Reliability- and Latency-Aware ATSSS Framework for URLLC Services

Ultra-reliable and low-latency communication (URLLC) services require strong latency and reliability performance under dynamic network conditions. However, achieving robust URLLC performance under single access is challenging due to user mobility, time-varying traffic loads, and fluctuating access availability. Multi-access operation provides an opportunity to enhance service robustness by exploiting heterogeneous connectivity. Access Traffic Steering, Switching, and Splitting (ATSSS), introduced in 3GPP Release 16, enables operator-controlled multi-access traffic management within the fifth-generation (5G) core network. In this paper, we propose the Reliability and Latency-Aware ATSSS Framework (RL-AF) for URLLC services, which incorporates latency and reliability awareness into adaptive traffic steering and redundant transmission decisions. RL-AF formulates the joint control problem as a constrained Markov decision process and employs a primal–dual deep reinforcement learning approach to balance latency–reliability performance and the network usage cost. Trace-driven simulations benchmark RL-AF against Cell-Only, Cost-Min (Wi-Fi-only), Redundant, RTT-Min, and Loss-Min schemes. At a residual RTT threshold of 30 ms and a packet loss constraint setting of 10−3, RL-AF achieves an approximately 94.4% RTT success rate and an empirical packet loss ratio of 1.0×10−3 with an average network usage cost of 4.14, corresponding to a 17.2% cost reduction relative to always-redundant transmission.