Multi-Task Learning with Convolutional Neural Network Approach for Packet Collision Avoidance in 802.11 WLAN

Abstract

Packet collision can degrade wireless network performance. The IEEE 802.11 Wireless Local Area Network (WLAN) uses the Clear Channel Assessment (CCA) mechanism to monitor channel availability to avoid interference of the presence signal. CCA successfully detects 802.11 signals if it obtains the packet preamble information or detects the threshold ambient power on the channel to determine the channel state. This paper proposes multi-task learning (MTL) with convolutional neural network (CNN) approach to detect WLAN packet formats and modulation types without preamble part information as a supplement to enhance CCA sensitivity. The main advantages of this method over single-task training are high classification accuracy and rapid learning with a lightweight neural network model. Shared knowledge of representation layers, such as model weights or gradients, improves the efficiency of training data and reduces redundancy. WLAN signals generated by the Matlab waveform simulator are used to verify the accuracy of the proposed method, which is then implemented on a real-time SDR-based hardware testbed. Although different time offsets affect the classifications, the proposed method proves superior in classifying the packet format and modulation of WLAN signals with an accuracy of 98.93% and 88.53% at SNR = 24 dB, respectively. The proposed method improves channel utilization and throughput of the WLAN network, as demonstrated by an NS-3 simulation.