FPGA-based Real-Time Compressed Sensing of Multichannel EEG Signals for Wireless Body Area Networks
Dan Liu,Qisong Wang,Yan Zhang,Xin Liu,Jingyang Lu,Jinwei Sun
DOI: https://doi.org/10.1016/j.bspc.2018.12.019
IF: 5.1
2019-01-01
Biomedical Signal Processing and Control
Abstract:The purpose of this study is to solve the issues in reconstruction computation complexity of the current method in wireless body area network (WBAN) through developing compressed sensing (CS) for multichannel electroencephalogram, performing model optimization, designing a system for compressing and collecting electroencephalogram (EEG) signals, and implementing real time compression and collection of multichannel signals. Firstly, based on the distributed compressed sensing theory, we analyze the sparsity of EEG signal, screen digital sensing matrix models, design multichannel joint reconstruction algorithm, and perform optimization analysis as well as simulation verification at each step. Secondly, based on field programmable gate array (FPGA), we realize real time collection, storage, compression, and transmission of multichannel EEG by setting up a compression and collection system. Lastly, each system function module is inspected, and the performance of the compressed multichannel EEG system is evaluated from the perspective of computation complexity, reconstruction accuracy, instantaneity, etc. Evaluation results show that the improvement of real-time performance is contributed by the application of binary permutation block diagonal matrix (BPBD), which converts CS multiplications into additions with a simple circuit and reduces the computational time drastically. The average signal to noise distortion ratio for signal reconstruction reaches 21.74 dB under the compression ratio of 2, which also meets the requirement of WBAN. The proposed method has faster computation, better accuracy, and simpler coding, can be utilized in a variety of applications related to multichannel EEG, especially in situations where the system power consumption and real-time performance are critical. (C) 2018 Elsevier Ltd. All rights reserved.