A normalization scheme to reduce numerical errors in inverse tangent computations on a fixed-point CORDIC processor

An analysis of fixed-point CORDIC (COordinate Rotation DIgital Computer) in the Y-reduction mode, which allows computation of the inverse tangent function, has shown that unnormalized input values can result in large numerical errors. The complexity of a floating-point implementation of the entire system can be avoided by locally normalizing the values before using CORDIC. The authors have implemented the normalization using a method that integrates it with the CORDIC iterations, resulting in an elegant solution to the problem. The method requires only O(n^1.5) extra hardware and does not affect the latency. It is believed that this scheme can be extended to the other modes of CORDIC