Throughput Maximization of HARQ-IR for ISAC

Abstract

In this letter, hybrid automatic retransmission request with incremental redundancy (HARQ-IR) is applied to assist integrated sensing and communication (ISAC). The long term average throughput (LTAT) of HARQ-IR-assisted ISAC is maximized via power allocation while ensuring both communication and sensing reliability as well as total average power budget. Since the LTAT maximization is a non-convex problem, the asymptotic outage approximation is leveraged for problem relaxation. Subsequently, successive convex approximation (SCA) is exploited to convert it into a successive geometric programming (GP) problem. However, the GP-based solution underestimates the LTAT due to the large approximation error of the asymptotic outage probability at a low signal-to-noise ratio (SNR). To address this issue, the original problem is transformed to a Markov decision process, which can be solved with deep reinforcement learning (DRL), e.g., deep deterministic policy gradient. Numerical results show that the DRL-based method delivers comparable performance to the GP-based one at high SNR while performing much better than the GP-based method at low SNR, albeit at the cost of higher complexity.