Improved Deep Neural Network for OFDM Signal Recognition Using Hybrid Grey Wolf Optimization

Yan Zhang,Dejun Liu,Yixuan Xian,Jialin Liu,Xu Wang

doi:10.1109/access.2020.3010589

Abstract

Orthogonal Frequency Division Multiplexing (OFDM), as the core technology in mobile communications, is a multi-carrier modulation technology with high frequency spectrum utilization, which has strong anti-multipath interference and anti-fading ability. The significant advantage of OFDM signals lies in the anti-multipath effect, so its application environment is mostly multipath fading channels. Therefore, it is of great significance to study the identification of OFDM signals in multipath channels. Deep learning, with superior big data processing and classification capabilities, is a potential solution to these problems. Based on the problem of OFDM signal recognition in complex signals in multi-path channel, an OFDM signal recognition method based on hybrid grey wolf optimization algorithm to optimize deep neural network model is proposed. Because the basic grey wolf optimization algorithm (GWO) is easy to fall into a stasis state when attacking prey, differential evolution algorithm (differential evolution algorithm) is integrated into GWO to force GWO to jump out of the stasis state with its strong search ability. The convergence speed and recognition performance of the proposed algorithm are greatly improved. The experimental results show that under the condition of low SNR, the recognition accuracy of proposed algorithm is 9.95% higher than the traditional DNN method, and nearly 4.5% higher than the other two intelligent optimization methods, and the values of Precision and Recall increase obviously, which indicates that the hybrid algorithm not only improves the accuracy of recognition, but also makes the search more complete and accurate. Compared with classical particle swarm optimization (PSO) and whale algorithm optimization algorithm (WOA), the hybrid algorithm has strong competitiveness both in recognition performance and optimization stability, which provides a new, simpler and more effective method for modulation recognition of OFDM signals in wireless communications.

Highlights

OFDM (Orthogonal Frequency Division Multiplexing), a multi-carrier (MC) digital modulation technology of great spectrum efficiency and anti-multipath ability, is widely used
In all the experiments in this paper, the feature vector is marked with six labels, these labels correspond to the following modulation modes: OFDM8-1, OFDM16-2, OFDM32-3, OFDM64-4, WPM signal contains {WPM16 WPM32, WPM64}, marked as WPM-5, single-carrier (SC) signal including {PSK4 PSK8, QAM16, QAM32, FSK2, FSK4}, tag to SC-6, so as to various actual output model
In this paper, an OFDM signal recognition method based on hybrid grey wolf optimization algorithm to optimize deep neural network model is proposed, because the basic grey wolf optimization (GWO) is easy to fall into stagnation when it carries out the operation of attacking prey, and differential evolution (DE) is integrated into GWO in order to force GWO to jump out of the stagnation with DE’s strong searching ability

Summary

INTRODUCTION

OFDM (Orthogonal Frequency Division Multiplexing), a multi-carrier (MC) digital modulation technology of great spectrum efficiency and anti-multipath ability, is widely used. The traditional DNN model needs to manually determine the network structure and parameters to obtain the optimal results and the intelligent optimization algorithm has a strong ability of automatic optimization, so combining feature extraction with deep neural network and using intelligent algorithm to optimize neural network structure can provide new ideas for modulation recognition [9]. The HGWO algorithm is used to optimize the input weights and thresholds of DNN model This method can identify the OFDM signal from the complex signals including SC signal, WPM signal and OFDM signal. Simulation results show that compared with the traditional DNN recognition method and the DNN model optimized by GWO algorithm and DE algorithm, the recognition performance of this method is greatly improved under different SNR. Where S and N are corresponding to the effective power of signal and noise

SIGNAL MODEL

SEPARATION FEATURE EXTRACTION OF SC SIGNAL AND MC SIGNAL

PERFORMANCE MEASUREMENT

PERFORMANCE EVALUATION

CONCLUSION