Measurement Based Delay and Jitter Constrained Wireless Scheduling With Near-Optimal Spectral Efficiency

We introduce two classes of measurement-based wireless schedulers. The Opportunistic Guaranteed Rate Scheduler (OGRS) meets a user’s delay constraints by opportunistically allocating the user the equivalent of a fixed service rate, which for a leaky-bucket constrained traffic ensures the delay requirements are met. By contrast, the Opportunistic Guaranteed Delay Schedulers (OGDS) schedules data transmissions when the current channel is better than what is expected in the time window before packet deadlines expire. Meeting such delay requirements requires a complementary admission control policy. We exhibit a simple measurement based policy, that indirectly accounts for heterogeneity in traffic, channel, and delay constraints by monitoring the statistics of user’s aggregate resource usage. We show that the spectral efficiency of our proposed approach is stochastically better than a wireless guaranteed rate scheduler. We bound spectral efficiency by considering an optimal offline policy with access to future channel rates and show via extensive simulations that OGRS can be within 10%-40% of the bound whereas OGDS is within 10% of the bound for a range of delay constraints. Additionally, we demonstrate that OGDS can exhibit better spectral efficiency at higher delay deadlines than schedulers leveraging neural network based predictions for future channel rates.