应用于脑机接口系统的动态稀疏矩阵压缩算法

Dynamic Sparse Matrix Compression Algorithms for Brain-Computer Interface Systems

摘要:

在脑机接口系统中，高通道数神经信号采集是一个核心功能模块。在高通道数神经信号采集中，因其原始数据量巨大，直接传输和处理产生的原始数据会消耗极大的功耗并增加硬件设计上的难度。为解决这个问题，一个有效的方法是在数据传输和处理前对原始数据进行压缩。神经元动作电位信号具有不应期性的特点，文中利用此特点，将多通道神经信号的数字标记输出在一定时间范围内定义为一个稀疏矩阵，并对此稀疏矩阵进行特征提取，根据其特征动态地采用优化算法进行数据压缩。文中的算法在使用FPGA作为中控硬件的32通道神经采集硬件系统上通过实时验证，实验证明文中提出的动态稀疏矩阵压缩算法可实现83.4%的数据压缩率。

Abstract:

: In multi-channel neural signal acquisition systems, direct transmission and processing of raw data can be extremely power intensive and make hardware design more difficult due to the sheer volume of raw data. An effective solution to this problem is to compress the raw data prior to transmission and processing. The neuronal action potential signal has the characteristic of having a refractory period. In this paper, the digital marker output of the multichannel neural signal is defined as a sparse matrix in a certain time range, and the features of this sparse matrix are extracted, and the data is compressed dynamically using an optimized algorithm according to its features. The algorithm in this paper was verified in real time on a 32-channel neural acquisition hardware system using an FPGA as the central control hardware, and it was experimentally demonstrated that the dynamic sparse matrix compression algorithm proposed in this paper can achieve 83.4% data compression rate.