Parallel independent FFT implementation on intel processors and Xeon phi for LTE and OFDM systems

Abstract

Fast Fourier Transform (FFT) is a key element for wireless applications based on the OFDM (Orthogonal Frequency Division Multiplexing) and challenging for implementing on processor multi-cores/many-cores. As an example, the Long Term Evolution (LTE) protocol establishes a requirement for processing, whereby many independent FFTs must be calculated within a limited time slot. By using Intel Math Kernel Library (MKL), in our approach to Xeon phi, we managed to reduce the maximum execution time of many independent FFTs. We proposed an implementation on processors multi-cores/many-cores using OpenMP (Open Multi-processing) reducing the mean time latency to 124 µs on native mode after 1300 µs with the offload. This is a challenge for shared memory projects. This paper describes how this level of performance can be obtained with multi-core Intel i7, Xeon processors and a many-core Xeon Phi. The best results were obtained with the Xeon Phi, which outperformed the Xeon Sandy-Bridge.