RBM Based Cooperative Bayesian Compressive
Spectrum Sensing with Adaptive Threshold
Xuekang Sun, Li Gao, Xudong Luo, Kun Su
School of Telecommunication Education
Beijing University of Posts and Telecommunications
Beijing, China 100876
Email: sunxuekang6395@126.com
Abstract—Cooperative spectrum sensing schemes can enable
cognitive radio (CR) users to efficiently identify the unoccupied
channels or spectrum holes, as well as overcome the impact
of shadowing and fading. Considering the hardware limitation,
compressive sensing (CS) is a solution scheme to alleviate the
requirements on the receiver hardware, which can recover the
wideband sparse signal sampled at sub-Nyquist rate. In this
paper, Restricted Boltzmann Machine (RBM) based cooperative
Bayesian compressive spectrum sensing with adaptive threshold
(RC-ABCS) is proposed for block sparse wideband signal. In this
scheme, we use the Bayesian compressive sensing (BCS) model
to sense the wideband sparse signal and report the results to a
fusion center. In this fusion center, a proposed iterative algorithm
of Relevance Vector Machine (RVM) with adaptive threshold is
used to increase the recovered accuracy of block sparse wideband
signals. And then we employ RBM learning to achieve the fusion
decision based on the recovery signals of the multi-CR users. The
simulation results show that the proposed scheme can increase
the detection accuracy, enhance the ability of anti-interference
and improve the convergence rate.
Index Terms—cooperative spectrum sensing; Bayesian com-
pressive sensing; recovery; fusion decision; adaptive threshold
I. INTRODUCTION
Spectrum sensing is a critical operation in cognitive radio
network (CRN), which enables CR users to identify unused
spectrum spaces and use these spaces optimally without harm-
ful interference to the licensed users [1]. However, it has
considerable technical challenges on the sensing speed and
accuracy especially in the wideband system that requires high
sampling rates due to the limitations of hardware operational
bandwidth of most receivers.
To alleviate the requirements on the receiver hardware, the
researchers propose various CS reconstruction methods such
as basis pursuit (BP) [2], orthogonal matching pursuit (OMP)
[3] and Bayesian compressive sensing (BCS) [4][5] for the
sparse signal. As Bayesian method can make use of previous
data to estimate the posterior probability of distributions of
the interested parameters as well as have faster convergence,
BCS approach is introduced to wideband spectrum sensing
for computing the parameter estimation recursively. However,
the spectrum sensing of a CR user is quite susceptible to
shadowing and multi-path fading, especially in a lower signal-
to-noise ratio (SNR) wireless environment. Cooperative spec-
trum sensing (CSS) provides an effective solution to improve
the detection performance by exploiting the spatial diversity
among the different CRs. In wideband CRN, CSS consists of
∗
This work is supported by Chinese National Nature Science Foundation
(61372116).
three stages: the local signal compression, the signal recon-
struction and the fusion decision. On the basis of different
control manners, the data fusion is generally classified into
two categories: centralized fusion and decentralized fusion.
In our paper, we focus on the centralized control network
which requires a fusion center (FC) to collect the sensing
measurements from multiple CR users and make the fusion
decision. Obviously, the selection of data fusion rules plays
an important role in the global decisions on the spectrum
occupation.
In recent years, several data fusion schemes for centralized
control network are investigated to increase the detection
accuracy corresponding to different sampling strategies. Ref-
erence [6] proposes a weighted orthogonal matching pursuit
(W-OMP) algorithm, in which a weight dictionary W that
represents the state of the channel affected by shadowing
and fading is introduced to describe the cooperative effect
of channel gains. In the hybrid distributed sensing matrix
(HDSM) algorithm [7], the cooperative CRs are firstly clas-
sified into several sensing groups based on the state of the
channel between the CR user and its fusion center. Each CR
uses its local sensing matrix which is assigned to its sensing
group by FC to obtain its sampling data. The FC can use these
data from different CRs to make a global decision on spectrum
occupation, while it can decrease the hardware complexity
of FC and CR users. In [8], a cooperative spectrum sensing
scheme based on Back Propagation (BP) neural network is
proposed to improve the detection accuracy and efficiency. In
this scheme, the author uses a new fusion rule on reliability
forecasted by adjusting the parameters of BP neural network
iteratively under various environmental conditions.
In this paper, we study how to use Restricted Boltzmann
Machine (RBM) framework to deduce the statistic relation
weights among CR users as a basis of the fusion decision
[9]. Furthermore, RBM based cooperative Bayesian compres-
sive spectrum sensing with adaptive threshold (RC-ABCS) is
proposed to improve the detection performance of the signal
affected by shadowing and fading. In the RC-ABCS, the
CR individually uses the local Bayesian compressive sensing
matrix to sense the PU signals and send the compressive data
to FC. In FC, we use the improved iterative algorithm of RVM
with adaptive threshold to recover spectrum signals in term of
the compressive data from different CRs [10]. And then we
employ RBM framework to make the global decision on the
spectrum occupation.
The rest of this paper is organized as following: Section II
presents the system model. Section III introduces our proposed