Role of latent source dependencies in cBSS-based motor neuron identification

Determine the influence of latent dependencies among motor neuron spike trains—arising from delayed sources introduced by the extension step and common synaptic inputs—on the identifiability and performance of convolutive blind source separation (ICA-based) methods for EMG-based motor neuron identification.

Background

Independent component analysis (ICA) approaches to convolutive blind source separation (cBSS) for motor neuron identification typically assume statistical independence of sources, yet motor neuron spike trains exhibit dependencies due to common synaptic inputs and delayed copies introduced in the extension step. Although ICA can be robust to modest dependence and sparsity can help separation, the precise impact of these latent dependencies on source recovery and identifiability in EMG-based settings has not been fully characterized.

Clarifying how such dependencies affect optimization landscapes, projection vectors, and separability would improve theoretical guarantees and guide algorithm design and parameter selection, especially for clinical and non-EMG modalities.