Efficient packet error rate estimation in wireless networks

Wireless links exhibit diverse link quality (e.g., in terms of error rate, latency, and required transmit power), and inferring the link cost correctly is crucial to finding an efficient path in multihop wireless networks. In this paper, we consider the problem of estimating packet error rate (PER) in wireless networks. Most existing schemes use observed error rates of probe messages as representative packet error rates for wireless links. However, since typical data packets are larger than probe messages, the actual data packet error rate is often higher than the observed error rate of probe messages. Instead of directly using measurement statistics, our approach is to analyze them to determine the link characteristics and estimate the link quality (e.g., PER) based on the understanding. We describe a simple estimation scheme based on a well-known two-state Markov bit error model for wireless channels. We perform various experiments on two wireless testbeds and show the proposed scheme can successfully estimate PERs for packets of arbitrary size in diverse environments. When compared to other alternatives, the proposed estimation scheme is accurate and incurs small control overhead. We also integrate the scheme into an existing routing scheme. Our experiment results show that the combined scheme can reduce the transmission overhead by 60% when compared to a naive estimation strategy based on one type of probe messages.