Performance of gradient-based OFDMA subcarrier schedulers with TCP traffic

We study interaction between gradient-based subcarrier allocation algorithms and TCP traffic sources in a downlink single-hop OFDMA wireless system. Specifically, we are interested in evaluating the maximum-rate (Max-SNR), Proportional Fair (PF), and queue-based Max-Weight (QBMW) subcarrier schedulers, in both non-iterative and iterative versions, with long-lived TCP fluid traffic. Using system throughput and the Jain's fairness index as benchmarks, our simulation results show that even when the users are homogeneous, QBMW leads to suboptimal throughput and extreme unfairness in the presence of TCP traffic. On the other hand, when the users are heterogeneous, Max-SNR is also unfair and gives suboptimal system throughput. However, in both cases, the iterative PF gives the best throughput as well as the best fairness among all considered schedulers, due to its slow variation in the subcarrier allocation, which is preferred by TCP.