Stochastic Model-Based Deep Learning for Constrained Optimization of IEEE 802.11 Vehicular Communication Systems

IEEE 802.11 communication systems have been extensively investigated for improving vehicular road safety. However, dynamic vehicular environments and various safety applications with different quality of service (QoS) requirements cannot be accommodated by a fixed set of communication parameters and network configuration. This paper proposes and exploits a real-time constrained optimization platform that leverages the combination of a fast stochastic model with a carefully configured regression deep learning neural network (DLNN) to achieve an optimal balance between the QoS and the channel spectrum efficiency. The stochastic model is utilized to predict the QoS of IEEE 802.11 broadcast vehicular ad hoc networks given a selected group of communication parameters and analytical equations or measured data about the communication channels. The data provided by the stochastic model is sorted based on the QoS requirements for a given safety service and is preprocessed to keep enough distance between the training data patterns. The DLNN is trained by a randomly sampled data set to accomplish the inverse mapping (from the QoS to the corresponding parameter sets) to facilitate real-time optimization. In the process of optimization, by working in tandem, the DLNN and stochastic model synergistically identify the optimal parameter set that maximizes channel efficiency while adhering to QoS constraints in a fast and precise manner. The computation complexity of the optimization is analyzed and estimated. The effectiveness and robustness of the proposed optimization system have been demonstrated through experiments conducted on Google Colab using TensorFlow and Python, showcasing its superiority over the alternative optimization algorithms.