Iterative constrained least squares design of near perfect reconstruction pseudo QMF banks

This paper presents a simple and efficient method to design M-band near perfect reconstruction (NPR) pseudo quadrature mirror filter (QMF) banks. This method does not rely on a conventional computationally intensive nonlinear optimization method but rather on an iterative constrained least squares algorithm. The algorithm is rapidly converging, simple to implement and flexible. Its convergence does not depend on the starting point. Moreover an iteratively calculated weighting function can be used to shape the stopband of the prototype filter and perform the minimax or the gain constrained least squares approximation. Design examples and a MATLAB program implementing the proposed algorithm are included.