Adaptive channels’ selection for hierarchical cluster based cognitive radio networks

Abstract

Cognitive Radio (CR) usually performs an analysis of the existing environment, enabling collection of all information on current spectrum use and available resources, to decide on its own transmission parameters to optimize communication performance. From practical point of view, the principal task of spectrum sensing is an efficient detection of different types of signals. This paper presents behavior of Sensing Client (SC) as a logical block of Supervisor (SP) in CR node, to obtain information about backup channels' ranking list. Methodology of creation, refreshing the ranking list of backup channels and change of used channel by SP are the main goals of this research. The ranking list of backup channels is based on spectrum sensing results collected by all nodes according to predefined sensing policy, enabling different sensing schemes like detector type, number of slots for sensing, sensing period etc. Channels with the highest positions on the ranking list are potential backup channels, used in case of jamming, interferences from legacy systems, decreasing link quality caused by nodes positions changes etc. To analyze behavior of SC, several simulation scenarios were developed. All tests were performed for CR network topology with presence of jammers and interferences from other users. Presented results are based on CORASMA simulator.

Full text

Authors: Paweł Skokowski, Krzysztof Malon, Jan M. Kelner, Jerzy Dołowski,
Jerzy Łopatka, Piotr Gajewski
Title: Adaptive channels’ selection for hierarchical cluster based cognitive
radio networks
Proceedings: Proceedings of ICSPCS-2014
Volume: –
Pages: 1-6
Conference: 8th International Conference on Signal Processing and Communication
System, ICSPCS-2014
Location: Gold Coast, QLD, Australia
Date: 15-17 Dec. 2014
DOI: 10.1109/ICSPCS.2014.7021123
INSPEC Accession Number: 14881824
Print ISBN: –
Publisher: IEEE
Original Source: http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=7021123
Copyright Notice
©2014 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this
material for advertising or promotional purposes or for creating new collective works for resale or
redistribution to servers or lists, or to reuse any copyrighted component of this work in other works
must be obtained from the IEEE.

Adaptive Channels’ Selection for Hierarchical Cluster
Based Cognitive Radio Networks
Pawel Skokowski, Krzysztof Malon, Jan M. Kelner, Jerzy Dolowski, Jerzy Lopatka, Piotr Gajewski
Institute of Telecommunications, Faculty of Electronics
Military University of Technology
Warsaw, Poland
{pskokowski, kmalon, jkelner, jdolowski, jlopatka,pgajewski}@wat.edu.pl
Abstract — Cognitive Radio (CR) usually performs an
analysis of the existing environment, enabling collection of all
information on current spectrum use and available resources, to
decide on its own transmission parameters to optimize
communication performance. From practical point of view, the
principal task of spectrum sensing is an efficient detection of
different types of signals. This paper presents behavior of
Sensing Client (SC) as a logical block of Supervisor (SP) in CR
node, to obtain information about backup channels' ranking list.
Methodology of creation, refreshing the ranking list of backup
channels and change of used channel by SP are the main goals of
this research. The ranking list of backup channels is based on
spectrum sensing results collected by all nodes according to
predefined sensing policy, enabling different sensing schemes like
detector type, number of slots for sensing, sensing period etc.
Channels with the highest positions on the ranking list are
potential backup channels, used in case of jamming, interferences
from legacy systems, decreasing link quality caused by nodes
positions changes etc. To analyze behavior of SC, several
simulation scenarios were developed. All tests were performed
for CR network topology with presence of jammers and
interferences from other users. Presented results are based on
CORASMA simulator.
Keywords — cognitive radio, spectrum sensing, detection,
sensing client, adaptive channels’ selection
I. I
NTRODUCTION
Cognitive Radio (CR) topic is one of the most investigated
one in the wireless communications field over the last few
years, and spectrum sensing is its key functionality [1][2]. The
first step in the cognitive cycle, is an analysis of the
environment, enabling CR collection of all information useful
for determination of current spectrum use and available
resources, to decide on own transmission parameters to
optimize communication performance. Spectrum sensing can
be determined as the basic tool for creation of modern
cognitive radio networks and it involves all activities that
allow CR to obtain situation awareness. From a practical point
of view, the principal task of spectrum sensing is signal
detection, because monitoring of spectrum occupancy allows
to use radio resources more efficiently and to modify network
behavior depending on radio conditions change. Sensing
algorithms are expected to be capable to detect the presence of
signals at very low signal to noise ratio (SNR) levels within a
short period of time. Moreover, it is necessary to develop
methods of channels’ selection which are robust to physical
impairments, parameter uncertainties, interferences and
intentional jamming.
This article is divided into 6 sections. The second section
presents short description of spectrum sensing methods to
introduce reader into discussed topic. The third one introduce
Sensing Block used in CORASMA [3] simulator, whereas the
forth one presents a concept of adaptive channels’ selection
for CR. Simulation results are presented in the fifth part of this
article. A brief summary at the end of the article presents
conclusions obtained during the simulation.
II. S
IGNAL
S
ENSING
From the classical detection theory it is known that the
optimum signal detector is the matched filter (MF). It provides
the best detection performance but it requires a precise
knowledge of the signal to be detected - a condition that in
practice is hard to satisfy. If the signal is unknown, the
optimum detector is the energy detector (ED), that estimates
the received energy in the band of interest and compares it to a
threshold that is related to the average noise power level. The
ED has a low computational complexity and is widely used
because it has a simple implementation, but its main
disadvantage is that it requires knowledge of the noise power
to properly set the detection threshold. This requirement is
often critical, especially for low SNR, where incorrect
determination of noise power can cause significant
performance losses. Moreover the ED cannot distinguish
between interference and useful signal [3][4][5].
The ED detector can be considered as a blind detection
algorithm, in the sense that it does not require any knowledge
of the signal to be detected, but it requires a knowledge of the
noise power, which depends on the environment properties
and can vary in time and the receiving node position.
Completely blind detection algorithms can be developed by
analyzing the autocovariance properties of the received signal.
These algorithms do not require any a priori knowledge, but
are based on the observation of some correlation properties in
the received signals. Typically these algorithms imply a high
computational complexity [6][7][8][9][10].
Previously described sensing methods, except the MF, are
detection algorithms. Their objective is to check the presence
of signals in the observed spectrum. If the goal of sensing is
signal classification, the cyclostationary analysis is a
technique that can be adopted, because from a statistical point

of view all signals used for communication nowadays are
cyclostationary, hence they are generated through specific
operations that contain inner periodicities. The cyclostationary
analysis is an extension of the traditional spectral analysis,
enabling detection of signals’ characteristic features, that can
be used to recognize the particular transmission and even
extract its parameters. These techniques enable separation
between signal and noise components. Unfortunately these
promising sensing methods have high computational and
implementation requirements [11][12][13].
The algorithms mentioned above are the most popular
algorithms proposed in the recent literature for CR
applications, and this is a reason why they were selected for
implementation in simulator and further investigation.
III. S
ENSING
B
LOCK
I
N
CORASMA
S
IMULATOR
CORASMA is the cognitive radio simulator, which was
developed, from 2010 to 2013, in the "Cognitive Radio for
Dynamic Spectrum Management" project conducted under the
auspices of European Defence Agency (EDA). This project
was realized by an international consortium, which included
the following companies and institutions: Thales
Communications from France (leader), Italy and Belgium,
Thales Defence Deutschland from Germany, SAAB from
Sweden, SELEX from Italy, Tekever from Portugal, and
Military University of Technology (MUT) from Poland.
CORASMA simulator was developed to evaluate the cognitive
solutions in tactical MANETS, and its more detailed
description can be found in [15].
Sensing Block (SB) as a part of the simulator was
developed to support situation awareness and spectrum
monitoring. Spectrum sensing algorithms selected for the
CORASMA simulator belong to two classes: the energy
detector (ED) and the eigenvalues based detectors (EVD).
These algorithms are general-purpose detectors: they can be
applied to any kind of transmissions and do not require
a’priori knowledge of any signal parameter. Additionally the
ED and the EVD based algorithms exhibit a very good trade-
off between detection rate and observation time. Moreover,
the SB of the CORASMA simulator implements two
additional algorithms that support the sensing functionalities:
long term noise estimation (LTNE) and signal to interference
plus noise (SINR) estimation.
The following algorithms have been implemented in the
SB of CORASMA simulator:
• ED - Energy Detector Sensing Algorithm.
• EVD - Eigenvalue-based Sensing Algorithms.
• LTNE - Long Term Noise estimation.
• SINR - estimation.
These algorithms are able to detect the energy or estimate
the noise, to verify the frequency occupancy, and they can be
used to satisfy the needs of the different cognitive solutions.
IV. A
DAPTIVE
C
HANNELS
’
S
ELECTION
A. Network topology
CORASMA simulator enables static and dynamic Ad-Hoc
networks analysis [15]. All information about simulation
parameters, nodes positions, traffic generators, available
frequency channels etc. are included in the XML scenario.
In the CORASMA simulator, networks has functional and
hierarchical structure. The whole network is divided into
smaller parts called clusters. In such network there are four
types of nodes:
• Cluster Head (CH), which is responsible for resources
management within its cluster,
• Gateway (GW), which provides communication
between neighboring clusters,
• Regular Node (RN).
External components, simulating other systems and
sources of interferences are called NcN (Non-cooperative
Nodes). Example network topology is presented in Fig. 1
Fig. 1. Example of network topology used for simulation.
B. Supervisor with Sensing Client
This section presents behavior of Supervisor (SP) as a
block of Cognitive Manager cooperating with SB. The
Supervisor module is implemented in each host and controls
the following cognitive modules: Clustering, Cluster Graph
Coloring and Sensing Block.
Functionality of creation and refreshing the ranking list of
backup channels and changing used data channel in CH’s
Supervisor are the main goals of performed research. Channels
with the highest positions on the ranking list are potential
backup channels in emergency situation like: jamming, high
interferences (including legacy systems/PUs), decreasing link
quality because of nodes positions changes etc.
During creation and refreshing ranking list, SP takes into
account the following coefficients:
• results from Sensing Block (SB),

• spectrum distance between actually used and
monitored channels,
• averaged and normalized power level for each
monitored channel from all regular nodes.
Monitored channels are selected from all available
channels, excluding own channel and channels used by
neighbor clusters. Sensing Client, as a part of SP, is
responsible for managing SB and can use different sensing
schemes/politics e.g.: detector selection, number of slots for
sensing, sensing cyclicality etc. All regular nodes perform
sensing periodically, with predefined sensing cyclicality and
send gathered results to the CH on request, depending on the
parameter sensing results interval. These results are valid until
parameter sensing results time of life elapse. The SC
cooperates with SB in each node.
To decide that emergency situation occurred, Supervisor in
CH takes into account a number of sensing responses from all
regular nodes in its cluster. If particular RN does not send
sensing results to the CH during predefined period, it is
suspected that this node is lost. The period is calculated as
follows: sensing results interval * active time factor. The
Activity check interval parameter defines how often each CH
compares the number of lost nodes and the total number of
nodes in the cluster. If the result of this comparison exceeds
the lost nodes threshold, SP in CH decides that the channel
change is necessary. The expected results of sensing in CH is
presented in TABLE I.
TABLE I. T
HE EXPECTED RESULTS OF SENSI NG IN
CH:
Channel
(colour) Node 1 ... Node N
--- timestamp_active … timestamp_active
1
RES11 ... RES1N
PD11 … PD1N
PL11 … PL1N
timestamp_sens … timestamp_sens
… … … …
M
RESM1 ... RESMN
PDM1 … PDMN
PLM1 … PLMN
timestamp_sens … timestamp_sens
M – number of free channels (colors)
N – number of actual sensing results from regular nodes
RESkn – detection result (0 – free channel, 1 – occupied channel) on k
channel k from node n
PDkn – probability of detection on channel k from node n
PLkn – power level on channel k from node n
C. Ranking list
Supervisor based on the results from cooperation work of
Sensing Client with Sensing Block updates permanently
ranking list of potentially available channels. The position of
the channel in the ranking list depends on the metrics values:
Fk and PLk. Fk is the k-channel fitness and is calculated using
formula (1):
NPLkwSDkcwOMkwFk
⋅
+
⋅
+
⋅
=
321
(1)
where:
• OMk – Occupancy Measure of channel k,
• SDkc – value of Spectrum Distance measure function
which depends on the spectrum distance between
channel k and actual used channel c,
• NPLk Normalized Power Level on channel k,
•
[
]
10321 ,,w,ww ∈
- weights for particular metric to
calculate fitness,
1321
=
+
+
www
.
Firstly, channels are sorted according to Fk values in
descending order (higher values of Fk denotes higher position
on the ranking list). In case of the same Fk values for some
channels, these channels should be additionally sorted
according to PLk values in ascending order (smaller values of
Fk denotes higher position on the ranking list). The highest
position on the ranking list is potentially the best backup
channel. Occupancy metric calculation is based on (2):
N
N)PDk_RESk_PDk_(RESk_
OMk
2
1100
+
⋅
−
⋅
=
(2)
where:
0
0
0
1
RESk_
PDki
PDk_
RESk_
i
∑
=
=
(3),
1
1
1
1
RESk_
PDki
PDk_
RESk_
i
∑
=
=
(4),
N
PLki
PLk
N
i
∑
=
=
1
(5)
• RESk_0 – number of “0” detection results on channel k
• RESk_1 – number of “1” detection results on channel k
• RESk_0 + RESk_1 = N
• PDk_0 – average probability of detection for “0”
detection results on channel k
• PDk_1 – average probability of detection for “1”
detection results on channel k
• PLk – average power level on channel k
For spectrum distance calculation the adjacent bands are
potentially less favorable because of transmitter spectral mask,
receiver sensitivity curve, co-site effects and wideband
jamming. Spectrum distance measure function is calculated
using (6):







+>
<
+≤≤
=
+
−
1for12
1for1
2
1
11for0
1
1
c k ,+ -
c-k , + -
ck c-,
SDkc
- k+c
- k +c
(6)

Example SDkc values for
[
]
20,1k
∈
and c = 8 are
presented in Fig. 2.
0 2 4 6 8 10 12 14 16 18 20
0
0.2
0.4
0.6
0.8
1
Spectrum distance measure function
Channel number
Spectral distance measure value
SDkc values
c - actual used channel
Fig. 2. Spectrum distance measure function.
Normalized power level (NPLk) is determined by (7):





<
≤≤
−
−
=
B'PLk
APLkB'
B'A
PLkA
NPLk
for1
for
(7)
where:
• C – constant threshold,
•







∧
=PLk
Amax
;







∧
=PLk
Bmin
,
•
CBB'
+
=
.
V. S
IMULATION
R
ESULTS
To validate proper work of the Sensing Client, different
simulation scenarios were created. Depending on the scenario
case there might be many different Non-cooperative Nodes
(NcNs) like:
• jammers: Fixed Frequency (FF), Frequency Hopping
(FH), chirp,
• legacy systems: TETRA and Private Mobile Radio
(PMR).
For both scenario 1 and 2, exactly the same network
topology is used and contains 5 regular nodes and 1 CH. All
nodes have fixed positions, that means mobility is not taken
into account. For these cases one cluster is created and node 6
is a CH. Initial data channel assigned to this cluster is 7
(301,225 MHz). Simulations parameters are as follows:
• sensing cyclicality equals to 1 s,
• sensing results interval equals to 2 s,
• sensing results time of life equals to 10 s,
• scenario duration equals to 180 s,
• La Fleche digital terrain model,
• large scale propagation effect is on.
• active time factor equals to 2,
• lost nodes threshold equals to 0,5,
• weight coefficients for fitness calculation: w1 = 0,6;
w2 = 0,2; w3 = 0,2.
Case 1 shows the functionality of creating and refreshing
the ranking list of channels. Control channel has the same id
as data channel (id = 7), and neighbor clusters channels are: 2
(237,525 MHz), 5 (285,25 MHz), 9 (318,25 MHz) and 10
(320,75 MHz).
Case 2 shows the functionality of changing used data
channel in emergency situations. Control channel has an id = 1
and neighbor clusters channels are: 5 (285,25 MHz) and 6
(287,75 MHz), 9 (318,25 MHz) and 11 (323,25 MHz).
The third scenario defines two clusters. Each cluster
contains 5 regular nodes. Initial data channels are set to 6 for
CH1 (287,75 MHz) and 3 (270,025 MHz) for CH2 and
control channel is set to 1 for both clusters. The scenario
duration equals to 120 s in each case.
A. Case 1
Scenario number 1 assumes that there are eight NcNs in
the network radio operational area. NcNs are defined as FF
Jammers, TETRA, PMR and CHIRP Jammer with parameters
presented in Fig. 3:
FF Jammer1 (node id = 1000, power = 47 dBm)
00:00:40 - 00:02:40
Channel 1
FF Jammer2 (node id = 1001, power = 47 dBm)
FF Jammer3 (node id = 1002, power = 47 dBm)
TETRA1 (node id = 1003, power = 32.5 dBm)
TETRA2 (node id = 1004, power = 32.5 dBm)
PMR1 (node id = 1005, power = 34 dBm, channel = 15)
PMR2 (node id = 1006, power = 34 dBm, channel = 15)
00:00:00 00:03:00
CHIRP Jammer (node id = 1007, power = 0 dBm, sweep time = 100 ms,
sweeping bandwidth = 1.25 MHz)
00:01:00 - 00:02:00
Channel 6
00:01:20 - 00:02:20
Channel 14
00:01:50 - 00:02:41
Channel 4
00:00:30 - 00:01:31
Channel 4
00:01:50 - 00:02:41
Channel 4
00:00:30 - 00:01:31
Channel 4
00:02:15 - 00:02:3000:00:55 - 00:01:05
00:01:30 - 00:01:5000:00:30 - 00:00:40
00:02:25 - 00:02:4500:01:10 - 00:01:25
00:00:45 - 00:00:50 00:01:55 - 00:02:05
00:00:50 - 00:02:20
Channel 12
Fig. 3. Non-cooperative Nodes parameters and their activity in case 1.
NcNs are activated on selected channels with different
power levels, time and distances from CORASMA nodes. In
this case created ranking list of monitored channels should
change during the simulation time and channels used by NCNs
should change their positions on the ranking list. Topology of
the simulated network is presented in Fig. 1.
Results in Fig. 4 show the ranking list of channels with
expected changes.

Ranking list of channels for CH6
Channel number
Simulation time [s]
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
20
40
60
80
100
120
140
160
180
-2
0
2
4
6
8
10
Fig. 4. Simualtion results with ranking list of channels for case 1.
The graph presents:
• bottom X axis – channel id number,
• Y axis – simulation time (top - down),
• color bar – position on the ranking list according to the
following colors:
o black – own data and hello channel,
o white – channels used by neighbor clusters,
o grey – control channel,
o blue – channel for which there are no actual sensing
results,
o green scale – three channels with the highest
positions on ranking list,
o red scale – rest of monitored channels.
This scenario was prepared to perform tests in case of
“high interferences” presence. Rank positions for jammed and
interfered channels changes respectively to different NcNs
activities, and after NcNs deactivation, the ranking list of
channels returns to the initial order.
B. Case 2
Scenario number 2 assumes that there is one NcN in the
network radio operational area.
Fig. 5. Network topology for simulation case 2.
Network topology is presented in Fig. 5 and NcN is
defined as a FF Jammer with parameters presented in Fig. 6:
Fig. 6. Non-cooperative Nodes parameters and activity for case 2
Results in Fig. 7 show a ranking list of channels with
expected changes. One can see that NcN activation on used
data channel (id=7) causes frequency/channel change. New
data channel is set to 2 because it is selected from the channel
ranking list as the best one.
Ranking list of channels for CH6
Channel number
Simulation time [s]
1 2 3 4 5 6 7 8 9 10 11 12 13
20
40
60
80
100
120
140
160
-4
-2
0
2
4
6
Fig. 7. Simualtion results with ranking list of channels for case 2.
C. Case 3
Scenario number 3 assumes that the radio network consist
of two clusters. Topology of the simulated network is
presented in Fig. 8.This scenario is prepared to show behavior
of CH’s SPs (channels ranking list refresh and used data
channels change) including influences not only from NcNs but
also from neighboring clusters.
Fig. 8. Network topology for simaultion case 3
Dynamic change of the used data channel in one cluster is
noticed in the neighborhood CHs. There are two NcNs in the
network radio operational area. NcNs are defined as a FF
Jammers with parameters presented in Fig. 9.
Results in Fig. 10 and Fig. 11 show ranking list of
channels for both clusters. Jammer1 (id = 1000) activation in
30s on channel with id = 6 causes CH1 used data channel

Fig. 9. Non-cooperative Nodes parameters and activity for case 3.
jamming and consequently this cluster change channel to 14
(first position on ranking list) as a new one.
This change is visible in CH1 (as a new used data channel
– black color in Fig. 10) and CH2 (as a new neighbor’s cluster
channel – white color in Fig. 11) respectively. Similar
situation occurred after Jammer2 (id = 1001) activation in 50s.
Analysis of this event shall be performed analogically as
described above. More interesting case one can see in 80s
(both NcNs are active on new data channels). In this moment
each cluster uses channel 2, as the best one in the ranking list.
CH1 changed used data channel faster than CH2. This event is
noticed by CH2, which modifies channels ranking list and
selects new best channel with id = 3.
Ranking list of channels for CH1
Channel number
Simulation time [s]
1 2 3 4 5 6 7 8 9 10 11 12 13
30
40
50
60
70
80
90
100
110
-4
-2
0
2
4
6
8
10
Fig. 10. Simualtion results with ranking list of channels for case 3 (CH1)
Ranking list of channels for CH2
Channel number
Simulation time [s]
1 2 3 4 5 6 7 8 9 10 1 1 12 13
40
60
80
100
120 -4
-2
0
2
4
6
8
10
Fig. 11. Simualtion results with ranking list of channels for case 3 (CH2)
VI. C
ONCLUSIONS
This paper presents a novel approach to adaptive selection
of backup channels, for cognitive MANETs working in harsh
conditions. This functionality is realized by Supervisor with
Sensing Client, and the idea of their operation is to prepare a
list of potential backup channels and manage the channel
change in emergency situations. Capabilities of developed
methods enable to fulfil different requirements by tuning
parameters defining behavior of this blocks. In order to
validate its operation, several tests were performed. Achieved
tests results for all prepared scenarios (defining different radio
environment conditions) show process of creation and
refreshing the ranking list of channels (enabling
electromagnetic situation awareness) and adaptive changes of
used data channel. Both SC and SP methods work properly. It
should be also noted that every sensing process and results
exchange decrease number of available resources for user’s
transmissions, and final decision on these parameters selection
for cognitive cycle is a compromise between current
electromagnetic situation awareness and desired throughput
and required QoS.
A
CKNOWLEDGMENT
This research work was carried out in the framework of the
CORASMA EDA Project B-0781-IAP4-GC.
R
EFERENCES
[1] S. Haykin, "Cognitive radio: brain-empowered wireless
communications," IEEE Journal on Sel. Areas in Commun., vol.23,
no.2, pp. 201- 220, Feb. 2005.
[2] F. R. Yu, "Cognitive radio mobile ad-hoc networks". New York:
Springer, 2011.
[3] “CORASMA Project (Contract N° B-781-IAP4-GC) COgnitive RAdio
for dynamic Spectrum MAnagement,” D3.2.5: Final report for Task
T3.2. Sensing.
[4] J. G. Proakis, Digital communications, McGraw-Hill, 2001, 4th edition.
[5] H. L. van Trees, Detection, estimation and modulation theory – Part I:
Detection, Estimation and Linear Modulation Theory. New York,
U.S.A., Wiley, 2001.
[6] H. Urkowitz, “Energy detection of unknown deterministic signals”,
Proc. IEEE, vol 55, no. 4, pp. 253-531, April 1967.
[7] A. Mariani, A. Giorgetti, and M. Chiani, “Effects of noise power
estimation on energy detection for cognitive radio applications,” IEEE
Trans. Commun., vol. 59, no. 12, pp. 3410–3420, Dec. 2011.
[8] Y. Zeng, Y. C. Liang, “Eigenvalue Based Spectrum Sensing Algorithms
for Cognitive Radio”, IEEE Trans. and Commun., vol. 57, no. 6, pp.
1784-1793, 2009.
[9] F. Penna, R. Garello, D. Figlioli, M. Spirito, “Exact Non-asymptotic
Threshold for Eigenvalue-based Spectrum Sensing”, in Proc. Int. Conf.
on Cognitive Radio Oriented Wireless Network and Commun., 2009.
[10] M. Chiani, M. Z. Win, “Estimating the Number of Signal Observed by
Multiple Sensor”, IEEE Int. Workshop on Cognitive Inf. Process. (CIP
2010), June 2010.
[11] W. A. III Brown and H. H. Loomis, “Digital implementations of spectral
correlation analyzers,” IEEE Trans. on Signal Process., vol. 41, no. 2,
pp. 703-720, Feb 1993.
[12] W. A. Gardner. Introduction to Random Processes with Applications to
Signals and Systems. McGraw-Hill, 2nd ed. edition, 1990.
[13] R. S. Roberts, W. A. Brown, and H. H. Loomis, “Computationally
efficient algorithms for cyclic spectral analysis,” IEEE Signal Process.
Mag., vol. 8, no. 2, pp. 38-49, Apr 1991.
[14] K. Malon, P. Skokowski, P. Marszalek, J. M. Kelner, and J. Lopatka,
“Cognitive Manager for Hierarchical Cluster Networks Based on Multi-
Stage Machine Method” in IEEE Military Communications Conference,
MILCOM-2014, Baltimore, USA, 2014, article after review - accepted.
[15] L. Rose, R. Massin, L. Vijayandran, M. Mérouane, and C. J. Le Martret,
“CORASMA program on cognitive radio for tactical networks: High
fidelity simulator and first results on dynamic frequency allocation,” in
IEEE Military Communications Conference, MILCOM-2013, San
Diego, USA, 2013, pp. 360–368.