Real-Time Data Compression of Neural Spikes

This paper presents an analysis of the effectiveness of delta compression (DC) with additional entropy encoding, referred to as modified delta compression (MDC), as real-time compression scheme for neural spikes. Neural data compression which does not result in major distortion of the data is an attractive option to lower the transmitter power consumption. Since a lossy compression will result in a potentially undesirable loss of information the above mentioned compression scheme is analyzed regarding its ability to preserve the spike sorting relevant information content of the neural signals. Our analysis shows that MDC with thresholding is suitable for the compression of low-noise synthetic neural spike signals, but that its performance significantly degrades in the presence of higher noise levels or with recorded neural signals. In our simulations, the lossless MDC scheme achieves a mean compression rate of 2 without signal distortion. An example circuit realization for the compression of 100 channels synthesized in a 180nm CMOS technology occupies a chip area of 0.72 mm(2) and consumes 0.97mW of power. Based on these results, it was found that the MDC scheme is capable of lowering the overall power consumption when the utilized wireless transmitters consume more than 121 pJ/bit, which applies to most state of the art transmitter implementations.