Spectrum Sensing & Signal Identification
In this project we consider a scenario where one or more sensing nodes observe a frequency band possibly used by radio transmitters forming packet based radio networks such as 802.11a/b/g, Bluetooth, Zig-bee, cordless phones, etc. Role of the sensing network is to perform analysis of the received signals and provide an appropriate characterization of the transmitters using the observed frequency band. Our main objective in this project is to develop signal processing algorithms needed to perform this task.
In one application scenario sensing nodes form a network dedicated for spectrum sensing. This concept has been adopted in the IEEE P1900.6 standard where there are three types of systems: legacy systems using static radio spectrum allocation, dedicated spectrum sensing network and new systems using dynamic spectrum allocation. The systems with dynamic spectrum allocation use information on spectrum availability provided by the spectrum sensing network to access spectrum without causing interference to the legacy systems. Thus, the spectrum sensing network provides the crucial information needed for enabling efficient radio spectrum utilization.
Most existing approaches to spectrum sensing are based on binary hypothesis testing where the goal is to determine if the observed frequency band is occupied or not or test if a certain signal is present or not in the observed frequency band. This problem is solved using well known tools such as energy detectors, matched filters, cyclostationary detectors, etc  . In our application we deal with multiple packet based radio transmitters where each transmitter produces a signal with non-persistent excitation. Thus, each signal is characterized with its spectra and an on/off activity sequence in time. Our approach is to develop algorithms for estimation of spectral and temporal parameters for each of the signals present. In other words, we wish to localize the packet based signals in time and frequency. Thus, our approach is significantly different from the existing spectrum sensing methods.
Since each transmitted signal consists of on and off transmission periods the received signal at each sensing node consists of a certain number of statistically homogeneous segments. First important task in the analysis of the received signals is to localize statistically homogeneous segments in time. We have developed a segmentation algorithm for this problem . Once the statistically homogeneous segments have been localized in time they can be analyzed using various statistical methods. We propose one such method based on fourth order spectrum . Our ongoing work is developing extensions of these algorithms for multiple sensors.