A More Rational and Efficient Kalman Filter Design for Motor Brain-Machine Interfaces

The Kalman Filter has long been one of the most widely used models in motor brain-machine interface (BMI) research due to its noise handling capabilities and real-time adaptability. However, as a model originally developed for traditional control systems, its underlying assumptions of Markov property and the designs of observation models may not always hold true in the context of BMI applications, potentially leading to oversimplifications. This paper examines the limitations that arise when applying the Kalman Filter to BMI, and proposes the Dilated Kalman Filter, which performs Gaussian multiplication between state transition distribution and observation-mapped state distribution in state space, thereby combining observation noise with BMI-specific observation model noise, and consequently incorporates historical information from both states and observations. The proposed method improves the accuracy of Kalman Filter while significantly enhancing computational efficiency, particularly when processing data from large numbers of neurons.