Optimal Prototype Filters for Near-Perfect-Reconstruction Cosine-Modulated Filter Banks.

In this paper, we propose a simple method for designing M-channel near-prefect-reconstruction (NPR) cosine-modulated filter banks (FBs). By employing the Parks-McClellan algorithm in constructing the prototype filter, an ideal magnitude response of the filter is achieved. Furthermore, the transition band of the prototype filter is constrained in such a way that it follows a cosine function. As a result, the FBs are approximately power complementary and therefore possess the NPR property. There are two main advantages in our proposed method. The first one is that no objective function in the optimization is explicitly required and the second one is that the resulting prototype filter is a global optimal solution. Compared with the traditional design method using general optimization methodology, the proposed method here is very simple and efficient. In addition, the stopband attenuation of the resulting FB is significantly higher than those offered by the traditional methods. [ABSTRACT FROM AUTHOR]