OFDM Receiver Design With Learning-Driven Automatic Modulation Recognition

The orthogonal frequency-division multiplexing (OFDM) is widely used in modern radio communications because of its efficient spectrum utilization. As we know, the adaptive modulation can efficiently improve the spectrum utilization of OFDM systems, in comparison with the non-adaptive modulation. For this, we design a learning-driven automatic modulation recognition (AMR) receiver for the OFDM system in this paper. The AMR receiver is composed of the robust channel estimation neural network (RCE-NET), the modulation recognition neural network (MR-NET) and the decoding network (DE-NET). Specifically, the RCE-NET estimates a channel state information by the least squares estimation, and the MR-NET realizes the modulation recognition based on the constellation diagram of the equalized signals. After that, the corresponding DE-NET is used for decoding according to the recognition result of the MR-NET. In order to verify the robustness of the designed receiver, we conduct the simulations in different channel models. Simulation results demonstrate the efficiency of our proposed AMR receiver on the channel estimation, modulation recognition and decoding.

I. Introduction

The orthogonal frequency-division multiplexing (OFDM) is one of the most widely used multicarrier modulation and demodulation systems because of its high spectrum utilization. However, the transmission signal is severely subject to the channel fading, leading to the low quality of service (QoS) [1]. In order to reduce the influence of channel fading on the signals, the channel estimation and equalization techniques are essential to compensate the interfered signals at the OFDM receiver [2], [3], [4], [5]. It implies that the design of OFDM receiver largely depends on the channel estimation and equalization modules. For this, a large number of scholars are devoted to the design of reliable channel estimation and equalization methods for improving the performance of OFDM receivers [6], [7].