RFID Counting over Unreliable Radio Channels - the Capture-Recapture Approach

Recently, Kodialam et al. [1], [2] have proposed a number of probabilistic counting algorithms with statistical guarantees for RFID. Those algorithms do not require explicit tag identification, and can provide good estimates of tag-set cardinality with low latency. However, these algorithms all assume unrealistically that the communication channel between the reader and the tags is perfect. Direct application of these algorithms in practice can lead to severe under-counting. In [7], we made the first attempt to modify the algorithm in [2] to account for the effects caused by unreliable communications with the tags. However, the scheme proposed in [7] requires the knowledge of the first two moments of the successful responding probability distribution of the tag-set of interest. In other words, a priori (indirect) channel calibration/ profiling is needed. In this paper, we introduce a probabilistic tag counting scheme which does not require prior knowledge of the communication channel between the tags and the reader and yet can provide a good estimate of the cardinality of the tag-set of interest while taking the imperfect, unknown radio channel conditions into account. Our proposed scheme is based on a novel interpretation of the "Capture-Recapture" techniques developed by the ecology/biostatistics community which can adapt automatically to different channel conditions. Our simulation results demonstrate the accuracy of the proposed scheme and its reduced latency when compared to other rudimentary approach relying on simple repeated probing by the reader.