Efficient Multi-Task Learning in Multi-User Multiple Input Multiple Output Systems Integrated Orthogonal Frequency Division Multiplexing Systems: A Hybrid Amalgamated Convolutional Neural Network-Bidirectional Long Short-Term Memory Approach

Abstract

Therefore, in today’s wireless communication systems and in particular, the Multi-User-Multiple Input Multiple Output-Orthogonal Frequency Division Multiplexing (MU-MIMO-OFDM) systems, channel estimation, the detection, and mitigation of the attack are important to ensure the safe operation of a system. Current approaches use distinct procedures for completing these jobs, and this causes high computational expenses, longer response times, and decreased performance of the system. In this work, a multi-task learning (MTL) framework is introduced to develop a new end-to-end deep learning solution of an Amalgamated Convolutional Neural Network (ACNN) for spatial feature extraction and a Bidirectional Long Short-Term Memory (Bi-LSTM) for temporal attack detection. The proposed system is effective in handling these tasks together because that would mean maximum efficiency and accuracy. To enhance the model’s efficiency, a Green Anaconda Optimization (GAO) algorithm is used to solve the multi-task loss function and enhance convergence rate and solution quality. The presented GAO approach provides a good balance between channel estimation, attack detection, and mitigation since GAO adapts the model parameters in the training process. Most of the current methods give slow convergence rates, and high computational costs, and are not very suitable for scale-up, especially in dynamic systems. These limitations make them unadoptable for real-time operations and analysis. The challenges described above are addressed by the proposed hybrid model with GAO, which is therefore ideal for modern secure wireless communication systems due to the reduced computational overhead and faster response time. The model reaches a first-level accuracy of 99% and costs 70 GFLOPs and 35 ms latency.