Implementation of compressed sensing in telecardiology sensor networks

Abstract

Mobile solutions for patient cardiac monitoring are viewed with growing interest, and improvements on current implementations are frequently reported, with wireless, and in particular, wearable devices promising to achieve ubiquity. However, due to unavoidable power consumption limitations, the amount of data acquired, processed, and transmitted needs to be diminished, which is counterproductive, regarding the quality of the information produced. Compressed sensing implementation in wireless sensor networks (WSNs) promises to bring gains not only in power savings to the devices, but also with minor impact in signal quality. Several cardiac signals have a sparse representation in some wavelet transformations. The compressed sensing paradigm states that signals can be recovered from a few projections into another basis, incoherent with the first. This paper evaluates the compressed sensing paradigm impact in a cardiac monitoring WSN, discussing the implications in data reliability, energy management, and the improvements accomplished by in-network processing.

Figure 1

Evolution of the BCG (red), ECG (blue), and PPG (green) signals during 10 seconds, with QRS complex and I valley marked.

Figure 2

Original (continuous blue), sampled (continuous red), and reconstructed (dashed black) ECG signal from 64 samples. Depiction of time signal (a) and Daubechies 4 wavelet representation with level 4 of decomposition (b).

Figure 3

Depiction of SNR and M influence in the nRMSD of the ECG waveform reconstruction, for Daubechies 4 wavelet transform and τ _coeff = 0.24.

Figure 4

Dependence on number and position of lost packets, for Daubechies 4, M = 32, and τ _coeff = 0.24, of the nRMSD of a reconstructed ECG waveform.

Figure 5

Dependence on number and position of lost packets, for Daubechies 4, M = 32, and τ _coeff = 0.14, of the nRMSD of a reconstructed BCG waveform.

Figure 6

PPG nRMSD dependence on wavelet transform and number of packets lost, for τ _coeff = 0.14.

Figure 7

Signals of different cardiac arrhythmias produced by a MPS450 simulator, compared with a normal ECG.

Figure 8

Example of reconstructed signals and the respective nRMSD.

Figure 9

Example TSN, with three intermediate nodes between the source and the sink.

Figure 10

Number of transmissions required for a packet to cross a WSN, with variable length and success probability.