Estimating transmitter activity patterns: An empirical study in the indoor environment

We empirically study problems of online estimation and modeling of radio activity (ON/OFF) patterns and their duty cycles. We consider performance of two algorithms that analyze power spectrum samples in time domain as sensed by WARP SDR boards placed at different indoor locations. The first algorithm utilizes density function of the marginal distribution of the received readings to perform clustering of the power samples into the classes that correspond to ON/OFF activity states. The second algorithm uses hidden Semi-Markov processes and the Viterbi algorithm to model and estimate the radio activity patterns of signal sources. Unlike the first algorithm it can also be used for both signal classification and estimation of the state duration that corresponds to the certain power level. We study signals obtained from one or two transmitters. Our results indicate the first technique can often provide accurate estimations. However, if sensor nodes receive low RSSI values or modeling of the temporal structure of the activity patterns is required then the second technique is preferable. This method also has a potential to benefit from cooperation between nodes.