Introduction is expected to increase eightfold by 2020.

Introduction

 

The
immense growth in the use of smart devices hosting convoluted application has
led to an increase in traffic in wireless network technology. According to
CISCO Visual Networking Index(VNI), due to tremendous increasing of wireless applications
and data rates, the data traffic is expected to increase eightfold by 2020.
However, thespectrum scarcity problem cannot be solved due to currentfixed
spectrum allocation policy used by the government. Recent advancement in
wireless technology is creating a spectrum shortage problem on a daily basis.
Cognitive radio, a novel technology, attempts to solve these problems by
dynamically using the free spectrum in wireless communication. Basically,
Cognitive Radio is an intelligent wireless communication device that can change
its operating parameters dynamically based on interaction with the environment
in which it operates.

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

 

The
experiments by FCC show that at any given time and location, much (between 80%
and 90%) of the licensed spectrum is underutilized. Such temporarily unused
spectrum slots are called spectrum holes, resulting in spectral inefficiency.
Thus not only is spectrum usage low in some licensed bands, but also true
scarcity of radio spectrum compounds the problem. Consequently, the growth of
wireless applications may be hindered. Key features of a cognitive radio
transceiver thus include radio environment awareness and spectrum intelligence.
The latter refers to an ability to learn the spectrum environment and adapt
transmission parameters. For instance, two types of cognitive radio networks
are distinguished based on the spectrum bands:

• On
unlicensed bands: These include ISM (industrial, scientific and medical) bands
such as 902–928 MHz, 2.4–2.5 GHz, and 5.725–5.875GHz. ISM bands are also shared
with non-ISM applications, e.g., Bluetooth, IEEE 802.11/WiFi. These bands can
be utilized by cognitive radio.

• On
licensed bands: The spectrum is licensed into different applications, e.g.,
aeronautical and maritime communications, AM radio etc. But there is
significant under-utilization of licensed spectrum which can be overcome with
the help of  cognitive radio. For
instance, the wireless regional area network (WRAN) standard operates in unused
television (TV) channels in 698–806 MHz.

 

The CR
requires four main functions spectrum sensing, spectrum management, spectrum
sharing, andspectrum mobility to dynamically access both licensed andunlicensed
spectrum bands.

 

IEEE
802.22 standard is known as CR standard because of cognitive features it
contains. The standard is still in development stage. One of the most
distinctive features of IEEE 802.22 standard is it’s spectrum sensing
requirement.

 

Energy
detection is a anticipating low-complexity and low-cost
spectrum sensing technique.This measures the received signal energy within the
pre-defined bandwidth and time period. The measured energy is then compared
with a threshold to determine the status (presence/absence) of the transmitted
signal. Not requiring channel gains and other parameter estimates, the energy
detector has low implementation cost.

 

 

 

The
detection of spectrum holes is a difficult signal processing problem. This
problem is made much more difficult due to signal fading that manifests itself
in 2 ways: multipath fading( Rayleigh fading) and shadowing(large scale
fading).

 

 

In this
paper, we propose novel CSS schemes based onmachine learning techniques.Machine
learning algorithms have been widely used forthe pattern classification
problems, where the feature vectorextracted from the training data is fed into
the classifier tocategorize the pattern into a certain class. Spectrum sensing
can be thought of as a binary-class classification problem.For CSS, we consider
an “energy vector”, each component ofwhich is the energy level determined at
individual secondarynode, as a feature vector. The classifier
categorizes this featurevector either into the “channel available class” or the
“channelunavailable class”. The classifier has to undergo through atraining
(learning) phase beforethe online classification starts.In general, two types
of learning algorithms exist, namely the”supervised” and the “unsupervised”. In
case of the supervisedlearning, the training feature vectors are fed to the
classifierwith their actual labels; while in case of the unsupervisedlearning,
the same are fed without any label.

We propose
to use unsupervised learning approachessuch as the K-means clustering and the
GMM for CSS.The K-means clustering algorithm partitions the featuresinto K clusters.
Each cluster is mapped to either thechannel available class or the channel
unavailable class.