A Simulation Study on Performance of Decoders in Chronic Recording
Zijun Wan, Xingchen Ran, Weidong Chen, Shaomin Zhang · 2019
In a chronic recording, the non-stationarity of neural signals is an inevitable problem which is one of the critical problems in neural decoding. In order to explore the effects of three main recording factors including signal noise ratio (SNR) decrease, recording channel loss and variation in neural preferred direction (PD) on the decoding accuracy separately and the adaptation ability of different decoders, a simulation study was applied to generate the neural signals under these conditions. Kalman Filter (KF), Unscented Kalman Filter (UKF) and Recurrent Neural Network (RNN) are used to decode the simulated signals. We found that RNN has the best performance in all conditions and that KF and UKF have no significant difference. Besides, as the changes of conditions accumulate, the performance of all three decoders will gradually decrease.