Content uploaded by Varun Kumar Kakar
Author content
All content in this area was uploaded by Varun Kumar Kakar on May 14, 2019
Content may be subject to copyright.
Prediction Model Based Hybrid Routing Protocol For
Cognitive Radio Ad-hoc Network
Gaurav Singh Negi
Department of Electronics and Communication Engineering
B.T. Kumaon Institute of Technology
INDIA
gauravsnegi747@gmail.com
Varun Kumar Kakar
Department of Electronics and Communication Engineering
B.T. Kumaon Institute of Technology
INDIA
Abstract— Cognitive is a new paradigm and now seen as a
solution to the ‘spectrum scarcity’ problem in wireless
communication system. It is necessary to use spectrum efficiently
due to its limited availability. Cognitive radio network reuses the
spectrum environment to detect the white spaces in the frequency
spectrum. White spaces are then allocated to secondary user to
carry out the transmission and keeping in check the interference
to primary user. Various prediction models and algorithms based
on probability theory have been used to predict the channel states
to ensure efficient use. In this paper, we propose a Prediction
Model Based Hybrid Routing protocol For Cognitive Radio Ad-
hoc Network. Proposed method is compared with technique used
previously. Simulating the scenario results, conclusions are
drawn and discussed.
Keywords- Channel sensing, spectrum sensing,cognitive radio,
Hidden Markov Model(HMM).
I.
I
NTRODUCTION
According to a federal communication commission, the
spectrum bands allotted on the basis of static allocation
strategies are used over a bounded geographical area or for a
limited period of time resulting in underutilization of resources
[2]. Cognitive radio is acclaiming to be the solution of scarcity
problem. Cognitive radio is a concept combining both the
adaptive signal processing and software defined radio [3].
Cognitive radio senses the spectrum environment where it
searches for the vacant space after which, optimal band is
selected according to availability and type of application. In
cognitive radio network secondary users have to use spectrum
allotted to the licensed or primary user [4]. So whenever the
primary user tries to access the channel for the transmission
secondary user has to vacate the channel for preventing the
interference that will occur by the transmission.
Routing plays a crucial role in the performance of a
cognitive radio network. Most of the works remain focussed on
Physical and Medium access layer as the network layer remain
unexplored. To achieve high performance cross-layer design is
needed which takes into account the parameters affecting both
layers. Designing routing protocols for cognitive radio network
is different from conventional wireless networks. Some of the
problems faced in designing routing strategy for CRN’s [9] are:
¾ Change in topology of the network due to activity of
PU’s.
¾ Various delays like switching, back off, queuing.
¾ Optimal channel access mechanisms.
¾ Large number of handoffs due to the frequent
change in operating frequency of PU.
¾ Limitations on power supply and computational
time.
Research in the fields of spectrum occupancy can be
characterized on the fundament of approaches used to model
the traffic usage patterns of primary user. One approach is to
use real time environment where the measurements are
collected by using an antenna. In this approach density
functions are generated by statistically analyzing data collected
by practical circuitry. Imperfect sampling makes it more
difficult to capture the accurate measurements. Though this
type of modeling used in real time scenarios but the results are
hindered due to the spectral and temporal dependence.
Second approach is to find a stochastic process based
model to describe the primary user’s behavior. Prediction of
presence of primary user is done on the basis of statistical
modeling [14]. These Models are based on mathematical
processes for example Markov process based models, queuing
theory based model, On/off models, and dictionary based
process. As the complexity of the model increases, ability to
construct the pattern sequence from the training sequence
improves.
In this paper, we propose Prediction model based hybrid
routing protocol for cognitive radio network. We have
considered a situation where the network consists of node
continuously sensing the spectrum environment. All the
transmission takes place through the cluster head and finally
base station. We have considered the secondary users to be
immobile.
The organization of the paper is as follows: Section II
reviews related work. Section III discuss about the network
model and assumptions which is going to be used. Section IV
discusses the flow chart and detailed architecture of improved
protocol. Performance evaluated by simulation is described in
section V and finally, Section VI concludes the paper.
II. R
ELATED
W
ORK
In this work, binary time series [6] is used to model the
spectrum occupancy patterns of a cognitive radio network. This
system performs well in the case where the provided data is
deterministic in nature even if the model is not updated but
fails where the data is nondeterministic. The author has
proposed a memory-based model which saves the previous
values of the channel occupancy to predict the future states.
In this [5] author proposed Markov based channel state
prediction algorithm where the primary user’s channel usage
patterns are Poisson distributed. Hidden Markov model based
channel access scheme is compared with CSMA based scheme
in terms of signal to interference ratio. MCPA shows
significant improvement in the signal to interference ratio,
which leads to decrement in the bit error rate of the network.
Analysis of error in the prediction of channel availability
[11] is done as a function of state transitions and channel
conditions. Relation between prediction accuracy and number
of samples used to predict the next state has been drawn. Error
which occurs in the prediction due to the error in the initial
probabilistic values provided during training been discussed
with miss detection probability.
SEARCH [7] is a routing protocol which collectively takes
decision about the route and channel to be used during the
transmission. Each node is aware of the location of neighbor
nodes. Beacons are exchanged between the neighboring nodes
to exchange information about the location. It uses Greedy
forwarding technique to minimize the number of hop counts
needed to cover the path from source to destination. Route
request is broadcasted to every available channel to find the
most efficient route. Disadvantage of this protocol is that
routing control overhead is very high due to the flooding of
message into every channel.
III. N
ETWORK
M
ODEL AND
A
SSUMPTIONS
A cognitive radio network consists of two types of users:
primary and secondary user. Primary user uses the spectrum
directly, however, cognitive user accesses the spectrum bands
in opportunistic mode. Cognitive users are equipped with
single transceiver capable of half duplex communication. C
n
is
the number of available channels which are orthogonal and
non-overlapping in nature. A common control channel is
devoted for exchanging the control information between the
nodes. Every channel is identified by a unique identification
number.
Clustering is done in such a way that the nodes forming
the network become cluster heads. Cluster heads governs all
the communication that takes place between the nodes in a
cluster or between the nodes representing different cluster.
Cluster head remains synchronized with the nodes with the rest
of the nodes using periodic beacon signaling.
IV. P
ROPOSED
S
YSTEM
M
ODEL
Primary user activity is observed on channels and the
observations are stored to model the pattern. Logic value ‘1’
represents the busy state of the channel while Free State is
shown by logic ‘0’. Sequence is stored in the form of binary
time series which is later used to train the prediction model.
Spectrum predicting model estimates its parameters and then
predicts the value of the next state[12]. Assumption has been
taken that the channel state is stable for one time slot and
spectrum sensing error is not taken into account.
It is a hybrid protocol which uses proactive routing for
intra-cluster communication and reactive routing for inter-
cluster communication. Clustering is done by taking into
account the spectrum available to the nodes. To achieve
communication between the nodes in a cluster, it is necessary
that they should have idle channels in common. Clusters are
modeled as biclique graphs. Our goal is to maximize the sum
of total number of nodes in a cluster and available channels
which are common between the nodes.
Fig 1. Flow diagram of the Proposed Approach
In this protocol, clustering is defined as maximum vertex
Biclique problem [10] and a cluster head election value is
defined to select optimal cluster head. Cluster head is selected
taking into account the mobility of the node and total number
of neighbors. This is done to avoid frequent clustering due to
varying network topology, which provides stability to the
cluster. Cluster head election value is calculated for every
node as in [15] is defined as follows:
ܥܪܧܸ
ೕ
ൌܹ
ೕ
ൈܰ
ೕ
ೕ
Where ݅
represent node i in cluster j, ܹ
ೕ
is normalization
factor takes into account the node instantaneous power and
Neighbor
Discover y
Route
Request
Parameter
Calculation
Optimi zation
Channel list
Formation
Cluster
Formation
Phase
Cluster head
S
election
Initializing the
parameters of Network
Route Repl
y
Channel selector
TransmissionEnd
Start
Channel free
Channel
free
Predicting Channel
State Based On CSI
P(O,1|H)шP(O,0|H)
Observe Channel
Usage Patterns
Update Model
Parameters
True
False
mobility of that individual node, ܰ
ೕ
is total number of
neighbor of node i in cluster j, ݄ܿ
ೕ
is the total number of
common channels that node i has in cluster j.
Where Source nodes send the CHRReq(cluster head route
request) packet to cluster head, which in response sends
CHRRep(cluster head route reply) to the source if the
destination is in the same cluster. Otherwise, it broadcast the
request packet attaching its id, hop count, achievable data rate,
density and cluster channel with it. After destination receives
CHRReq packet it will calculate the weight for available
routes and will choose the path with minimum value of M(r).
Path optimization is done by using Evolutionary dynamic
weighted aggregation method and optimal route selection is
formulated as:
ܯሺݎሻൌሺݓ
ଵ
Ǥߤ
்
ሺݎሻݓ
ଶ
ሺݎሻǤͳ
ܴ
்
ሺݎሻݓ
ଷ
ሺݎሻǤ ܪ
்
ሺݎሻሻ
s.t. w
i
(r) 0 where σ࢝
ୀ
ሺ࢘ሻൌ
Where w
1
(r), w
2
(r), w
3
(r) are the weights assigned to density
(ߤ
்
ሺݎሻ), achievable data rate (ܴ
்
ሺݎሻ) and hop count (ܪ
்
ሺݎሻ)
respectively. Values of these weights change periodically
between 0 to 1 resulting in a set of Pareto optimal solutions
instead of a single solution. Values of the weights can be
calculated using following equations:
ݓ
ଵ
ሺݐሻൌቚ
ୱ୧୬ଶగ௧
ி
ቚ
ݓ
ଶ
ሺݐሻൌͳെݓ
ଵ
ሺݐሻቚ
ୱ୧୬ଶగ௧
ி
ቚ
ݓ
ଷ
ሺݐሻൌͳെݓ
ଵ
ሺݐሻെݓ
ଶ
ሺݐሻ
Where t is generation index and F is the frequency at which
weights vary. F is chosen in a range where it is high enough to
make optimizer move from one stable point to another or it
should not be too large because algorithm may fail to
converge. Modulus returns the absolute value of the weights.
Hidden Markov Model [1] is based on the two properties:
(1) Current state (s
n
) of the model depends on the
previous state (s
n-1
) of the model.
(2) States of the Markov model will remain hidden.
Observer can only see observation sequence available
at time n.
Hidden Markov Model is modeled by the following:
(1) Number of Symbols emission, M
(2) Number of states of HMM, N
(3) Sequence of Observation, Ob={0b
1
, 0b
2,
0b
3,…………
0b
n-1,
0b
n
}
(4) State transition probabilities, a
ij
= P(q
n
= S
j
|q
n-1
= S
i
)
Subject to conditions a
ij
0 and σ ൌ ͳ
ே
ୀଵ
(5) System emission probabilities, b
j
(V
m
) = P(Ob
t
=V
m
|q
t
= S
j
)
Subject to conditions b
j
(V
m
) 0 and σሺሻൌ
ெ
ୀଵ
ͳ, 1jN
(6) Initial state distribution, ʌ = {ʌ
1
,ʌ
2
,ʌ
3
,…..ʌ
N-1
ʌ
N},
where ʌ = P(q
i
= S
i
)
And satisfies the equation ʌ
i
0 and σߨ
ൌͳ
ே
ୀଵ
.
Hidden Markov Model can be represented by H = (ʌ, A,
B), where ʌ is initial states distribution, A is a N × N matrix
consisting of the state transition probabilities of the model a
ij
where I and j represents the number of rows and column
respectively and B is the N × M matrix containing the
probabilities of emission of symbols b
j
(V
m
) where j denotes
number of rows and m represents the number of column.
There are three types of problems associated with Hidden
Markov Model:
(1) Evaluation Problem: probability of producing the
given observation Ob= ob
1
ob
2
ob
3
ob
4
………ob
T
by
the model is calculated?
(2) Decoding Problem: In this type of problem, state
sequence of the model H is evaluated which produces
Ob= ob
1
ob
2
ob
3
ob
4
………ob
T
?
(3) Learning Problem: How should we adjust the model
parameters {A, B, ʌ} in order to maximize
P{O|H},whereas a model H and a sequence of
observation Ob= ob
1
ob
2
ob
3
ob
4
………ob
T
are given?
Consider the sequence of observation {Ob=
ob
1
ob
2
ob
3
ob
4
………ob
T
} representing state of the channel.
Occupied channel state is denoted by 1 while an unoccupied
state is represented by 0. Objective of the predictor model is to
anticipate the next observation on the basis of past history of
observation occurrence. Prediction of the next state will be
possible if hidden Markov model will be generate the past
sequences with maximum likelihood. Maximum likelihood
ensures the high probability of observation sequence and the
next correct observation prediction. To achieve the maximum
likelihood Hidden Markov model is needed to be trained by
the same set of observation sequence. Training process is done
using Baum Welch algorithm. Training a Hidden Markov
model is done in the following way:
Step 1: Initializing the Hidden Markov model's parameters H
0
and compute Probability of observations occurrence on the
given parameters.
Step 2: Observation sequence is given and the parameters
representing the Hidden Markov model H
k-1
estimate H
k
, k is
the number of iterations.
Step 3: If P(Ob| H
k
) ޓ P(Ob| H
k-1
). Repeat the same process
from Step 2. Otherwise, terminate the procedure as H
k-1
is
taken as optimal parameters
of the model.
Once training is completed, the joint probability of observing
the sequence O followed by a busy slot or an idle slot at
instant T + 1 is calculated. In other words, the joint
probabilities P(O, 1|H) and P(O,0|H) are calculated. The slot
occupancy at instant T + 1 is predicted according to decision
rule given by
If P(Ob, 1|H) P(Ob, 0|H) then Ob
T+1
= 1
If P(Ob, 1|H) < P(Ob, 0|H) then Ob
T+1
= 0
where Ob
T+1
is the predicted value.
After the route reply has been reached to the sender node.
Then it senses all of the channels that are available for the
transmission process and selection is done randomly from
them. Here, the channel which is predicted to be free at the
time of the transmission request is send to the channel selector
block of the node. After that the channel is searched and
selected. This is the starting step necessary for transmission
phase setup. Cognitive users have to search for the vacant
channel if the channel is requested by any of the primary user.
If any of the primary user gets activated at the time or in the
middle of transmission process, the model will predict the
available channel on the basis of the previous observations and
joint probabilities of the occurrence of previous observation
with the next state of the channel. This decreases the
possibilities of packet drop during the transmission and
increases the packet delivery ratio of the network.
V. S
IMULATION
R
ESULTS
MATLAB2014a is used to simulate the Prediction Model
Based Hybrid Routing protocol For Cognitive Radio Ad-hoc
Network. Simulation area is taken as 300×300 m
2
which
contains sixteen primary users distributed non-uniformly.
Source data rate varies from 100 Kb/s to 700 Kb/s.
Performance metrics used for evaluation are delivery ratio
and average delay. Average delay is a sum of transmission
delay, back off delay, queuing delay, switching delay.
Delivery ratio is the ratio between number of received packets
and number of transmitted packets.
Fig 2. Effect on delivery ratio in terms of number of PU
Fig.2 shows the comparison of delivery ratio achieved by
the prediction model based routing protocol with the OCHR in
the case where source data rate is fixed and number of primary
user’s are increasing. Source data rate is fixed to be 100Kb/s.
This is due the consideration of data rate in the path
optimization problem where we choose the path of low
network density from the optimal routes list.
Fig 3. Effect on average delay in terms of number of PU
Fig 3 shows the comparison of effect of increasing the
number of primary user’s on Average delay. Average delay
gets reduced due to reduction in the switching and
transmission time.
Fig 4. Effect on delivery ratio in terms of number of increase
in source Data rate
Fig.4 shows the comparison of delivery ratio achieved
when the source data rate is varied and number of primary
user’s are fixed. Better delivery ratio is achieved because of
increase in the achievable data rate of the link.
Fig 5.Effect on average delay in terms of varying source
Data rate
Fig.5 shows the comparison of effect of increasing the source
data rate on Average delay.
VI. CONCLUSION
In this work, we presented a method for prediction of
channel occupancy using the Hidden Markov Model for a
cognitive radio ad-hoc network. Proposed Prediction model
based Routing protocol outperforms the On Demand Hybrid
Routing Protocol for Cognitive Radio Ad-Hoc Networks in
the regime of low average delay and is more robust to several
effect such as increase no. of primary user. It considers the
history of the previously measured values in predicting the
future values. Hidden Markov Based predictor predicts the
channel with low probability of false alarm and lowers the
sensing time. Hidden Markov model based spectrum sensing
increases the achievable data rate of the link used for the
transmission. Performance of the routing protocol is measured
in terms of the Average delay in transmission and Packet
delivery ratio. Further analysis can be done taking into
account effects of sensing error and using the artificial
intelligence systems.
R
EFERENCES
[1] L. Rabiner, “A tutorial on hidden markov models and selected applica-
tions in speech recognition", Proceedings of the IEEE, vol. 77, no. 2,
pp.257-286, Feb 1989.
[2] F.C.Commission, “Fcc adopted rules for unlicensed use of television
white spaces, tech.rep.et docket no.04-186, second report and order and
memorandum opinion and order,” 2004.
[3] S.Haykin, “Cognitive radio: brain empowered wireless
communications,” IEEE Journal on selected areas in communications,
vol.23, no.2, pp.201-220, 2005.
[4] I.F.Akyildiz, W.Y.Lee, M.C.Vuran, and S.Mohanty, “NeXt
Generation/Dynamic Spectrum Access/Cognitive Radio Wireless
Networks: A Survey”, Elsevier Computer Networks Journal, September
2006.
[5] I. Akbar and W. Tranter, “Dynamic spectrum allocation in cognitive
radio using hidden markov models: Poisson distributed case,” in
SoutheastCon, Proceedings. IEEE, pp. 196 –201, march 2007.
[6] S. Yarkan and H. Arslan, “Binary time series approach to spectrum
prediction for cognitive radio network”,in vehicular technology
IEEE
66
th
conference, pp. 1563-1567, Sept 2007.
[7] K.chowdhury and M.felice, “SEARCH: A Routing Protocol for Mobile
Cognitive Radio Ad-hoc Networks”, computer communication, pp.1983-
1997, 2009.
[8] Chittabrata Ghosh, Carlos Cordeiro, M. Bhaskara Rao, Dharma P.
Agrawal,” Markov Chain Existence and Hidden Markov Models in
Spectrum Sensing”,PERCOM, Proceedings. IEEE 2009.
[9] Cesana, M., Cuomo,et.al., “Routing in Cognitive Radio Networks:
challenges and solutions”, Ad Hoc Networks, pp.228-248 ,2011.
[10] M. Bradonji and L. Lazos, “Graph-Based criteria for spectrum-aware
clustering in cognitive radio network,”Ad Hoc Networks, vol.10, no.
1,pp.75 – 9,2012.
[11] E. Chatziantoniou, B. Allen, V. Velisavljevic, “HMM-based
spectrumoccupancy predictor for energy efficient cognitive radio", in
International Symposium on Personal Indoor and Mobile
Communications (PIMRC), pp. 601-605, Sep. 2013.
[12] Zhe Chen, Nan Guo, Zhen Hu, and Robert C. Qiu, ”Channel State
Prediction in Cognitive Radio, PartII: Single-User Prediction”
SoutheastCon, Proceedings. IEEE,2014.
[13] Jason Jacob, B. R. Jose and Jimson Mathew, “Spectrum prediction in
cognitve radio networks: a bayesian approach,” NGMAST, April 2014.
[14] Jyu-Wei Wang and Ramzi Adriman, “Analysis of opportunistic
spectrum access in cognitive radio networks using hidden Markov model
with state prediction”, EURASIP Journal on Wireless Communications
and Networking,2015.
[15] Mahdi Zareei et.al., “On-demand Hybrid Routing for Cognitive Radio
Ad-Hoc Network”, IEEE Access, 2016.