Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Design and analysis of energy harvesting wireless communication systems Ahmed, Imran 2017

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


24-ubc_2018_february_ahmed_imran.pdf [ 921.11kB ]
JSON: 24-1.0360776.json
JSON-LD: 24-1.0360776-ld.json
RDF/XML (Pretty): 24-1.0360776-rdf.xml
RDF/JSON: 24-1.0360776-rdf.json
Turtle: 24-1.0360776-turtle.txt
N-Triples: 24-1.0360776-rdf-ntriples.txt
Original Record: 24-1.0360776-source.json
Full Text

Full Text

Design and Analysis of EnergyHarvesting Wireless CommunicationSystemsbyImran AhmedB.Sc. Hons., University of Dhaka, Bangladesh, 2008M.Sc., University of Dhaka, Bangladesh, 2010A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFMASTER OF APPLIED SCIENCEinTHE COLLEGE OF GRADUATE STUDIES(Electrical Engineering)THE UNIVERSITY OF BRITISH COLUMBIA(Okanagan)November 2017c© Imran Ahmed, 2017The undersigned certify that they have read, and recommend to theCollege of Graduate Studies for acceptance, a thesis entitled: Design andAnalysis of Energy Harvesting Wireless Communication Systemssubmitted by Imran Ahmed in partial fulfilment of the requirements of thedegree of Master of Applied ScienceDr. Md. Jahangir Hossain, Faculty of Applied Science/School of EngineeringSupervisor, Professor (please print name and faculty/school above the line)Dr. Thomas Johnson, Faculty of Applied Science/School of EngineeringSupervisory Committee Member, Professor (please print name and faculty/school abovethe line)Dr. Jonathan F. Holzman, Faculty of Applied Science/School of EngineeringSupervisory Committee Member, Professor (please print name and faculty/school abovethe line)Dr. Solomon Tesfamariam, Faculty of Applied Science/School of EngineeringUniversity Examiner, Professor (please print name and faculty/school above the line)November 27, 2017(Date Submitted to Grad Studies)iiAbstractRecent advancement in wireless communication networks expect expo-nential growth of smart phones, diverse wireless services and Internet ofThings (IoT) applications. The extensive growth of wireless devices cansignificantly increase energy consumption, and therefore, creates environ-mental pollution. An urge for green communication is building up day byday. As a matter of fact, there is a need to design environment friendlywireless communication technologies and energy efficient resource alloca-tion solutions, which will potentially drive the next generation of wirelesscommunication. In this thesis, we design and analyse energy harvestingwireless communication system by considering both renewable energy andRF energy sources. For a relayed communication system, we develop jointoptimal power and power-splitting ratio allocation scheme where relay doesnot have its own energy supply. The relay harvests energy from interferenceand the signal received from the source. We consider both half-duplex andfull-duplex relaying. For half–duplex relaying, we analyse the amount ofharvested energy by controlling the amount of incoming interference usingpower splitting ratio. In addition, we show the impact of self–interferencefor full–duplex relay and compare the results with half–duplex cooperativeenergy harvesting wireless networks. Next, we consider multi-relay network,where relays harvest renewable energies from the surrounding environment.For this setup, we propose the optimal policy for relay selection and powerallocation under unknown statistics of the channel fading and energy arrivalprocesses. In particular, we develop an efficient low-complexity learning al-gorithm that does not require the statistics of the channel fading and energyharvesting processes to be known.iiiPrefaceThis thesis is based on the research work conducted in the School ofEngineering at The University of British Columbia, Okanagan Campus, un-der the supervision of Prof. Md. Jahangir Hossain. Published works arecontained in this thesis.Chapter 3 of this thesis has been partially published in the Proceed-ings of IEEE International Conference on Ubiquitous Wireless Broadband(ICUWB). The analytical evaluations and numerical solutions related to thispaper are the results of my own independent work. Dr. Ahmed providedme with some valuable suggestions to improve the system model. Prof.Hossain and Dr. Ahmed helped me prepare the manuscripts for scholarlypublication by checking the validity of analytical and numerical results andproofreading.Chapter 4 of this thesis has been published in IEEE Wireless Commu-nication Letters. I am the principle contributor for this work. Dr. Ahmedand Prof. Hossain helped me to check the validity of the analytical andnumerical findings and to proofread the manuscript.A list of my papers published at The University of British Columbia,Okanagan Campus is provided below.Conference Paper Published1. I. Ahmed, M. J. Hossain and I. Ahmed, ”Resource Allocation for Decode-and-Forward Relayed System with Interference Aided Energy Harvesting,”2015 IEEE International Conference on Ubiquitous Wireless Broadband(ICUWB), Montreal, QC, 2015, pp. 1-5 (Invited paper).Journal Paper Published1. I. Ahmed, I. Ahmed and M. J. Hossain, ”Optimal Stochastic PowerAllocation and Relay Selection for Energy Harvesting Systems,” in IEEEWireless Communications Letters, vol. 6, no. 4, pp. 546-549, Aug. 2017.ivTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . vList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ixList of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . xAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . xiiChapter 1: Introduction . . . . . . . . . . . . . . . . . . . . . . . 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Literature Review and Motivation . . . . . . . . . . . . . . . 21.2.1 Resource Allocation for Decode-and-Forward RelayedSystem with Interference Aided Energy Harvesting . . 21.2.2 Power Allocation and Relay Selection for Energy Har-vesting Systems . . . . . . . . . . . . . . . . . . . . . . 51.3 Thesis Outline and Contributions . . . . . . . . . . . . . . . . 7Chapter 2: Background . . . . . . . . . . . . . . . . . . . . . . . 82.1 Energy Harvesting . . . . . . . . . . . . . . . . . . . . . . . . 82.1.1 EH sources . . . . . . . . . . . . . . . . . . . . . . . . 102.2 RF-EH Receiver Module . . . . . . . . . . . . . . . . . . . . . 112.2.1 Simultaneous wireless information and power transfer(SWIPT) . . . . . . . . . . . . . . . . . . . . . . . . . 122.3 Transmission Modes . . . . . . . . . . . . . . . . . . . . . . . 142.3.1 Benefits of using FD operations . . . . . . . . . . . . . 15vTABLE OF CONTENTS2.3.2 SI cancellation . . . . . . . . . . . . . . . . . . . . . . SI suppression . . . . . . . . . . . . . . . . . Analog cancellation . . . . . . . . . . . . . . Digital cancellation . . . . . . . . . . . . . . 172.4 Fading Channels . . . . . . . . . . . . . . . . . . . . . . . . . 182.4.1 Rayleigh fading . . . . . . . . . . . . . . . . . . . . . . 182.4.2 Rician fading . . . . . . . . . . . . . . . . . . . . . . . 192.5 Interference . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20Chapter 3: Resource Allocation for Decode-and-Forward Re-layed System with Interference Aided Energy Har-vesting . . . . . . . . . . . . . . . . . . . . . . . . . . 213.1 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . 213.2 Problem Formulation and Solution . . . . . . . . . . . . . . . 243.2.1 Problem formulation . . . . . . . . . . . . . . . . . . . 243.2.2 Solution of the optimization problem . . . . . . . . . . 253.2.3 Baseline methods . . . . . . . . . . . . . . . . . . . . . 303.2.3.1 Exhaustive search scheme . . . . . . . . . . . 303.2.3.2 Heuristic scheme . . . . . . . . . . . . . . . . 303.2.4 Complexity . . . . . . . . . . . . . . . . . . . . . . . . 303.3 Full Duplex . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303.3.1 System model . . . . . . . . . . . . . . . . . . . . . . . 303.3.2 Receiver architecture . . . . . . . . . . . . . . . . . . . 313.3.3 Problem formulation and solution . . . . . . . . . . . 323.3.3.1 Problem formulation . . . . . . . . . . . . . . 323.3.3.2 Solution of the optimization problem . . . . 333.4 Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . 333.4.1 Throughput performance for different interference lev-els for HD system . . . . . . . . . . . . . . . . . . . . 343.4.2 Performance comparison with the baseline schemes forHD system . . . . . . . . . . . . . . . . . . . . . . . . 353.4.3 Throughput analysis for different levels of signal pro-cessing power at the relay for FD system . . . . . . . 363.4.4 Throughput analysis for different SI levels . . . . . . 373.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38Chapter 4: Power Allocation and Relay Selection for EnergyHarvesting Systems . . . . . . . . . . . . . . . . . . 394.1 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . 394.2 Problem Formulation and Solution . . . . . . . . . . . . . . . 41viTABLE OF CONTENTS4.3 Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . 464.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48Chapter 5: Conclusions . . . . . . . . . . . . . . . . . . . . . . . 505.1 Summary of Contributions . . . . . . . . . . . . . . . . . . . . 505.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . 51Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53viiList of TablesTable 2.1 Examples of different EH sources and their perfor-mances [1, 2]. . . . . . . . . . . . . . . . . . . . . . . . 12Table 3.1 Simulation parameters and statistical model setting . . 34Table 4.1 Simulation parameters and statistical model setting . . 46viiiList of FiguresFigure 1.1 Next generation of wireless communications [3]. . . . 2Figure 2.1 A typical block diagram of power receiver module. . . 9Figure 2.2 Block diagram of PS receiver module. . . . . . . . . . 13Figure 2.3 Block diagram of TS receiver module. . . . . . . . . . 13Figure 2.4 Block diagram of antenna switching receiver module. 14Figure 2.5 Block diagram of spatial switching receiver module. . 14Figure 2.6 SI cancellation techniques. . . . . . . . . . . . . . . . 16Figure 3.1 Illustration of the system model. A source, S com-municates with a destination, D via a DF relay, R.Interferring nodes, I1, I2, · · · , IM , are other sourcesand D1, D2, · · · , DM are their corresponding destina-tions. . . . . . . . . . . . . . . . . . . . . . . . . . . . 22Figure 3.2 Throughput vs. interference power, PI (Watts) fordifferent values of ES . . . . . . . . . . . . . . . . . . . 35Figure 3.3 Performances of throughput and optimal ρ. UpperFigure: Throughput (bits/s/Hz) vs. PI (Watts). LowerFigure: ρ vs. PI (Watts). . . . . . . . . . . . . . . . . 36Figure 3.4 Performances of throughput vs. ES (Watts) for dif-ferent signal processing powers at relay, R. . . . . . . 37Figure 3.5 Performances of throughput vs. ES (Watts) for dif-ferent levels of self-interferences at R. . . . . . . . . . 38Figure 4.1 Illustration of the system model consists of one sourceS, one destinationD, andN relaysRn, n ∈ {1, 2, · · · , N}. 40Figure 4.2 Throughput vs. average harvested energy for differ-ent values of PS for N = 3. . . . . . . . . . . . . . . . 47Figure 4.3 Convergence behavior of proposed scheme. . . . . . . 48Figure 4.4 Throughput vs. total number of transmission timeintervals for N = 5. . . . . . . . . . . . . . . . . . . . 49ixList of AbbreviationsAF Amplify–and–ForwardAWGN Additive White Gaussian NoiseBER Bit Error RateCCI Co–Channel InterferenceCPU Central Processing UnitDC Direct CurrentDF Decode–and–ForwardDP Dynamic ProgrammingEH Energy HarvestingGBD Generalized Benders Decompositioni.i.d. Independent and Identically DistributedIoT Internet of ThingsLTE Long-Term EvolutionLOS Line–of–SightMAC Medium Access ControlMDP Markov Decision ProcessMILP Mixed Integer Linear ProblemmmW Millimeter WaveMINLP Mixed Integer Non–Linear ProgrammingOFDMA Orthogonal Frequency Division Multiple Accesspdf Probability Density FunctionPS Power SplittingPHY PhysicalQoS Quality of ServiceQCQP Quadratically Constrained Quadratic ProgramRF Radio FrequencyRFEH Radio Frequency - Energy HarvestingSDP Semi Definite ProgramSI Self–InterferenceSINR Signal–to–Interference–plus–Noise RatioSNR Signal–to–Noise RatioSDMN Software Defined Mobile NetworkingxList of AbbreviationsSWIPT Simultaneous Wireless Information and Power TransferTS Time SwitchingWiFi Wireless Fidelity3GPP 3rd Generation Partnership Project5G Fifth GenerationxiAcknowledgementsI am deeply grateful to my supervisor Prof. Md. Jahangir Hossain forhis enthusiasm, guidance, advice, encouragement, and support. He grantedme a great flexibility and freedom in my research work. He taught me theacademic knowledge and research skills. I will continue to be influenced byhis rigorous scholarship, clarity in thinking, and professional integrity. It ismy honor to study and do research under his supervision.I owe many people for their generosity and support during my at The University of British Columbia, Okanagan Campus. I wouldlike to thank my dear colleagues for sharing their academic experiences andconstructive viewpoints generously with me during our discussions.Finally, I would like to thank my father Dr. Farruk Ahmed and mymother Mrs. Shirin Ahmed for their support and blessings and all otherfamily members for their patience, understanding and support all theseyears. Specially, I would like to thank my daughter Arisha and my wifeNadia for their constant patience and support. Last but not the list, mybrother Imtiaz for his immense support and encouragement. All my achieve-ments would not have been possible without their constant encouragementand support.xiiChapter 1Introduction1.1 IntroductionThe evolution of different generations of the cellular network is mainlyinfluenced by the rapid growth in wireless end user devices, data usage, andthe need for better quality of services. More than 50 billion network de-vices are expected to utilize the cellular network services by the end of theyear 2020 [4]. This feature will in turn increase the data rate requirement,see Fig. 1.1. In addition, one of the main highlights of fifth generation(5G) cellular networks is to provide energy-efficient, low-cost, and securedcommunication systems [3]. Emerging high data rate applications in 5Grequire careful allocation of limited wireless communication resources, e.g.,the transmit power and the channel bandwidth, such that the total systemcapacity is enhanced and the error rate is minimized. However, increasedpower consumption by wireless communication systems not only puts bur-den on the service providers but also provides detrimental impacts on theenvironment, e.g., greenhouse effect, etc [5]. This increasing energy demandrequires careful wireless system design and optimizations. To address thisfundamental challenge, the academic and industrial community has comeforward to put valuable contributions in developing energy efficient 5G sys-tems. Energy harvesting (EH) from ambient energy sources (sun, light, etc.)can be a perfect example of green communication that can potentially re-duce the dependency on the supply of grid or battery energy. In particular,EH nodes harvest energy from the renewable sources of their surroundingenvironment, convert it to electrical energy, and store the electrical energyin batteries in order to carry out their functions. In general, energy can beharvested from unused ambient renewable energy sources using solar, ther-moelectric, and motion effects, or through other physical phenomena. Oneof the main problems of these renewable energy sources is the randomnessin energy availability [6]. This randomness in energy availability is not ob-served in conventional energy supply. Therefore, it is imperative to handlethis random incoming energy in such a way so that system utility can bemaximized while ensuring minimal energy outage.11.2. Literature Review and MotivationFigure 1.1: Next generation of wireless communications [3].1.2 Literature Review and Motivation1.2.1 Resource Allocation for Decode-and-Forward RelayedSystem with Interference Aided Energy HarvestingGreen communication has received a great deal of attention from academiaand industry due to the environmental concerns raised by the rapidly in-creasing energy consumption of wireless communication systems [7]. To thisend, different energy–saving solutions have been proposed in the literature[7]. One potential solution is environmental EH from renewable sources topower the communication nodes due to its minimal ecological impact [8].EH nodes harvest renewable energies from their surrounding environmentand convert them into electrical energy. This electrical energy is then storedin the batteries, super–capacitors, etc. for future use, if required.Traditionally, renewable energies are harvested from natural resources,e.g., solar, wind, mechanical, thermal effects, etc. [8]. Apart from the con-ventional energy sources, background radio frequency (RF) electromagneticsignal is also a potential source of energy to harvest [9]. EH from the tradi-tional sources is not always reliable whereas the energy level of RF signalsis relatively stable and can thus be controllable partially [9]. Moreover, animportant advantage of RF–EH system is that the RF signals can carry theenergy and the information at the same time; hence, the energy–constrainednodes can harvest energy and can process the information simultaneously[10]. However, the simultaneous wireless information and power transferrequires a careful design at the receiver end so as to extract the informationwith required reliability and to harvest energy successfully [10, 11].Co–channel interference (CCI) is one of the major concerns in wirelesscommunication systems [12, 13]. Therefore, interference management poses21.2. Literature Review and Motivationa new challenge on the design of fifth 5G wireless communication systemsdue to the increasing number of uncoordinated low–power nodes, femto–cellnodes to improve the coverage and capacity [14]. As a matter of fact, dif-ferent techniques, e.g., inter–cell interference coordination and coordinatedmulti–point communication, have been introduced in the standards of LTE–Advanced [14]. In [15], the role of interference has been re–examined andseveral techniques have been proposed to exploit the interference in a con-structive manner. Among the proposed techniques, harvesting RF energyfrom the interference while decoding information from the transmitting nodecan help an energy–limited node to participate in data transmission. How-ever, a careful trade–off has to be made to receive information with requiredreliability and to extract the energy from the interference. A notable contri-bution on harvesting energy from the interference is depicted in [16], wherethe energy efficiency of data transmission is maximized for an orthogonal fre-quency division multiple access (OFDMA)–based non–cooperative commu-nication system. A cooperative communication system has been consideredin [17], where the energy–limited relay harvests energy from the receivedsignal and interference, and then exploits the energy for relaying operation.Although authors in [17] developed closed–form expression for throughputfor the cooperative communication system with EH capability, they did notprovide any resource allocation framework for such systems.On the other hand, full–duplex (FD) wireless-powered amplify-and-forward(AF) relay with self-energy recycling is discussed in [18] where at the relaynode, time switching or power splitting is not required for harvesting energy.Yong et al. also propose self-energy recycling protocols in [18] in which partof loop energy is used for data transmission by the relay node can be har-vested and stored for future use. Unlike in [19] where the FD relaying isstudied which requires additional energy consumption because of the severeself-interference (SI) at the relay node. The authors provide an analyti-cal framework for three different communication categories: instantaneoustransmission, delay-constrained transmission, and delay tolerant transmis-sion for wireless powered FD relaying. Instead of using single antenna atthe relay node, they have employed two antennas to harvest energy. Time-switching protocol has been adopted in-spite of having two antennas at therelay node. Chen et al. discuss cooperative simultaneous wireless infor-mation and power transfer (SWIPT) network for FD relays in [20]. Unlikeexisting literature, the authors consider two wireless powered relays in a dualhop network configuration. AF protocol has been proposed for successiverelaying where two relays alternatively take initiative to forward the sourcemessage to the destination. Outage probability has been analyzed with a31.2. Literature Review and Motivationclosed form expression for AF relaying protocol. A dynamic hybrid relayingtechnique has also been introduced to avoid zero diversity gain. Dynamichybrid scheme helps to switch between FD and half–duplex (HD) relayingto select best transmission mode. Kieu et al. discuss EH architecture of FDrelaying networks in [21]. Time-switching based protocol has been studiedand closed form expression of outage probability has been derived for thesystem. Two-way dual-hop relay network is considered as system model.The communication process is based on two phase protocol. Numerical re-sult suffices that throughput is maximum for a particular optimal value.The throughput increases upon reaching the optimal value and starts de-creasing after reaching the optimal value. In [22], the authors propose aninterference aided relaying network where they provide optimal solution forjoint optimization of relay nodes and power–splitting ratio allocation. Theyexploit the interference as a source of energy for HD relaying network.Motivated by the works in [15, 17], in Chapter 3 we consider an interference–limited single–link cooperative communication system with EH capability,and we develop joint optimal power and the power-splitting ratio alloca-tion scheme. In particular, in order to maximize the end–to–end systemthroughput by joint optimal allocation of the transmit powers of the sourceand relay nodes and the power–splitting ratio, we formulate an optimiza-tion problem. The optimization problem results in a non–convex polynomialprogram, which cannot be solved optimally in polynomial time. Instead,we develop a low–complexity suboptimal scheme to solve the optimizationproblem. For this purpose, at first, we transform the formulated polyno-mial program into a quadratically constrained quadratic program (QCQP),which is then relaxed into a semi–definite program (SDP). We solve therelaxed problem optimally and then adopt the well–known randomizationtechnique to find a very tight solution to the optimal one. We show numer-ically that the developed suboptimal scheme performs close to the optimalscheme, which employs exhaustive search technique.Motivated by the work in [18, 20], in Chapter 3, we also consider aninterference-limited single-link cooperative communication system for FDrelay network with EH capability. FD relays create self–interference whichcan also be used as a source of energy. In particular, we formulate anoptimization problem similar to HD system mentioned above to jointly op-timize the power allocation of wireless nodes and power–splitting ratio. Theoptimization problem cannot be solved optimally in polynomial time thatresults as a non–convex problem. Hence, this finding motivates us towardsdeveloping a low-complexity suboptimal resource allocation algorithm.41.2. Literature Review and Motivation1.2.2 Power Allocation and Relay Selection for EnergyHarvesting SystemsDeploying EH nodes in cooperative communications has been envisionedas a promising approach for energy-limited and densely deployed networksin 5G wireless communication systems [23, 24]. In particular, the sensornodes in Internet-of-Things (IoT) networks being capable of scavenging re-newable energies e.g., in the form of light, heat, pressure, RF can ensurea longer lifetime. On the other hand, multi-hop communication has alsobeen addressed as a potential technology. It is crucial to carefully allocatethe limited resources e.g., energy stored in the battery in order to prolongsystem life-time, minimize the system outage, and maximize the end-to-enddata rate [25].In this work, we consider a two-hop multi-relay cooperative communica-tion network, where the relays operate in HD AF mode. The relays do nothave conventional energy supply; instead, they harvest renewable energiesin the form of light, heat, thermal, electromechanical, etc., store the energyin the batteries, and use it for transmission. We develop optimal joint relayselection and transmit power allocation scheme that can be implementedonline. In particular, the online scheme exploits only the causal informa-tion of channel and energy states (in the batteries) and does not require thechannel and energy statistics or any non-causal information to be known.We formulate an optimization problem as a Markov decision process (MDP)and solve the problem with time-averaging algorithm by incorporating post-decision state-value approach [26, 27]. In [28], a two-way multi-relay systemhas been considered, where system outage probability has been analyzedand joint relay selection and power allocation scheme has been proposed.The relays harvest energy from the signals received from the communicat-ing nodes. Similar type of system model has been considered in [29]. Pleasenote that we cannot apply the resource allocation scheme considered in [28]to the system model adopted in our work for the following notable reasons:− [28] and [29] assume that energy can only be harvested from radiofrequency signals, whereas our model is applicable for any renewableenergy arrival process.− We assume that energy can be stored in a battery of infinite capacityand energy update model results in a Markov decision process.− We optimize the relay transmit power, whereas [28] and [29] optimizethe transmit powers of two transmitting nodes.51.2. Literature Review and Motivation− The receiver at the relay node considered in [28] and [29] is differentfrom that in our model. We consider that the received signal at therelay node is solely exploited for information detection. Moreover, wehave an EH module that scavenges renewable energy, converts it intoelectrical energy, and stores it in a battery of infinite size.On the other hand, in [30], a joint link selection and relay power alloca-tion scheme has been proposed for a single-relay cooperative communicationsystem by minimizing the system outage probability. Please note that thedeveloped model cannot be extended to our system in a straight-forwardapproach as this method requires the statistical information of the channeland EH processes to be known. Please note that the statistical informationof the channel and EH processes are not known in our work.Our developed scheme is discussed in Chapter 4, which is applicable fora large network, where nodes enter and leave in a dynamic manner. Hence,acquiring the statistical information about the channel and EH processesmay not always be possible. Please note that similar type of system modelhas been considered and analyzed in [26]. However, our work is significantlydifferent from the work in [26] as follows:− In [26], an optimal online scheme was developed by formulating ajoint relay selection and power allocation optimization problem. Apartfrom the optimal online scheme, two suboptimal online schemeswere proposed. On the contrary, in this work, we developed an optimalonline scheme that was not considered in [26].− In [26], causal and non-causal information about the channel and EHprocesses were assumed to be known for optimal online scheme. More-over, for suboptimal online scheme, causal and statistical informationwere considered to be available. In contrast, in this work, we assumethat neither non-causal nor statistical information about the channeland EH processes are known for resource allocations. Therefore, wecannot apply the schemes developed in [26] in our considered scenario.− Our proposed scheme provides optimal results for large number of timeinterval, whereas the online schemes developed in [26] yield suboptimalresults even for large number of time intervals.− Our developed scheme is applicable for any environment, where the EHsources rapidly change over time and hence estimating the probabilitydensity functions of the energy arrival process is quite cumbersome.61.3. Thesis Outline and Contributions1.3 Thesis Outline and ContributionsThe thesis is organized into four chapters. Chapter 1 presents the his-tory and development of the EH based wireless communication systems. Inaddition, this chapter provides a detailed literature review related to therest of this thesis.Chapter 2 provides detailed technical background for the entire thesis.First, the basic system architecture of EH has been discussed and classifieddifferent sources of EH i.e., light, thermal, wind, RF, etc. Second, RF-EHreceiver architecture is discussed which is different from the receiver whichharvests energy from renewable sources. Third, the basic characteristicsand challenges of HD and FD transmission modes are reviewed that areconsidered in this thesis. Finally, the related channel fading models thatwill be adopted in this thesis are presented followed by different causes ofinterference.In Chapter 3, joint optimal power and the power-splitting ratio alloca-tion scheme is proposed for single–link interference–limited EH cooperativecommunication systems. The resource allocation problem is formulated forboth HD and FD DF relay to maximize the system throughput. The relayharvests energy from RF which carries both information and energy. Thepower–splitting receiver architecture is adopted where a portion of the powerwill be used for EH and rest of the portion will be used for information decod-ing. The proposed scheme is compared with two different baseline schemesfor HD system. The proposed system’s behaviour has also been observedfor FD system where SI plays a significant role.In Chapter 4, an efficient learning algorithm has been proposed for al-locating resources that incurs lower computational complexity which doesnot require channel states and EH profile to be known. The optimizationproblem is formulated to maximize the system’s throughput. A multi-relaybased cooperative communication is considered to evaluate system perfor-mance where a single relay is selected for transmitting signal from sourceto destination. Unlike Chapter 3, in this model, relay harvests energy fromrenewable sources (light, wind etc.) The performance of the proposed onlinelearning is compared with optimal offline and two sub-optimal online algo-rithm. The computational complexity of the proposed algorithm is muchless compare to the existing approaches.Chapter 4 concludes the thesis. Future works are also suggested.7Chapter 2BackgroundIn this thesis, we explore different wireless communication systems thatare empowered by randomly available renewable energies and are impairedby channel fading and additive white Gaussian noise (AWGN). The diversi-fied nature of renewable energy sources pose different challenges on wirelesssystem designers. Therefore, in this chapter, at first, we discuss the funda-mental characteristics and properties of EH nodes and the underlying chal-lenges of using these nodes in wireless communication systems. Moreover,in our research, we deploy half and full-duplex communication nodes andcompare their performance when these nodes are energized by EH nodes. Inthis chapter, we discuss the fundamental differences between half and full-duplex communications and their performance-complexity trade-offs. Afterthese discussions, a generic description on wireless fading channel concludesthis chapter.2.1 Energy HarvestingEH has attracted considerable attention from academia and industry asan environmentally friendly supply of energy for the wireless communicationnodes. In practice, the transmitter and receiver nodes in communicationsystems expend their energy for processing and transmitting data. Con-necting these transceiver nodes to the power grid is cumbersome for someapplications and in other cases, is not even possible. For instance, it is in-convenient to run cables in order to power the sensor nodes in wireless sensornetworks. Pre-charged batteries can be a viable solution to this problem.In practice, the limited storage capacity of the batteries and high transmitpowers may result in the quick drainage of the batteries. As a result, thebatteries need to be periodically replaced or recharged, which can be im-practical. For example, wireless sensor nodes can be placed in a toxic orhazardous environment that prohibits human intervention. On the otherhand, conventional energy, such as thermal powers (from coal, petroleum,and natural gas), hydel power (from high velocity of running water) aretapped and used abundantly at present. Their uses are practised for a long82.1. Energy Harvestingtime. Using conventional constant energy sources can raise environmentalconcern. For example, oil burning leads to carbon emissions which createsenvironmental and ecological damage. According to International EnergyAgency, with January 2015 the world consumes more than 94 million bar-rels per day [31]. The deployment of EH nodes is an alternative solution ofgreen energy consumptions. EH nodes harvest energy from the renewablesources of their surrounding environment, convert it to electrical energy, andstore the electrical energy in batteries in order to carry out their functions.Figure 2.1: A typical block diagram of power receiver module.In general, the energy can be harvested from the unused ambient re-newable energy sources using solar, thermoelectric, and motion effects, orthrough other physical phenomena. A list of various EH sources and theirperformances is provided in Table 2.1. Deploying EH nodes leads to main-tenance free operation of the system and ensures a perpetual life time of theEH devices. Moreover, the operating cost is much lower than the conven-tional power supply. In addition to that, these types of energy can be calledas Green energy which is available everywhere and creates no ecologicaldamage. However, using EH nodes for conventional communication systemsis not straightforward and requires a careful attention to design communica-tion devices and to develop communication protocols and algorithms. Themain challenge of maintaining EH nodes is careful allocation of resourcei.e., energy. In general, energy is random in nature which can not be pre-dicted. This energy can be stored in a battery for the future use, if required.However, using battery can lead to more involved system optimizations asthe battery states need to be incorporated over transmission time intervals.These considerations altogether makes system design problem a MDP. Thus,allocating resource (energy) to achieve the maximum system performance92.1. Energy Harvestingsuch as throughput is either a short term or long term optimization prob-lem. Energy can be harvested from RF as well which is semi-controllable innature. RF-EH can be employed where the wireless communication systemcannot depend on the weather or in indoor environment where accessingnatural sources are impossible. RF-EH nodes charge their batteries fromelectromagnetic radiation. The system can harvest energy from ambientsignals opportunistically or from a dedicated source (e.g., base stations) ina fully controlled manner. The average amount of harvesting energy in RF-EH remains almost constant for a fixed distance. A typical block diagram ofEH module is shown in Fig. 2.1. In practice, such EH systems are availablein the market such as ECO 200, ECS 200, powercast etc [2, 32]. In both re-newable and RH-EH process, the amount of energy which can be harvestedis low but sufficient enough to run low power sensor devices which is themain focus for next generation energy efficient wireless systems (e.g., IoTdevices). Our main goal in this thesis work is to design such wireless com-munication systems which is energy efficient in terms allocating renewableenergy and also RF energy in such a way that total system throughput canbe increased.2.1.1 EH sourcesEH from light: The power density of the raw solar energy (directed to-wards the sun) at midday is approximately 100mW/cm2. There are manycommercialized products of wireless sensor nodes that use miniature solarmodules to harvest solar energy. Among others, popular products of EnO-cean GmbH, e.g., ECS 310, ECS 300, STM 110, STM 250 harvest solarenergy and deploy it for data transmission and reception [32].Thermal EH: Electricity can be produced from heat emitted from machin-ery parts, radiators or even from the human body, for instance. Harvestersthat employ thermoelectric effect for EH usually work on the principles oftemperature gradient or fluctuations. These thermoelectric generators canproduce constant output power as long as there is a temperature gradientmaintained at the junctions.EH from wind: Harvesting energy from wind is a popular technology forhigh power application. For example, large wind turbine-generators are usedfor supplying power to remote loads, grids, and cellular base stations. EHfrom wind is considered as one of the fastest growing electricity generationtechnologies in the world [33].102.2. RF-EH Receiver ModuleEH from RF: In addition to well-known alternative energy sources, suchas solar, wind, geothermal, and mechanical, ambient RF signals present an-other promising source that can be exploited in transporting energy alongthe information. RF signal is a viable source of energy as the available fre-quency spectrum is very crowded due to the various applications of radiowaves. In RF-EH, radio signals with frequency range from 3 kHz to 300GHz are used as a medium to carry the energy in a form of electromagneticradiation. A clear advantage of this technique, compared to other alterna-tive energy sources, are as follows:− The ambient RF sources can be consistently available regardless oftime and location in urban areas.− RF-EH provides the advantage of delivering controllable and efficienton-demand wireless information and energy concurrently.− The radio signals can carry the information and the energy simulta-neously; hence, the system does not have to rely on external energysources.− RF-EH offers a low-cost option for sustainable operations of wirelesssystems without hardware modification on the transmitter side.− RF-EH systems can be built cheaply in small dimensions, which couldbe a significant advantage in manufacturing small and low cost com-munication devices such as sensor nodes.There are few products available for practical implementations to harvestenergy from RF. For example, the Powercast transmitter is commerciallyavailable as a RF source [2].2.2 RF-EH Receiver ModuleA wireless power receiver consists of the following components: a receiverantenna or antenna array, a matching network, a RF to direct current (DC)converter or rectifier, a power management unit and the energy storage unit[34]. Upon the successful charging of the energy storage unit, the storageunit which usually considered as a rechargeable battery or a super capacitor,112.2. RF-EH Receiver ModuleTable 2.1: Examples of different EH sources and their performances [1, 2].EH source PerformanceAmbient light (direct sunlight) 100mW/cm2Ambient light (illuminated indoor) 100µW/cm2Thermal (5 K gradient) 60µW/cm2Ambient airflow 1mW/cm2Isotropic RF transmitter with4W transmission (902-928 MHz)5.5µWat15mTX91501 Powercast transmitter with3W transmission (915 MHz)189µW at 5mwill provide power to the central processing unit (CPU), sensors or anylow powered devices. Efficient wireless power transfer can be determinedbased on the Friis free space equation. The received RF power depends onthe available power density and antenna’s effective area, Ae. The receivedpower, PR can be expressed asPR = cos2 φPTGT4piR2Ae (2.1)where Ae =λ2GR4pi and PT is transmitted power. GT and GR are the trans-mitter and receiver gains respectively. λ and cosφ denotes wavelength andpolarization loss factor of angle φ. From 2.1 we can state that the receiverantenna needs to have high gain to ensure maximum and successful wirelesspower transmission.2.2.1 Simultaneous wireless information and power transfer(SWIPT)In SWIPT, the same signal can carry both energy and information. Inpractice, this SWIPT is not possible as the information that is transmittedcan be fully destroyed in the RF domain unless we have a carefully designedRF receiver. The received signal needs to be split in two different parts,one for information decoding and another for EH. In the following, differentsignal splitting techniques in different domains are discussed.Power Splitting: The power splitting (PS) technique achieves SWIPT bysplitting the received signal into two different power levels; one signal level122.2. RF-EH Receiver Moduleis sent to the rectenna circuit for EH, and the other is converted to base-band for information decoding as depicted in Fig. 2.2. The PS techniqueuses a PS factor ρ, which needs to be optimized to get better results. Thistechnique also involves higher design complexity and suitable for delay con-strained critical information or energy processing application.Figure 2.2: Block diagram of PS receiver module.Time Switching: The time switching (TS) technique involves simple hard-ware implementation where accurate time synchronization is necessary forsimultaneous information decoding or EH process. In TS technique, thereceiver switches its operation in time to decode information and harvestenergy. Unlike PS technique, the receiver separates each frame into twotime slots for RF-EH and information decoding, α and 1 − α are, respec-tively, the ratio of RF-EH time slot and the information decoding time slot.Figure 2.3: Block diagram of TS receiver module.Antenna Switching: The antenna switching technique RF-EH receiveris equipped with an energy harvester and information detector with inde-pendent antennas. The antennas are divided into two groups; one for EHand the other for information decoding. Thus, the antenna switching tech-nique enables to perform the RF-EH and ID simultaneously by dynamicallyswitches each antenna element.132.3. Transmission ModesFigure 2.4: Block diagram of antenna switching receiver module.Spatial Switching: The spatial switching technique can be applied toachieve SWIPT in the spatial domain using multiple degrees of freedom ofthe interference channel without violating any constraints. Parallel eigen-channels are used to carry both information and energy. There is a switch atthe output of each eigenchannel, which helps to run either decoding circuitor rectification circuit.Figure 2.5: Block diagram of spatial switching receiver module.2.3 Transmission ModesIn general, simplex is a communication mode in which only signal istransmitted at the same time in the one direction. The transmitter and thereceiver operate on the same frequency. On the other hand, in a duplexcommunication system, both points (devices) can transmit and receive in-formation. For example, telephones and walkie-talkies are duplex devices.There are two types of duplex system i.e., HD and FD. FD communicationbetween two points (devices) means that both can transmit and receive in-formation between each other simultaneously at the same time and usingsame frequency. On the other hand, in HD systems, the transmission andreception of information must happen alternately i.e., not at the same timeinterval or not at the same frequency. While one point is transmitting, theother must only receive.Now-a-days most of the communication system usually employ devices142.3. Transmission Modesthat use either a time division or frequency division duplex for the signal’stransmission and reception. Thus, in practice only HD operations have beenemployed which leads to a erosion of the resource allocation. On the otherhand, FD operations promise to achieve high spectral efficiency by avoid-ing the utilization of two independent channels for bi-directional end-to-endtransmission which is the main motivation of using HD operations. Thebiggest impairment to achieving FD capability is SI, which is another rea-son of using HD system. FD also involves high practical implementationcomplexity. In terms of benefits, FD operations promise of nearly doubledchannel capacity compared to the conventional HD system. There are dif-ferent SI cancellation techniques proposed in the literature. According toZhang et al. in [35], SI can be cancelled by employing approaches like passivesuppression, analog cancellation and digital cancellation.Various research institutes have been developed the commercial FD plat-forms. Wireless fidelity (WiFi)-like platforms are the most prominent one,which operates at 2.4 GHz of bandwidth. In addition to that, FD platformsare also developed for some media access control layer protocols.2.3.1 Benefits of using FD operationsThe main attraction behind the advancement of FD communications isthe characteristics of nearly doubled channel capacity compared to conven-tional HD operations. The increased complexity in FD practical architecturemay be tolerated by enhancing the SI cancellation and reducing bit errorrate at the same time. The packet loss ratio which occurs due to simultane-ous transmission and reception can be reduced by facilitating larger buffersize in the FD architecture. In terms of performance comparison betweenHD and FD operations, Sahai et al. show that over 70 percent throughputgains from using FD operations over HD in practical scenarios [36].2.3.2 SI cancellationThe main goal of using FD operations is to transmit and receive signalsimultaneously within the same frequency band. As a matter of fact, FDnodes not only receive the desired signal, but also its own transmitting signal,which introduces SI. According practical research outcome, this SI signalstrength may be 50-100 dB higher than the desired receive signal [35]. As aresult, high quantization noise occurs at the receiver, which in turn degradessignal-to-interference-plus-noise ratio (SINR). To overcome this problem,sufficient reduction of SI strength before decoding the desired signal at the152.3. Transmission Modesreceiver is necessary. On the other hand, in practical systems, FD nodes mayface both linear and non linear distortions. Experimental results in [37], for atypical WiFi radio using transmit power of 20dBm, 80MHz bandwidth and areceiver noise floor of -90dBm show that the SI signal contains 20dBm linearnoise component and -10 dBm of non linear component. Linear componentof 20dBm strength reflects to the noise floor of above 110dB where as -10dBmmeans 80dB above the noise floor. Thus, SI suppression or mitigation isnecessary to keep SI signal strength below the noise floor. In general, theFD nodes must be capable of atleast 60dB and 50dB SI power cancellation inboth analog-domain and digital-domain respectively to meet the decodingrequirement [37]. SI cancellation techniques can be classified into passiveand active suppression. Active suppression technique is further classifiedinto analog and digital cancellations.Figure 2.6: SI cancellation techniques.Fig. 2.6 shows the block diagram of a FD transceiver. Two types ofantenna interfaces can be used for FD operation i.e., 1) a circulator and2) an antenna pair. If we use a circulator then a single antenna is sharedbetween transmitter and receiver. SI must be performed in both analogand digital domain for active SI cancellation to achieve in excess of 100dBSI suppression. Typically, 20-30dB of SI reduction is required from analogdomain if transmitter and receiver antenna has isolation of 20-30dB. In thatcase, a total 50-60dB of transmitter and receiver isolation is achieved beforedigital cancellation technique is applied. But, this amount of transmitterand receiver isolation is critical to maintain receiver’s linearity. Digital can-cellation is further employed to cancel the linear SI and other non-lineardistortions.162.3. Transmission Modes2.3.2.1 SI suppressionPassive SI suppression represents the attenuation of the SI signal dueto the path-loss effect between transmitter and receiver at FD node. Thereare different passive SI suppression techniques discussed in the literature,i.e., antenna separation, antenna cancellation etc. A larger transmitter andreceiver antenna separation means higher SI suppression capability. Hanedaet al. shows that 70dB SI power cancellation can be achieved by separationof 5m distance between transmitter and receiver antenna [38]. It can be fur-ther increased by employing an interference canceller. The authors studiedoutdoor-to-indoor wireless networks where compact relay acts as a repeaterbetween outdoor base stations and indoor users. On the other hand, an-tenna cancellation based passive suppression can be achieved by employingtwo transmitter antennas and one receiver antenna. Two transmitter an-tennas are placed at d distance and (d + λ/2) away from receiver antennawhere λ represent wavelength. Analog cancellationUsually, the analog cancellation is performed either at the analog base-band level or RF level. It can be accomplished by creating SI-inverse signal.In general, an accurate signal inversion can be achieved at the central fre-quency. However, the inverted signal can be deviated at each sides of thecentral frequency, which causes phase-distortion. A careful hardware designneeds to be employed to address this issue while using SI-inverse signal tomitigate SI component. In theory, a perfect SI-inverse signal will result inzero SI level at the output. Delay and attenuation adjustment also plays animportant factor in analog cancellation technique. The QHx220 noise can-cellation chip [39] can be used to adjust delay and attenuation. Balun-aidedcancellation can also be used to achieve significant SI reduction, in whichboth the phase and the amplitude of the inverted SI signal must be set prop-erly. Typically a balun is type of transformer which can be used betweenvarious parts of wireless communication systems. It is used to convert anunbalanced signal to a balance signal or vice versa. Digital cancellationThe aim of digital SI cancellation is to mitigate SI after cancelling SIin the analog-domain. The main concept of digital SI cancellation is re-constructing SI and subtracting it from the original received signal. Toassure guaranteed throughput within a given time, digital SI cancellation is172.4. Fading Channelsperformed on any residual SI. Handshake protocols, shift registers, sched-uled accessible shared registers and dedicated buffers are used to reconstructSI signal. The perfect timing between SI and the received signal must beensured at the moment of decoding. Thus, synchronization and channelestimator need to be designed perfectly.2.4 Fading ChannelsIn this section, we discuss briefly about different fading channel modelsthat are common in wireless communication systems. In general, small scalefading refers to random variation of the amplitude of wireless channel gain.Usually, wireless channels can experience multipath fading and shadowing.Multipath fading occurs due to the constructive and destructive combina-tion of randomly reflected, scattered, and diffracted signal components. Themain reasons for fading are the different obstacles in the transmission pathof signals. Fading can be classified into slow fading and fast fading based onthe symbol duration of the transmitted signal and coherence time of fadingchannels. Fast fading arises when the coherence time of channel is smallerthan the symbol duration and slow fading arises when the coherence time ofchannel is large than the symbol duration. Usually, fast fading is representedby the rapid fluctuations of the signal over small distance. In slow fading en-vironment, the channel conditions change at much slower rate as comparedto the transmitted signal. Similarly, fading can be classified into flat fadingand frequency-selective fading based on the bandwidth of transmitted sig-nal and channel coherence. Coherence bandwidth means the bandwidth orfrequency interval over which two frequencies of a signal are likely to expe-rience correlated amplitude fading. If the transmitted signal bandwidth ismuch smaller than the channel coherence bandwidth, the fading is consid-ered to be flat and if it is stated otherwise then frequency-selective fadingoccurs.There are a number of fading models in the literature. We reviewonly those models that are considered in this thesis in the following.2.4.1 Rayleigh fadingThe Rayleigh distribution is the radial component of the sum of two un-correlated Gaussian random variable. This fading model is frequently usedto characterize time varying nature of the received wireless signal when thereis no direct line-of-sight (LOS) path between the source and destination.Wireless communication channels that experience highly scattered environ-ments such as urban area, etc., we usually follow this distribution. The182.4. Fading Channelsprobability distribution function (PDF) of the Rayleigh distributed randomvariable X is given by [40]fX(x) =xσ2exp(− x22σ2) (2.2)where E[x2] = 2σ2 is the mean square value of the received signal amplitudeand x ≥ 0. In Rayleigh distribution, the power is exponentially distributedand phase is uniformly distributed that is independent from the amplitude.This fading model is widely used in wireless communication systems.2.4.2 Rician fadingIf there is a dominant LOS path exists between source and destinationalong with weaker random multipath signal components, the fading modelis assumed to have a Rician distribution. Rayleigh distribution is a specialcase of Rician distribution. Here, channel is complex Gaussian with non-zero mean and the signal envelope x = |h|. We denote, h = αjφ+vejθ whereα follows the Rayleigh distribution and v ≥ 0 and v2 is the power of theLOS signal component. θ and φ are the angles and assumed to mutuallyindependent and uniformly distributed on [−pi, pi). The PDF of the Riciandistributed random variable X is given by [40]fX(x) =xσ2exp(−x2 + v22σ2)I0(xvσ2) (2.3)The average received power, P¯r in the Rician fading is P¯r = v2 + 2σ2.The Rician distribution is often represented in terms of fading parameter Kwhich is defined as K , v22σ2. K is the relation between the power of the LOScomponent and the power of the Rayleigh component. When K →∞ thenthere is no LOS component and when K = 0, the Rician PDF simplifies toa Rayleigh PDF. If we substitute v2 = KP¯rK+1 and 2σ2 = P¯rK+1 , we can writeRician PDF asfX(x) =2x(K + 1)P¯rexp[−K − (K + 1)2x2P¯r]I0(2x√K(K + 1)P¯r)(2.4)192.5. Interference2.5 InterferenceWireless transmissions have to counter interference from a wide varietyof sources. Two main forms of interference are - i) adjacent channel inter-ference and ii) CCI. In adjacent interference, signals in nearby frequencieshave components outside their allocated ranges, which may interfere withon-going transmission in the adjacent frequencies. It can be avoided by in-troducing guard bands between the allocated frequency ranges. CCI, some-times referred to as narrow band interference, occurs due to other nearbysystems that using the same transmission frequency band. Inter-symbol in-terference is another type of interference, where distortion happens in thereceived signal. This distortion is caused by the temporal spreading and theconsequent overlapping of individual pulses in the signal.20Chapter 3Resource Allocation forDecode-and-Forward RelayedSystem with InterferenceAided Energy HarvestingIn this chapter, we consider a single–link cooperative communicationsystem, where decode–and–forward (DF) relay does not have its own en-ergy supply. Instead, it harvests energy from the interference and the signalreceived from the source and uses that harvested energy for relaying oper-ation. We consider both half–duplex and full–duplex DF relay to evaluatesystem performance. In both cases, for such systems, we develop joint opti-mal allocation of the transmit powers of the source and relay nodes and thepower–splitting ratio allocation scheme. In particular, in order to maximizethe end–to–end system throughput by joint optimal allocation of the trans-mit powers of the source and relay nodes and the power–splitting ratio, weformulate optimization problems for both half–duplex and full–duplex sys-tems. The resulting problems are non–convex polynomial programs, whichare hard to solve optimally and efficiently. Therefore, we develop an effi-cient solution by transforming, relaxing, and modifying the original problem.Numerical results show that our developed scheme for half–duplex systemperforms very close to the optimal solution with reduced complexity andprovides an intuitive direction on how to exploit the interference so as toachieve a good throughput performance. We also compare the end–to–endsystem throughput between proposed half–duplex and full–duplex system.3.1 System ModelWe consider a cooperative communication system, where a source, Scommunicates with a destination, D via a DF relay, R, see Fig. 3.1. Apartfrom S, there are M other sources, I1, I2, · · · , IM , that communicate with213.1. System ModelFigure 3.1: Illustration of the system model. A source, S communicateswith a destination, D via a DF relay, R. Interferring nodes, I1, I2, · · · , IM ,are other sources and D1, D2, · · · , DM are their corresponding destinations.their corresponding destinations D1, D2, · · · , DM . We assume that all thenodes operate in the same frequency band and perfect synchronization hasbeen made among the nodes for transmission. The communications betweenIm and Dm, m ∈ {1, 2, · · · ,M} create interference on the received signal atR. We assume that D is far apart (e.g., a cell–edge mobile node) from Imnodes and hence, these nodes create negligible interference on D1. If weincorporate the effect of interference on D, then the interference power willbe added in the de-numerator of the throughput expression of the R − Dlink. As a matter of fact, the end-to-end throughput would be much lessbecause of the interference at destination node. A notable example of theconsidered interference model can be an underlay cognitive radio network,where a dual–hop secondary link is affected by the interference created bythe primary networks. We assume that S and Im,m ∈ {1, 2, · · · ,M} arepowered by conventional constant energy supplies. On the contrary, R doesnot have a constant energy supply, rather it harvests energy from the RFsignals transmitted by S and Im nodes. We consider a power–splittingreceiver [11] at the relay node. It is worth mentioning that the consideredmodel can represent different low–power wireless networks, e.g., a multi–hopsensor network, where an energy–deficient sensor node harvests RF energyfrom other sensors.1 We intentionally assume this network configuration to show the impact of interferenceon data detection and EH at R. However, we can still incorporate the effect of interferenceon D with necessary modifications without changing the developed solution techniquedescribed in Section System ModelWe assume that the relay operates in a half–duplex mode, and thus theend–to–end transmission is accomplished over two consecutive time slots.We assume that the duration of each slot is T = 1 s. During the firstphase of the transmission, S transmits and R receives. The power–splittingreceiver at R exploits 0 ≤ ρ ≤ 1 portion of the received power at R for EHand 1 − ρ portion of the power for the signal detection [11, 41]. Therefore,the interferers are leveraged as a source of power, albeit they deteriorate theperformance of data detection. The power–splitting receiver at R strikes abalance between data detection and EH by tuning its power–splitting ratio,ρ [11]. We assume a perfect passive power splitter unit which does notconsume any power. Moreover, we assume that the receiver is equippedwith a battery of finite capacity that can temporarily store the harvestedenergy2. The received signal yR,k at R can be expressed asyR = (1−ρ)(√PSgShSxS+M∑m=1√Pmgmhmxm+nA)+ nS , (3.1)where PS and Pm represent the transmit powers of S and Im, respectively.hS (gS) and hm (gm) represent the complex–valued channel gains (real–valued path loss factors) of S–R and Im–R links, respectively. Note thatthe path loss factor takes into account the effect of transmit and receiveantenna gains, distance between the transmitter and receiver, and the pathloss exponent [41]. However, xS and xm represent the transmitted datasymbols from S and Im, respectively. For future reference, we define γS =|hS |2 and γm = |hm|2 as the squared channel–gains for S–R and Im–R links,respectively. nA and nS represent the antenna and signal processing noises,respectively [16]. Note that we use the same power–splitting receiver modelas depicted in [16]. Both nA and nS are modeled as AWGN with zeromean and variances σ2A and σ2S , respectively. We assume R performs error–free detection in order to achieve the maximum end–to–end capacity [42,Proposition 1] and hence the detected signal, xˆS = xS . Therefore, duringthe first phase of transmission, the transmission rate, ξS over the S–R linkcan be written asξS = W log2(1+(1− ρ)gSγSPSσ2S + (1− ρ)(σ2A +M∑m=1γmgmPm)), (3.2)2A battery with relatively higher capacity yields an option to store the energy for futureuse. In this case, our proposed scheme would provide a suboptimal result as storing theenergy for future and using it afterwards result in a Markov decision process, and adynamic programming approach would provide an optimal result in this case.233.2. Problem Formulation and Solutionwhere W represents the channel bandwidth. The total amount of harvestedenergy, HR from the received signal at R can be calculated asHR = ηρ(γSgSPS +M∑m=1γmgmPm), (3.3)where ηργSgSPS Joules/s of energy is harvested from S and ηρM∑m=1γmgmPmJoules/s of energy is harvested from the interferers. Here, η representsthe RF to DC energy conversion efficiency. During the second phase oftransmission, R transmits xˆS and the received signal at D can be representedasyD =√PRgRhRxˆS + nD, (3.4)where hR (gR) represents the complex–valued channel gain (real–valued pathloss factor) of R–D link and nD denotes AWGN with zero mean and σ2Dvariance. We define γR = |hR|2 as the squared channel gain of R–D link.Hence, ξR = W log2(1 +γRgRPRσ2D) bits/s of data are transmitted via R–Dlink. Therefore, the end–to–end throughput of S–R–D link is given by12 min{ξS , ξR} bits/s.3.2 Problem Formulation and SolutionIn this section, we formulate an optimization problem, which maximizes12 min{ξS , ξR} bits/s by jointly allocating PS , PR, and ρ.3.2.1 Problem formulationWe assume that the nodes Im, m ∈ {1, 2, · · · ,M} transmit with powerPm = (2Rm − 1)/γ′m to satisfy their own data rate requirements Rm, whereγ′m represents the squared channel gains of Im–Dm links. Note that withoutloss of generality, we assume the fixed–rate transmission scheme for Im–Dmlinks. As mentioned earlier, S and Im operate in the same frequency band,and hence the data transmission by Im nodes creates interference on R. Sand R have individual power budgets, and we ensure that there is no dataloss at R. Our objective is to calculate the optimal transmit powers for S andR and the power splitting ratio. We, therefore, formulate an optimization243.2. Problem Formulation and Solutionproblem as follows:P1: max{PS ,PR,ρ}0min{ξS , ξR} (3.5)s.t. C1: PS + PS,C ≤ ES (3.6)C2: PR + PR,C≤ηρ(PSγSgS +M∑m=1γmPmgm) (3.7)C3: ξS = ξR (3.8)C4: ρ ≤ 1, (3.9)where PS,C and PR,C represent constant powers required for signal process-ing operations at S and R, respectively. Note that in this work, we confineourselves towards developing a centralized scheme to solve P1. Without lossof generality, we assume that S is the central node, which has the perfectchannel state information of all the links, calculates the optimal resources ofthe whole network, and informs R about the optimal resource allocations.Note that the power required for channel estimation and signal processingtasks are incorporated in PS,C and PR,C . C1 and C2 incorporate the maxi-mum individual power budget for S and R nodes, respectively. It is worthconsidering C3 in P1 as otherwise ξS = ξR will not be satisfied at the op-timal point, and hence there will be data loss at R. C4 appears from thedefinition of the power splitting ratio receiver at R.3.2.2 Solution of the optimization problemBefore discussing the main solution strategy, we provide some intuitiveobservations made from P1 as follows:Observation 1: The constant value of PS,C (PR,C) represents the fact thatS (R) cannot be active if ES < PS,C (ηρ(PSγSgS +M∑m=1γmPmgm) < PR,C).Therefore, C2 implicitly ensures that ηρ(PSγSgS +M∑m=1γmPmgm) > PR,C isa necessary condition for a successful relaying operation and hence for datatransmission. Moreover, ES > PS,C has to be satisfied to obtain a feasiblesolution.Observation 2: To satisfy C3, either C1 or C2 meets with equality.Observation 3: P1 is a non–convex optimization problem due to the productform of ρ and PS in the objective function and C2.253.2. Problem Formulation and SolutionAs log(·) is an monotonic increasing function of its argument, we can rewritean equivalent problem of P1 (i.e., both the problems have same optimalsolution) as follows:P2: max{PS ,PR,ρ}0min{ΩS ,ΩR} (3.10)s.t. Constraints C1, C2, and C4 (3.11)C5: ΩS = ΩR, (3.12)where ΩS =(1−ρ)γSgSPSσ2S+(1−ρ)(σ2A+M∑m=1γmgmPm) and ΩR = γRgRPRσ2D . Using (3.12)in (3.10) and (3.11) and considering GS = γSgS , GR = γRgR, and G′m =M∑m=1γmgmPm yields the following optimization problem:P3: max{PR,ρ}0GRPRσ2D(3.13)s.t. Constraint C4 (3.14)C6:−G′mGRρPR + (σ2A + σ2S +G′m)GRPR+ (GSσ2D(ES − PS,C)− σ2A)ρ+ σ2A−GSσ2D(ES − PS,C) ≤ 0 (3.15)C7:− (GSσ2D + ηGSGRσ2S + ηGSGRσ2A+ ηG′mGRGS)ρPR + (ηGSGRσ2A + ηG′mGRGS)ρ2PR +GSσ2DPR − (GSσ2DPR,C + ηG′mGSσ2D)ρ+ ηG′mGSσ2Dρ2 +GSσ2DPR,C ≤ 0. (3.16)Observation 4: P3 is a polynomial program, which is a non–convex andNP–hard problem. Note that a special class of polynomial programmingproblems, called geometric programs, are tractable [43]. However, P3 is nota geometric program.It is worth mentioning that any polynomial program can be representedas a QCQP to achieve an insightful and efficient solution[44, Sec. 2]. Forthis purpose, we introduce a new optimization variable, u = ρ2 for P3. Byincorporating the new optimization variable u, we can convert the originalpolynomials into quadratic objective and constraints, thus turning the orig-inal polynomial problem into a QCQP with an additional variable and a263.2. Problem Formulation and Solutionconstraint. The transformed problem can be represented as follows:P4: max{PR,ρ,u}0GRPRσ2D(3.17)s.t. Constraint C4 and C6 (3.18)C8:− (GSσ2D + ηGSGRσ2S + ηGSGRσ2A+G′mGRGS)ρPR + (ηGSGRσ2A + ηG′mGRGS)uPR+GSσ2DPR − (GSσ2DPR,C +G′mGSσ2D)ρ+ ηG′mGSσ2Du+GSσ2DPR,C ≤ 0 (3.19)C9: u = ρ2. (3.20)For the purpose of exposition, we now represent P4 in the standard QCQPformulation. Let us denote the vector of the optimization variables as z =[PR ρ u]T . By using simple calculus, we represent the objective function ofP4 as follows:zTΛ0z + ΦT0 z + r0, (3.21)where Λ0 = O3×3 and r0 = 0. Note that Om×n represents an m× n dimen-sional null matrix. Φ0 is a 3× 1 vector with Φ0(2) = GR/σ2D. To this end,we denote A(n) to represent the nth element of A. The rest of the elementsof Φ0 are zero. Now, C4, C6, C8, and C9 can be represented as follows:zTΛiz + ΦTi z + ri ≤ 0, i ∈ {1, 2, · · · , 5}, (3.22)where Λi and Φi are 3 × 3 and 3 × 1 dimensional matrices and vectors,respectively. In particular, Φ1(2) = 1, r1 = −1, Λ2(1, 2) = −G′mGR,Φ2(1) = (1 + σ2S + G′m)GR, Φ2(2) = GSσ2D(ES − PS,C) − σ2A, and r2 =σ2A −GSσ2D(ES − PS,C).Moreover,Λ3(1, 2) = −(GSσ2D + ηGSGRσ2S + ηGSGRσ2A +G′mGRGS) (3.23)Λ3(1, 3) = ηGSGRσ2A + ηG′mGRGS (3.24)Φ3(1) = GSσ2D (3.25)273.2. Problem Formulation and SolutionΦ3(2) = −(GSσ2DPR,C +G′mGSσ2D) (3.26)Φ3(3) = ηG′mGSσ2D (3.27)r3 = GSσ2DPR,C (3.28)Λ4(2, 2) = 1 (3.29)Φ4(3) = −1 (3.30)Λ5(2, 2) = −1 (3.31)Φ5(3) = 1 (3.32)The rest of the elements of Λi and Φi, i ∈ {1, 2, · · · , 5} are zero. Note thatC9 is represented by two inequality constraints in (3.22). Now, for the sakeof clarity, we rewrite P4 in the standard form of QCQP as follows:P5: maxzzTΛ0z + ΦT0 z + r0 (3.33)s.t. zTΛiz + ΦTi z + ri ≤ 0, i ∈ {1, 2, · · · , 5}. (3.34)Observation 5: From the aforementioned transformations, we can concludethat Λi is not a positive semi–definite matrix, and hence P5 is a non–convexoptimization problem. Thus, we cannot provide an optimal solution of P5in a straight–forward manner.Let us denote zTΛz = Tr(Λ(zzT )) = Tr(ZΛ), where Tr(X) denotes thetrace operation on matrix X, and Z is a symmetric positive semi–definitematrix. Therefore, P5 can be equivalently represented as follows:P6: maxz,ZTr(ZΛ0) + ΦT0 z + r0 (3.35)s.t. zTΛiz + ΦTi z + ri ≤ 0, i ∈ {1, 2, · · · , 5} (3.36)Z = zzT . (3.37)We observe that the non–convexity arises from constraint (3.37), which ne-cessitates that the rank of matrix Z is one. We now relax P6 into a convex283.2. Problem Formulation and Solutionproblem by replacing Z = zzT with a convex positive semi–definite con-straint Z  zzT . Therefore, the relaxed problem can now be representedas follows:P7: maxz,ZTr(ZΛ0) + ΦT0 z + r0 (3.38)s.t. zTΛiz + ΦTi z + ri ≤ 0, i ∈ {1, 2, · · · , 5} (3.39)Z  zzT . (3.40)Observation 6: P7 is an SDP problem [45]. SDP is a generalization of linearprogramming (LP) over matrices. Several solvers, e.g., SeDuMi [46] canefficiently solve the SDP problems. Relaxing P6 implicitly drops the rankone property of Z, and hence solving the relaxed SDP problem, P7 does notsolve the original problem. We adopt randomization technique to achieveclose to the optimal solution from the relaxed problem, P7.Randomization technique: Let Zopt denotes the optimal solution to P7.Note that if Zopt is a rank one matrix, then the optimal z can be obtained byfinding the principle eigenvector corresponding to only non–zero eigenvalue.However, because of the SDP relaxation step, i.e., relaxation of the rank–oneconstraint, Zopt may not necessarily be a rank–one matrix.We adopt the first randomization method described in [47] to obtainan efficient solution as follows. We calculate the eigenvalue decompositionof Zopt as Zopt = UΞUH , where U represents a unitary matrix and (·)Hdenotes the Hermitian operation. We then choose the candidate solution zˆnsuch that zˆn = UΞ1/2z˜n, n ∈ {1, 2, · · · , N} where N denotes the numberof realizations and z˜n contains uniformly distributed independent randomvariables that lie on the unit circle in the complex plane. This considerationensures that zˆHn zˆn = trace(Zopt), irrespective of the value of z˜n for anyparticular realization. Therefore, when rank(Zopt) > 1, at least one of theconstraints in (3.39) will be violated. A feasible solution can be found bysimply scaling zˆn so that all the constraints are satisfied. Finally, among allthe feasible candidates zˆn, the one with the smallest norm is chosen as thesolution vector.It has been shown in [48] that the rank–relaxation technique of a QCQPproblem yields the closest convex problem to the original NP–hard problem.Moreover, the complexity of the randomization technique is small comparedto solving an SDP, and hence our proposed scheme exhibits polynomial–timecomputational complexity.293.3. Full Duplex3.2.3 Baseline methodsTo evaluate the performance of our proposed scheme, we consider twobaseline schemes, namely ‘exhaustive search’ scheme and ‘heuristic’ schemeas follows. Exhaustive search schemeIn this scheme, we quantize ρ over [0, 1] in steps of  ≤ 0.0001 andsolve P1 for each of the considered values of ρ. Note that for a given ρ,P1 is a convex optimization problem of PS and PR and hence, we can solvethe problem optimally and efficiently [49]. Note that the exhaustive searchmethod incurs high computational complexity as we have to search over allthe possible values of ρ. Heuristic schemeIn this method, we fix ρ = 0.5 to impose an equal impact on EH anddata detection and then solve P1 to compute PS and PR for the given valueof ρ. However, compared to exhaustive search scheme, this method has alower complexity at the expense of the performance degradation.3.2.4 ComplexityThe computational complexity of the proposed scheme is polynomial.In the proposed scheme, the optimization variables are PS , RR and ρ. Onthe other hand, for exhaustive search scheme, the computational complex-ity is polynomial times the number of step sizes. In heuristic scheme, P1is computed for a fixed row, which incurs much less complexity than theexhaustive search scheme.3.3 Full DuplexIn this section we will discuss about the methodology of solving the jointoptimization problem for full–duplex system.3.3.1 System modelWe consider a cooperative communication system, where a source, Scommunicates with a destination, D via a DF relay, R, see Figure 3.1[22].We assume that R operates on the full–duplex mode and introduce SI into303.3. Full Duplexthe system. In particular, R receives information from S in a current timeinterval, and forwards the information, which was received in the last timeinterval, to D simultaneously using the same carrier frequency.3.3.2 Receiver architectureWe consider power–splitting receiver at the FD relay node, R like HDsystem. We consider T = 1 s as a normalized transmission time slot du-ration. The power–splitting receiver at R exploits 0 ≤ ρ ≤ 1 portion ofreceived power at R for EH and 1 − ρ portion of the power for the signaldetection [50]. The power–splitting receiver at R maintains a balance be-tween data detection and EH by tuning its power–splitting ratio parameterρ. We adopt a perfect passive power splitter unit that is not required toconsume any power. Further, the receiver is equipped with an energy stor-age having finite capacity that can temporarily store the harvested energy.The received signal yR at R can be expressed asyR = (1−ρ)(√PSgShSxS +√PRgLhLyLxR +M∑m=1√Pmgmhmxm + nA)+ nS ,(3.41)where PS , Pm, and PR represent the transmit powers of S, Im, and R,respectively. hS (gS), hR (gR), and hL (gL) represent the channel gains(complex valued) of S–R link, R–D link, and self–interference loop at R,respectively. Note that the channel gain takes into account the effect of theantenna gains, distance between the the transmitter and receiver nodes, andthe path loss exponent [51]. Moreover, xS , xR, and xm represent the datasymbol transmitted from S, R, and Im, respectively, while yL representsthe attenuation factor along the self–interference path. In order to ensurea perfect detection of the received signal, the self–interference has to becancelled completely. In other words, the power of the self–interferencesignal cannot be greater than the power of the noise. For future reference, wedefine γS = |hS |2, γR = |hR|2, and γL = |hL|2 as the squared channel–gainsfor S–R link, R–D link, and self–interference loop, respectively. nA and nSrepresent the antenna and signal processing noises, respectively [52]. Thenoise components are modeled as AWGN with zero mean and variances σ2Aand σ2S , respectively. R is assumed to perform error–free detection in orderto achieve the maximum end–to–end capacity [53] and hence the detectedsignal, xˆS = xS . Therefore, during the first phase of transmission, the313.3. Full Duplextransmission rate ξFS (bits/s) along S–R link can be written asξFS = W log2(1+(1− ρ)gSγSPSσ2S + (1− ρ)(σ2A + γLgLyLPR +∑Mm=1 γmgmPm)),(3.42)where W represents the channel bandwidth. The total amount of harvestedenergy HR from the signals received at R can be calculated asHR = ηρ(γSgSPS + γLgLPR +M∑m=1γmgmPm), (3.43)where ηργSgSPS , ηργLgLPR, and ηρM∑m=1γmgmPm Joules/s of energies areharvested from S, the self–interference signal, and the interferers, respec-tively. Here, η represents the RF to DC energy conversion efficiency. Rtransmits xˆS during the second phase of transmission. Hence, the signalreceived at D can be represented as 3.4 and ξFR = W log2(1 +γRgRPRσ2D) bits/sof data are transmitted via R–D link. Considering all the notations men-tioned above, the end–to–end throughput of S–R–D link can be expressedas min{ξFS , ξFR} bits/s.3.3.3 Problem formulation and solutionFor the sole purpose of allocation of PS , PR, and ρ, we formulate anoptimization problem. We consider all the limitations and available re-sources and solve the problem optimally and efficiently in order to deployour method in a practical real-time environment. In this section, we for-mulate and derive an optimization problem, which maximizes min{ξFS , ξFR}bits/s by jointly allocating PS , PR, and ρ. Problem formulationWe formulate an optimization problem as follows:P1F: max{PS ,PR,ρ}0min{ξFS , ξFR} (3.44)s.t. C1F: PS + PS,C ≤ ES (3.45)C2F: PR + PR,C≤ηρ(PSγSgS + γLgLyLPR (3.46)323.4. Numerical Results+M∑m=1γmPmgm) (3.47)C3F: ξFS = ξFR (3.48)C4F: ρ ≤ 1, (3.49) Solution of the optimization problemSolution approach of the optimization problem is similar to the half-duplex problem discussed in section Numerical ResultsIn this section, we evaluate the performance of our developed resourceallocation schemes by computer simulations. For the purpose of illustration,we adopt the path–loss model standardized for IEEE 802.11 Wireless LANs[27]. In particular, we assume 20 dB directional transmit and receive an-tenna gains and 1 m reference distance. The system bandwidth is W = 200kHz and the carrier frequency is 470 MHz. The signal processing noise con-sists of quantization noise and thermal noise, where we consider the thermaland quantization noise powers are −111 dBm/Hz and −23 dBm/Hz, respec-tively [16]. Moreover, the antenna noise power and the energy conversionefficiency, η are assumed as −128 dBm/Hz and 0.8, respectively. As men-tioned earlier, the path loss components gS , gR, and gm incorporate theeffect of antenna gains, distance between the nodes, and path loss exponentin all our simulation results. We set the path loss exponent, ζ = 3.5. Forthe purpose of exposition, we assume that hS , hR, and hm are indepen-dent and identically distributed (i.i.d.) Rayleigh fading channels with zeromean and unit variance. The signal processing powers at S and R are as-sumed to be 10 dBm each. We run our simulations over 106 independentrealizations of channel gains and provide the average throughput in bits/s.However, we model hL as i.i.d. Rician fading channel with Rician parameterK = 2. Throughout the simulations, we set M = 2, dm = 3m, m ∈ {1, 2},dS = dR = 5m, and dL = 2cm. The signal processing powers at S and R areassumed to be 10 dBm each. Moreover, we set PS,C = 1 mW. Please notethat in case of FD relaying, PR,C requires higher value than that is requiredfor HD relaying. We run our simulations over 106 independent realizationsof channel gains and provide the average throughput in bits/s. Table 3.1lists all the values of simulation parameters.333.4. Numerical ResultsTable 3.1: Simulation parameters and statistical model settingAntenna gain (transmit and receive) 20dBSystem bandwidth, W 200kHzCarrier frequency 470MHzThermal noise power −111dBm/HzQuantization noise power −23dBm/HzAntenna noise power −128dBm/HzEnergy conversion efficiency, η 0.8Path loss exponent 3.5Number of interfering nodes, M 2Fading channel model RayleighFading channel model (SI loop) Rician (K = 2)Distance between S −R and R−D link 5mDistance between Im −Dm link 3m3.4.1 Throughput performance for different interferencelevels for HD systemIn Fig. 3.2, we show optimized throughput vs. interference power fordifferent values of ES . In particular, we consider two interferers, i.e., M = 2,with d1 = d2 = 3m and P1 = P2 = PI Watts and show the throughputperformance for ES ∈ {2, 5, 10, 15, 20, 25} Watts and dS = dR = 5m. Weobserve that for all the considered values of ES , the throughput increases,reaches at the maximum value, and then decreases with increasing PI . Thedemonstrated feature of the throughput curve is due to the fact that withthe increased value of PI , more energy is harvested at R and hence R canrelay more bits to D. However, a large value of PI also creates interference atR and degrades the data detection performance. Therefore, the throughputreaches at a peak value, where the interference power can optimally controlthe data detection and EH processes. Furthermore, we observe that byincreasing ES , the throughput curve shifts upwards, i.e., the throughputincreases. Increasing ES helps R harvesting more energy and compels R totransmit more data bits to D as well. We also observe that the peak valueof throughput shifts towards right by increasing ES , i.e., for a higher valueof ES , the peak throughput is observed for a higher value of interferencepower. For instance, if the interference power changes from 4 Watts to 8Watts, the highest value of throughput can be attained by increasing ESfrom 5 Watts to 25 Watts.343.4. Numerical Results0 10 20 30 40 50050001000015000PI (Watts)Throughput (bits/s)  ES = 2ES = 5ES = 10ES = 15ES = 20ES = 25Figure 3.2: Throughput vs. interference power, PI (Watts) for differentvalues of ES .3.4.2 Performance comparison with the baseline schemesfor HD systemFig. 3.3 shows optimal throughput and power splitting ratio, ρ for theproposed, exhaustive search, and heuristic schemes. In particular, the up-per figure shows optimal throughput vs. PI and the lower figure showsoptimal ρ vs. PI for the same system configurations. We set ES = 10 Wattsand consider the same values for other parameters as adopted in Fig. 3.2.For the exhaustive search scheme, we quantize the value of ρ ∈ {0, 1} witha step–size of 0.0001. For all the considered schemes, the throughput in-creases, reaches at the maximum value, and then decreases with increasingPI . Moreover, it is evident that interference (PI > 0) helps to achieve higherthroughput compared to non–interference system (when PI = 0). We ob-serve in the lower figure in Fig. 3.3 that the optimal value of ρ decreasesfrom a high value (0.73) to a low value (0.001) with increasing values of PI .When PI is small, a high value of ρ helps to harvest more energy from S,I1, and I2. On the other hand, when PI is large enough, a low value ofρ is sufficient enough to harvest energy and a large power is required fordata detection to maximize throughput. We also observe that the through-put (and optimal3 ρ) curves for the proposed scheme and the exhaustivesearch scheme are very close to each other. However, the exhaustive search3For the heuristic scheme, ρ = 0.5 is fixed for all considered values of PI .353.4. Numerical Results0 10 20 30 40 500200040006000800010000PI (Watts)Throughput (bits/s)  ProposedExhaustiveHeuristic0 10 20 30 40 5000. (Watts)ρ  ProposedExhaustiveHeuristicFigure 3.3: Performances of throughput and optimal ρ. Upper Figure:Throughput (bits/s/Hz) vs. PI (Watts). Lower Figure: ρ vs. PI (Watts).scheme provides the global optimal value, albeit high computational com-plexity due to exhaustive search operation over ρ. We observe that for lowto mid–range values of interference power, our scheme yields much betterthroughput compared to the heuristic approach, which does not take intothe account the optimal value of ρ. However, when the interference poweris very high, the throughput difference between the heuristic and proposedschemes is small. This finding exemplifies the fact that at a very high level ofinterference power, it is better to consider ρ = 0.5 instead of optimizing ρ, asthe joint optimization in the proposed scheme incurs higher computationalcomplexity.In the following subsections, we discuss simulation results for the con-sidered full-duplex system.3.4.3 Throughput analysis for different levels of signalprocessing power at the relay for FD systemWe observe in Fig. 3.4, that the throughput increases with increasingES . This outcome is expected as the higher energy availability at S helpsto transmit more data bits from S to D by supplying more energy to Sexplicitly and to R implicitly. We also observe that increasing PR,C yieldsdegraded throughput performance as higher requirement of power for signalprocessing reduces the energy availability for data transmission. We have363.4. Numerical Results2 4 6 8 10 12 14 16 18 20ES (W)103104105Throughput (bits/sec)HDFD: PR,C=0.5 WFD: PR,C=1.0 WFD: PR,C=2.0 WFD: PR,C=3.0 WFD: PR,C=4.0 WFigure 3.4: Performances of throughput vs. ES (Watts) for different signalprocessing powers at relay, R.also included the throughput performance of a similar network but witha HD relay [22]. In general, the required signal processing power for FDrelaying is higher than that of HD relaying. However, increasing PR,C forFD relaying reduces the throughput gap between FD and HD relaying andfor a very high value of PR,C (for FD relaying) the throughput performancesof FD and HD relaying are same.3.4.4 Throughput analysis for different SI levelsIn Figure 3.5, we show throughput vs. ES for different values of SI levels.We set c = ζσ2S and changed the values of ζ from low level to high level.For the purpose of comparison, we included the throughput performanceof HD relaying as well [22]. We observe that if we cannot cancel the self-interference for FD relaying efficiently, it will lead to degraded performance,and for instances, worse performance than the HD relaying. Therefore, it ishighly imperative to cancel the self-interference level for FD duplex relayingin order to achieve high system throughput.373.5. Summary2 4 6 8 10 12 14 16 18 20ES (W)102103104105Throughput (bits/sec)HDFD:  = 0.01FD:  = 0.1FD:  = 1.0FD:  = 10FD:  = 100Figure 3.5: Performances of throughput vs. ES (Watts) for different levelsof self-interferences at R.3.5 SummaryIn this chapter, we have developed a resource allocation scheme, whichjointly optimizes the source and relay transmit powers along with the power–splitting ratio to maximize the end–to–end system throughput for an inter-ference aided EH cooperative communication system. We have formulateda polynomial optimization problem and have developed an efficient solutionscheme that provides very close to the optimal scheme. Numerical resultsshow that our developed scheme performs close to the optimal exhaustivesearch scheme and much better than the trivial heuristic scheme. Further-more, the results provide us with the insights on how to optimally allocatetransmit powers and power–splitting ratio for different levels of interference.38Chapter 4Power Allocation and RelaySelection for EnergyHarvesting SystemsIn this chapter, we develop a stochastic joint relay selection and transmitpower allocation scheme for a two-hop amplify-and-forward multi-relay net-work, where relays harvest renewable energies from the surrounding environ-ment. We study the optimal policy for relay selection and power allocationunder unknown statistics of the channel fading and energy arrival processes.In particular, we formulate a stochastic optimization problem as an infinite-horizon Markov decision process. Then we develop a low-complexity onlinealgorithm that does not require the statistics of the channel fading and en-ergy harvesting processes to be known. The considered multi-relay systemmodel is a classical example of cooperative communication systems. Ourdeveloped optimal stochastic online algorithm is a completely new approachto jointly select the best relay and allocate its transmit power when thestatistics of EH and channel processes are unknown. Please note that thisapproach is a unique way to optimize system resources for cooperative com-munication systems. We developed an efficient learning approach that incurslower computational complexity as it does not require channel states to beknown. Simulation results reveal that the developed algorithm performsclose to the offline scheme proposed in the literature.4.1 System ModelLet us consider an EH AF relay system with one source S, one destina-tion D, and N relays Rn, n ∈ {1, 2, · · · , N} , see Fig. 4.1. We assume that Sand D are not EH nodes and have continuous supply of power. On the con-trary, Rn are EH nodes and rely on the harvested energies to communicate.These nodes are equipped with batteries with sufficiently large capacities.We assume the data transmission happens in equal duration time intervals,394.1. System ModelFigure 4.1: Illustration of the system model consists of one source S, onedestination D, and N relays Rn, n ∈ {1, 2, · · · , N}.where each interval consists of two time slots of duration t. In the following,we set t = 1s.We define γSRn,k and γRnD,k as the squared channel gains of the S–Rnand Rn–D links, respectively in interval k ∈ {1, 2, · · · }. The S–Rn (Rn–D) channel processes {γSRn,k} ∈ GSRn,k ({γRnD,k} ∈ GRnD,k) are ergodic(the time average is equal to the ensemble average) and independent andi.i.d. with pdf pGSRn,k(γSRn,k) (pGRnD,k(γRnD,k)) over the channel state spaceGSRn,k (GRnD,k). The transmit powers of S and Rn are denoted by PS,k andPRn,k, respectively. We assume the channels are quasi–static within eachinterval k. Moreover, γSRn,k (γRnD,k) is identically distributed over the timeintervals. We denote the average channel SNRs of the S–Rn and the Rn–Dlinks as γ¯SRn and γ¯RnD, respectively. In each time interval k ∈ {1, 2, · · · },a single relay RΞ, where Ξ ∈ {1, 2, · · · , N} is selected out of the N relaysthat provides the highest end-to-end SNR to forward the signals receivedfrom S to D. During the first time slot, S transmits and RΞ receives, andduring the second time slot, RΞ transmits and D receives. The end-to-end(received at D) SNR SNReq,Ξ,k can be stated as [26]SNReq,Ξ,k =PS,kγSRΞ,kPRΞ,kγRΞD,kPS,kγSRΞ,k + PRΞ,kγRΞD,k + 1, (4.1)where the noise power spectral densities at RΞ and D are assumed as unity.Thus, the end–to–end (S–D) system throughput (normalized to channelbandwidth) in interval k is given by 0.5 log2(1 + SNReq,Ξ, k) bits/s/Hz. Wealso define wn,k as the relay selection variable which can be either 0 or 1. In404.2. Problem Formulation and Solutionparticular, wn,k = 1 (wn,k = 0) indicates that Rn is selected (not selected)for transmission in time interval k.The energy harvester at node N ∈ {1, 2, · · · , N} collects HN ,k Joules (J)of energy from its surroundings by the end of the kth interval. {HN ,k} ∈ HNis modeled as a stationary ergodic random process with energy state spaceHN and pdf pHN (H). HN ,0 denotes the amount of energy stored beforethe transmission begins. We assume that the power required for signalprocessing remains constant over time intervals. Let us denote the signalprocessing power for relay N as PCN . The energy stored in the battery atthe beginning of time interval k is represented by BN ,k, which is updatedas follows:BN ,k+1 = [BN ,k − wN ,kP˜N ,k]+ +HN ,k, ∀k, (4.2)where [x]+ = max{x, 0} and P˜N ,k = PN ,k + PCN . Thus, {BN ,k} ∈ BNfollows a first–order MDP that depends only on the present and immediatepast conditions. The state of the battery at relay N is represented by BN .4.2 Problem Formulation and SolutionIn this section, we describe how we develop joint relay transmit powerand relay selection scheme. Our goal is to develop an algorithm that can beimplemented online without requiring the statistical information about thechannels and EH processes to be known.Problem Formulation:We aim to develop an optimal online algorithmthat provides optimal utility over infinite-horizon for the considered systemmodel. The randomness in the energy arrival process and the energy causal-ity make the design and the analysis of EH relay systems more challengingcompared to conventional (non-EH) systems. Energy causality constraintensures that energy expenditure is always less than the stored energy at anygiven time. We formulate an optimization problem P1 as follows:P1: maxPN ,k≥0,wn,k ∀k,∀nE{0.5 log2(1 + SNReq,k)}(4.3)s.t.:N∑n=1wn,k = 1, ∀k (4.4)wn,k ∈ {0, 1}, ∀k (4.5)Constraint (4.2), (4.6)414.2. Problem Formulation and Solutionwhere for sufficiently high SNR, SNReq,k can be represented as [54] SNReq,k =N∑n=1wn,kPS,kγSRn,kPRn,kγRnD,kPS,kγSRn,k+PRn,kγRnD,k. Here, E{·} denotes statistical expectation. InP1, we maximize the end-to-end expected throughput while incorporatingthe battery constraints in each relay (4.6). However, only one relay is se-lected in each time interval. This feature is incorporated in constraints (4.4)and (4.5). The formulation of the optimization problem is similar to theframework considered in [26]. However, unlike [26] we do not optimize PS .Problem Solution: We observe that P1 is an infinite–horizon MDP. Wepresent an approach to solve P1 analytically and efficiently in the following.As part of developing the solutions of P1, let us denote N–tuples ofthe battery and harvested energy states in a given time interval for all re-lays as B = (B1 · · · , BN ) ∈ B =N∏n=1Bn and H = (H1 · · · , HN ) ∈ H =N∏N=1Hn, respectively. Furthermore, we define 2N–tuples of channel statesas γ = (γSR1 , · · · , γSRN , γR1D · · · γRND) ∈ G =2N∏n=1Gn. Moreover, N–tuplesof transmit powers and N–tuples of relay selection variables can be denotedas P = (P1 · · · , PN ) ∈ R+N and w = (w1, · · · , wN ) ∈ {0, 1}N , respec-tively, where R+ represents the set of non–negative real numbers. Similarly,we denote P˜ = (P˜1 · · · , P˜N ) ∈ R+N for notational convenience. In eachtime interval, the controller at the central node (e.g., S) observes the state(B,γ) and determines the actions P and w simultaneously. We define astationary policy pi for this scheme as a sequence of all feasible decisionrules. The stationary policy pi can be represented by a 2N–tuple function(P ,w) : B × G → R+N × {0, 1}N . According to [55], the optimal solutionof an MDP can be obtained by defining the state-value function followingBellman’s optimality equation.Following the reasons above, we now denote U(B,γ) as the state–valuefunction for P1. In particular, U(B,γ) is the optimal value of P1 with theinitial state (B,γ). The Bellman’s optimally equation for P1 can be writtenas follows [56]:U(B,γ) = max{0.5 log2(1 + SNReq)+∑γˆ∈G, ˆH∈H2N∏n=1pGn(γˆn)N∏n=1pHn(Hˆn)U(B −w  P˜ + Hˆ, γˆ)}− U(B0,γ0), (4.7)for a fixed state (B0,γ0). Here,  denotes the point–wise product operation.424.2. Problem Formulation and SolutionMoreover, pGn(γˆn) and pHn(Hˆn) denote the pdfs of the channel SNRs andthe harvested energies, respectively.Remark 1: The state-value function in (4.7) has two parts; the first part0.5 log2(1 + SNReq) reflects the immediate (current) outcome, whereas thesecond part∑γˆ∈G, ˆH∈H2N∏n=1pGn(γˆn)N∏n=1pHn(Hˆn)U(B−w P˜ +Hˆ, γˆ) reflectsthe long-term (future) effect.Remark 2: Solving (4.7) requires the pdfs of the channel SNR and har-vested energies to be known. As mentioned earlier, these statistics may notbe accurately predicted ahead of time. Therefore, in this work, we developthe scheme in such a way so that we do not need to incorporate these pdfsto solve the problem.Remark 3: Moreover, we observe that the state-value function depends onB and γ. With the increase of N , the size of the problem states increase.Therefore, our objective is to reduce the computational complexity by re-ducing the number of states.Remark 4: We also observe that the expectation in (4.7) is embedded intothe term of maximization that makes the problem less amenable for onlineimplementation using time-averaging.Based on above mentioned remarks, we define the ‘post–decision’ state–value function V (Bˇ) by following the similar approach made in [27]. V (Bˇ)is denoted as follows:V (Bˇ) =∑γˆ∈G, ˆH∈H2N∏n=1pGn(γˆn)N∏n=1pHn(Hˆn)U(Bˇ + Hˆ, γˆ) (4.8)for post–decision states Bˇ ∈ B. Notice that the dynamics of the battery ofnode N in time interval k can be represented as BˇN ,k = [BN ,k−wN ,kP˜N ,k]+.From (4.7) and (4.8), we can write the optimality equation as follows:V (Bˇ) =∑γˆ∈G, ˆH∈H2N∏n=1pGn(γˆn)N∏n=1pHn(Hˆn)max{0.5 log2(1 + SNReq)+V (Bˇ + Hˆ −w  P˜ )}− V (Bˇ0) (4.9)for some arbitrary but fixed state Bˇ0.Lemma 4.1. For a selected relay RΞ, the post-decision state-value functionV (Bˇ) is a concave non-decreasing function of Bˇ.434.2. Problem Formulation and SolutionProof. We can follow the steps described in the proof of [57, Theorem 1] toprove Lemma 1.Remark 5: V (Bˇ) is a representation of U(B,γ) in different format. Bothof the value functions provide the same optimal results.Remark 6: Introducing the post-decision state and corresponding valuefunction (4.9) allows us to perform the maximization without computing theexpectation as the expectation operator is now outside the maximizationoperator. We can therefore apply time-average algorithm. In particular,we can optimize the relay transmit power and select the best relay andupdate the state-value function based on the information obtained from theoptimized results in each time intervals. These findings are exploited foronline learning V (Bˇ) using time averaging.Remark 7: Using post-decision approach helps reducing the number ofstates to compute the state-value function as we do not need to keep trackof the channel states in V (Bˇ) to achieve the optimal state-value function.Based on Remark 7, we now describe an online algorithm in the follow-ing to solve the optimization problem and hence to jointly select relay andallocate their transmit powers over transmission time intervals.Online Algorithm: We now propose an online scheme to obtain the op-timal policy pi∗. First, notice that, V (Bˇ) can be computed using the time-averaged recursive value iteration algorithm as follows for k = 1, 2, · · · :Vk+1(Bˇ) =∑γˆ∈G, ˆH∈H2N∏n=1pGn(γˆn)N∏n=1pHn(Hˆn)max{0.5 log2(1 + SNReq)+Vk(Bˇ + Hˆ −w  P˜ )}− Vk(Bˇ0) (4.10)with an initial value of Vk(Bˇ0). The convergence of (4.10) to V (Bˇ) can befound by following the similar steps described in [58]. We now describe theimplementation strategy of the proposed online algorithm as follows:− Optimization stage: Let us initialize V1(Bˇ) and fix Bˇ0 ∈ B for all therelays. For k = 1, 2, · · · , the relay powers are determined (as if all ofthem are selected) byPn,k = arg maxPn,k∈[0,Bn,k]{0.5 log2(1 + SNReq) + Vk(Bn,k − Pn,k)}. (4.11)Then a relay is selected which provides the maximum end-to-endthroughput. The transmit powers of the non-selected relays are as-signed to zero.444.2. Problem Formulation and Solution− Statistical learning stage: We update the post–decision state–valuefunction as:Vk+1(Bˇ)=(1−φk)Vk(Bˇ)+φk(max{0.5 log2(1 + SNReq)+Vk(Bˇ + Hˆ − w  P˜ )}− Vk(Bˇ0)), (4.12)where φk represents the step-size parameter for value–iteration func-tion [58].Algorithm 1 is presented here to solve the above mentioned optimizationproblem.Algorithm 1: Time Averaging Algorithm1. Initialize V1(Bˇ0) and fix Bˇ0 ∈ B for all the relays2. Initialize maximum tolerance 3. repeat {Main Loop}4. Solve 4.11 and select the best relay n andits optimal power Pn.k for time slot k5. Update Vk+1(Bˇ) by using 4.126. if Vk+1(Bˇ)− Vk(Bˇ) ≤  for all the relays then7. Convergence = true8. else9. Convergence = false10. end if11. until Convergence = trueRemark 8: The optimization problem (4.11) is mixed-integer non-linear(MINLP) problem, which can be solved by Generalized Bender’s decomposi-tion (GBD) method as deployed in [26]. However, the complexity in solving(4.11) is linear with N . Therefore, we evaluate (4.11) by solving Pn,k for allcombinations of relay selection variables and selecting that particular relay,which provides the optimal throughput. Please note that we formulated aMDP and deduced a state-value function by following Bellman’s equation.It is worth mentioning that this is a standard approach to solve a MDP.However, we observed that the computational complexity to solve the prob-lem by the standard state-value approach is reasonably high as we need tokeep track of channel states of all the links (in addition to energy states ofall the relay nodes) in the function for each iteration. Therefore, we adoptedpost-decision approach and hence deduced a modified state-value function454.3. Numerical ResultsTable 4.1: Simulation parameters and statistical model settingFading channel model RayleighAverage channel SNR 10dBEH profile Uniformly distributedSignal processing power at relay, PCN 0.01Wthat yields a lower computational complexity. The modified post-decisionstate-value function in (4.11) does not require the channel state to be con-sidered in each iteration, and hence the overall computational complexity isreduced.4.3 Numerical ResultsIn this section, we demonstrate numerical results obtained from our con-sidered model. We evaluate the performance of our proposed scheme byMonte-Carlo simulations in MATLAB.Implementation strategy: At first, we describe the implementation strat-egy of our online algorithm. We employ the online algorithm proposed inthe previous section by following optimization and learning stages. We ini-tialize V1(Bˇ) for all the relays at the beginning of transmission. In eachtime interval, we calculate the possible transmit powers for all the relaysaccording to (4.11) as if all of them are selected for transmission. We thenselect that relay for transmission that yields the highest end-to-end systemthroughput. The transmit powers of rest of the relays are reset to zero.These steps accomplish the optimization stage. We then update Vk(Bˇ) forall the relays according to (4.12). Please note that Vk(Bˇ) represents thelong-term effect on throughput. Moreover, the proposed algorithm does notrequire the pdfs of channel SNR and incoming energy to be known to cal-culate Vk(Bˇ). Intuitively, we keep learning Vk(Bˇ) over transmission timeintervals while updating/adjusting their values. Throughout the simulationresults, we assume i.i.d. Rayleigh fading channels for all the links with anaverage channel SNR of 10 dB. Moreover, we consider uniformly distributedEH profile across each relay node with average harvested energy H¯n and setPCN = 0.01W. Table 4.1 lists all the simulation parameters.Throughput vs. average harvested energy: In Fig.4.2, we show theaverage number of transmitted bits obtained from our proposed schemeas a function of average harvested energy (H¯1 = H¯2 = · · · = H¯N ) fordifferent values of PS . We consider N = 3 and 106 time intervals in this464.3. Numerical Results2 4 6 8 10Average harvested energy (J/s)1.822. throughput (bits/s/Hz) PS = 2WPS = 4WPS = 6WPS = 8WPS = 10WFigure 4.2: Throughput vs. average harvested energy for different values ofPS for N = 3.setup. We observe that the system throughput increases with increasingharvested energy as more energy helps to relay more data bits from S to D.Moreover, increasing PS helps to increase the system throughput as well bytransmitting more data bits from S.Throughput vs. number of time intervals: In Fig.4.3, we show theconvergence behavior of the proposed time-averaging scheme for three dif-ferent scenarios by setting N = 3 and PS = 10W. We assume H¯1 = H¯2 =· · · = H¯N = H¯ and consider H¯ = 4 J/s, H¯ = 7 J/s, and H¯ = 10 J/s forScenarios 1, 2, and 3, respectively. We observe that the algorithm convergesto the steady-state within finite number of iterations for all the consideredscenarios. In Fig.4.4, we compare the performance of our proposed algo-rithm with that of the existing algorithms in the literature. We evaluate theperformance of the algorithms in [26] and compare them with our one. Wecalculate the average throughput as a function of number of transmissiontime intervals. Note that the algorithms in [26] are proposed for finite num-ber of time intervals. Therefore, for the purpose of comparison, we evaluatethe performance of our developed algorithm for finite number of time slots.We consider N = 5, PS = 8W, and H¯1 = H¯2 = · · · = H¯N = 10J for thissetup. We observe that the offline scheme performs close to our proposedscheme for higher number of time intervals. The optimal offline scheme de-scribed in [26] assumes that both causal and non-causal information aboutthe channel and energy states are known a priori and thus provides the474.4. Summary0 200 400 600 800 1000 1200 1400 1600 1800 2000Iteration Index00.511.522.533.54Throughput (bits/s/Hz)Scenario 1Scenario 2Scenario 3Figure 4.3: Convergence behavior of proposed scheme.optimal throughput by exploiting all these information. In fact, this of-fline scheme provides the upper bound of throughput irrespective of anysystem parameters, e.g., considered time horizon. In particular, this offlinescheme yields the best performance for small and large number of time in-tervals. However, the assumptions behind developing the offline scheme arenot completely realizable in practice; it is very unlikely to accurately predictthe future incoming energy and future channel states as these processes arerandom in time. This observation reflects the fact that our algorithm per-forms close to optimal result for large number of time intervals. Moreover,the proposed algorithm performs much better than the suboptimal onlinealgorithms proposed in [26] that requires the average energy harvesting rateto calculate the transmit power and to select the best relay. The onlinescheme we developed considers only the causal information about the chan-nel and EH processes and provides optimal result when we calculate thethroughput over a large number of time intervals. For the small number oftime intervals, the scheme cannot provide throughput close to offline schemein [26] as our scheme learns statistics of the random processes over time andhence provides better throughput for large number of time intervals.4.4 SummaryWe have developed a stochastic joint relay selection and transmit powerallocation scheme for a two-hop amplify-and-forward multi-relay network484.4. Summary10 20 30 40 50 60 70 80 90 100Number of transmission time intervals2345678Avergae throughput (bits/s/Hz)Offline [4]HR assisted [4]Naive [4]ProposedFigure 4.4: Throughput vs. total number of transmission time intervals forN = 5.where relays harvest renewable energies from the surrounding environment.We have developed a low-complexity optimal online (real-time) joint relayselection and its transmit power allocation scheme that maximizes the end-to-end system throughput. The proposed algorithm does not require thestatistics of the channel and energy harvesting processes to be known. Pre-sented simulation results exhibited that our developed algorithm performsclose to the existing offline scheme even without requiring the channel andenergy statistics to be known.49Chapter 5ConclusionsIn this chapter, we summarize the contributions of this dissertation, andpropose some future work related to resource allocation for energy harvestingsystem.5.1 Summary of ContributionsIn this thesis, we studied two different optimization frameworks for allo-cating resources in EH wireless communication systems. First, we considerinterference aided RF-EH for single link communication system. Second, wedeveloped an efficient online learning algorithm for multi relay based net-work where EH node harvests energy from renewable sources. The technicalcontributions of this thesis can be summarized as follows.− In Chapter 3, we developed an efficient optimization framework for aninterference limited relay network for both HD and FD relays, wherethe relay harvests energy from the signal received from the source andfrom the interference i.e., RF-EH. Our objective is to show the con-structive role of interference for the relay network. We do not controlthe interference power, but we can control the amount of incominginterference by tuning the power splitting ratio ρ. In particular, whenρ is large, R harvests more energy from the interference (and from thesources signal), but detects less amount of data that are transmittedby the source. On the other hand, when ρ is small, R detects moredata but harvests less amount of energy. So, we can say that for agiven level of interference power, we can control the usage of interfer-ence by tuning up ρ. As ρ is an optimization variable along with PSand PR, our optimization algorithm can correctly choose the value of ρso that the end-to-end throughput is maximized. In addition, for FDrelays, we observe that, if we cannot cancel the self-interference effi-ciently, it will lead to degraded performance, and for instances, worseperformance than the HD relaying.505.2. Future Work− In Chapter 4, the considered multi-relay system model is a classicalexample of cooperative communication systems where relay harvestsenergy from renewable sources. We performed joint relay selectionand power allocation by developing efficient algorithm. However, ourdeveloped optimal stochastic online algorithm is a completely new ap-proach to jointly select the best relay and allocate its transmit powerwhen the statistics of EH and channel processes are unknown. Thisapproach is a unique way to optimize system resources for coopera-tive communication systems. In offline schemes, we assume that thenoncausal information regarding the channel SNR, the harvested en-ergy, etc. are known a priori. In general, offline schemes provide use-ful performance upper bounds for the more practical online schemes.Moreover, these schemes can provide design insights that can help usto design an efficient online scheme. on the other hand, in practice,the amount of harvested energy, the channel SNR are random in na-ture and cannot be predicted in advance. Therefore, we developed anefficient online learning approach for resource allocation that incurslower computational complexity as it does not require channel statesto be known.5.2 Future WorkResource Allocation for Energy Harvesting Massive IoT Net-work: One of the envisioned features for 5G is massive IoT network, wheredifferent standards will have a common framework to communicate witheach other [59]. Billions of devices will be connected in the network thatwill lead to handling of large amount of data over the network. Manag-ing this large amount of data over a massive IoT network is challenging forboth physical (PHY) and higher layer protocol designers. For the PHY andmedium access control (MAC) layers, cross layer optimizations with limitedavailable information about resources can be designed. For higher layers,e.g., network and application layers, different big data analysis tools canbe exploited in order to investigate the performance of IoT system. Forinstance, software defined mobile networking (SDMN) approach can be en-visioned to be deployed to design radio access networks for different IoTstandards [60]. Large amounts of wireless data can be efficiently handled bydecoupling the control and data planes for EH-IoT networks according tothe SDMN methodology.Radio Frequency Energy Harvesting in millimeter wave sys-515.2. Future Worktems: In this thesis, we have considered harvesting energies from RF elec-tromagnetic signals with frequency lower than 3GHz. However, the primecandidate for 5G cellular networks is millimeter wave (mmW) bands, wherethe carrier frequency can span from 3GHz to 300GHz [61]. The propagationcharacteristics of higher frequency signal is different from that of sub-3GHzsignals. At higher frequencies, the path loss is much higher than that inlower frequencies. Morover, the penetration loss is high in higher frequen-cies. To encounter these effects, large number of antenna arrays should beused at base station. Considering all these facts, the EH capability of a sys-tem, e.g., average EH rate at a given distance, energy coverage probability,etc. is worth analyzing from analytical perspective.Communications with Energy Harvesting Drone Base Stations:Mobile base stations for small cells have recently attracted by academia andindustry. In particular, standardization body, 3rd Generation PartnershipProject (3GPP) has started its study items on drone communications forpublic safety and general emergency service [62]. Harvesting renewable en-ergies, e.g., solar energy, etc., at the drone can provide a supplemental powersupply to any existing conventional energy source. However, allocating theenergy for transmission power of a moving node, e.g., drone, is more chal-lenging. Following the strategies developed in Chapter 4, we will developresource allocation algorithms that will jointly allocate transmit power anddrone altitude.Energy Conversion Efficiency for RF-EH Systems: In this thesis,we consider 80 percent of energy conversion efficiency for RF-EH commu-nication systems. Improvement of this energy conversion efficiency is de-pendent on the underlying hardware. In general, if harvested RF power issmall, RF-to-DC energy conversion efficiency is low. Higher conversion effi-ciency can be achieved by diodes with lower built-in voltage. If the SWIPTreceiver exhibits variable percentage of conversion efficiency, the system’sperformance will be varying too. Thus, resource allocation by taking intoconsideration of variable energy conversion efficiency can be an interestingproblem for researchers in RF-EH wireless communication systems.52Bibliography[1] J. A. Paradiso and T. Starner, “Energy scavenging for mobile and wire-less electronics,” IEEE Pervasive Computing, vol. 4, no. 1, pp. 18–27,Jan 2005. → pages viii, 12[2] Powercast. [Online]. Available: → pagesviii, 10, 11, 12[3] Towards 5g. [Online]. Available: → pages ix, 1, 2[4] Ericsson: Connected devices. [Online]. Available: 1686565587 c →pages 1[5] M. A. Khan, M. Z. Khan, K. Zaman, and L. Naz, “Global estimatesof energy consumption and greenhouse gas emissions,” Renewable andSustainable Energy Reviews, vol. 29, no. Supplement C, pp. 336 –344, 2014. [Online]. Available: → pages 1[6] O. Ozel, K. Tutuncuoglu, S. Ulukus, and A. Yener, “Fundamental lim-its of energy harvesting communications,” IEEE Commun. Magazine,vol. 53, no. 4, pp. 126–132, April 2015. → pages 1[7] Z. Hasan, H. Boostanimehr, and V. Bhargava, “Green cellular networks:A survey, some research issues and challenges,” IEEE Commun. SurveysTutorials, vol. 13, no. 4, pp. 524–540, Fourth 2011. → pages 2[8] O. Ozel, K. Tutuncuoglu, J. Yang, S. Ulukus, and A. Yener, “Transmis-sion with energy harvesting nodes in fading wireless channels: Optimalpolicies,” IEEE J. Sel. Areas Commun., vol. 29, no. 8, pp. 1732–1743,September 2011. → pages 2[9] Q. Zhang and S. Kassam, “Finite-state markov model for rayleigh fad-ing channels,” IEEE Trans. Commun., vol. 47, no. 11, pp. 1688–1692,Nov 1999. → pages 253Bibliography[10] T. Zhu, Z. Zhong, Y. Gu, T. He, and Z. L. Zhang, “Leakage-aware En-ergy Synchronization for Wireless Sensor Networks,” Proc. of the 7thInternational Conference on Mobile Systems, Applications, and Ser-vices, pp. 319–332, 2009. → pages 2[11] H. Wang and N. B. Mandayam, “A simple packet-transmission schemefor wireless data over fading channels,” IEEE Trans. Commun., vol. 52,no. 7, pp. 1055–1059, July 2004. → pages 2, 22, 23[12] J. Cui and A. Sheikh, “Outage probability of cellular radio systemsusing maximal ratio combining in the presence of multiple interferers,”IEEE Trans. Commun., vol. 47, no. 8, pp. 1121–1124, Aug 1999. →pages 2[13] D. Tse and P. Viswanath, Fundamentals of wireless communication.Cambridge university press, 2005. → pages 2[14] W. Nam, D. Bai, J. Lee, and I. Kang, “Advanced interference man-agement for 5G cellular networks,” IEEE Commun. Magazine, vol. 52,no. 5, pp. 52–60, May 2014. → pages 3[15] K. Zheng, S. Ou, J. Alonso-Zarate, M. Dohler, F. Liu, and H. Zhu,“Challenges of massive access in highly dense lte-advanced networkswith machine-to-machine communications,” Wireless Communications,IEEE, vol. 21, no. 3, pp. 12–18, June 2014. → pages 3, 4[16] D. Ng, E. Lo, and R. Schober, “Wireless information and power trans-fer: Energy efficiency optimization in OFDMA systems,” IEEE Trans.Wireless Commun., vol. 12, no. 12, pp. 6352–6370, December 2013. →pages 3, 23, 33[17] I. Ahmed, A. Ikhlef, R. Schober, and R. Mallik, “Power allocationfor conventional and buffer-aided link adaptive relaying systems withenergy harvesting nodes,” IEEE Trans. Wireless Commun., vol. 13,no. 3, pp. 1182–1195, March 2014. → pages 3, 4[18] Y. Zeng and R. Zhang, “Full-duplex wireless-powered relay with self-energy recycling,” IEEE Wireless Commun. Lett., vol. 4, no. 2, pp.201–204, April 2015. → pages 3, 4[19] C. Zhong, H. A. Suraweera, G. Zheng, I. Krikidis, and Z. Zhang, “Wire-less information and power transfer with full duplex relaying,” IEEETrans. Commun., vol. 62, no. 10, pp. 3447–3461, Oct 2014. → pages 354Bibliography[20] Z. Chen, P. Xu, Z. Ding, and X. Dai, “The application of swipt to acooperative full duplex network,” in 2015 International Symposium onWireless Communication Systems (ISWCS), Aug 2015, pp. 426–430.→ pages 3, 4[21] T. N. Kieu, D.-T. Do, X. N. Xuan, T. N. Nhat, and H. H.Duy, Wireless Information and Power Transfer for Full DuplexRelaying Networks: Performance Analysis. Cham: SpringerInternational Publishing, 2016, pp. 53–62. [Online]. Available: 5 → pages 4[22] I. Ahmed, M. J. Hossain, and I. Ahmed, “Resource allocation fordecode-and-forward relayed system with interference aided energy har-vesting,” in 2015 IEEE International Conference on Ubiquitous Wire-less Broadband (ICUWB), Oct 2015, pp. 1–5. → pages 4, 30, 37[23] C. Huang, R. Zhang, and S. Cui, “Throughput maximization for thegaussian relay channel with energy harvesting constraints,” IEEE J.Sel. Areas Commun., vol. 31, no. 8, pp. 1469–1479, Aug. 2013. →pages 5[24] A. Kansal, J. Hsu, S. Zahedi, and M. B. Srivastava, “Power manage-ment in energy harvesting sensor networks,” ACM Trans. EmbeddedComp. Syst., vol. 6, no. 4, p. 32, Sep. 2007. → pages 5[25] L. Jiang, H. Tian, Z. Xing, K. Wang, K. Zhang, S. Maharjan, S. Gjess-ing, and Y. Zhang, “Social-aware energy harvesting device-to-devicecommunications in 5g networks,” IEEE Trans. Wireless Commun.,vol. 23, no. 4, pp. 20–27, Aug. 2016. → pages 5[26] I. Ahmed, A. Ikhlef, R. Schober, and R. K. Mallik, “Joint power allo-cation and relay selection in energy harvesting af relay systems,” IEEEWireless Commun. Lett., vol. 2, no. 2, pp. 239–242, Apr. 2013. → pages5, 6, 40, 42, 45, 47, 48[27] K. Phan, T. Le-Ngoc, M. van der Schaar, and F. Fu, “Optimal schedul-ing over time-varying channels with traffic admission control: Struc-tural results and online learning algorithms,” IEEE Trans. WirelessCommun., vol. 12, no. 9, pp. 4434–4444, Sep. 2013. → pages 5, 33, 43[28] Y. Zhang, J. Ge, J. Men, F. Ouyang, and C. Zhang, “Joint relay se-lection and power allocation in energy harvesting af relay systems with55Bibliographyicsi,” IET Microwaves, Antennas Propagation, vol. 10, no. 15, pp. 1656–1661, 2016. → pages 5, 6[29] J. Men, J. Ge, C. Zhang, and J. Li, “Joint optimal power allocationand relay selection scheme in energy harvesting asymmetric two-wayrelaying system,” IET Commun., vol. 9, no. 11, pp. 1421–1426, 2015.→ pages 5, 6[30] Y. Mao, J. Zhang, S. H. Song, and K. B. Letaief, “Joint link selectionand relay power allocation for energy harvesting relaying systems,” in2014 IEEE Global Communications Conference, Dec 2014, pp. 2568–2573. → pages 6[31] International energy agency. [Online]. Available: → pages 9[32] Enocean gmbh: Solar cells for energy harvesting wire-less sensors. [Online]. Available: → pages 10[33] National wind technology center. [Online]. Available: → pages 10[34] Z. Popovic, “Cut the cord: Low-power far-field wireless powering,”IEEE Microwave Magazine, vol. 14, no. 2, pp. 55–62, March 2013. →pages 11[35] Z. Zhang, K. Long, A. V. Vasilakos, and L. Hanzo, “Full-duplex wirelesscommunications: Challenges, solutions, and future research directions,”Proceedings of the IEEE, vol. 104, no. 7, pp. 1369–1409, July 2016. →pages 15[36] A. Sahai, G. Patel, and A. Sabharwal, “Pushing the limits offull-duplex: Design and real-time implementation,” CoRR, vol.abs/1107.0607, 2011. [Online]. Available: → pages 15[37] D. Bharadia, E. McMilin, and S. Katti, “Full duplex radios,” inProceedings of the ACM SIGCOMM 2013 Conference on SIGCOMM,ser. SIGCOMM ’13. New York, NY, USA: ACM, 2013, pp. 375–386.[Online]. Available: →pages 1656Bibliography[38] K. Haneda, E. Kahra, S. Wyne, C. Icheln, and P. Vainikainen, “Mea-surement of loop-back interference channels for outdoor-to-indoor full-duplex radio relays,” in Proceedings of the Fourth European Conferenceon Antennas and Propagation, April 2010, pp. 1–5. → pages 17[39] Qhx220: Active isolation enhancer and interference canceller. [On-line]. Available: →pages 17[40] M. K. Simon and M. S. Alouini, Digital Communications Over FadingChannels:A Unified Approach to Performance Analysis. Wiley, 2000.→ pages 19[41] C.-S. Chang, “Stability, queue length, and delay of deterministic andstochastic queueing networks,” IEEE Trans. Auto. Cont., vol. 39, no. 5,pp. 913–931, May 1994. → pages 23[42] A. Host-Madsen and J. Zhang, “Capacity bounds and power allocationfor wireless relay channels,” IEEE Trans. Inf. Theory, vol. 51, no. 6,pp. 2020–2040, 2005. → pages 23[43] J. Tang and X. Zhang, “Quality-of-service driven power and rate adap-tation over wireless links,” IEEE Trans. Wireless Commun., vol. 6,no. 8, pp. 3058–3068, August 2007. → pages 26[44] N. Salodkar, A. Karandikar, and V. Borkar, “A stable online algorithmfor energy-efficient multiuser scheduling,” IEEE Trans. Mobile Com-puting, vol. 9, no. 10, pp. 1391–1406, Oct 2010. → pages 26[45] J. Tang and X. Zhang, “Cross-layer-model based adaptive resource al-location for statistical QoS guarantees in mobile wireless networks,”IEEE Trans. Wireless Commun., vol. 7, no. 6, pp. 2318–2328, June2008. → pages 29[46] I. Polik. Sedumi user guide. [Online]. Available: Accessed: 2016-11-23. → pages 29[47] N. Sidiropoulos, T. Davidson, and Z.-Q. Luo, “Transmit beamformingfor physical-layer multicasting,” IEEE Trans. Signal Process., vol. 54,no. 6, pp. 2239–2251, June 2006. → pages 2957Bibliography[48] Y. Nesterov, H. Wolkowicz, and Y. Ye, “Semidefinite programmingrelaxations of nonconvex quadratic optimization,” in Handbook ofsemidefinite programming. Springer, 2000, pp. 361–419. → pages29[49] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, UK:Cambridge University Press, 2004. → pages 30[50] L. Liu, R. Zhang, and K.-C. Chua, “Wireless information and powertransfer: A dynamic power splitting approach,” IEEE Trans. Com-mun., vol. 61, no. 9, pp. 3990–4001, September 2013. → pages 31[51] A. A. Nasir, X. Zhou, S. Durrani, and R. A. Kennedy, “Relaying proto-cols for wireless energy harvesting and information processing,” IEEETrans. Wireless Commun., vol. 12, no. 7, pp. 3622–3636, July 2013. →pages 31[52] D. Ng, E. Lo, and R. Schober, “Wireless information and power trans-fer: Energy efficiency optimization in OFDMA systems,” IEEE Trans.Wireless Commun., vol. 12, no. 12, pp. 6352–6370, December 2013. →pages 31[53] A. Host-Madsen and J. Zhang, “Capacity bounds and power allocationfor wireless relay channels,” IEEE Trans. Inf. Theory, vol. 51, no. 6,pp. 2020–2040, 2005. → pages 31[54] M. O. Hasna and M.-S. Alouini, “End-to-end performance of transmis-sion systems with relays over rayleigh-fading channels,” IEEE Trans.Wireless Commun., vol. 2, no. 6, pp. 1126–1131, 2003. → pages 42[55] E. Altman, Constrained Markov decision processes. CRC Press, 1999,vol. 7. → pages 42[56] M. L. Puterman, Markov Decision Processes: Discrete Stochastic Dy-namic Programming. Wiley-Interscience, 2005. → pages 42[57] I. Ahmed, K. T. Phan, and T. Le-Ngoc, “Optimal stochastic powercontrol for energy harvesting systems with delay constraints,” IEEE J.Sel. Areas Commun., vol. 34, no. 12, pp. 3512–3527, Dec. 2016. →pages 44[58] N. Salodkar, A. Bhorkar, A. Karandikar, and V. Borkar, “An on-linelearning algorithm for energy efficient delay constrained scheduling over58Bibliographya fading channel,” IEEE J. Sel. Areas Commun., vol. 26, no. 4, pp. 732–742, May 2008. → pages 44, 45[59] J. Andrews, S. Buzzi, W. Choi, S. Hanly, A. Lozano, A. Soong, andJ. Zhang, “What will 5G be?” IEEE J. Sel. Areas Commun, vol. 32,no. 6, pp. 1065–1082, June 2014. → pages 51[60] T. Muciaccia and V. M. N. Passaro, “Future scenarios for software-defined metro and access networks and software-defined photonics,”Photonics, vol. 4, no. 1, 2017. → pages 51[61] Millimeter waves: How we got here, the physical challenges, and5g opportunities. [Online]. Available:→ pages 52[62] A. M. Hayajneh, S. A. R. Zaidi, D. C. McLernon, and M. Ghogho, “Op-timal dimensioning and performance analysis of drone-based wirelesscommunications,” in 2016 IEEE Globecom Workshops (GC Wkshps),Dec 2016, pp. 1–6. → pages 5259


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items