Design of high throughput 2D FIR filters using singular value decomposition (SVD) and genetic algorithms

Many researchers have studied the application of SVD in 2D digital filter design. It has been shown that filters designed using this methodology generally have several parallel sections of two cascaded 1D filters in z/sub 1/ and z/sub 2/ variable. Such filters require fewer multipliers than a direct 2D implementation. As well, the parallel structure of the designed filter using SVD lends itself easily to a pipelined, high throughput implementation. The implementation of 1D filters using canonical signed digit coefficients has been shown to decrease calculation time resulting in an increased throughput. We present a method for the design of 2D filters using cascaded 1D filters in a parallel structure. The 1D filters have coefficients in canonical signed digit format to improve further the throughput rate of the designed 2D filters. The 1D design is performed using genetic algorithms.