Phase-only adaptive filters: theory and algorithms

This paper presents algorithms and techniques for designing adaptive finite impulse response (FIR) filters that maintain nearly lossless constraints on their input-output characteristics. Both single-channel all-pass and multichannel paraunitary adaptive filters are considered. Our methods make use of a simple, iterative procedure for adjusting the coefficients of a single or multichannel FIR filter to impose all-pass or paraunitary filter constraints periodically during coefficient adaptation. The technique is easily extended to multidimensional convolution form and we illustrate such an extension via a simple image sequence encoding application.