Wideband Spectrum Sensing Using Compressive Sensing
Spectrum sensing Number of wireless devises is growing every day spectrum has become valuable The most of allocated bands are unoccupied more than 90% of the time PUs pay for the spectrum bands and no SU can use those bands to avoid interference CR maximize the efficiency of a wireless system by detecting ad transmitting on underused bands while avoiding interference No need to any pre-assigned Frequencies to operate numerous commercial and military benefits
Challenges Sampling and hardware sampling CRs are used in a restricted frequency range limits in usefulness If wideband sampling at Nyquist rate In practice, impossible to operate over wideband frequency band Compressive sensing
Compressive Sensing in Spectrum Sensing Recover certain signals from far fewer samples Conditioned on : signal should be sparse in a particular domain Principle of sparsity: information rate of a continuous time signal is much smaller than suggested by its bandwidths. Sampling in Sub-Nyquist rate AIC (ADC+CS) AIC x(t)y[m]y[m]
Analog to Information Cnonverter x(t)x(t) p(t)p(t) y(t)y(t)y[m]y[m]
Compressive sensing x is in time domain and sparse in frequency domain l0 minimization (NP-hard) L1-minimization: – measurement matrix should have RIP –
l1/l2 minimization The received signal is block sparse For a licensed user, its operating spectrum is a certain band Using the priori information of the spectrum boundaries between PUs, we can use l1/l2 minimization instead of l1 minimization
Threshold-Iterative Support Detection (Threshold-ISD) Reduced requirement on the number of measurements compared to the classical l1 minimization Recovers the signal iteratively (but a few iterations) Algorithm farmework: 1. s=0: l1-minimization, 2. while stopping criteria isn’t met: – where – Solve truncated BP : – j j+1 Truncated BP
n=256 m=140 sparsity ratio = 30% (# of non-zeros = 77) # of iterations=3 Beta=5
n=256 m=120 sparsity ratio = 30% (# of non-zeros = 77) # of iterations=3 Beta=5
n=256 m=85 sparsity ratio = 30% (# of non-zeros = 77) # of iterations=3 Beta=5
Performance of 4 algorithms n = 256 sparsity ratio = 30% (# of non-zeros = 77) # of runs=75 per each m
Future works Spectrum has correlation in consequent time slots, so we can search over the frequency bands which are more likely to be unoccupied Achieve better performance and less complexity -Adaptive ISD : each iteration of ISD can be done in each time slot. (threshold should be constant) -Extend to the distributed cognitive radio networks -Using sparse measurement matrix to be able to use verification based algorithms whose complexity is much smaller than classic l1- minimization
References First work on spectrum sensing using compressive sensing: Zhi Tian; Giannakis, G.B., “Compressed Sensing for Wideband Cognitive Radios”, ICCASP, June 2007 ISD : Yilun Wang and Wotao Yin “Sparse Signal Reconstruction via Iterative Support Detectio”, SIAM Journal on Imaging Sciences, pp , August 2010 L1-l2 : Mihailo Stojnic, Farzad Parvaresh, and Babak Hassibi “On the reconstruction of block-sparse signals with an optimal number of measurements”, IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 57, NO. 8, August 2009 Applying 11/l2 in spectrum sensing contex: Yipeng Liu and Qun Wan, “Compressive Wideband Spectrum Sensing for Fixed Frequency Spectrum Allocation”, arxiv, 2010