Tracking Intermittent Tremor Frequency with a Particle Filter

Our previouswork has demonstrated that the extendedKalman filter (EKF) is a suitable method to track tremor frequencies embedded in spike trains, whose firing rate can be modeled as a sinusoid contaminated with noise. However, when tremor is intermittent, the EKF frequency tracker takes a long time to regain its track of tremor frequencies or never locks on to tremor frequencies even when tremor reappears in spike trains. This is mainly due to the linearization error of nonlinear state space processes. A particle filter (PF) can overcome this issue and track intermittent tremor frequencies more accurately. We applied the EKF and PF on both synthetic and real data to show the superior performance of the PF to that of the EKF.