This technology is a configurable, energy-efficient Kalman filtering platform for real-time motion prediction in brain-computer interfaces.
Current brain-computer interfaces (BCI) rely on computational methods that prioritize decoding accuracy and often assume access to abundant power, memory, and delay-tolerant hardware. As neural interfaces scale to higher electrode counts and larger neural datasets, these assumptions break down, limiting real-time performance, mobility, and long-term implantability. Existing signal estimation methods are not well-suited to the constraints of embedded BCI systems, where power consumption, latency, and adaptability must be balanced. Addressing these limitations is essential to enable practical, real-world BCIs that can operate continuously and reliably outside of research environments.
This technology enables real-time neural decoding by implementing state-based interference methods within a configurable embedded architecture designed to operate under strict power and latency constraints. Neural signals are processed sequentially using predictive models that combine incoming measurements with prior system states to continuously estimate control or motion intent. The system leverages spatiotemporal correlations in neural data to reduce computational load while maintaining decoding performance, and supports runtime adjustments of accuracy, latency, and energy trade-offs for various BCI applications.
This technology has been validated through FPGA-based prototyping, demonstrating improved energy efficiency and real-time performance on high-dimensional neural datasets.
Patent Pending
IR CU25167
Licensing Contact: Greg Maskel
