Signal processing techniques to reduce energy consumption in EEG acquisition and transmission for WBAN

Abstract

Recently there has been a lot of interest in health tele-monitoring. Some vital heath indicators like EEC, ECG, MEG, Pulse Plethysmograph etc. are collected by sensor nodes; the acquired signal is transmitted to a remote healthcare facility for monitoring and analysis. In effect, this comprises a Wireless Body Area Network (WBAN). Such a system allows for health monitoring of the elderly in western countries without the intruding presence of a paramedic. In India, such systems would bene t healthcare penetration into remote rural areas. In a WBAN, the main challenge is to preserve energy. There are three main energy sinks namely : sensing, processing and transmission. The transmission energy is the highest, therefore the priority is to reduce it. This requires signal compression; however since the computational capacity at the sensor nodes is limited; this precludes application of sophisticated transform coding techniques. Recent studies proposed an alternate Compressed Sensing (CS) based approach wherein the sampled signal is projected onto a lower dimension by a random matrix, thereby effecting compression. The compressed signal is transmitted to the base station, where it is recovered by sophisticated CS recovery algorithms. Since computational power at the base station is not at a premium, this is a plausible model. In this work, we are particularly interested in EEG. However the contribution of this work extends to any biomedical signal in general. EEG signals are acquired via multiple probes. All the probes monitor the same brain activity; therefore the EEG signals from the multiple probes/channels are correlated. We reviewed prior studies in this area and observed that all these studies concentrated on piecemeal recovery of the compressed EEG signal - they did not exploit the inter-channel correlations. For the first time, we exploited inter-channel correlations for EEG reconstructions. We proposed three methods - group-sparse recovery; Kronecker CS recovery and low-rank matrix recovery. The fundamental idea in all of the three remain the same, these are three different approaches to model the inter-channel correlations. We were able to show that the recovery results improved considerably compared to previous CS based methods. The second contribution of this work is even more significant. We showed that by exploiting the inter-channel correlations we can actually reduce sensing energy and eliminate the requirement for processing / compression altogether. This simply requires randomly under-sampling the EEG signals. Such an operation reduces sensing cost and as the signal is compressed during acquisition, it eliminates processing costs. We found that our recovery techniques works well in this framework and yields good recovery, i.e. accuracy from random under-sampling is the same as accuracy from full sampling followed by compression. The final contribution of this work is towards the signal processing community. We have derived algorithms for solving group-sparse recovery problems, low-rank matrix recovery problems and rank-aware sparse / group-sparse recovery problems. These algorithms are derived on the Split Bregman approach. These general purpose algorithms will find other areas of signal processing like multi-spectral imaging, X-Ray CT imaging and Magnetic Resonance Imaging to name a few.