Neural Network Based Clipping and Filtering Scheme for Cubic Metric Reduction in OFDM Systems

Abstract

Orthogonal frequency division multiplexing (OFDM) has been widely used in modern communication systems. One drawback of OFDM signals is that they have large envelope fluctuations, which limits the transmitter power efficiency. Peak-to-average power ratio (PAPR) and cubic metric (CM) are two metrics commonly used to quantify the envelope fluctuations, and the latter is supposed to be more accurate when used to predict the power de-rating of power amplifier. Traditional iterative clipping and filtering (ICF) algorithm is a simple method reducing envelope fluctuations. However, due to the iterations, it is difficult for ICF to satisfy the requirement of real-time communication. As is well known, neural network can be regarded as a universal approximator of functions. By using this property, this paper constructs a simple neural network to approximate the traditional ICF algorithm to reduce the CM. The structure of network as well as the training and testing procedures are elaborated. Simulation results show that, compared with the traditional ICF algorithm, the neural network based scheme can achieve very close performance in both CM reduction and bit error rate (BER) but with much less execution time. Thus, the proposed scheme is promising to satisfy the requirement of real time.