Algorithm and hardware considerations for real-time neural signal on-implant processing.

Objective.Various on-workstation neural-spike-based brain machine interface (BMI) systems have reached the point of in-human trials, but on-node and on-implant BMI systems are still under exploration. Such systems are constrained by the area and battery. Researchers should consider the algorithm complexity, available resources, power budgets, CMOS technologies, and the choice of platforms when designing BMI systems. However, the effect of these factors is currently still unclear.Approaches.Here we have proposed a novel real-time 128 channel spike detection algorithm and optimised it on microcontroller (MCU) and field programmable gate array (FPGA) platforms towards consuming minimal power and memory/resources. It is presented as a use case to explore the different considerations in system design.Main results.The proposed spike detection algorithm achieved over 97% sensitivity and a smaller than 3% false detection rate. The MCU implementation occupies less than 3 KB RAM and consumes 31.5 µW ch-1. The FPGA platform only occupies 299 logic cells and 3 KB RAM for 128 channels and consumes 0.04 µW ch-1.Significance.On the spike detection algorithm front, we have eliminated the processing bottleneck by reducing the dynamic power consumption to lower than the hardware static power, without sacrificing detection performance. More importantly, we have explored the considerations in algorithm and hardware design with respect to scalability, portability, and costs. These findings can facilitate and guide the future development of real-time on-implant neural signal processing platforms.

Development of an Ultra Low-Cost SSVEP-based BCI Device for Real-Time On-Device Decoding.

This study details the development of a novel, approx. £20 electroencephalogram (EEG)-based brain-computer interface (BCI) intended to offer a financially and operationally accessible device that can be deployed on a mass scale to facilitate education and public engagement in the domain of EEG sensing and neurotechnologies. Real-time decoding of steady-state visual evoked potentials (SSVEPs) is achieved using variations of the widely-used canonical correlation analysis (CCA) algorithm: multi-set CCA and generalised CCA. All BCI functionality is executed on board an inexpensive ESP32 microcontroller. SSVEP decoding accuracy of 95.56 ± 3.74% with an ITR of 102 bits/min was achieved with modest calibration.

The significance of neural inter-frequency power correlations.

It is of great interest in neuroscience to determine what frequency bands in the brain have covarying power. This would help us robustly identify the frequency signatures of neural processes. However to date, to the best of the author's knowledge, a comprehensive statistical approach to this question that accounts for intra-frequency autocorrelation, frequency-domain oversampling, and multiple testing under dependency has not been undertaken. As such, this work presents a novel statistical significance test for correlated power across frequency bands for a broad class of non-stationary time series. It is validated on synthetic data. It is then used to test all of the inter-frequency power correlations between 0.2 and 8500 Hz in continuous intracortical extracellular neural recordings in Macaque M1, using a very large, publicly available dataset. The recordings were Current Source Density referenced and were recorded with a Utah array. The results support previous results in the literature that show that neural processes in M1 have power signatures across a very broad range of frequency bands. In particular, the power in LFP frequency bands as low as 20 Hz was found to almost always be statistically significantly correlated to the power in kHz frequency ranges. It is proposed that this test can also be used to discover the superimposed frequency domain signatures of all the neural processes in a neural signal, allowing us to identify every interesting neural frequency band.

Predicting Single-Unit Activity from Local Field Potentials with LSTMs.

This paper investigates to what extent Long Short-Term Memory (LSTM) decoders can use Local Field Potentials (LFPs) to predict Single-Unit Activity (SUA) in Macaque Primary Motor cortex. The motivation is to determine to what degree the LFP signal can be used as a proxy for SUA, for both neuroscience and Brain-Computer Interface (BCI) applications. Firstly, the results suggest that the prediction quality varies significantly by implant location or animal. However, within each implant location / animal, the prediction quality seems to be correlated with the amount of power in certain LFP frequency bands (0-10, 10-20 and 40-50Hz, standardised LFPs). Secondly, the results suggest that bipolar LFPs are more informative as to SUA than unipolar LFPs. This suggests common mode rejection aids in the elimination of non-local neural information. Thirdly, the best individual bipolar LFPs generally perform better than when using all available unipolar LFPs. This suggests that LFP channel selection may be a simple but effective means of lossy data compression in Wireless Intracortical LFP-based BCIs. Overall, LFPs were moderately predictive of SUA, and improvements can likely be made.

Lossless Compression of Intracortical Extracellular Neural Recordings using Non-Adaptive Huffman Encoding.

This paper investigates the effectiveness of four Huffman-based compression schemes for different intracortical neural signals and sample resolutions. The motivation is to find effective lossless, low-complexity data compression schemes for Wireless Intracortical Brain-Machine Interfaces (WI-BMI). The considered schemes include pre-trained Lone 1st and 2nd order encoding [1], pre-trained Delta encoding, and pre-trained Linear Neural Network Time (LNNT) encoding [2]. Maximum codeword-length limited versions are also considered to protect against overfit to training data. The considered signals are the Extracellular Action Potential signal, the Entire Spiking Activity signal, and the Local Field Potential signal. Sample resolutions of 5 to 13 bits are considered. The result show that overfit-protection dramatically improves compression, especially at higher sample resolutions. Across signals, 2nd order encoding generally performed best at lower sample resolutions, and 1st order, Delta and LNNT encoding performed best at higher sample resolutions. The proposed methods should generalise to other remote sensing applications where the distribution of the sensed data can be estimated a priori.