Data Compression Techniques for Brain Machine Interfaces

In last decades development of Brain Machine Interfaces (BMIs) became more and more widespread. The technology allows recording of electrical activity in the brain, to further understand its complex dynamics. But not just for pure research, BMIs offer a bridge between thoughts and actions, robotic prosthesis activated by “thoughts” have already been tested on human patients.

BMI scheme
Figure 1 - BMI scheme showing an implanted side for data recording and an external unit for data processing. Ideally the link between internal and external unit would be wireless.

BMIs need to have a high spatial resolution and be able to detect a large number of neurons, this is one of the greatest challenges for data transmission in the implant. A ≥100 channel recorder can easily generate tens of Mbit/s of data rate.

The focus of this project is to provide data reduction techniques in order to transmit neural data wirelessly through the skull and skin. Since there are many different ways of elaborating neural signals we aim to provide lossless or almost lossless compression.  To reduce data rate both spatial and temporal redundancy of neural signals are exploited. A manifold of techniques is being investigated to find the best candidate for an implanted system. This includes delta compression with entropy coding, compressed sensing and non-uniform quantization.

Example of delta compression scheme
Figure 2 - Example of delta compression scheme that exploits temporal redundancy to achieve compression (Courtesy of: U. Bihr et al., NEWCAS 2014)
Example of delta compression scheme
Figure 3 - Example of delta compression scheme that exploits spatial redundancy to achieve compression

Project member

M.Sc. Matteo Pagin