Ultra-reliable low-latency industrial wireless communications: optimization and implementation

Abstract

Ultra-reliable and low-latency communications (URLLC) are crucial for mission-critical services in fifth-generation (5G) mobile networks. Due to limited time and frequency resources in URLLC, decoding packet error probability (PEP) is unavoidable, which makes meeting the reliability constraint extremely challenging. A key challenge is that current wireless systems use pilot symbols for channel estimation, sharing channel resources with data symbols. In this thesis, I first design unsupervised learning algorithms to estimate a resource allocation policy for independent and identically distributed fading channels, including a model-based algorithm for measurable PEP and a model-free method for discrete PEP observations. Results demonstrate that my methods can significantly enhance resource utilization efficiency, even with a lower signal-to-interference-plus-noise ratio. Furthermore, I focus on temporally correlated channel realizations and propose deep reinforcement learning algorithms to determine the resource allocation policy that can maximize long-term resource utilization efficiency. I formulate the optimization problem as a partial observation Markov decision process and develop a cascaded-action Twin Delayed Deep Deterministic policy (CA-TD3) to address it. I introduce a primal CA-TD3 algorithm and compare it with a primal-dual method. The results indicate that the primal CA-TD3 converges more efficiently than the primal-dual method. Lastly, I aspire to design a hardware platform to implement wireless time-sensitive networking, which is one of the most vital scenarios of URLLC. I select a commercial 802.11-based platform and utilize a time division multiple access (TDMA) mechanism to schedule transmissions through novel real-time quality of service and fine-grained aggregation schemes. Experimental results demonstrate the superiority of my proposed protocol compared to existing TDMA-based 802.11 and legacy 802.11 systems.