This paper presents a novel two-phase design method for nearly perfect reconstruction
cosine-modulated filter banks, aimed at reducing filter order while preserving signal
fidelity. Traditional approaches increase the prototype filter order to minimize distortion,
which in turn increases complexity due to more adders and multipliers. The proposed
method overcomes this trade-off by first optimizing filter banks with a small number of
channels, and then extending the design to a larger number of channels using a
Farrow structure. The initial coefficients of the Farrow structure are derived from the
prototype filter designed for the small number of channels, and then optimized for filter
banks with a large number of channels. This structured approach s particularly suitable
for offline application, also enabling an efficient scaling while maintaining low amplitude
and aliasing distortion. Implemented in MATLAB with minimal user input, the method
provides improvements over the existing techniques in terms of filter order and
reconstruction quality.