AccuRate: Constellation Based Rate Estimation in Wireless Networks.

This paper proposes to exploit physical layer information towards improved rate selection in wireless networks. While existing schemes pick good transmission rates, this paper takes a step further towards computing the optimal bit rate. The main idea is to capture the channel behavior through symbol level dispersions, and "replay" these dispersions on different rate encodings of the same packet. The "replay" action can be emulated at the receiver without requiring the transmitter to send the packet at every other rate. The maximum successful rate is likely to be the optimal rate of the received packet, and assuming that the channel remains coherent, the same rate can be prescribed for the next transmission. We design, implement, and evaluate this idea over a small testbed of USRP hardware and GNURadio software. Our proposal, called AccuRate, predicts a packet's optimal rate 95% of times when the packet is received correctly. When the packet is received in error, AccuRate computes its optimal rate with 93% accuracy. In terms of throughput, we show that AccuRate improves over the state-of-the-art scheme SoftRate by around 10%, and is reasonably close to the optimal.