基于深度学习的多模OFDM索引调制检测器

本文提出了一种基于深度学习（Deep Learning， DL）的多模正交频分复用索引调制（Multi-Mode Orthogonal Frequency Division Multiplexing with Index Modulation， MM-OFDM-IM）检测器。在该检测器中包括两个子卷积神经网络（Sub-Convolutional Neural Network， SCNN）并行对MM-OFDM-IM信号的索引位和载波位进行检测，接收符号在经过迫零（Zero Force， ZF）均衡后再预处理生成二维矩阵，同时输入到子卷积网络中学习信号的内在特征。经过离线训练，该检测器可以实现MM-OFDM-IM符号的在线检测。仿真结果表明，该检测器在瑞利衰落信道条件下能以较低的计算复杂度获得近似最大似然（Maximum Likelihood， ML）检测性能。通过对已训练后的模型进行剪枝操作，能在保证检测误码率（Bit Error Rate， BER）的前提下大幅度减少模型的参数量，达到了性能与计算复杂度的有效平衡。 

Deep Learning Based Detector for Multi-Mode OFDM Index Modulation

? ?In this paper， a Deep Learning （DL） based Multi-Mode Orthogonal Frequency Division Multiplexing with Index Modulation （MM-OFDM-IM） detector is proposed. In this detector， two Sub-Convolutional Neural Network （SCNN） detects the index and carrier bits of MM-OFDM-IM signal in parallel， and the received symbols are preprocessed after Zero Force （ZF） equalization to generate a two-dimensional matrix， which is fed to the SCNN in parallel to learn the signal's intrinsic features. After offline training， the detector can achieve online detection of MM-OFDM-IM symbols. Simulation results show that the detector achieves approximate Maximum Likelihood （ML） detection performance with low computational complexity under Rayleigh fading channel conditions. By pruning the trained model， the number of parameters of the model can be significantly reduced while the Bit Error Rate （BER） is guaranteed， achieving an effective balance between performance and computational complexity.