Ultra-reliable low-latency communication is a key enabling technology for next-generation industrial services, such as smart manufacturing, autonomous systems, and immersive experiences. These services require stringent communication guarantees with high reliability and low delay, posing significant resource management challenges. This paper proposes the joint resource optimization framework for delay-sensitive services over generalized fading channels (JO-DSGF), which maximizes throughput by flexibly adjusting bandwidth, transmission time interval, and semantic compression ratios, while meeting the constraints of delay and reliability requirements. In terms of theory, based on the key theory of stochastic network calculus (SNC), this paper derives the upper bound for the delay violation probability under generalized -- fading channels, considering the characteristics of both service and arrival processes. Two cases are comprehensively considered: Case 1 solves the delay violation probability given the decoding error rate, and Case 2 solves the decoding error rate given the transmission rate. On the algorithmic aspect, we propose the adaptive experience-driven long short-term memory (LSTM) advantage actor–critic (AE-LA2C) algorithm, which achieves efficient resource allocation by deeply coupling SNC and DRL, with the help of LSTM-based feature extraction. Simulation results demonstrate the effectiveness of the proposed JO-DSGF framework and AE-LA2C algorithm.
Zhang et al. (Tue,) studied this question.