Effective data reduction for wireless transmission of neural activity (EDnA)

• Applicant: Professor Dr.-Ing. Maurits Ortmanns
• Subject Area: Electronic Semiconductors, Components and Circuits, Integrated Systems, Sensor Technology, Theoretical Electrical Engineering
• Term: from 2013 to 2018
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 230027621

• Final Report Year: 2019

Final Report Abstract

Multichannel neural recorders aim to record brain activity from an ever increasing number of channels and thereby generate a amount of data, which - for patient safety and avoidance of discomfort - has to be transmitted wirelessly to a basestation. Typical resolutions for neural signal recordings range in literature from 12 to 16 bits and sampling frequency of 20KHz. Thus, tens to hundreds of Mbytes/s need to be transmitted wirelessly, while the wireless transceiver may only consume mW of power - obviously a contradiction. Also storage of long-term measurement of such experiments is challenging, since within hours, Terabyte of data needs to be stored. The goal of this project was to investigate on digital and mixed signal techniques in order to reduce the amount of recorded data before wireless transmission; this data reduction should be done without loosing the essential information of the neural signals. For this, neural spike compression was investigated based on delta compression, compressed sensing as well as based on a mixed-signal frequency separation. It was found that many published results only perform as good as claimed on low-noise synthesized neural data or rely on predetecting the temporal position of spikes. In this project, we have developped a framework that allows a direct comparison of the information loss by quantization and compression by comparing the spike sorting results of the original and the compressed signals. We have investigated the effectiveness of compressed sensing and found it insufficient for highnoise, in-vivo raw data. Delta compression, both temporal as well as spatial, is easy to implement and performs a signficant lossless data reduction; it was found feasible for on-chip implantable implementation. A boost in compression rates, but also in implementation complexity, was found with neural networks, which achieve high lossless compression rates; they were evaluated more suitable for rack-based or not fully implanted systems in order to reduce e.g. data storage.

Publications

• “A neural recorder IC with HV input multiplexer for voltage and current stimulation with 18V compliance”. In: ESSCIRC 2014 - 40th European Solid State Circuits Conference (ESSCIRC). Sept. 2014, pp. 103–106
  U. Bihr et al.
  
• “Delta compression in time-multiplexed multichannel neural recorders”. In: 2016 12th Conference on Ph.D. Research in Microelectronics and Electronics (PRIME). June 2016, pp. 1–4
  M. Pagin et al.
  
• “A neural data lossless compression scheme based on spatial and temporal prediction”. In: 2017 IEEE Biomedical Circuits and Systems Conference (BioCAS). Oct. 2017, pp. 1–4
  M. Pagin and M. Ortmanns
  
• “Evaluation of logarithmic vs. linear ADCs for neural signal acquisition and reconstruction”. In: 2017 39th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). July 2017, pp. 4387–4390
  M. Pagin and M. Ortmanns
  
• “Evaluation of spike sorting and compression for digitally reconfigurable non-uniform quantization”. In: 2017 15th IEEE International New Circuits and Systems Conference (NEWCAS). June 2017, pp. 177–180
  M. Pagin, J. Becker, and M. Ortmanns
  
• “Study of Compressed Sensing and Predictor Techniques for the Compression of Neural Signals under the Influence of Noise”. In: 2018 40th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). July 2018, pp. 1102–1105
  M. Pagin and M. Ortmanns