Noise Variance Estimation Through Penalized Least-Squares for ED-Spectrum Sensing

Cognitive Radio (CR) is an auspicious solution to current problem of spectrum scarcity due to evaluation of new technologies. These techniques are useful in detecting spectral holes, and allocating them to secondary users. Energy Detection is a predominant method for spectrum sensing due to its low computational complexity and capability of detecting spectrum holes without requiring apriori knowledge of primary signal. The energy based spectrum detectors depends on the precision of threshold chosen to distinguish signal and noise. But, energy detection needs to estimate the noise variance for finding the detection threshold. Most of the conventional techniques use fixed threshold with known noise variance. In practical scenarios noise variance is unknown, so we are proposing a fast computational noise variance estimation algorithm for spectrum sensing using Penalized Least Squares (PLS). We have introduced a smoothing parameter which is determined by Discrete Cosine Transform (DCT) as the penalizing factor. The amount of smoothing is determined by minimizing Generalized Cross Validation (GCV). Simulations were carried out in AWGN and Rayleigh fading channels for the proposed noise variance estimation through which Receiver Operating Characteristics (ROC) are obtained.