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 forBrain-computer Interface Systems

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,能够为外部计算机设备采集大量人脑中的神经信息;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依据原始神经信号数据的特点对其进行压缩;神经元动作电位信号具有不应期性即有效信号的时域宽度与信号重复周期之比很小;利用此特点,能够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;所提出的算法在Xilinx平台使用FPGA进行设计与实现,并且将其作为中控硬件在32通道神经信号采集硬件系统上通过实时验证,实验证明提出的动态稀疏矩阵压缩算法可实现83.4%的数据压缩率。

In brain-computer interface systems,multi-channel neural signal acquisition is a core functional module,which can collect much neural information in human brain for external computer equipment.In the multi-channel neural signal acquisition,because of the huge amount of original data,the generated original data is directly transferred and processed to take huge power consumption and increase the difficulty of hardware design.To solve this problem,an effective method is to compress the original neural signal data according to its characteristics before the data transmission and processing.The neuron action potential signal has the refractory period property,that is,the effective signal has the very low ratio of time domain width to repetition period.This paper adopts this feature to define the digital label output of the multi-channel neural signals as a sparse matrix in a certain time range,extracts the feature of this sparse matrix,and dynamically uses the optimization algorithm to compress the data according to its feature.The proposed algorithm is designed and implemented on a Xilinx platform using FPGA,and taking it as the central control hardware to pass the real-time verification on the 32 channels neural signal acquisition hardware system.Experiments show that the proposed dynamic sparse matrix compression algorithm can achieve a data compression rate of 83.4%.