Compressive Covariance Sampling for Spectrum Sensing (CoCoSa)

点击翻译按钮获取中文摘要

The available radio spectrum has become a scarce resource despite the fact that large parts of the licensed spectral bands are underutilized. Approaches to make more efficient use of the available spectrum have been developed within the framework of cognitive radio.The key idea is to let unlicensed radios access free spectral resources as long as they can ensure not to interfere with licensed usage. The methods enabling reliable licensed user detection, and thus safe secondary utilization, go by the name of spectrum sensing.The problem at hand is the detection of signals in very low signal to noise ratio (SNR) regimes. Multiple approaches have been put forward, some of which exploit the presence of inherent stochastic features in communication signals, e.g., properties of a signal's covariance matrix. However, to detect characteristic stochastic features of communication signals reliably, a large amount of measurement data has to be processed.To this end, we aim at developing methods and algorithms for licensed transmitter detection from a drastically reduced number of samples. Different types of covariance estimation shall be analyzed with respect to their error performance, and new detectors are to be developed. In particular, we will make a rigorous mathematical analysis on the minimal number of samples required for accurate covariance estimation under realistic assumptions on its structure. Typically, estimation of the covariance matrix and the choice of a test statistic for the binary hypothesis test (channel free or occupied) are treated independently. By interlocking estimation and detection approaches and associated test statistics new insights are to be expected due to the intended cooperation.In concrete terms, we plan on developing customized sparse rulers for the lossless recovery of the covariance matrix of different signal types. Furthermore, we will theoretically analyze the number of samples necessary for estimating a covariance matrix under different error guarantees. Finding a minimal sparse ruler can only be accomplished by exhaustive search. To tackle this problem and to attain real-time capability, we will develop smart search heuristics. Moreover, we intend to improve upon known detectors by finding new test statistics. Since better estimation of test statistic parameters leads to improved detection performance, a large part of the cooperative effort will be placed on this topic. Error bounds for the estimated parameters of the test statistics based on the signal covariance matrix will be derived. This will lead to more effective test statistics. As a final step, we will implement the new methods on a software defined radio testbed in order to evaluate their performance in the real world.