Flow-Level Performance of Device-to-Device Overlaid OFDM Cellular Networks

Spectrum partition and mode selection are two fundamental issues in D2D overlaid cellular networks, whose performance has been studied at the packet level for a static user population assuming infinite backlogs. In this paper, we aim at the flow-level performance with user dynamics governed by the arrival and completion of random service demands over time. We demonstrate the flow-level performance can be evaluated based on a queuing model with a dispatcher that assigns the service demands to two independent Processor Sharing servers. The queue model provides closed-form expressions for the mean number of users, the mean delay and the mean throughput, based on which the optimal spectrum partition and mode selection parameters can be derived and useful insights for the design principle can be obtained. Finally we perform numerical experiments to validate our model.