Design considerations for massively parallel channel estimation algorithms

Abstract

Accurate channel estimation may require complex algorithms for effective results, especially in the case of a multiuser detector. The introduction of Graphic Processing Units (GPUs) has opened up new possibilities for the implementation of numerically intensive channel estimation algorithms. This paper studies the implementation on GPUs of channel estimation algorithms for channels affected by strong phase noise. While classic Maximum Likelihood estimation is still the most competitive in terms of throughput and memory bandwidth, Steepest Ascent algorithms show the largest speed improvement due to their structure, which is the most suitable for implementation on a parallel processor like the GPU.