Single sensor radio scene analysis for packet based radio signals

Abstract

Abstract-We consider the problem of RF signal analysis where one sensing node observes a frequency band possibly used by multiple packet based radio transmitters. Analysis of the received signal consists of two steps. In the first step we use spectrogram to perform temporal segmentation of the received piecewise statistically homogeneous signal. This task is formulated as a clustering problem. In the second step we compute a certain 2-D slice of the fourth order spectrum for each of the segments found in the first step. The fourth order spectrum slices are arranged in a three-way array. Key idea of the second step is to use uniqueness properties of the low rank decomposition of the three-way array to recover spectra and associated activity sequences of individual components in the received signal. We derive a numerical algorithm for the low rank decomposition, which computes estimates of the spectra and activity sequences by optimizing a weighted least squares criterion under application specific constraints. The approach is illustrated with simulation examples involving signals used in 802.11a/b/g and Bluetooth networks. The proposed algorithm can be used as a spectrum analysis tool, providing crucial information needed for achieving efficient utilization of radio spectrum and elimination of mutual interference between the coexisting systems.

Unsupervised Radio Scene Analysis Using Neural Expectation Maximization

“…Blind source separation (BSS) techniques based on subspace structures were employed in the CR setups [3], [4]. The power spectrum of the received signal was clustered over time and the 4th-order spectrum was analyzed using three-way tensor decomposition in [5]. Mixed signals were classified using dictionaries trained on the singletransmitter scenes in [6].…”