Title: Leveraging Interference-Based Wave Modeling for Continuous Temporal Representations of Event-Based Signals

Abstract:

Conventional discrete or strictly real-valued models often struggle to capture the asynchronous, highly structured activation patterns inherent in spatio-temporal signals generated by event-driven biological processes, such as surface electromyography (sEMG). To address this limitation, we introduce a continuous temporal modeling framework grounded in interference-based wave representations. This methodology transforms event-like inputs into a complex-valued latent wave field, encoding temporal dynamics through phase modulation and interactions among latent components.

By projecting this wave field into an energy domain, the model generates structured activation patterns that effectively capture both temporal localization and relational dependencies within finite observation windows. Notably, this approach achieves these results without depending on explicit recurrence or causal state propagation. The formulation is specifically tailored for event-driven biosignals, as continuous representations facilitate robust feature extraction and efficient gradient-based optimization.

The method is primarily designed to facilitate learning from sEMG data for downstream control applications in biomechanical systems, including prosthetics and exoskeletons. Our experimental findings indicate that the proposed interference-based wave model yields superior representation quality relative to purely real-valued alternatives, all while preserving the computational efficiency necessary for practical, real-world deployment.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC