A Novel Nonuniform Filter Bank Design Method

Filter banks are widely used in signal processing, especially if both time and frequency domain processing is required. Filter banks can be reviewed for various features, two important ones are the ability for perfect reconstruction and the uniformity of the widths of the frequency bands. Nonuniform filter banks are especially important, because of several practical applications, such as audio signal processing. However, most design techniques impose serious constraints on the specification or rely on computationally expensive algorithms to ensure the perfect reconstruction property. At the cost of introducing errors into the reconstruction, some of the constraints can be lifted. Nevertheless, the majority of non-perfect reconstruction design techniques are still reliant on complex optimization problems. The paper introduces a novel nonuniform filter bank design method with minimal constraints and low computational demand. The resulting filter banks are built by finite impulse response filters. The calculation of the filter coefficients can be implemented easily and it is based on the fast Fourier transform so it can be carried out even in real time.