Wiener filter (linear FIR decoder)

The Wiener filter is the classical least-mean-square linear estimator for stationary signals; in BCI it appears as the tapped-delay-line linear decoder that predicts a kinematic variable from a window of recent binned firing rates across channels. In effect it is regularized linear regression with time lags: the output at time t is a weighted sum of neural activity at t, t−1, …, t−k, with weights solving the Wiener–Hopf normal equations (the OLE generalized over a temporal window).

Its strengths are simplicity, closed-form fitting, and a memory of recent activity that captures conduction and processing delays without an explicit dynamics model. Its weaknesses are exactly the absence of that model: it has no principled uncertainty, treats the signal as stationary, and tends to produce jittery output unless heavily smoothed or regularized (e.g. ridge). Historically the Wiener/linear decoder was the workhorse of early intracortical BMI reaching studies and remains a strong, honest baseline; the velocity Kalman filter largely displaced it in cursor control by adding an explicit smoothness prior.