion phanom Figure 4. Measuremen seup for conrolled environmen 2.3. Mehods The hear layer, which is he las layer in his model, is considered as he only moveable layer. Here i is assumed ha hear has linear one-dimensional movemen which leads o generaes linear Doppler. In his sudy we evaluae echniques based on variance, Fourier ransform, wavele ransform and power specral densiy, PSD for phanom moion rae deecion. In all echniques body movemen cancellaion and removing background cluer is he same. The five seps of algorihm o deec hear rae are described in he following as shown in Fig 6: (1) Each received waveform is recorded in one row of a marix called he received marix, R. (2) For removing he body movemen, he cross correlaion beween each row of marix R, (each received waveform) and he firs row, which is considered he base waveform in he ime domain, is calculaed. Then, each row is circularly shifed o he poin where he amoun of is

4 22 Amad Hashemi e al.: An Efficien Algorihm for Remoe Deecion of Simulaed corresponding cross correlaion is maximized. (3) The moion filer is applied on he marix. This filer calculaes he average of each column and subracs i from each sample of ha column (7), R m (i. ) = R(i, ) 1 N R(i, ) N I=1 (14) where R is he received marix and N is he oal number of rows of marix R. (4) Afer applying he moion filer, he saic background is removed and only he dynamic par of he channel daa remains. This par consiss of daa relaed o he hear bea of he arge. The goal of his sep is o find a column from marix R m, in which he hear moion daa is appeared. I is shown in[14] ha he oal energy of he samples for each column is maximized in a column, which conains moion daa. Figure 6. Block diagram of he proposed algorihm Various mehods o evaluae (a) Variance mehod: To perform his, he variance of all columns was calculaed, and he column wih maximum variance is seleced as he reference line called p. (b) Fourier ransform mehod: Discree Fourier Transform (DFT) of marix R was calculaed and recorded in marix RR FFFFFF. In he frequency domain, for finding he column ha has he maximum energy, he proporion of peaks RR FFFFFF (ii, ) 2 o he oal energy of signals in each column was calculaed by (15) and recorded in array E ff, EE ffffff () = max (RR FFFFFF (ii, ) 2 ) (15) ii=ll II= RR FFFFFF (ii, ) 2 Where is he column index, i is he row index, and L is he number of rows in marix RR FFFFFF. The column, in which E has he maximum value, is he arge poin and is called p. The frequency of he peak in he frequency specrum of column p is he frequency of moion. (c) Discree wavele ransform: The procedure of muli-resoluion decomposiion of a signal R[n] is implemened using a complemenary low-pass/high-pass filering followed by down-sampling by 2. Each sep of his scheme consiss of wo digial filers and wo down samplers by 2. The oupus of he high-pass filers are deails and hose of he low-pass filers are approximaions. The wavele ransform, which is used in his paper for respiraion rae calculaions, is as follows: + TT mm,nn = RR() 1 ψ(aa mm nnbb )dddd (16) mm aa 2 where ψ is he moher wavele, n is he ranslaion, m is he dilaion, a is he dilaion sep parameer, and b is he locaion parameer. The energy of he wavele is as follows: EE mm = nn= (TT mm,nn ) 2 (17) In his sudy he Daubechies 8 wavele filer was chosen as he moher wavele. Wavele ransform was applied on all of he columns of marix R. Then, he energy of all of he frequency inervals is calculaed and heir maximum is saved in an array, E wavele. The column, in which E wavele has is maximum value, presens he locaion of he moion ha is occurred. (d) Power specrum densiy (PSD): The main sep in he deecion algorihm is o find he column ha conains moion daa. In real condiions, applying he variance mehod does no give a correc answer when he experimen is being repeaed. The reason is ha he received waveform consiss of many mulipahs from differen obecs, which are correlaed o each oher. Moreover, he noise affeced daa more han he siuaion when he wave absorber was used. These problems yields maximizing of he variance a false poins; hence, we have improved he signal processing algorihm in[18, 19] and proposed an algorihm based on he disribued energy in he frequency domain and specral analysis of he daa. The deails of he algorihm is described as follows: Afer applicaion of he moion filer, PSD is applied on each column and he frequency specrum of each column is recorded in marix R PSD. This feaure compues he PSD of each column hrough he Fourier ransform of he auocorrelaion of each column. PSD illusraes how he power of a signal is disribued over he differen frequencies. The specral represenaion is a very useful device for describing saionary random processes in he frequency domain. The formula is as follows: RR PPPPPP = FFFFFF{RR xxxx nn } ff=mmmm (18) 2MM mm where RR xxxx nn is he auocorrelaion funcion of he nn h column. NN

5 American Journal of Biomedical Engineering 213, 3(6): For finding he phanom moion, he column ha has he maximum energy has o be deeced. The proporion of he PSD signal in each column o he oal PSD of signals was calculaed using (19) and recorded in array E psd, EE pppppp () = max (RR PPPPPP (ii, ) 2 ) (19) ii=ll II= RR PPPPPP (ii, ) 2 where,i, and L are he column index, row index, and number of rows in marix R PSD, respecively. The column, in which E psd has he maximum value, is he arge poin and is called p. By applying Fourier ransform on he p column, he frequency of moion was deeced. The complee flowchar of he algorihm is shown in Fig Resuls In order o evaluae he robusness of he proposed algorihm o deec he hear rae, experimens were performed wih six differen subecs in differen propagaion and noisy environmens. The resuls were obained in an environmen ha conains oher obecs, which ac as reflecors. Daa had pulse periods equal o 1 and 2 millisecond. Resuls for he case of pulse periods equal o 1 and 2 ms are shown in Figs. 7 and 8, respecively. The moion rae of phanom is deeced wih a maximum error of 2.2%. The resuls associaed wih oher subecs are shown in Table 2. Subec Table 2pecificaion and he parameers of ransceiver Pulse period Moion Rae PSD Mehod Variance Mehod FFT Mehod Wavele Mehod To evaluae our algorihm in a noisy environmen, we added whie Gaussian noise o he signal. The phanom moion rae was undeecable wih SNR=1 and less. Our algorihm based on he PSD algorihm in a noisy environmen wih SNR=15 has he same resul as ha obained in a clean environmen. Figures 9 and 1 show noisy signals. 4 Hear Rae Deecion Based on PSD Signal Ampliude Time(S) Magniude of FFT Frequency (Hz) Figure 7imulaion experimenal resuls of conrolled condiions in he frequency and ime domain wih a pulse period of 1 ms

6 24 Amad Hashemi e al.: An Efficien Algorihm for Remoe Deecion of Simulaed 4 Hear Rae Deecion based on PSD Signal Ampliude Time(S) 8 Magniude of FFT Frequency (Hz) Figure 8imulaion experimenal resuls of conrolled condiions in he frequency and ime domain wih a pulse period of 2 ms 1 x 14 Hear Rae Deecion based on PSD in Noisy Enviromen Signal Ampliude Time(S) 6 Magniude of FFT Frequency (Hz) Figure 9imulaion experimenal resuls of conrolled condiions in he frequency and ime domain wih a pulse period of 1 ms in a noisy environmen (SNR=15)

7 American Journal of Biomedical Engineering 213, 3(6): Hear Rae Deecion based on PSD in Noisy Enviromen Signal Ampliude Time(S) 1 Magniude of FFT Frequency (Hz) Figure 1imulaion experimenal resuls of conrolled condiions in he frequency and ime domain wih a pulse period of 2 ms in a noisy environmen (SNR=15) 4. Conclusions In his paper we have proposed a UWB-based monioring echnique for non-invasive deecion of moion phanom raes for subecs locaed wihin a few meers of he radar. A flexible UWB framework is presened, which simulaes an environmen in which UWB signals can be observed and evaluaed for wireless measuremen of moion phanom raes and diagnoses of hear arrhyhmias. Error Moion Rae Deecion Error per Subec Variance Mehod FFT Mehod Wavele Mehod PSD Mehod Figure 11. Error calculaed in variance, FFT, wavele, and PSD algorihms in six subecs

8 26 Amad Hashemi e al.: An Efficien Algorihm for Remoe Deecion of Simulaed In his sudy, some well-known mehods based on variance, Fourier ransform, wavele ransform and PSD for deecion of phanom moion rae were applied. In all echniques body movemen cancellaion, removing background cluer was same. By employing cross correlaion beween each row of daa marix R and he firs row of he marix R, body movemen was calculaed, and hen circularly shifed o he poin where he amoun of is corresponding cross correlaion is maximized, all daa were synchronized. Background cluer was omied by subracing he average of each column from all samples in ha column. In he proposed algorihm, afer body movemen cancellaion and background cluer removal, an algorihm based on PSD in frequency domain for hear rae deecion is applied. The resuls of PSD algorihm in his sep proved an accuracy of 97.8% in simulaed daa. Figure 11 shows error in variance, FFT, wavele, and PSD mehods. Error averages obained in hese algorihms are 6.35%, 3.46%, 4.4%, and 2.2%, respecively. This resul shows ha a moion phanom algorihm based on PSD performs he bes resuls. This framework can be used as a basis o design/evaluae pracical UWB sysems wih proper parameers for moion phanom rae deecion. This provides us wih he expeced resuls of employing such a sysem in real siuaions. The new efficien algorihm proposed here can be used o accuraely esimae he moion phanom rae in a noisy environmen. The presened framework is poenially applicable in remoe deecion of abnormaliies in respiraory and hear raes. ACKNOWLEDGEMENTS This research was suppored by Tehran Universiy of Medical Sciences and Insiue for Advanced Medical Technologies (IAMT), Research Cenre for Biomedical Technology and Roboics. REFERENCES [1] E,Pancera. X, Li. M, Jalilvand. T, Zwick. W, Wiesbeck. UWB medical diagnosic: in-body ransmission modeling and applicaions. in Anennas and Propagaion (EUCAP), Proceedings of he 5h European Conference on. IEEE [2] A, Taparugssanagorn. A, Rabbachin. J, Salorana. J, Iinai, A review of channel modelling for wireless body area nework in wireless medical communicaions. 28. [3] G, Varoo and E.Maderini, On he UWB medical radars working principles. Inernaional Journal of Ulra Wideband Communicaions and Sysems, 2(2): p , 211. [4] G, Breed. A summary of FCC rules for ulra wideband communicaions. High Frequency Elecronics. 4(1): p [5] S, Hanna. Regulaions and sandards for wireless medical applicaions. in Proceedings of he 3rd inernaional symposium on medical informaion and communicaion echnology. 29. [6] A.M, Abbosh,., M.E. Bialkowski, and S. Crozier. Invesigaions ino opimum characerisics for he coupling medium in UWB breas cancer imaging sysems. in Anennas and Propagaion Sociey Inernaional Symposium, IEEE. 28. [7] M, Klemm. I,Craddock. J, Leenderz. A, Preece. R, Benamin. Experimenal and clinical resuls of breas cancer deecion using UWB microwave radar. in Anennas and Propagaion Sociey Inernaional Symposium, 28. [8] M, Guardiola, Capdevila.S,Blanch. J, Romeu L, Jofre. UWB high-conras robus omographic imaging for medical applicaions. IEEE Inernaional Conference in Elecromagneics in Advanced Applicaions, 29. [9] W, Zhiguo. Xi, L and F. Yuanchun. Moving arge posiion wih hrough-wall radar. IEEE Inernaional Conference in Radar, 26. [1] S. Ergu, R. R. Rao, O. Dural, and Zahinoglu, "Localizaion via TDOA in a UWB Sensor Nework using Neural Neworks," IEEE Inernaional Conference in Communicai ons, pp , 28. [11] M, Jalilvand, Li, X. T, Zwick,W, Wiesbeck. E Pancera. Hemorrhagic sroke deecion via UWB medical imaging. IEEE Proceedings of he 5h European Conference in Anennas and Propagaion, 211. [12] G.A,Zio, E.Maderini, and S. Pisa, A win spiral planar anenna for UWB medical radars. Inernaional Journal of Anennas and Propagaion, 213. [13] G. Ossberger, T. Buchegger, Echimback, Aelzer, and R. Weigel, "Non-invasive respiraory movemen deecion and monioring of hidden humans using ulra wideband pulse radar," Inernaional Workshop in Ulra Wideband Sysems, pp , 24. [14] S. Venkaesh, C. R. Anderson, N. V. Rivera, and R. M. Buehrer, "Implemenaion and analysis of respiraion-rae esimaion using impulse-based UWB," IEEE Conference in Miliary Communicaions, MILCOM, pp [15] M. Baboli, Aharafi, A. Ahmadian, and S. KarimiFard, "A framework for simulaion of UWB sysem for hear rae deecion," Inernaional Conference in Biomedical and Pharmaceuical Engineering, pp [16] M. Baboli, S. A. Ghorashi, Naniei, and A. Ahmadian, "A new wavele based algorihm for esimaing respiraory moion rae using UWB radar". Inernaional Conference in Biomedical and Pharmaceuical Engineering, pp [17] S. N. Pavlov and S. Vamkov, "Algorihm of signal processing in ulra-wideband radar designed for remoe measuring parameers of paien's cardiac aciviy," Second Inernaional Workshop in Ulrawideband and Ulrashor Impulse Signals, pp , 24. [18] M. Baboli, Aharafi, A. Ahmadian, and M. Nambakhsh, "An accurae and robus algorihm for deecion of hear and respiraion raes using an impulse based UWB signal," Inernaional Conference in Biomedical and Pharmaceuical Engineering, pp [19] Sharafi, A., M. Baboli, and M. Eshghi. A new algorihm for

9 American Journal of Biomedical Engineering 213, 3(6): deecion moion rae based on energy in frequency domain using UWB signals. Inernaional Conference in Bioinformaics and Biomedical Engineering (icbbe), 21. [2] C, Bilich, UWB radars for Bio-Medical Sensing: Aenuaion Model for Wave Propagaion in he body a 4GHz. 26.