Design and evaluation of RFID counting algorithms under time-correlated channels

Several new RFID counting algorithms have recently been proposed based on the probabilistic counting schemes introduced by Kodialam et al. These existing algorithms took into account the unreliability of the communication channels between the RFID reader and the tags, and are capable of providing accurate tag-count estimates. However, all algorithms were designed and evaluated based on a simplistic packet loss model. It assumes that the loss probability of a packet between the reader and the tag-set follows an independent, identical distribution. As presented by some empirical measurements, movements of personnel or equipments in a building can generate Doppler effect, which introduces time correlations to the fading signal. Thus, the realistic packet loss of the wireless channels is temporally correlated due to the frequent change of the nearby environment. Depending on specific implementations of each algorithm, temporally correlated packet loss might have significant impact on the tag-set cardinality estimation. In this paper, we evaluate the performance of the aforementioned RFID counting algorithms under a more sophisticated time-correlated channel fading model. In particular, we focus on investigating how temporal correlations would influence the accuracy of these existing algorithms. Based on the experimental statistics that characterized the indoor channels, we refine the channel model to describe the time-correlation. Comparisons of the performance of the counting schemes under the simplistic uncorrelated packet loss channel model and the refined correlated channel model are conducted. We also propose extensions for these RFID counting schemes to mitigate the estimation inaccuracy generated by the correlated packet loss.