Compression of Spike Data Using the Self-Organizing Map

Motivated by current attempts to use wireless in brain-machine interfaces (BMIs), this paper presents a method for the compression of spike data. Supported by vector quantization (VQ) theory, we use a 1-dimensional self-organizing map (SOM) to quantize vectors of input samples. The indices are entropy coded to further reduce the necessary bandwidth, taking advantage of the non-uniform frequency of firing of the SOM processing elements (PEs). The complexity of the use of the SOM is also considered and addressed. After training several SOMs, the method was simulated with real data achieving compression ratios as high as 185.7:1, i.e. a bitrate of 862 bits-per-second-per-channel, assuming sampling at 20 kHz with 8 bits-per-sample (bps)