On-line approach for fast convolution over sensor networks

Abstract

It is assumed that at some time moment, in wireless sensor networks the new set of current samples of input and system impulse response enter a digital memory replacing the previous samples. It is urgent for each new sample or for a small part of new samples to update a convolution as well. Therefore, a recursive fast convolution algorithm is proposed here to solve a linear filtering problem for a nonstationary system. The calculation operations are reduced because most columns of Fourier code matrices and respective rows of the right-hand side vectors were deleted for equal previous and current samples. An example with the ordinary and modified 8-point Fourier code matrices is presented for a nonstationary linear system. The amounts of operations, necessary for recursive and fast Fourier transform algorithms, are calculated. Results are discussed, and the conclusion is given.