A new method for designing farrow filters based on cosine basis neural network

In this paper, a cosine basis neural network was proposed to design Farrow filters. Traditionally, Farrow filters are designed in a least-square sense by formulating an error function which reflects the difference between the desired variable bandwidth filter and the practical filter. The filter coefficients are obtained by solving linear equations. Consequently, complex matrix inversion is inevitable and it leads to high complexity when the order of the matrix is high. This problem is solved by the proposed simple and effective method based on cosine basis neural network which convert the problem of coefficient solving into weights training problem.