L Priya1*, A Kandaswamy2, C D Lakshmi3
*1,3Department of Biomedical Engineering, PSG College of Technology, Coimbatore, Tamil Nadu, India.
2Industry Sponsored Research and Consultancy, PSG College of Technology, Coimbatore, Tamil Nadu, India.
Received 03 February 2015; revised 14 October 2015; accepted 28 June 2016
Direct transmission of Electrocardiogram (ECG) signal through wireless network is susceptible to Inter Symbol Interference (ISI). Transmission of ECG signal over a band limited channel or through a multipath propagation causes ISI.
The proposed work mitigates the Inter symbol Interference using blind equalization algorithm and estimate the transmitted ECG signal without transmitter assistance. This paper discuss the importance of blind equalization for physiological signal transmission over existing wireless networks such as 3G mobile networks and Wireless Body Area Networks (WBAN’s).This method has been validated Swith existing conventional equalization techniques based on computational complexity, correlation coefficient, convergence rate and Mean Square Error (MSE). Results indicated that the performance of blind equalizer same as that of the non blind equalizer, but they do not rely on training signals is the major advantage for deploying this blind receiving scheme in biotelemetry applications.
Keywords: ECG, ISI, MSE, Correlation Coefficient, Convergence Rate, Computational Complexity, Biotelemetry, WBAN’s.
Introduction
Biotelemetry is used to remotely track the vital signs like heart rate, respiration rate, body temperature, blood pressure and muscle movement of ambulatory patients through wireless networks such as Global system for mobile communication (GSM), Wireless Body Area Network (WBAN), General Packet Radio Service(GPRS), Third Generation Universal Mobile Telecommunications System (3G UMTS) and Bluetooth/Zigbee1. Wireless body area network consist of independent nodes that continuously monitor human’s physiological signals and activities. These real time vital signals are transmitted to the nearby body node coordinator (BNC) (e.g. smart phone) via Zigbee / Bluetooth and then to a remote heath care station via the internet2. One of the major challenge in the development of body area network is that the role of environment fading characteristics. The nature of local environment is responsible for diffraction, reflection and scattering effect of the transmitted wave with varying amplitude and phase resulting in transmission errors. The transmitted symbol overlaps with the adjacent symbols and tend to introduce ISI in the transmitted ECG
signal3. The motivation behind the proposed work is to achieve the error free ECG transmission in human body communication channel. In this work, an efficient blind equalization strategy that equalizes the received signal and endeavors to recover the information from the channel output.
Proposed system model
This section discusses the transmitted signal (Normal ECG), human body communication channel and blind equalization algorithm. Figure.1 shows the ECG transmitter and receiver baseband communication system. Electrocardiogram is the electrical manifestation of the contractile activity of the heart, and can be recorded fairly easily with surface electrodes on the limbs or chest. These signals are transmitted wirelessly for the detection of cardiovascular disease for early diagnosis and treatment. In this work, ECG test data from record number 16265 of the MIT-BIH Normal sinus rhythm database has been used. This Discrete ECG signal is downloaded from the Physiobank ATM of one minute duration9. It has sampled at a rate of 128 Hz. In this work, we are using 1200 samples for processing. The sampling interval is 0.0078 s. The maximum amplitude of an ECG signal ranges from -1.5 mV to 2.5 mV. This work is also suitable to patient with malignant ventricular arrhythmia.
——————
*Author for correspondence E-mail: lpm@bme.psgtech.ac.in
On-body fading channel
In typical radio channels, multipath propagation occurs because of many reflections of the transmitted signal. The various propagation paths are characterized by different delays, and this leads to a time dispersive behavior of the channel and induces the Intersymbol Interference. The measures that have been taken depend on the data rate to be transmitted or, equivalently, on the bandwidth processed by the transmission system.
The root mean square (rms) delay spread is a measure of multipath spread within the on–body channel, an important parameter for characterizing time dispersion or frequency selectivity, and provides a figure of merit for estimating data rates for multipath channels 4. The rms delay spread τrms versus antenna separation along the arm, leg, back, torso and whole body is reported in3.The values of rms delay spread increase with antenna separation. The rms delay spread is modeled as follows
τrms(d) = C(eDd − 1) [ns], for d ≤ dbp
= E + F ln dbp[ns], for d > dbp … (1) dbp is the breakpoint and the distance d between transmitter and receiver in cm and C [ns], D [1/cm], E [ns], and F [ns] the parameters of the model.
A measure of the data rate that can be supported over the channel without additional receiver techniques is determined by the root mean square multipath delay spreads. The maximum value of delay spread as 17ns which increase the bit error rate in WBAN’s.
Therefore the blind equalizers are used to achieve the bit error rate greater than 10-3. Rayleigh distributions results for the on-body communication channel. The measurements have taken along the front, side, and back of the body. The large number of multipath effects contributes to the attenuation of the signal including diffraction, reflection, energy absorption, antenna losses, etc. along the front of the body. This Rayleigh distribution is observed when several paths of equal amplitude and random phase combine at the receiver. In, 6 indicates the Rayleigh distribution provides the best fit to model the on-body fading channel. This paper proposes the novelty in wireless body area networks as to analyze Rayleigh fading model at 2.45GHz band using IEEE 802.15.4 standard.
In this work, proposed model was simulated with the measurement data reported in3. The sequence of ECG data with T symbol period is transmitted through the on-body channel H. At the receiver end, FIR filter is applied to equalize the received signal which is corrupted by Intersymbol Interference and Additive White Gaussian Noise (AWGN). The received signal is given by
0 0 1 0 0
1 1 1
2 2 2
0 1
x k h h s k v k
x k s k v k
x k s k v k
xLk h h sLk vLk
é ù é ù é ù é ù
ê ú ê ú ê ú ê ú
ê ú ê ú ê ú + ê ú
ê ú ê ú ê ú ê ú
ê ú ê ú ê ú ê ú
ë û ë û ë û ë û
… (2)
The matrix H denotes the time decimated channel convolution of the on-body fading channel model5.
Fig.1—ECG transmitter and receiver block diagram of communication System
modulus criterion in blind equalization technique is easy to implement , which is very independent of the initial value of the equalizer tap vector .The received signal weight vector (yk) is constructed from the Intersymbol Interference corrupted ECG signal(xk).
This method updates the weight (Wk+1) at each iteration based on gradient estimation of input │yk│2.
The step size µ is obtained from the eigen value λ. The CMA algorithm is stable if and only if 1/λmax >
µ>0. Then the equalizer filter output signal (yk) is
J(w)=E{│1-yk
2 │} … (3)
The perfect (zero forcing) equalization of the received signal is obtained by minimizing the CM cost function. When the J(w) converges to minima , equalizer with CMA algorithm is very efficient in removing the effect of ISI and estimate the source signal.
Results and Discussion
Figure.3 (a) shows the source ECG signal which is downloaded from the Physionet Normal sinus rhythm database (nsrdb/16265). This Lead 1 ECG signal is recorded for 4sec duration and its amplitude ranges from -1.5 mV to 2.5 mV. The sampling rate is 128Hz and sampling interval is 0.0078125 s. The gain of the
Fig.2—Flowchart for Blind Adaptive Equalization Technique
Fig.3—(a) Source ECG Signal (b) Channel Output Signal (c) Blind Equalizer Output Signal
(a)
(b)
(c)
amplifier is 200. After amplification, the ECG signal amplitude ranges from -150mv to 500mV.The simulation is averaged over 1280 samples. The time dispersive on-body fading channel model was simulated according to measurement data reported
3.This time dispersive channel obviously gives rise to ISI in the transmitted ECG signal which is shown in Figure.3 (b). Figure.3 (c) shows the Blind equa lizer output in time domain. This figure shows that the equalizer output ECG signal with respective samples is identical with that of Source ECG signal. The proposed method is used to minimize the transmission error induced by the channel better than that of zero forcing equalization method. Because zero forcing equalization technique needs pilot training signal to track the channel variation. So using blind system the continuous monitoring of patient’s ECG signal will be possible at the remote health care station. Table.1 shows the comparison performance of CMA algorithm with other non blind linear equalization algorithms such as Least Mean Square (LMS) and Recursive Least square algorithm (RLS). The Signal to Noise ratio (SNR) is calculated using the formula
│S/N│=│mean of filter output signal │/ │standard deviation of filter output signal │. SNR of filter output is greater than channel output, which indicates that the equalizer results less Intersymbol interference. Table.1 shows that the LMS algorithm used for the channel equalizer has better signal to noise ratio compared with CMA and RLS algorithms, but it requires prior knowledge about the source signal. From Table.2, the LMS is found to be the most computationally efficient among the algorithms.
LMS algorithm’s multiply operation is proportional to 2N+1, RLS algorithm’s is proportional to 2.5N2+4.5N and CMA’s is proportional to 2M(3M + 4N + 1) −2.LMS algorithm can be implemented in a practical system without squaring or averaging and is elegant in its simplicity and efficiency10. We noticed that CMA has low MSE compared to the LMS and RLS. Therefore the blind technique outperforms the non blind technique in terms of convergence rate. The cross correlation coefficient value is close to one which gives the perfect match between filter output and input signal. When the cross correlation coefficient value is less than one, it indicates filter output matches less with the input signal. The cross correlation coefficient varied between zeros to one indicates that the filter output is matched with that of input signal with respect to different percentages. Hence it is observed that the least mean square produces 73% similarity between input and filter output, recursive least square produces 68%, constant modulus algorithm produces 39%. Therefore the LMS gives better correlation than other algorithms, but it requires the training data.
Conclusion
Results indicated that, with exception of computational complexity, the blind technique using CMA receivers performs almost as well as non blind/trained method using LMS algorithm. Cross correlation results show that the blind filter output signal is highly correlated with original ECG signal which means that the transmitted ECG signal is
Table 1—Comparison of Signal to noise ratio of CMA with conventional equalization Algorithms Signal under
consideration
Constant Modulus Algorithm Conventional Linear equalization Algorithms
Least Mean Square Algorithm Recursive Least Square Algorithm Mean
(mV)
Standard Deviation
(mV)
Signal to Noise
ratio
Mean (mV)
Standard Deviation
(mV)
Signal to Noise Ratio
Mean (mV)
Standard Deviation (mV)
Signal to Noise
ratio
Channel output signal 13.7 54.6 0.25 13.7 54.6 0.25 13.7 54.6 0.25
Filter output signal 5.2x1013 2.4x1014 0.21 12.8 44 0.29 6.8 54.6 0.12
Table 2—Performance metrics of CMA and conventional equalization algorithms
Parameters Constant Modulus Algorithm Conventional Linear equalization Algorithms
Least Mean Square Algorithm Recursive Least Square Algorithm Computational complexity Multiply Operations : 10246 Multiply Operations : 2561 Multiply Operations : 4101760
Mean square error -5x1015 -1x1015 -2x1015
Auto Correlation Coefficient 1 1 1
Cross Correlation Coefficient 0.59 0.73 0.68
References
1 Jasemian Y, Design and implementation of a wireless telemedicine system applying Bluetooth technology and cellular communication network: new approach for real time remote patient monitoring, PhD Thesis, Aalborg University, Denmark, (2005).
2 Ullah S, Higgins H & Braem B, A Comprehensive Survey of Wireless Body Area Networks On PHY, MAC, and Network Layers Solutions, J Med Syst, 36 (2012) 1065-1094.
3 Simon C L & William S G, An Experimental Investigation into the Influence of User State and Environment on Fading Characteristics in Wireless Body Area Networks at 2.45 GHz, IEEE Trans on Wirless Commn, 8 (2009) 6-12.
6 Jaffery Z, Adaptive Equalization of Digital Communication Channel Using Feed–Forward Neural Network, Jour of Commn and Comp , 8 (2011) 404-409.
7 Mahmoud S S, Fang Q, & Hussain Z M, A blind equalization algorithm for biological signal transmission, Else, DSP, 22 (2012) 114-123.
8 Baweja R, Sharma A & Shukla P, Simulation of Constant Modulus Algorithm Equalizer for Human Body Communication Channel, Special Issue of Int Jour of Comp and Commn Engg , 6 (2011) 0975-8887.
9 Physiobank, Physionet, Physiologic signal archives for biomedical research, http://www.physionet.org/physiobank/, viewed Jul (2013).
10 Rupp M, Convergence Properties of Adaptive Equalizer Algorithms, IEEE Trans on Sig Proc, 59 (2011) 2562-2574.
