Abstract—Since the radio channel between the base-station and a User Equipment (UE) is stochastic, semi-persistent scheduling (SPS), commonly used for Voice over New Radio (VoNR), may be suboptimal in the utilisation of time-frequency resources, thus providing lower residual bit-rates for enhanced mobile broadband (eMBB) flows. We consider optimal scheduling of voice packets so that their resource utilisation is minimised, without violating their transmission deadlines. Assuming continuous time-frequency resources, we compare the performance of the following algorithms: SoA: Schedule on Arrival, which is one version of semi-persistent scheduling; RDR: Rate to(Residual) Deadline ratio based policy, a natural heuristic; OST: Optimal Stopping Time scheduling, which uses the finite horizon backward dynamic programming approach. We compare the above policies with GAS: Genie-Aided Scheduling, a bounding approach, in which the scheduler, non-causally, gets the resource requirements in every slot up to the deadline of the packet. We emperically evaluate the impact of these scheduling policies on VoNR resource utilisation and packet loss, and on the eMBB throughput region. We also study the effect of advancing the deadline for packet delivery, to reduce the downlink scheduling overhead and UE power consumption. For a 100 MHz system, i.i.d. channel over slots, independent across users, 100 VoNR calls, and call-by-call scheduling, we find that, for a packet deadline of 20 ms (40 slots), state dependent policies can provide up to 80% reduction in resource utilisation by packet-voice, whereas a deadline of 1.5 ms (3 slots) already yields a reduction of up to 60%. When there are slot-to-slot correlations in the channel, there is reduction in the above gains for coherence times 5 ms (10 slots) and 10 ms (20 slots), but there is still room for optimisation beyond SPS.
SCHEDULING POLICIES