Fast convolution filter banks for satellite payloads with on-board processing

Filter banks able to demultiplex channels of different sizes are necessary on board the satellite when a single bit-rate channel is assigned to each user. In this paper, the method referred to as the fast convolution filter bank is analyzed for regenerative payloads. The size of the input fast Fourier transform (FFT) depends on the narrowest channel, while the inverse fast Fourier transform (IFFT) number of points is relative to the decimation ratio associated with each channel. In order to reduce the complexity, most of the frequency coefficients are forced to zero, leading to a violation of the overlap-and-save (OS) condition. It is shown that the resulting error can be modeled as a cyclostationary process, and an optimization method based on minimization of the associated mean-square error (MSE) is proposed. Fast convolution is then compared with multistage implementation in terms of flexibility, reconfigurability, and complexity.