Compressive Wideband Frequency Spectrum Sensing Based On MUSIC

Locating multiple active subbands in wide frequency range is an important application of spectrum sensing in cognitive radio systems. While a high Nyquist sampling rate required by an analog-to-digital converter (ADC) devices becomes a common burden in this field, inaccurate frequency band estimation also becomes a severe problem leading to interference between a licensed and unlicensed user (secondary user). In this paper, a high resolution estimation algorithm, i.e., multiple signal classification (MUSIC), and sub-Nyquist rate sampler, i.e., non uniform multi-coset sampler, are adopted to alleviate those two facts appearing in the wideband frequency spectrum sensing. Specifically, the basic ingredients of our proposed method involve splitting the entire band into uniform bins, solving the coset correlation matrix, followed by the MUSIC pseudospectrum reconstructions. The circulant structure embedded in the coset correlation matrix allows the strong compression ratio which can be achieved by selecting the coset patterns based on the circular sparse ruler. An averaging procedure followed by frequency-smoothing method are employed to eliminate the redundant elements and construct the full rank correlation matrix, respectively. The simulation in this study shows that it is possible to locate the frequency band of active users clearly (clear boundaries between signal and noise) and accurately based on the reconstructed MUSIC pseudospectrum. This will be beneficial later for particular cognitive radio application, i.e., wireless regional area network (WRAN) which has very low received signal-to-noise ratio (SNR) as the standard requirement.