HarmonICA: Neural non-stationarity correction and source separation for motor neuron interfaces

TL;DR

HarmonICA tackles non-stationarity in neural signal source separation for motor neuron interfaces by combining a quasi-linear ICA framework with a neural compensation module that learns time-varying mixing $H(t)$ and preserves identifiability via alternating linear and nonlinear updates. It introduces a two-term loss scheme for independence and non-stationarity, a compensation network with sinusoidal temporal encoding, and a practical training regimen that maintains identifiability while adapting to nonstationary latents. The method is validated on real intramuscular EMG with synthetic drift, simulated non-isometric surface EMG, and real surface EMG during dynamic contractions, showing improved unit recovery and the ability to perform direct source separation without windowing. This work advances robust motor neuron interfaces by enabling accurate, nonstationarity resilient decomposition across modalities, with potential impact on prosthetics and wearable neural interfaces.

Abstract

A major outstanding problem when interfacing with spinal motor neurons is how to accurately compensate for non-stationary effects in the signal during source separation routines, particularly when they cannot be estimated in advance. This forces current systems to instead use undifferentiated bulk signal, which limits the potential degrees of freedom for control. In this study we propose a potential solution, using an unsupervised learning algorithm to blindly correct for the effects of latent processes which drive the signal non-stationarities. We implement this methodology within the theoretical framework of a quasilinear version of independent component analysis (ICA). The proposed design, HarmonICA, sidesteps the identifiability problems of nonlinear ICA, allowing for equivalent predictability to linear ICA whilst retaining the ability to learn complex nonlinear relationships between non-stationary latents and their effects on the signal. We test HarmonICA on both invasive and non-invasive recordings both simulated and real, demonstrating an ability to blindly compensate for the non-stationary effects specific to each, and thus to significantly enhance the quality of a source separation routine.

Figures (1)

• Figure 1: HarmonICA in operation. a After an initial pretraining phase, the linear separation vector is converged on a single source. However it lacks the capacity to account for the non-stationary latents in its source prediction. b Using alternating backpropagation, HarmonICA cycles between improving the separation vector and training a neural network to adapt the vector such that it accounts for non-stationarities at each time point. c At the end of training, the source prediction is no longer modulated by the non-stationary latents.