Abstract: One of the main drawbacks of Orthogonal Frequency Division Multiplexing (OFDM) signals is the large signal envelope fluctuation. Peak to Average Power Ratio (PAPR) is a commonly used metric for quantifying the envelope fluctuations of OFDM signals. However, recent researches have shown that Cubic Metric (CM) is a more accurate metric when it is used to quantify the envelope fluc- tuation. Clipping and filtering technique can be employed to reduce the CM. The filtering operation in traditional clipping and filtering technique cannot lead the signal to the optimal performance. Therefore, an optimal clipping and filtering algorithm for CM reduction is proposed. The key idea is to consider the impact of filtering operation on in-band and out-of-band components of signals and model the filter design as an optimization problem. The problem is solved to obtain an optimal filter, which is combined with clipping to reduce CM efficiently. Due to the high complexity of solving the optimization problem, a deep neural network based realization scheme of the optimal clipping and filter algorithm is further proposed. Simulation results show that both the proposed algorithm and corresponding neural network scheme have close performance, but the latter has much lower complexity. Compared with some existing algorithms, the proposed algorithm and scheme exhibit better performance.

Key words: cubic metric, deep learning, filter design, OFDM, peak to average power ratio, clipping and filtering