Controls

Step size mu

Filter length N

Reference delay

Noise level

Samples per frame

How it works

Adaptive noise canceller (Widrow-Hoff LMS). The primary input d = s + h*x contains the wanted signal s plus noise passed through an unknown path h; the reference x is correlated with the noise.

y(n) = w . x(n-D..n-D-N+1)
e(n) = d(n) - y(n)   (estimate of s)
w <- w + mu e(n) x_vec
stable for mu < 2 / (N * power(x))

Reduction = 10 log10(P_d / P_e). Reference delay D misaligns the reference and limits cancellation; too large mu diverges.

Noise reduction-

Residual power-

Weight error-

Noise type-

Noise Cancellation: 2D LMS Adaptive Filter

Adaptive noise canceller using the LMS algorithm with live waveforms, weights and learning curve.