Adaptive rate control with Quality of Service guarantees in wireless broadband networks

In this paper, we develop a framework for adaptive learning algorithms for rate prediction with Quality of Service (QoS) guarantees in broadband wireless communications systems such as LTE and LTE-Advanced. These approaches are motivated by real-world requirements for online rate adaptation for scheduling packet data transmissions and ensuring a good trade-off between maximizing rate and delay constraints. We pose the problem of temporal learning of the achievable rate as a function of a vector of noisy observations available from the receiver in terms of channel quality information (CQI). We show that a requirement to maintain a certain target frame error rate (FER) for a packet transmission translates to a problem of rate estimation using the notion of quantile regression with an asymmetric ί_χ norm error criterion. The algorithm uses the existing feedback from the receiver in the form of Ack/Nack bits of packet decoding, which provide a natural measure of error gradient in the proposed formulation. This framework of rate prediction is shown to have many applications within the wireless networks, including rate prediction for MIMO rank switching, adapting to noisy CQI data for frequency selective scheduling (FSS) or inter-cell interference co-ordination (ICIC). Simulation results are shown for a couple of such applications that establish the efficacy of the proposed framework and algorithms.