Session S33.3

Recovering of ECG Missing Samples in Wireless Transmission

A Prieto-Guerrero, C Mailhes*, F Castanié

ENSEEIHT
Toulouse, France

Considering the emergence of telemedicine applications, different links such as fixed access network (PSTN), mobile access network (GSM/GPRS and future UMTS) or satellite interfacing (DVB-RCS technology) are involved in e-health applications. These are liable to induce errors and/or missing packets on the received data. These errors/losses can occur anytime and anywhere (according to the channel availability, memory overflows, protocols, etc) during a transmission process. Therefore the recovering of missing samples for biomedical signals is of great interest.
This paper proposes a reconstruction method which is a combination of a left-sided and right-sided autoregressive (AR) model, and the well-known Gerchberg-Papoulis (GP) method for band-limited signals. The latter involves iterations between time and frequency domains: gaps in the signal are filled by zeros and then transformed by FFT. The spectrum is then limited in band and the inverse FFT is applied to obtain the reconstructed signal. The proposed interpolation algorithm uses the samples before and after the sequence of missing samples as follows: first, a forward and a backward AR models are estimated using the known samples before and after the gap. Second, the linear prediction error (LPE) is obtained in both cases. Then, the time-reversed version of the LPE is filtered to generate a model for the gap in both ways, forward and backward. Third, a cross-fading window is applied to each obtained model. Finally, this estimate will replace the zeroes in the original GP method.
The proposed method was tested in the MIT Arrhythmia Database. 3700 gaps of length going up to 30 samples were generated in a random way over each ECG record. An objective measure based on a local SNR (Signal to Noise reconstruction Ratio) was used to assess the method performance. Results were plotted (length of the gap vs. SNR in dB) to compare three methods: the method based only on the left-sided and right-sided AR model, the classic GP method and the modified method (GP with the AR initialization). Results show that the AR initialization on the Gerchberg-Papoulis method gives the best results in all cases. Indeed, this implementation gives a gain from 2 to 8 dB (depending on the gap length) over the original method. Results show that this interpolation method represents a really suitable technique to ECG signal reconstruction in a possible corrupted transmission.

(Abstract Control Number: 274)