802.11 Link Quality and Its Prediction - An Experimental Study

Reliable link quality prediction is an imperative for the efficient operation of mobile ad-hoc wireless networks (MANETs). In this paper it is shown that popular link quality prediction algorithms for 802.11 MANETs perform much more poorly when applied in real urban environments than they do in corresponding simulations. Our measurements show that the best performing prediction algorithm failed to predict between 18 and 54 percent of the total observed packet loss in the real urban environments examined. Moreover, with this algorithm between 12 and 43 percent of transmitted packets were lost due to the erroneous prediction of link failure. This contrasts sharply with near-perfect accuracy in corresponding simulations. To account for this discrepancy we perform an in-depth examination of the factors that influence link quality. We conclude that shadowing is an especially significant and hitherto underestimated factor in link quality prediction in MANETs.