Mark Weichold Mounir Hamdi Muhammad Zeeshan Shakir Mohamed Abdallah George K Karagiannidis Muhammad Ismail (Eds.) 156 Cognitive Radio Oriented Wireless Networks 10th International Conference, CROWNCOM 2015 Doha, Qatar, April 21–23, 2015 Revised Selected Papers 123 Lecture Notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering 156 Editorial Board Ozgur Akan Middle East Technical University, Ankara, Turkey Paolo Bellavista University of Bologna, Bologna, Italy Jiannong Cao Hong Kong Polytechnic University, Hong Kong, Hong Kong Falko Dressler University of Erlangen, Erlangen, Germany Domenico Ferrari Università Cattolica Piacenza, Piacenza, Italy Mario Gerla UCLA, Los Angels, USA Hisashi Kobayashi Princeton University, Princeton, USA Sergio Palazzo University of Catania, Catania, Italy Sartaj Sahni University of Florida, Florida, USA Xuemin (Sherman) Shen University of Waterloo, Waterloo, Canada Mircea Stan University of Virginia, Charlottesville, USA Jia Xiaohua City University of Hong Kong, Kowloon, Hong Kong Albert Zomaya University of Sydney, Sydney, Australia Geoffrey Coulson Lancaster University, Lancaster, UK More information about this series at http://www.springer.com/series/8197 Mark Weichold Mounir Hamdi Muhammad Zeeshan Shakir Mohamed Abdallah George K Karagiannidis Muhammad Ismail (Eds.) • • • Cognitive Radio Oriented Wireless Networks 10th International Conference, CROWNCOM 2015 Doha, Qatar, April 21–23, 2015 Revised Selected Papers 123 Editors Mark Weichold Texas A&M University at Qatar Doha Qatar Mohamed Abdallah Texas A&M University at Qatar Doha Qatar Mounir Hamdi Hamad Bin Khalifa University Doha Qatar George K Karagiannidis Aristotle University of Thessaloniki Greece and Khalifa University United Arab Emirates Muhammad Zeeshan Shakir Texas A&M University of Qatar Doha Qatar Muhammad Ismail Texas A&M University at Qatar Doha Qatar ISSN 1867-8211 ISSN 1867-822X (electronic) Lecture Notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering ISBN 978-3-319-24539-3 ISBN 978-3-319-24540-9 (eBook) DOI 10.1007/978-3-319-24540-9 Library of Congress Control Number: 2015950861 Springer Cham Heidelberg New York Dordrecht London © Institute for Computer Science, Social Informatics and Telecommunications Engineering 2015 This work is subject to copyright All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed The use of general descriptive names, registered names, trademarks, service marks, etc in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication Neither the publisher nor the authors or the editors give a warranty, express or implied, with respect to the material contained herein or for any errors or omissions that may have been made Printed on acid-free paper Springer International Publishing AG Switzerland is part of Springer Science+Business Media (www.springer.com) CROWNCOM 2015 Preface 2015 marks the 10th anniversary of the International Conference on Cognitive Radio-Oriented Wireless Networks (Crowncom) Crowncom 2015 was jointly hosted by Texas A&M University at Qatar and Hamad Bin Khalifa University in Doha, Qatar, April 21–23, 2015 The event was a special occasion to look back at the contribution of Crowncom toward the advancements of cognitive radio technology since its inaugural conference in 2006 in Mykonos, Greece, as well as to look forward to the decades ahead, the ways that cognitive radio technology would like to evolve, and the ways its emerging applications and services can ensure everyone is connected everywhere Evolution of cognitive radio technology pertaining to 5G networks was the theme of the 2015 edition of Crowncom The technical program of Crowncom 2015 was structured to bring academic and industrial researchers together to identify and discuss recent developments, highlight the challenging gaps, and forecast the future trends of cognitive radio technology toward its integration with the 5G network deployment One of the key topics of the conference was cognition and self-organization in the future networks, which are now widely considered as a striking solution to cope with the future ever-increasing spectra demands Going beyond the theoretical development and investigation, further practical advances and standardization developments in this technology could provide potential dynamic solutions to cellular traffic congestion problems by exploiting new and underutilized spectral resources One of the challenging issues that Crowncom 2015 brought forward was to facilitate the heterogeneous demands of users in heterogeneous-type environments — particularly in the 5G network paradigm, where the networks are anticipated to incorporate the provision of high-quality services to users with extremely low delays and consider these requirements without explicit demand from users Machine-type communications and Internet of Everything are now representing emerging use cases of such ubiquitous connectivity over limited spectra Crowncom 2015 strongly advocated that the research community, practitioners, standardization bodies, and developers should collaborate on their research efforts to further align the development initiatives toward the evolution of emerging highly dynamic spectrum access frameworks The biggest challenge is to design unified crosslayer new network architectures for successful aggregation of licensed and unlicensed spectra, addressing the spectrum scarcity problem for ubiquitous connectivity and preparing the ground for “The Age of the ZetaByte.” Crowncom 2015 received a large number of submissions, and it was a challenging task to select the best and most relevant meritorious papers to reflect the theme of the 2015 edition of Crowncom All submissions received high-quality reviews from the Technical Program Committee (TPC) members/reviewers and eventually 66 technical papers (with an acceptance ratio of 56 %) were selected for the technical program of the VI CROWNCOM 2015 conference The technical program of Crowncom 2015 is the result of the tireless efforts of 14 track chairs, and more than 200 TPC members and reviewers We are grateful to the track chairs for handling the paper review process and their outstanding efforts, and to the reviewers/TPC for their high-quality evaluations We offer our sincere gratitude to the Advisory Committee, local Organizing Committee (especially colleagues at Texas A&M University at Qatar), and the Steering Committee members for their insightful guidance We would like to acknowledge the invaluable support from European Alliance for Innovation and the Qatar National Research Fund for the success of Crowncom 2015 2015 Mark Weichold Mounir Hamdi Muhammad Zeeshan Shakir Mohamed Abdallah George K Karagiannidis Muhammad Ismail Organization General Chair Mark Weichold Mounir Hamdi Texas A&M University at Qatar, Qatar Hamad Bin Khalifa University, Qatar Technical Program Chair Muhammad Zeeshan Shakir Mohamed Abdallah George K Karagiannidis Texas A&M University at Qatar, Qatar Texas A&M University at Qatar, Qatar Aristotle University of Thessaloniki, Greece, and Khalifa University, UAE Advisory Board Athanasios V Vasilakos Khalid A Qaraqe Jinsong Wu David Grace Naofal Al-Dhahir Kaushik Chowdhury Kuwait University, Kuwait Texas A&M University at Qatar, Qatar Bell Labs, China University of York, UK University of Texas, Dallas, USA Northeastern University, USA Special Session Chair Alhussein Abouzeid Rensselaer Polytechnic Institute, USA Panel Chair Maziar Nekovee Samsung, UK Publication Chair Muhammad Ismail Texas A&M University at Qatar, Qatar Tutorial Chair Mohamed Nafie Nile University, Egypt Exhibitions and Demos Chair Majid Butt Qatar University, Qatar VIII Organization Web Chair İslam Şafak Bayram Qatar Environment and Energy Research Institute, Qatar Local Arrangements Carol Nader Mohamed Kashef Texas A&M University at Qatar, Qatar Texas A&M University at Qatar, Qatar Track Chairs Track 1: Dynamic Spectrum Access/Management Mohammad Shaqfeh Texas A&M University at Qatar, Qatar Track 2: Networking Protocols for CR Tamer Khattab Amr Mohamed Qatar University, Qatar Qatar University, Qatar Track 3: Modeling and Theory Zouheir Rezki Syed Ali Raza Zaidi King Abdullah University of Science and Technology, Saudi Arabia University of Leeds, UK Track 4: HW Architecture and Implementations Ahmed El-Tawil Fadi Kurdahi University of California, Irvine, USA University of California, Irvine, USA Track 5: Next Generation of Cognitive Networks Muhammad Ali Imran Richard Demo Souza CCSR/5G Innovation Centre University of Surrey, UK Federal University of Technology - Paraná (UTFPR), Curitiba - PR - Brazil Track 6: Standards and Business Models Stanislav Fillin Stephen J Shellhammer Markus Dominik Mueck National Institute of Information and Communications Technology (NICT), Japan Qualcomm Technologies, Inc., USA INTEL Mobile Communications, Germany Track 7: Emerging Applications for Cognitive Networks Octavia A Dobre Hai Lin Memorial University, Canada Osaka Prefecture University, Japan Contents Dynamic Spectrum Access/Management Fractional Low Order Cyclostationary-Based Spectrum Sensing in Cognitive Radio Networks Hadi Hashemi, Sina Mohammadi Fard, Abbas Taherpour, and Tamer Khattab Achievable Rate of Multi-relay Cognitive Radio MIMO Channel with Space Alignment Lokman Sboui, Hakim Ghazzai, Zouheir Rezki, and Mohamed-Slim Alouini Effective Capacity and Delay Optimization in Cognitive Radio Networks Mai Abdel-Malek, Karim Seddik, Tamer ElBatt, and Yahya Mohasseb 17 30 Auction Based Joint Resource Allocation with Flexible User Request in Cognitive Radio Networks Wei Zhou, Tao Jing, Yan Huo, Jin Qian, and Zhen Li 43 Two-Stage Multiuser Access in 5G Cellular Using Massive MIMO and Beamforming Hussein Seleem, Abdullhameed Alsanie, and Ahmed Iyanda Sulyman 54 Detection of Temporally Correlated Primary User Signal with Multiple Antennas Hadi Hashemi, Sina Mohammadi Fard, Abbas Taherpour, Saeid Sedighi, and Tamer Khattab Non-uniform Quantized Distributed Sensing in Practical Wireless Rayleigh Fading Channel Sina Mohammadi Fard, Hadi Hashemi, Abbas Taherpour, and Tamer Khattab Downlink Scheduling and Power Allocation in Cognitive Femtocell Networks Hesham M Elmaghraby, Dongrun Qin, and Zhi Ding 66 78 92 CSS Using TAS for CRS 781 [5] has been recently proposed as an alternative framework to realize a cognitive radio network In cooperative spectrum sharing (CSS), secondary transmitter relays the data of primary system in order to get spectrum access over licensed band of primary user In this architecture, primary and secondary system consists of transmitter receiver pair known as primary transmitter (PT) - primary receiver (PR) and secondary transmitter (ST) - secondary receiver (SR) respectively, are allowed to coexist in the same frequency band with the assurance that secondary system will improve the performance of primary system Substantial amount of literature has demonstrated the performance of conventional CSS protocol under decode and forward relaying [6]-[7] In these schemes, whenever the instantaneous transmission rate of primary system drops below the target rate, it seeks cooperation from the neighbouring terminals which may help it in achieving the target rate Secondary transmitter (ST) “disguises” itself as a relay and collaborates with primary system by forwarding its data to the destination Primary system returns the favour by helping the secondary system with spectrum access However, the performance of CSS protocols is limited by the interference tolerable at PR from ST Moreover, most of these schemes have been confined to single antenna system Recently, some work has also been proposed where multiple antenna CR system have been used to enhance the performance of both systems [8]-[9] The authors in [8] proposed a scheme with multiple antennas at ST node which utilizes zero-forcing precoding technique in order to cancel the interference at PR caused due to presence of cognitive system But the application of this precoding technique requires perfect transmit channel state information (CSI) at ST Assuming that perfect transmit CSI is available at ST may not be practically feasible in the case of fading environment Moreover, in [8], as ST is working as an amplify and forward relay, therefore while forwarding the data from PT to PR, it will amplify both the required signal as well as noise received from PT In [9], authors have proposed a CSS scheme in which ST is equipped with two antennas Both the antennas receive primary’s data which is decoded at ST and then forward this data by selecting one of the two antennas randomly This will improve the performance of primary system when compared to conventional CSS scheme because of increase in probability of successful decoding of primary’s data However it still suffers from the drawback on the amount of interference at PR due to presence of secondary system which is same as conventional CSS system In this paper, we have proposed a transmit antenna selection [10] based scheme with multiple antennas at ST node which can alleviate the drawbacks of [6]-[7], [9] Moreover, unlike [8], proposed scheme doesn’t require perfect CSI, it just requires partial CSI feedback to select the best among the set of antennas at ST (that maximizes the post processing SNR at PR) This reduces the transmitter complexity and lowers the feedback bandwidth while preserving the gains from diversity [11]-[12] In the proposed scheme, once primary and secondary system enter into CSS, PT broadcasts its data in half of the overall time slot (represented as phase 1) which is received by all the present nodes i.e PR, ST and SR After receiving primary’s data ST will try to decode it In the 782 N Jain et al remaining half of the time slot (phase 2), ST chooses the antenna having larger instantaneous gain between ST and PR for primary’s data transmission and secondary’s data is transmitted via other antenna which has comparatively lower gain as shown in Fig Finally, the data received in both the phases, is decoded using maximum rate combining (MRC) at PR However, if ST fails to decode primary’s data, it will remain silent in phase This technique is advantageous in two ways; first, we can improve the performance of primary system by reducing the interference caused due to secondary’s data at PR Second, the performance of secondary system is unaffected because of interference cancellation at SR Moreover, when ST works as a pure relay and transfer only primary’s data, in such a scenario, ST can also be seen as a selection combiner [13] Consequently, PR will receive its signal from a selection combiner and a direct link (PT-PR) The performance of primary as well as secondary system has been analysed by deriving the closed form expressions for outage probability The results demonstrate the considerable improvement in the performance of primary system along with spectrum access for secondary system Throughout this paper, a complex Gaussian random variable (RV) Z with mean μ and variance σ is denoted as Z ∼ CN (μ, σ ) An exponentially distributed RV X with mean λ1 is denoted as X ∼ ε(λ) ∼ is used to indicate “has the distribution of” and i.i.d is used to represent independent and identically distributed The transpose of a matrix A is denoted by AT fX (x) symbolizes the probability density function (PDF) of RV X and fX,Y (x, y) symbolizes the joint PDF of RVs X and Y Moreover, FX (x) symbolizes the cumulative distribution function (CDF) of RV X and FX,Y (x, y) symbolizes the joint CDF of RVs X and Y The rest of the paper is organized as follows Section describes the proposed system model and obtains the analytical results for outage probability of primary and secondary systems Section discusses the simulation results and finally section concludes the paper Model Description with Performance Analysis 2.1 System Model The primary and secondary system consists of transmitter receiver pair known as PT-PR and ST-SR respectively We have considered multiple antennas at ST, named as ST1 and ST2.1 Channels between the links are modeled as Rayleigh flat fading channels and the channel coefficients between PT-PR, PT-SR, PTST(1), PT-ST(2), ST(1)-PR, ST(2)-PR, ST(1)-SR, ST(2)-SR is h1 , h2 , h3 , h4 , h5 , h6 , h7 , h8 respectively Here, hi ∼ CN (0, d−v i ) where, v is the path loss component and di is the normalized distance between the corresponding link The normalization is done with respect to the distance between PT-PR link there2 fore, d1 = The instantaneous gain of each channel is given as γi = |hi | where, γi ∼ ε(dvi ) For ease of analysis, we have assumed that ST is equipped with two antennas, however the results obtained can be easily extrapolated to scenarios where ST is equipped with multiple (>2) antennas CSS Using TAS for CRS 783 Fig Transmission Phases 2.2 System Equations In transmission phase 1, PT broadcasts primary signal i.e xp which is received by all the nodes Therefore, signal received at PR is given as (1) yP R = Pp xp h1 + n11 (1) where, Pp is the power assigned to PT and nij ∼ CN (0, σ ) is the AWGN in ith phase of transmission at j th receiver and j =1,2,3 corresponds to PR, SR, ST respectively The signal received at SR in phase is given by (1) ySR = Pp xp h2 + n12 (2) Since ST is equipped with two antennas, hence the signal received at ST can be given as (1) h yST Pp xp + n13 (3) (2) = h4 yST In transmission phase 2, ST decodes the primary signal (i.e xp ) and transmits it along with its own signal (i.e xs ) As ST has two antennas, in order to reduce interference at PR, it will transmit xp and xs from the antenna which provides maximum and minimum instantaneous gain between ST-PR respectively Therefore, signal received at PR in phase is given by (2) yP R = hmax hmin z + n21 (4) h5 if γ5 > γ6 h6 if γ5 > γ6 where, hmax = ,h = , h6 if γ5 ≤ γ6 h5 if γ5 ≤ γ6 √ T z= αPs xp (1 − α)Ps xs , α and (1 − α) is the fraction of power provided by the secondary transmitter to transmit primary signal and secondary signal 784 N Jain et al respectively Therefore the signal received at PR in the both phases can be written as (1) Pp h1 yP R = √ (2) αPs hmax yP R (1 − α)Ps hmin xp n + 11 xs n21 (5) Now, the signal received at SR in phase is given by (2) ySR = [h7 h8 ]z + n22 (6) √ T where, z = αPs xp (1 − α)Ps xs Using (2), SR will estimate the primary signal (i.e xˆp ) which helps in cancelling the xp signal received in phase and hence the overall signal received at SR after applying interference cancellation is given as (7) ySR = (1 − α) Ps h8 xs + n22 2.3 Outage Probability of Primary System Outage at primary system occurs when system fails to achieve the target transmission rate (Rpt ) There are two such cases: In first case, outage occurs if ST is unable to decode the primary signal in phase and along with this, the link between PT-PR also fails to achieve Rpt , or in second case, outage occur if ST successfully decodes xp but still overall rate achieved at PR is less than Rpt Therefore, the expression for outage probability at primary system is given as PR Pout = P [R11 < Rpt ]P [R13 < Rpt ] + P [R13 > Rpt ]P [RM RC < Rpt ] (8) where, R11 is the transmission rate achieved in phase between PT-PR link, R13 is the transmission rate achieved between PT-ST in phase and RM RC is the rate achieved at PR after applying MRC of both transmission phases Solving for (8), Pp γ1 R11 = log2 + (9) σ The factor phases is due to the fact that the whole transmission is divided into two P [R11 < Rpt ] = P γ1 < σ2 ρ −σ ρ = − e Pp Pp (10) as, ρ = 22Rpt − 1, γ1 ∼ ε(1) R13 = Pp γ4 Pp γ3 log2 + + 2 σ σ and P [R13 < Rpt ] = P γ3 + γ4 < σ2 ρ Pp (11) (12) CSS Using TAS for CRS 785 We assume that the distances between the antennas at ST is negligible as compare to distance between the nodes, hence d3 = d4 , d5 = d6 , d7 = d8 Therefore, γ3 and γ4 are i.i.d and hence fγ3 ,γ4 (γ3 , γ4 ) = fγ3 (γ3 )fγ4 (γ4 ) where, v dv3 e−d3 γ3 γ3 > 0 otherwise fγ3 = Therefore, σ2 ρ Pp P [R13 < Rpt ] = σ2 ρ Pp −γ4 0 =1− 1+ fγ3 ,γ4 (γ3 , γ4 )dγ3 dγ4 σ ρ v − σP2pρ dv3 d e Pp (13) Moreover, P [R13 > Rpt ] = 1+ σ ρ v − σP2pρ dv3 d e Pp (14) The rate at PR after MRC is obtained as RMRC = where, SNRMRC = min(γ5 , γ6 ) Therefore, P p γ1 σ2 + log2 (1 + SNRMRC ) αPs γmax (1−α)Ps γmin +σ , (15) γmax = max(γ5 , γ6 ), γmin = Pp γ1 αPs γmax + Rpt]P [RSR2 > Rst] Pout (21) where, R12 is the transmission rate achieved between PT-SR link in phase 1, R13 is the transmission rate achieved at ST in phase (given in (14)) and R2SR is the rate achieved at SR in phase Solving for (21), R12 = Pp γ2 log2 + 2 σ Therefore, (22) v P [R12 > Rpt ] = P γ2 > Moreover, R2SR = Ps (1 − α)γ7 log2 + σ2 Therefore, P [R2SR > Rst ] = P γ7 > d ρσ ρσ − = e Pp Pp (23) (24) dv ρs σ ρs σ = e− Ps (1−α) Ps (1 − α) (25) where, ρs = 22Rst − After substituting (23), (14) and (25) in (21), we get SR =1− Pout 1+ σ ρ v − σP2pρ dv3 d e Pp − e dv ρσ Pp dv ρs σ e− Ps (1−α) (26) CSS Using TAS for CRS 787 Simulation Results and Discussion In this section, we have discussed the analytical and simulation results for outage probability We have compared our results with the scheme in [9], where they randomly pick an antenna at ST for transmission Fig shows the simulation model of the proposed scheme, in which for the ease of analysis all nodes are assumed to be collinear The value of d (distance between PT-ST) is considered to be 0.5 and 0.8 The target rate chosen for primary and secondary system is P i.e Rpt = Rst = 1, and we have considered σp2 = 5dB Fig System Model 10 Sim d=0.5, α=0.5 Th d=0.5, α=0.5 Sim d=0.5, α=0.7 Th d=0.5, α=0.7 Sim d=0.5, α=1 Th d=0.5, α=1 Sim d=0.8, α=0.5 Th d=0.8, α=0.5 Sim d=0.8, α=0.7 Th d=0.8, α=0.7 Sim d=0.8, α=1 Th d=0.8, α=1 [7] d=0.5, α=0.5 [7] d=0.5, α=0.7 [7] d=0.8, α=0.5 [7] d=0.8, α=0.7 Direct −1 10 −2 10 −3 10 10 15 20 25 Fig Outage Probability of Primary System Fig and Fig shows the outage probability of primary and secondary s system respectively with respect to P σ From the plots it is quite obvious, that the outage probability of both primary as well as secondary system is continuously decreasing with the increase in power at secondary transmitter However this decrement gradually reduces after 10dB because the outage probability also 788 N Jain et al Outage Probability 10 Sim d=0.5, α=0.5 Th d=0.5, α=0.5 Sim d=0.5, α=0.7 Th d=0.5, α=0.7 Sim d=0.8, α=0.5 Th d=0.8, α=0.5 Sim d=0.8, α=0.7 Th d=0.8, α=0.7 [9] d=0.5, α=0.5 [9] d=0.5, α=0.7 [9] d=0.8, α=0.5 [9] d=0.8, α=0.7 −1 10 −2 10 −3 10 10 15 20 25 Ps/σ2 (dB) Fig Outage Probability of Secondary System depends on the successful decoding of primary’s data at ST in phase (from (8 and 21)) The results are shown for two different values of α i.e 0.5 and 0.7 By transmitting xs from channel having less instantaneous gain, interference level at PR get reduced which results in considerable improvement in the performance s of primary system (approximately 10 times at d = 0.5 and α = 0.7 for P σ2 = 5dB) compared to [9] Even when half of the power of ST (α = 0.5) is allocated to secondary signal, the performance of proposed scheme is still far better than that of [9] with an improvement of approximately times It is also obvious from Fig that notwithstanding the improvement in the performance of primary system, we are still able to retain the performance of secondary system as in [9] Furthermore, we also demonstrate the results for the case wherein ST acts as a pure relay (α = 1) i.e it is transmitting only primary’s data with the channel having larger instantaneous gain For such scenario the proposed scheme works as a selection combiner in phase Conclusion In this paper, two phase cooperative spectrum sharing scheme with decode and forward relay at secondary system has been proposed The proposed technique utilizes transmit antenna selection scheme at secondary transmitter in order to reduce interference at primary receiver due to presence of secondary signal The perfect agreement between the simulated results and the analytically obtained closed form expression for outage probability validated theoretical analysis presented in the paper Acknowledgments Authors would like to thank Dr Sanjit Kaul for helping us in deriving closed form expression for outage probability of primary system CSS Using TAS for CRS 789 References Haykin, S.: Cognitive radio: brain-empowered wireless communications IEEE Journal on Selected Areas in Communications 23(2), 201–220 (2005) Akyildiz, I.F., Lee, W.-Y., Vuran, M.C., Mohanty, S.: Next generation/dynamic spectrum access/cognitive radio wireless networks: A survey Computer Network Journal (ELSEVIER) 50, 2127–2159 (2006) Zou, Y., Yao, Y.-D., Zheng, B.: Cognitive transmissions with multiple relays in cognitive radio networks IEEE Transactions on Wireless Communications 10(2), 648–659 (2011) Tachwali, Y., Basma, F., Refai, H.: Cognitive radio architecture for rapidly deployable heterogeneous wireless networks IEEE Transactions on Consumer Electronics 56(3), 1426–1432 (2010) Nosratinia, A., Hunter, T., Hedayat, A.: Cooperative communication in wireless networks IEEE Communications Magazine 42(10), 74–80 (2004) Han, Y., Pandharipande, A., Ting, S.H.: Cooperative decode-and-forward relaying for secondary spectrum access IEEE Transactions on Wireless Communications 8(10), 4945–4950 (2009) Bohara, V.A., Ting, S.H.: Measurement results for cognitive spectrum sharing based on cooperative relaying IEEE Transactions on Wireless Communications 10(7), 2052–2057 (2011) Manna, R., Louie, R.H., Li, Y., Vucetic, B.: Cooperative spectrum sharing in cognitive radio networks with multiple antennas IEEE Transactions on Signal Processing 59(11), 5509–5522 (2011) Vashistha, A., Sharma, S., Bohara, V.: Outage & diversity analysis of cooperative spectrum sharing protocol with decode-and-forward relaying In: 2015 7th International Conference on Communication Systems and Networks (COMSNETS), pp 1–7, January 2015 10 Chen, Z., Yuan, J., Vucetic, B.: Analysis of transmit antenna selection/maximalratio combining in rayleigh fading channels IEEE Transactions on Vehicular Technology 54(4), 1312–1321 (2005) 11 Prakash, S., McLoughlin, I.: Predictive transmit antenna selection with maximal ratio combining In: Global Telecommunications Conference, GLOBECOM 2009, pp 1–6 IEEE, November 2009 12 Molisch, A.,Win, M.,Winters, J.: Capacity of mimo systems with antenna selection In: IEEE International Conference on Communications, ICC 2001, vol 2, pp 570–574 (2001) 13 Chen, Y., Tellambura, C.: Distribution functions of selection combiner output in equally correlated rayleigh, rician, and nakagami-m fading channels IEEE Transactions on Communications 52(11), 1948–1956 (2004) 14 Yates, R., Goodman, D.: Probability and stochastic processes: a friendly introduction for electrical and computer engineers, p 519 (2005) A Survey of Machine Learning Algorithms and Their Applications in Cognitive Radio Mustafa Alshawaqfeh1 , Xu Wang1 , Ali Rıza Ekti1(B) , Muhammad Zeeshan Shakir2 , Khalid Qaraqe2 , and Erchin Serpedin1 Department of Electrical and Computer Engineering, Texas A&M University, College Station, TX, USA {mustafa.shawaqfeh,xu.wang,arekti}@tamu.edu, serpedin@ece.tamu.edu Department of Electrical and Computer Engineering, Texas A&M University at Qatar, Doha, Qatar {muhammad.shakir,khalid.qaraqe}@qatar.tamu.edu Abstract Cognitive radio (CR) technology is a promising candidate for next generation intelligent wireless networks The cognitive engine plays the role of the brain for the CR and the learning engine is its core In order to fully exploit the features of CRs, the learning engine should be improved Therefore, in this study, we discuss several machine learning algorithms and their applications for CRs in terms of spectrum sensing, modulation classification and power allocation Keywords: Cognitive radio · Machine learning Spectrum sensing · Modulation classification · Learning engine · Introduction The evolution of wireless communications systems and many other devices is continuously subject to two major development trends: a) improvement of existing capabilities, and b) extension and insertion of new features into the existing structures In what concerns the first trend, one can notice that insertion of new features arises from the fact wireless systems progress very fast in accordance with the market demands Therefore, wireless systems always require new services and applications One of the most striking examples for such situations is cell phones Earlier cell phones were used only for voice transmissions along with limited text messaging applications however contemporary cell phones are capable of transmitting multimedia along with an operating system running on In what concerns the second trend, a continuous improvement of existing capabilities is a necessity since incorporating new features adds new dimensions that help improve the existing capabilities The above mentioned considerations suggest that adaptation and optimization should always be employed as key enabling technologies for the continuous update of communication systems to dynamically changing conditions c Institute for Computer Sciences, Social Informatics and Telecommunications Engineering 2015 M Weichold et al (Eds.): CROWNCOM 2015, LNICST 156, pp 790–801, 2015 DOI: 10.1007/978-3-319-24540-9 66 A Survey: Machine Learning Algorithms in Cognitive Radio 791 In this regard, the purpose of this study is to provide a conceptual description of machine learning algorithms used in the design of wireless communication systems in the light of a recently emerging technology called cognitive radio (CR) [1–5] The idea of CR was first presented by Joseph Mitola III and Gerald Q Maguire, Jr in [3] “The point in which wireless personal digital assistants and the related networks are sufficiently and computationally intelligent about radio resources and related computer-to-computer communication to detect user communications needs as a function of use context, and to provide radio resources and wireless services most appropriate to those needs” [1] There are many advantages offered by CRs in wireless communications A CR is basically an intelligent wireless device which is aware of the environment and spectrum and is able to adapt/optimize itself easily to the characteristics of the communication channel to satisfy the user needs The environment of a CR may include radio frequency (RF) spectrum, user behavior, transmission characteristics and parameters, multi-access interference, localization and data rates of users The key strengths of machine learning algorithms are their adaptive nature with respect to the dynamic changes of the channel and communication system parameters In addition, the ability to work without prior knowledge about the communication environment represents another important feature of CRs These considerations recommend machine learning as a promising technology for CRs In this paper, applications of machine learning for learning engine, spectrum sensing, modulation classification and power allocation in CRs are studied along with currently available methods and approaches to better adapt and optimize the overall system performance The rest of the paper is organized as follows The learning engine is presented in Section An overview of key machine learning techniques that can be implemented into the learning engine is presented in Section A review of machine learning applications in spectrum sensing, modulation classification and power allocation for CRs are presented in Section 4, Section and Section 6, respectively Concluding remarks are provided in Section Learning Engine The cognitive engine is the brain of a CR system and it enables the system to react intelligently to changes in the environment Basically, as shown in Figure 1, the CR extends a software-defined radio by adding an independent cognitive engine, which consists of a learning engine and reasoning engine [6] The learning engine lies in the core of the cognitive engine and it aims to build a model or an objective function based on the inputs that are to be used in taking the right decisions and making the correct predictions In the context of CRs, no simple relationship between the system inputs and the objective function is available due to the high complexity and degree of freedom of the software-defined radio (SDR) In this case, several channel statistics, such as transmit power, modulation scheme and sensing scheme, need to be adjusted simultaneously [7] In such scenarios, adopting a policy-based 792 M Alshawaqfeh et al Information Acquisition Learning Engine Reasoning Engine SDR Cognitive radio engine Fig The structure of cognitive radio engine decision making strategy is infeasible due to the large number of states that the cognitive radio networks (CRNs) and its radio frequency (RF) environment assume In addition, even if the resources are available, considering all the possible states and actions is impossible given the dynamic and random nature of CRNs Thus, the learning engine is crucial in the operation of the CR engine A learning engine is adopted to estimate the channel statistics The results are incorporated into a predictive calculus-based reasoning engine to make decisions and achieve certain objectives Several learning algorithms can be used to implement the learning engine For the sake of brevity, Table lists some of recent works involving the applications of machine learning algorithms in CR The recent literature shows extensive use of different learning algorithms in CRs which will be discussed in Section However, several factors influence the selection of the learning algorithm to implement the learning engine For example, one important factor is the availability of prior knowledge about the environment Supervised learning methods are applicable only if prior information about the environment is known to train the agent On the other hand, unsupervised learning methods are appealing for scenarios with lack of prior information The computational complexity of the algorithm is the main limiting factor especially for CRs with limited resources In general, CRNs and their RF environment exhibit the following characteristics [7]: (i) incomplete observation information about the state variable, (ii) incorporation of CRs into CRNs and (iii) unknown RF environment Consequently, the learning engine must be designed by taking into account the above characteristics such that the learning method efficiently and optimally adapt to the changes and the incompleteness of the observed information and RF environment A Survey: Machine Learning Algorithms in Cognitive Radio 793 Table Classification of Papers Exploiting Machine Learning Algorithms Supervised Learning Unsupervised Learning Reinforcement Learning SVM KNN Spectrum Sensing [8, 9] [8, 9] [8, 10] [11–13] Modulation Classification [14, 15] [16] Power Allocation [17–19] Machine Learning In literature, machine learning techniques can be categorized into three different types, namely, supervised learning, unsupervised learning and reinforcement learning (RL) 3.1 Supervised Learning Supervised learning is a machine learning approach that infers an objective function from a labeled training data Thus, this method requires prior information about the environment The training data consists of input-output pairs An inferred function is derived based on the samples to map the future input For instance, the training samples (xi , yi ) are given and it is assumed that (xi , yi ) are drawn from some distribution P (x) Classification is the main function for supervised learning and its goal is to find a classifier function f such that it fits and characterizes the training examples The classifier is used to map and classify the newcoming data One well-known example of supervised learning methods is referred to as the support vector machine (SVM) and it was first developed in [20] The original SVM approach builds a linear classifier that maps the input vectors to a high-dimensional space A nonlinear SVM classification method was proposed by Boser et al [21] using the kernel trick SVM is exploited in a wide range of machine learning applications due to its accurate predictions, fast evaluation of the targeted function and the robustness against noise and errors For more information about SVM, the reader is referred to [22,23] 3.2 Unsupervised Learning In contrast to the supervised learning, the unsupervised learning applies to an environment in which the prior knowledge is unknown Specifically, the unsupervised learning extracts hidden features from the unlabeled data Since the samples from unsupervised learning are unlabeled, unsupervised learning receives neither targeted outputs nor environmental rewards This fact distinguishes the unsupervised learning from the supervised learning and the reinforcement learning The main functions for unsupervised learning are clustering, dimensionality reduction and blind signal separation [24,25] In principle, a clustering algorithm aims to group objects into clusters such that the elements in the same cluster are similar to each other and different from the elements placed in any other clusters There are several clustering algorithms such as K-means or centroid-based clustering [26,27] and mixture models 794 3.3 M Alshawaqfeh et al Reinforcement Learning Reinforcement learning is an online learning method which lies in the middle between supervised and unsupervised learning The general idea behind the reinforcement learning is to maximize a specific reward function According to [28], the reinforcement learning consists of three main components: a policy, a reward function and value function Let S be the set of all possible states of the environment, and A be the set of all possible actions and n denote the time index A policy π : S × A → S is the rule that defines the selection of next state sn+1 based on the current state-action pair (sn , an ) The policy can be deterministic or stochastic In a deterministic policy, the agent selects the actions in a deterministic fashion based on the current state The reward function rn : A × S → is a scalar function that maps each state-action pair (sn , an ) into a single real number, reward, that indicates the reward obtained by selecting the action an at state sn to move into state sn+1 According to the knowledge of the reward function, reinforcement learning is classified into a model-based learning if the reward is known and a model-free learning otherwise Generally, the reward functions may be stochastic The reward function determines the immediate or short term reward of an action However, the agent is interested in the long-run total reward which is defined by the value function or return Starting from state sn , the return is the random variable Rn defined as: Rn = ∞ k k=0 γ rn+k+1 : N : k=0 rn+k+1 non-episodic model episodic model, (1) where γ ∈ [0, 1] is the discount factor The goal of the reinforcement algorithm is to find a policy that maximizes Rn In principle, the optimal policy can be found by exhaustive search of the policy space This solution is computationally infeasible due to the large (or even infinite) number of policies to be checked Hence, the core of reinforcement learning algorithms is to find an efficient method to calculate or approximate the function value One appealing method is to estimate the function value Estimation of function values in more details is commonly carried out within a Markov Decision Process (MDP), which represents a general framework for reinforcement learning MDP is a reinforcement learning environment in which states satisfy Markov property Markov property means that deciding the next state sn+1 depends only on the current state sn and action an In other words, the current state and actions contain all the required information about future state Mathematically, this condition can be expressed as follows: P r{sn+1 , rn+1 |sn , an , rn , sn−1 , , s0 , a0 , r0 } = P r{sn+1 = s, rn+1 = r|sn , an , rn } (2) The Markovian assumption simplifies the analysis by allowing prediction of future rewards based only on the current state and action A finite MDP means that state and action spaces are finite A natural way to estimate the value function is to take the sample mean of the received rewards Since the rewards depend A Survey: Machine Learning Algorithms in Cognitive Radio 795 on the selected action, the estimated value function depends on the selected policy Define the state-value function for π policy (V π ) as the expected value of return given that agent is in the sn state and follows the π policy For MDP, V π (sn ) is defined as: (3) V π (sn ) = Eπ [Rn |sn , π] Similarly, the action-value function for π policy, Qπ (sn , an ), is defined as the expected return starting from state sn and taking the action an and following the policy π In MDP, Qπ (sn , an ) can be defined as: Qπ (sn , an ) = Eπ [Rn |sn , an , π] (4) ∗ It is shown in [28] that the optimal action-value function Q (sn , an ) satisfies: Q∗ (sn , an ) = max Qπ (sn , an )· π P r[sn+1 |sn , sn ] [rn + γ max Q∗ (sn+1 , an+1 )] an+1 ∈A sn+1 ∈S (5) One way to maximize the action-value functions is the Q-learning algorithm [29] Q-learning follows a fixed state transition and does not require prior information about the environment The update for the one-step version is given by: Qn+1 (sn , an ) = Qn (sn , an ) + α[rn+1 + γ max Qn (sn+1 , an+1 ) − Qn (sn , an )] an+1 ∈A (6) The reinforcement learning is subject to a trade-off between exploration and exploitation This trade-off manifests through the fact that at each stage, the agent has to decide whether to exploit the current highest reward action or to explore new actions for higher rewards Two action selection methods for controlling the trade off between exploration and exploitation are the -greedy and softmax action [28,30] In -greedy, the next action is selected either at random with uniform probability or by selecting the optimal action a∗ = maxa Q(a, s) with probability − In the softmax method, the action a is selected with probability exp{Q(sn , an )/τ } , an+1 ∈A exp{Q(sn , an )/τ } (7) where τ is a positive weight factor for each action and is referred to as the temperature factor Reinforcement learning algorithms differ by how they efficiently compute the value function Reinforcement learning algorithms can be also divided into single agent reinforcement learning (SARL) and multiple agent reinforcement learning (MARL) In SARL, the learning process is local at each agent in the sense that rewards for each agent does not depend on the other agents In MARL, the reward depends on both, the environment and all agent policies and actions This dependence on other agents’ policies complicates the learning process The interested reader is referred to [28,31,32] for detailed information about the reinforcement learning ... (www.springer.com) CROWNCOM 2015 Preface 2015 marks the 10th anniversary of the International Conference on Cognitive Radio- Oriented Wireless Networks (Crowncom) Crowncom 2015 was jointly hosted... Karagiannidis Muhammad Ismail (Eds.) • • • Cognitive Radio Oriented Wireless Networks 10th International Conference, CROWNCOM 2015 Doha, Qatar, April 21–23, 2015 Revised Selected Papers 123 Editors... in Cognitive Radio Networks Mai Abdel-Malek, Karim Seddik, Tamer ElBatt, and Yahya Mohasseb 17 30 Auction Based Joint Resource Allocation with Flexible User Request in Cognitive Radio Networks