Two-dimensional transform domain adaptive filtering

A two-dimensional (2-D) orthogonal transform is incorporated into a 2-D FIR adaptive filter to improve the input autocorrelation matrix eigenvalue spread, thereby achieving improved convergence rates for 2-D adaptive filters operating in colored noise. Eigenvalues analyses are used to predict the relative merits of different transforms when operating on different colored noise inputs. Both theoretical predictions and experimental results are presented to demonstrate that the reduction in eigenvalue spread results in greatly improved convergence rates in 2-D adaptive filters, which suffer from inherently slow convergence due to the large number of coefficients required in two dimensions.