Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Efficient access control for power line communication networks Huo, Yinjia 2017

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

Item Metadata


24-ubc_2017_may_huo_yinjia.pdf [ 751.92kB ]
JSON: 24-1.0345630.json
JSON-LD: 24-1.0345630-ld.json
RDF/XML (Pretty): 24-1.0345630-rdf.xml
RDF/JSON: 24-1.0345630-rdf.json
Turtle: 24-1.0345630-turtle.txt
N-Triples: 24-1.0345630-rdf-ntriples.txt
Original Record: 24-1.0345630-source.json
Full Text

Full Text

Efficient Access Control for Power Line CommunicationNetworksbyYinjia HuoB.A.Sc., Zhejiang University, 2014A THESIS SUBMITTED IN PARTIAL FULFILLMENTOF THE REQUIREMENTS FOR THE DEGREE OFMaster of Applied ScienceinTHE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES(Electrical and Computer Engineering)The University of British Columbia(Vancouver)April 2017c© Yinjia Huo, 2017AbstractBroadband power line communications (BB-PLC) reuse power line infrastructure to provide high-speed and high-penetration data transmissions, which make BB-PLC an attractive solution to in-vehicle networks (IVNs) and home area networks (HANs). However, there still exist many researchissues that need to be addressed before we can readily apply BB-PLC to these two application sce-narios. To partially address these issues, the thesis proposes some efficient access control schemesby extending a popular BB-PLC protocol, HomePlug AV (HPAV).The latency performance of the original HPAV protocol is not satisfactory for IVNs. Thus, weintroduce the Virtual Collision (VC) mechanism as an enhancement to the HPAV random back-offprocedure to reduce the queuing latency. In addition, the relationships between transmissions ofdifferent classes of network traffic involved in an IVN are not well handled by the strict prior-ity transmission selection algorithm (TXSA) specified in the HPAV protocol. In this regard, wepropose to combine the strict priority TXSA with the Audio Video Bridging credit-based shaping(CBS) TXSA.The efficiency of the HPAV medium access control (MAC) layer is restricted by various MACoverheads, which makes the deployment of BB-PLC in a HAN less attractive. We use the emergingin-band full duplexing capability to enable two new techniques, Mutual Preamble Detection (MPD)and Contention Free Pre-sensing (CFP) to reduce these overheads. Then, aiming to provide aunified solution to support various HAN applications, we develop an interface with prioritizationand traffic shaping to accommodate the heterogeneous network traffic involved.We use OMNeT++, a discrete event simulator, to verify the effectiveness of our proposediischemes, by comparing the network performance with our proposed schemes to that of the orig-inal HPAV protocol. The simulation results show that VC satisfactorily reduces queuing latency,the new TXSA better handles different priorities, MPD works well with CFP to improve MACefficiency and the developed interface functions properly.While there still exist many research issues before we can exploit the full advantages of BB-PLC, our proposed schemes bring us one step closer towards that goal.iiiPrefaceThe efficient access control schemes developed for in-vehicle network in Chapter 3 of this thesisare based on the work conducted in UBC Wireless Networks and Mobile Systems Laboratorywith Qiang Zheng, Dr. Zhengguo Sheng and Professor Victor C.M. Leung. The work has beenpublished in paper #1 listed below. As the first author of this paper, I initiated and conductedmost of the research. I conducted background survey, contributed in research ideas, performed theanalysis and carried out simulations. Qiang Zheng and Dr. Zhengguo Sheng contributed in researchideas and algorithm formulation. Professor Victor C.M. Leung helped in research directions andmanuscript proof-reading.The efficient access control schemes developed for home area network in Chapter 4 of thisthesis are based on the work conducted in UBC Wireless Networks and Mobile Systems Laboratorywith Gautham Prasad, Professor Lutz Lampe and Professor Victor C.M. Leung. The work has beenpublished partially in paper #2. Remaining part of the work is included in paper #3, which has beensubmitted to a journal. As the first author of these papers, I initiated and conducted most of theresearch. I conducted background survey in related topics, contributed in research ideas, performedthe analysis and carried out simulations. Gautham Prasad contributed in mathematical modelling,simulation schemes and manuscript preparation. Professor Lutz lampe and Professor Victor C.M.Leung helped in research directions and manuscript proof-reading.1. Yinjia Huo, Qiang Zheng, Zhengguo Sheng and Victor C.M. Leung, “Queuing Enhance-ments for In-Vehicle Time-Sensitive Streams Using Power Line Communications”, in Pro-ceedings of IEEE International Conference on Communications in China (ICCC), Shenzhen,ivChina, November 2015.2. Yinjia Huo, Gautham Prasad, Lutz Lampe and Victor C.M. Leung, “Mutual Preamble De-tection for Full Duplex Broadband Power Line Communications”, in Proceedings of IEEEGlobal Communications Conference, Washington D.C., USA, December 2016.3. Yinjia Huo, Gautham Prasad, Lutz Lampe and Victor C.M. Leung, “Efficient Access Controlfor Broadband Power Line Communications in Home Area Network”, submitted to a journal,10 pages, double column, March 2017.vTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiList of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiiNotations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .xviiiDedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xix1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Motivations and Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Research Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.2.1 Research Problems in IVNs . . . . . . . . . . . . . . . . . . . . . . . . . . 31.2.2 Research Problems in HANs . . . . . . . . . . . . . . . . . . . . . . . . . 31.3 Summary of Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4vi1.4 Outline of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62.1 Development of Power Line Communications . . . . . . . . . . . . . . . . . . . . 62.2 In-Vehicle Networks and Power Line Communications . . . . . . . . . . . . . . . 72.2.1 Development of In-Vehicle Networks . . . . . . . . . . . . . . . . . . . . 72.2.2 Network Architecture for Future In-Vehicle Networks . . . . . . . . . . . . 72.2.3 HPGP Protocol in IVNs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82.2.4 Network Traffic Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.2.5 Packet Latency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.2.6 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.3 Home Area Networks and Power Line Communications . . . . . . . . . . . . . . . 102.3.1 Translate PHY Data Rate into MAC Layer . . . . . . . . . . . . . . . . . . 112.3.2 Improve MAC Efficiency using IBFD . . . . . . . . . . . . . . . . . . . . 112.3.3 Related Works on CSMA/CD using IBFD . . . . . . . . . . . . . . . . . . 112.3.4 An Interface to Support Various HAN Applications . . . . . . . . . . . . . 122.4 An Introduction to HPAV MAC Protocol . . . . . . . . . . . . . . . . . . . . . . . 132.4.1 CSMA/CA Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132.4.2 MAC Throughput and Efficiency . . . . . . . . . . . . . . . . . . . . . . . 162.4.3 Possible Improvement of MAC Efficiency . . . . . . . . . . . . . . . . . . 173 Queuing Enhancements for In-Vehicle Time-Sensitive Streams . . . . . . . . . . . . 183.1 Virtual Collision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183.1.1 Backoff Procedure in HPGP . . . . . . . . . . . . . . . . . . . . . . . . . 193.1.2 Problem Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193.1.3 Proposed Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213.2 Incorporating AVB Credit Based Shaping . . . . . . . . . . . . . . . . . . . . . . 223.2.1 Existing Transmission Selection Algorithms . . . . . . . . . . . . . . . . . 22vii3.2.2 Proposed Transmission Selection Algorithm . . . . . . . . . . . . . . . . . 233.2.3 Realization of CBS in HomePlug GP . . . . . . . . . . . . . . . . . . . . . 233.3 Performance Evaluations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243.3.1 Simulation Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . 243.3.2 Simulation Results and Analysis . . . . . . . . . . . . . . . . . . . . . . . 243.3.3 Discussion on Metrics Used in Evaluations . . . . . . . . . . . . . . . . . 273.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284 Efficient Access Control Schemes for Home Area Networks . . . . . . . . . . . . . . 294.1 Improving the MAC Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294.1.1 Contention Free Pre-sensing . . . . . . . . . . . . . . . . . . . . . . . . . 304.1.2 Mutual Preamble Detection . . . . . . . . . . . . . . . . . . . . . . . . . . 364.1.3 Implementation of CFP and MPD in a BB-PLC Device . . . . . . . . . . . 384.2 An Interface to Accomodate HAN Applications . . . . . . . . . . . . . . . . . . . 394.2.1 Network Traffic Generated by HAN Applications . . . . . . . . . . . . . . 394.2.2 Prioritizing HAN Traffic . . . . . . . . . . . . . . . . . . . . . . . . . . . 404.2.3 Traffic Shaping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414.2.4 Admission Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424.3 Performance Evaluations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424.3.1 Simulation Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . 424.3.2 Performance of CFP with Single Node Flooding . . . . . . . . . . . . . . . 444.3.3 Performance Evaluation with Multiple Active Nodes . . . . . . . . . . . . 454.3.4 Discussion on Metrics Used in Evaluations . . . . . . . . . . . . . . . . . 484.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505.2 Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51viiiBibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59ixList of TablesTable 2.1 In-Vehicle Network Traffic Latency requirements . . . . . . . . . . . . . . . . . 9Table 3.1 Modified Backoff Status Update . . . . . . . . . . . . . . . . . . . . . . . . . . 22Table 3.2 Simulation Parameters for the IVN . . . . . . . . . . . . . . . . . . . . . . . . 24Table 3.3 Maximum latency of transmitted packets over 30s simulations . . . . . . . . . . 27Table 4.1 PRS SNR under Varying Minimum Channel Gains . . . . . . . . . . . . . . . 35Table 4.2 Simulation Parameters for the HAN . . . . . . . . . . . . . . . . . . . . . . . . 43xList of FiguresFigure 2.1 An Example of In-Vehicle Network Topology . . . . . . . . . . . . . . . . . . 8Figure 2.2 MAC Frame Format of HomePlug AV in CSMA/CA mode . . . . . . . . . . . 13Figure 2.3 Activity on the Medium in Case of Collision . . . . . . . . . . . . . . . . . . . 13Figure 3.1 Simulation Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25Figure 3.2 Real-Time Control Maximum Latency over 30s Simulations . . . . . . . . . . 26Figure 3.3 Excellent Effort Maximum Latency over 30s Simulations . . . . . . . . . . . . 27Figure 4.1 MAC Frame Transmission with CFP with CFC successfully detected . . . . . . 30Figure 4.2 Activity on the Medium in Case of Collision with our Deployment of MPD . . 37Figure 4.3 Illustration of Prioritization and Traffic Shaping . . . . . . . . . . . . . . . . . 40Figure 4.4 Network Simulation Topology . . . . . . . . . . . . . . . . . . . . . . . . . . 43Figure 4.5 MAC efficiency of single node flooding in variable MPDU interval. . . . . . . 45Figure 4.6 MAC Efficiency with PTS in Variable MPDU Interval. . . . . . . . . . . . . . 47Figure 4.7 MAC Efficiency with PTS in Variable Active Nodes. . . . . . . . . . . . . . . 48xiList of AbbreviationsACK/NACK Acknowledgment/Negative AcknowledgmentAV Audio and VideoAVB Audio Video BridgingBB-PLC Broadband Power Line CommunicationsBC Back-off CounterBPC Back-off Procedure Event CounterCA Collision AvoidanceCBS Credit-Based ShapingCCo Central CoordinatorCD Collision DetectionCFC Contention Free ConditionCFP Contention Free Pre-sensingCIFS Contention Inter-Frame SpaceCN Contention NotificationCSMA Carrier Sense Multiple AccessCW Contention WindowDC Deferral CounterEC Echo CancellationECU Electronic Control UnitEIFS Extended Inter-Frame SpacexiiFC Frame ControlFIFO First-in-First-outHAN Home Area NetworkHD Half-DuplexHPAV HomePlug AVHPGP HomePlug Green PHYIBFD In-Band Full DuplexingIC Ideal CentralizedIVN In-Vehicle NetworkMAC Medium Access ControlMPD Mutual Preamble DetectionOFDM Orthogonal Frequency Division MultiplexingOOK On-Off KeyingPHY PhysicalPLC Power Line CommunicationsPRP Priority Resolution PeriodPRS Priority Resolution SlotPSD Power Spectral DensityPTS Poisson Traffic ShapingQoS Quality of ServiceRIFS Response Inter-Frame SpaceRSI Residue Self-InterferenceRTS/CTS Request-to-Send/Clear-to-SendSACK Selective AcknowledgmentSI Self-InterferenceSNR Signal-to-Noise-RatioSOF Start of FramexiiiSOI Signal-of-InterestTC Transmit CounterTXSA Transmission Selection AlgorithmVC Virtual CollisionVCS Virtual Carrier SensexivNotationsχi bit value of the i-th PRSη MAC efficiencyηmax Optimum MAC efficiencyγ signal-to-noise-ratioγmin minimum signal-to-noise-ratioκ bandwidth allocation control thresholdκk network resource consumed by priority k messageλn,k event rate for n-th node priority k messageµ approximated MAC frame intervalµn,k average MAC frame intervals for n-th node priority k messageψ(c) sub-carrier specific phase angleΦR power of received signalΨR( f ) PSD of received signal at frequency fΨRSI the average RSI PSD after non-ideal SI cancellationΨT( f ) PSD of transmit signal at frequency fΨT,max maximum transmit signal PSDτ a defined ratio of two network parametersb0 decision threshold of non-coherent OOK normalized to root-mean-square noisebopt optimum decision threshold normalized to root-mean-square noisefc central frequency of an OFDM subcarrierxvC set of OFDM subcarriers used for PRS and preamble transmissionCW contention windowCWmax maximum contention windowEb received bit energyE[nBF ] expected number of back-off time slots∆ f width of an OFDM subcarrierf1 lower frequency of all OFDM subcarriersf2 upper frequency of all OFDM subcarriers|Hmin|2 flat minimum channel gainH( f ) power line channel transfer function at frequency fIn(·) n-th-order modified Bessel function of the first kindj˜n the least significant priority resolution bit of value ’1’ of node njn a certain priority resolution bit of node nK total number of network nodes` number of a certain time sampleL total number of time samplesm supported maximum priority resolution bitsmax(·) maximumMaxFL maximum time duration of data payloadN total number of active nodesN set of active network nodesn number of a certain nodeN0 average PSD of the cumulative noise at the receiverN0,FD effective IBFD noise floor at the receivernACK number of successfully acknowledged packetsNpkt number of total packets transmittedP average physical layer data ratexviPDE detection error ratePe overall network node error ratePFA false alarm ratepn priority level of network node npstd supported maximum priority levelPtot normalized total errorQ(·) first-order Marcum-Q functionri physical data rate of i-th data packetS MAC throughputs1 PRS signals2 preamble signalSCINR signal to cancelled interference plus noise ratio after ECSNRHD SNR in half duplex modeT total time durationtCIFS time duration of the contention inter-frame spacetEIFS time duration of the extended inter-frame spacetFC time duration of a frame controltFL time duration of data payloadtpd time interval for PRS detectiontRIFS time duration of response inter-frame spacetSLOT time interval for a back-off time slotti time duration of i-th data packettp time duration of a preambleTS total simulation timeZ+ set of all positive integersxviiAcknowledgmentsI would like to take this chance to thank all of those who have helped me and supported me duringmy research work.I would like to thank Professor Victor C.M. Leung as my supervisor of my graduate study.He introduced me to a promising and interesting field of study. I would like to appreciate Dr.Zhengguo Sheng, whom I worked with during my first year of graduate study. I would also like tothank Professor Lutz Lampe, who helped me a lot in understanding the underlying technology.I would like to thank my colleague Gautham Prasad, who helped me a lot with theoreticalmodels and mathematical derivations. I would like to thank my colleague Yuanfang Chi, whohas offered a lot of help with research methods. I would like to thank Qiang Zheng, Xuan Dongand Wei Tu for discussing research issues and sharing opinions. I would like to thank RobertoAntonioli, Morgan Roff and Jia Liu who helped with the construction of simulation platform.I would like to thank my parents for their continuous support.This work has been supported by the Natural Sciences and Engineering Research Council ofCanada (NSERC) and AUTO 21, Canada’s automotive research and development program.xviiiDedicationThis thesis is dedicated to my beloved family.xixChapter 1Introduction1.1 Motivations and ObjectivesThis thesis studies broadband power line communications (BB-PLC) in two specific applicationscenarios, the in-vehicle network (IVN) [1] and the home area network (HAN) [2].As a mature technology, PLC was first employed by utility companies for monitoring, control-ling and maintaining the power grid [3]. Recent advances in signal processing techniques haveenabled PLC to provide high-speed data communications through BB-PLC [4]. Among variousapplications of BB-PLC, two promising fields are IVN and HAN.IVNs provide solutions to communications among various electronic control units (ECUs), sen-sors and actuators pervasive onboard automobiles. In the future intelligent transportation system,IVNs are crucial to realize many features of an automobile, including the safety critical featureslike x-by-wire systems, the intelligent driving features like advanced driving assistance systemsand the entertainment features like in-vehicle entertainment systems. All these features involvecommunications within the vehicle, which are supported by the IVN. When we consider usingEthernet as the backbone network to support new applications with high data rate requirements [1],there are places onboard an automobile where limited space is available for wiring or the Ethernetwires can not reach. As an attractive alternative, PLC uses the existing direct current power linesin an automobile as the communication medium. The complexity and amount of wiring in an au-1tomobile can thus be significantly reduced, which reduces weight and saves cost. The reduction oftotal wire weight in an automobile also contributes to fuel saving and CO2 reduction.HAN provides solutions to communications among various in-home and mobile devices. Inthe future Internet of Things, in-home multimedia applications and home automation applicationswill simultaneous run over the HAN. In-home multimedia applications, including video conferenc-ing and in-home gaming, provide both convenience and entertainment to the household residents.Home automation applications, including smart metering and smart controlling, brings us towardsa smart home. Over the past few decades, the data rates supported by BB-PLC systems have in-creased dramatically [5], which makes BB-PLC an attractive solution to HAN. In addition, the ac-cess points of BB-PLC, i.e. power outlets, are pervasive in an in-home environment and BB-PLCmake use of the existing in-home power line infrastructure, which avoids additional installationcosts compared to other solutions to HAN.We consider using a popular BB-PLC protocol, HomePlug AV (HPAV) [6], to provide solu-tions to IVN and HAN. Along with supporting multimedia traffic with high data rate transmissionrequirements [7], the HPAV standard also offers many favourable features like short frame trans-missions with robust modulation scheme (referred to as ROBO mode in [6]). Moreover, developedas a subset of HPAV, HomePlug Green PHY (HPGP) [8] supports reliable data transmission amonglow cost devices with reduced overhead. These are very beneficial to handle the heterogeneous net-work traffic with different transmission requirements in an IVN or a HAN. Due to the shared natureof power lines as a communication medium, carrier sense multiple access (CSMA) is very efficientas a multiple access scheme under light to medium traffic loads [9]. Therefore, in the remainingpart of this thesis, we consider the cases where only CSMA is applied.However, there are different characteristics and transmission requirements associated with eachof the specific application scenarios. Applying the current HPAV protocol without any adaptationswill cause some problems detailed in Section 1.2, which makes the deployment BB-PLC in thesetwo application scenarios less attractive. Targeting these problems, this thesis proposes some effi-cient access control schemes.21.2 Research ProblemsMany research problems exist in each of the two considered application scenarios. In this the-sis, we partially address these issues by handling four of these problems, two in each consideredapplication scenario.1.2.1 Research Problems in IVNsIn an IVN, automobile applications are classified into four classes by the Society of AutomobileEngineers. Each class has its own requirements of data rates, latency and reliability. While we canuse the ROBO modulation scheme to provide reliable data communication, it is very difficult foran IVN implemented with the current HPAV protocol to meet the stringent quality of service (QoS)requirements of the network traffic involved [1, 10]. In order to make BB-PLC a viable solution,we need to improve the latency behaviour of different classes of network traffic.In an HPAV network, each message is associated with a priority level and a total of four prioritylevels are supported by the HPAV protocol. We can exploit the prioritization of network traffic torealize the QoS differentiation in an in-vehicle PLC network. However, the HPAV protocol imple-ments the strict priority transmission selection algorithm (TXSA), which means the transmissionsof lower priority messages should always yield to those of higher priority messages. The resultantlong burst of higher priority messages significantly degrades the QoS of lower priority messages.A new TXSA, which better deals with the relationship between different priority levels, needsto be proposed.1.2.2 Research Problems in HANsIn a HAN, despite the high physical (PHY) layer data rate supported by the HPAV protocol, itremains a challenge to effectively translate this data rate into throughput in the medium accesscontrol (MAC) layer. The efficiency of HPAV MAC protocol is largely confined by various kindsof MAC overheads. The CSMA with collision avoidance (CSMA/CA) operation is associatedwith the lengthy collision recovery. Regardless of this, the HPAV protocol implements CSMA/CA3because collision detection (CD) generally requires in-band full duplexing (IBFD) operations [11,Ch. 5]. Moreover, in order to avoid collisions, each CSMA/CA transmission is associated with arandom back-off stage, which further restricts the achieved MAC throughput. We need to reducethese MAC overheads caused by contentions and collisions to improve the MAC efficiency.In order to accommodate heterogeneous network traffic generated by various applications run-ning over PLC HANs, an interface with QoS differentiation is required. The HPAV uses multiplepriority levels to prioritize the network traffic. Yet, a specific network traffic prioritization schemefor HANs needs to be proposed. To better manage the network resource utilized by each prioritylevel, in particular, to prevent lower priority starvation, a traffic shaping scheme is required. Thus,we need to propose an interface with network traffic prioritization and traffic shaping.1.3 Summary of ContributionsWe can summarize our contributions in this thesis as follows:• We introduce a Virtual Collision (VC) mechanism as an enhancement to the original HPAVrandom back-off procedure. The VC mechanism provides better queuing fairness in ac-cordance with the first-in-first-out (FIFO) principle. As a result, the latencies of differentnetwork traffic types are effectively reduced, which makes PLC more suitable for IVN uses.• We propose to combine the strict priority TXSA and Audio Video Bridging (AVB) credit-based shaping (CBS) TXSA as the new TXSA for in-vehicle PLC network. In additionto better dealing with the relationships between the transmissions of different classes ofnetwork traffic, safety critical feature of a vehicle is guaranteed with the highest priority.• Leveraging IBFD capability [12], we propose two schemes, Mutual Preamble Detection(MPD) and Contention-Free Pre-sensing (CFP), to improve MAC efficiency. MPD is the firstpractical scheme to realize CD in a BB-PLC network while CFP eliminates the superfluousrandom back-off stages under contention free condition (CFC). Considering the imperfectionof IBFD, we analytically show the feasibility of our solutions. These two schemes work well4with each other and achieve over 95% of the optimum MAC efficiency, which can only beattained in the idealized case of no contentions or collisions.• We provide an interface with network traffic prioritization and traffic shaping to accommo-date heterogeneous network traffic involved in a HAN, which is not available in the literature.The network traffic prioritization enables QoS differentiation while the traffic shaping helpssupport network resource allocation, stream admission control as well as prevents lower pri-ority starvation.1.4 Outline of the ThesisThe rest of the thesis is organized as follows: In Chapter 2, we provide some background informa-tion as well as list some related work. In Chapter 3, we introduce the VC mechanism as well as anew TXSA, as two queuing enhancements for in-vehicle time-sensitive streams, in order to providetimely delivery for different classes of network traffic with consideration of both contention andcongestion. We evaluate the effectiveness of our proposed schemes by presenting the simulationresults of message delivery latencies under different network settings. In Chapter 4, we explain theefficient access control schemes developed for HAN. We evaluate the effectiveness of our proposedschemes by presenting the simulation results of MAC efficiencies under different network settings.In Chapter 5, we draw conclusions as well as give some suggestions for future work.5Chapter 2Background2.1 Development of Power Line CommunicationsThe emergence of PLC technology can be dated back to the late 1800s when first applied to remotemeter reading. Since then, PLC technology has continuously evolved [5]. Before the late 1990s,most PLC systems developed fall under ultra-narrowband and low data rate narrowband PLC witha maximum data rate typically lying below a few kbps. Smart grid applications [3] have been oneof the main driving forces of PLC systems with higher data rates. Utility companies develop theseapplications for monitoring, controlling, and maintaining the power grid. Since the late 1990s, thederegulation of telecommunications and energy markets in Europe has fostered the developmentof BB-PLC systems with a data rate ranging from several Mbps to several hundred Mbps.Note that the power line medium was not originally designed for data transmission and thepower line is actually a very harsh environment for communications [4]. The power line mediumis known to be subject to varieties of noise including background noise, synchronous impulsenoise, asynchronous impulse noise and aperiodic impulse noise. To address this noise problem,various signal processing techniques have been developed, which are indispensable for the suc-cessful development of BB-PLC systems.Through the years, the data rate supported by BB-PLC systems has increased dramatically.With frequency band extension, IEEE 1901 supports a data rate as high as 500 Mbps [13]. Fur-6ther with multiple-input multiple-output transmission technology, HomePlug AV2 promises a peakPHY data rate up to 1012 Mbps [14].The high data rates supported by BB-PLC make it an appealing solution to IVN or HAN.2.2 In-Vehicle Networks and Power Line CommunicationsThe IVN involves communications among different ECUs, sensors and actuators. According tothe Society of Automotive Engineers, the number of ECUs in an automobile ranges from 30 forsimple cars to approximately 100 for luxury cars. The electronics accounts for 15 percent of totalvehicle cost in 2005, climbing from 5 percent in the late 1970s. Today, vehicles are increasinglybehaving as an intelligent computing system.12.2.1 Development of In-Vehicle NetworksAt the early stage of contemporary automobile designs, in-vehicle communications are commonlyrealized through point-to-point wiring between electronic components, resulting in bulky, expen-sive and complicated harnesses. With increasing scale and complexity, the in-vehicle networkgrows into a state where volume, weight and reliability becomes a real problem [15, 16]. As aresult, a more integral solution is called for and several automobile communication bus proto-cols have been developed. By connecting a number of the electronic components to the samein-vehicle communication bus, the communication medium can be shared and wiring can thus besaved. Meanwhile, the in-vehicle architecture can be more hierarchical and structural, which alsosimplifies the automobile designing procedure.2.2.2 Network Architecture for Future In-Vehicle NetworksDue to the increasing data rate requirements introduced by new applications, e.g., the latest camera-based advanced driver assistance systems, it becomes attractive to use Ethernet as the backbone ofnext generation in-vehicle networks [1]. However, most of the space of an automobile is occupiedby the cabin, which results in limited space available to accommodate wiring. In addition, there are1 r=07places where Ethernet wires cannot reach, e.g., where an aftermarket reversing radar is installed.On the other hand, all electronic devices need power supply to function properly. In-vehicle PLCuses the existing DC power line as the communication medium, which significantly reduces thecomplexity and the amount of wiring in an automobile. This leads to not only savings in thedesign and manufacturing costs, but more importantly fuel saving and CO2 reduction. In addition,in-vehicle PLC is extremely beneficial where cabling space is strictly restricted, which is not anuncommon scenario, since in automobile much of the physical space is taken up by the passengercabin. Therefore, PLC is an attractive supplement to the existing solutions for IVNs.Ethernet Backbone NetworkPLC Bridge & CCo of PLC NetworkECU3PLC BusLegacy Communication NetworkECU4ECU2ECU1Figure 2.1: An Example of In-Vehicle Network TopologyFig. 2.1 shows a typical IVN topology. In the figure, ECUs are interconnected to each otherthrough the PLC bus. Messages heading for the Ethernet backbone network are transmitted to thePLC bridge, which forwards arriving packets from the PLC network to the Ethernet backbone andvice versa. The bridge also acts as the central coordinator (CCo) of the PLC network to realizevarious network management functions. Other legacy networks may coexist with the PLC network.2.2.3 HPGP Protocol in IVNsWhile several protocols exist for off-the-shelf IVNs, most of these protocols, such as Local Inter-connect Network and Controller Area Network, are based on the wired communication mediumwith twisted pairs of wires. None of these protocols can be readily applied to the PLC networkbecause of the unique characteristics of the PLC channel [17].On the other hand, developed as a subset of HPAV, HPGP [8] is a reliable PLC protocol, which8Table 2.1: In-Vehicle Network Traffic Latency requirementsTraffic Class Max End-to-End DelayControl Data 2.5 msSafety Data (Video) 45 msInfotainment Data 150 msis popular due to its low cost and smaller overhead incurred. Thus, we consider HPGP for in-vehicle PLC protocol. However, some modifications are necessary because of the stringent QoSrequirements of IVN applications. Antonioli et al. have shown [10] that the latency of the currentHPGP protocol is too long for many applications, and may have difficulty meeting the requirementsof in-vehicle traffic as shown in Table 2.1 [1].2.2.4 Network Traffic TypesThere are several types of network traffic involved in an IVN [18]. We denote all the ECUs includ-ing the CCo as network nodes. Each message is sent as one or more data packets. We classify datapackets according to their QoS requirements.Four classes of data packets are considered [19]. Real-time control packets are usually small insize and have low data rate requirements, but they should be delivered with strict time constraintssince they are typically generated by safety critical applications, such as drive-by-wire. Data pack-ets generated by other systems, like the comfort system and the infotainment system, are classifiedinto critical application packets, excellent effort packets, and best effort packets. Critical applica-tion packets have higher QoS requirements than the excellent effort packets. Best effort packetsare sent without any QoS guarantee.We aim to design efficient access schemes for in-vehicle PLC networks which can meet thelatency requirements of all types of network traffic.2.2.5 Packet LatencyIn this thesis, the latency of a packet is measured from the instance when the packet is originatedand enters the queuing buffer at the source node to the instance when the source node receivesan acknowledgment (ACK) from the destination node indicating that the packet has been success-9fully received. Note that in case of a collision, the data packet is not received at the destinationnode and the ACK is not returned; thus the packet latency measurement continues and it includesthe retransmissions caused by a collision. However, we assume that the IVN traffic is transmittedusing a robust modulation scheme (referred to as ROBO mode in [6]); thus we do not considertransmission errors in an IVN and the packet latency measurement does not include the retrans-missions caused by transmission errors as we assume that such retransmissions never occur. Insuch a condition, the packet latency is measured with considerations of collisions and congestionsonly.2.2.6 Related WorkPreviously, Cano and Malone have carried out simulations [20] and experimental tests [21] onHPAV Protocol, focusing on the network throughput under saturated traffic, rather than networklatency. However, saturated traffic is not desired in IVNs because it induces much larger latency.Antonioli et al. have made some modifications to the MAC layer of HPGP [10]. However, theirwork overlooked the buffer queuing delay from upper layers. Sheng et al. have proposed a multi-channel MAC protocol for vehicular power line communication systems [22]. By resolving colli-sions in both the time and frequency domains, the collision overhead can be significantly reduced.However, the robustness of using frequency domain to resolve collisions is questionable.2.3 Home Area Networks and Power Line CommunicationsSince the introduction of the 10-Mbps class BB-PLC products using HomePlug 1.0, data rates pro-vided by BB-PLC have increased multi-folds [23]. Current HomePlug AV2 compliant devices usemultiple wires available in most in-home wiring installations to achieve multiple-input multiple-output operation, and offer data rates of up to 2 Gbps [7]. The gigabit range of throughput and thewidespread availability of access points, i.e., power outlets, makes BB-PLC an attractive solutionfor a backbone and/or stand-alone communication medium for HANs [5, 24].Due to the inherent upward and downward compatibility that HPAV provides with other Home-10Plug releases, as well as with IEEE 1901 standard, all solutions we propose can be easily extendedto these BB-PLC standards compliant devices as well [25].2.3.1 Translate PHY Data Rate into MAC LayerDespite the high data rates obtained at the PHY layer, it remains a challenge to translate this PHYdata rate efficiently into MAC throughput. In a CSMA/CA operation, overheads like inter-framespaces, transmission of priority resolution symbols (PRSs) and transmission of frame control (FC)messages consume additional time. Moreover, CSMA is implemented with CA using a randomback-off strategy to prevent collisions. When a collision occurs, a relatively long time is spent oncollision recovery [6]. Since no payload data is transferred over the medium during the randomback-off or the collision recovery, these MAC overheads caused by contentions and collisionsconsiderably reduce the achieved MAC throughput. Nevertheless, HPAV standard uses CSMA/CAin place of CSMA/CD, since CD requires network nodes to support full-duplex operation [11, Ch.5].2.3.2 Improve MAC Efficiency using IBFDIBFD has recently been successfully applied in a BB-PLC system [12]. The authors proposeda two-step Echo Cancellation (EC) scheme in order to cancel the self-interference (SI) which isa main hurdle in realizing IBFD. The IBFD implementation enables network nodes to sense themedium while simultaneously transmitting data. This inspires us to propose not only a practical CDscheme, similar to CSMA/CD in full-duplex wireless networks, but also to devise an IBFD-basedmethod to eliminate the redundant back-off stages. Although CSMA/CD is also implemented inearly Ethernet networks [26, Ch. 6], they are different from BB-PLC in many aspects.2.3.3 Related Works on CSMA/CD using IBFDCSMA/CD was considered infeasible in wireless networks due to the inability of network nodes totransmit and receive signals simultaneously in the same band [27]. Pseudo-CSMA/CD procedures,like CSMA with collision notification (CN), were instead proposed as a middle-ground solution11between CSMA/CA and CSMA/CD [27]. However, the introduction of IBFD has propelled feasi-ble CSMA/CD methods to be proposed for wireless networks. The authors in [28, 29] used IBFDto enable the receiver node to continuously transmit acknowledgments as collision-free indica-tors while receiving the data payload. Such a scheme not only deprives an IBFD system of thebidirectional data payload transmission, but also potentially causes multiple false alarms in condi-tions of packet errors. Furthermore, it introduces additional power consumption at the destinationnode for the continuous acknowledgment transmission [30]. Alternative CSMA/CD techniqueswere proposed for wireless networks in [31, 32] to detect a collision at the transmitter by sens-ing the medium during transmission without relying on the feedback from the destination node.However, [31, 32] provided analysis of CSMA/CD under the assumption of Rayleigh channel anda fixed self-interference cancellation performance. In contrast, we specify a complete detectionand reaction procedure to realize CSMA/CD in BB-PLC networks through the IBFD detection ofpreamble symbols, and prove the feasibility of our solution by analytically deriving the detectionerror and false alarm rates under a worst-case BB-PLC line attenuation condition. We use theself-interference cancellation performance reported in the literature [12, 33], which is shown to bedependent on the power line channel attenuation.2.3.4 An Interface to Support Various HAN ApplicationsWe aim to provide a unified HAN solution that supports both in-home multimedia traffic [34] aswell as home automation applications [35]. Many in-home multimedia applications, like audioand video (AV) streaming [35], require high data rate transmissions [7]. On the other hand, homeautomation applications, which include various smart home applications, such as monitoring andsmart controlling of home appliances, usually require in time robust message delivery [36]. Aninterface to accommodate the heterogeneous network traffic involved is thus required, which toour best knowledge is currently not available in the literature [13].122.4 An Introduction to HPAV MAC ProtocolIn this section, we briefly introduce the HPAV MAC protocol that we aim to refine in this thesis,and also define the MAC layer throughput and efficiency, which we adopt as the metrics to showthe performance enhancements of our proposed access control schemes of CFP and MPD.2.4.1 CSMA/CA OperationFig. 2.2 shows the time line of an MAC frame transmission under a standard CSMA/CA operation.In the following, we describe the parts of this operation that are relevant to the schemes we propose.Figure 2.2: MAC frame transmission of HPAV in CSMA/CA mode.Figure 2.3: Activity on the medium in case of a collision.Priority ResolutionAn MAC frame transmission is initiated with a priority resolution procedure (PRP). The HPAVprotocol specifies four priority levels that the network nodes can choose from using two prioritybits. The four priority levels are resolved bit-by-bit starting with the slot for PRS0, and followedby PRS1, with PRS0 indicating the most significant bit in the binary representation of the prioritylevel. Each MAC frame is associated with a priority level from 3 (highest) to 0 (lowest). Nodeswith the highest priority level in the network win the PRP, which ensures messages of higherpriority levels always get transmitted before those of lower priority levels.13Collision AvoidanceCollision avoidance in CSMA/CA is realized through the random back-off mechanism. Nodeswinning the PRP proceed to participate in the back-off stage, which consists of a variable numberof back-off time slots of equal time interval. At each back-off time slot, a participating node eithertransmits a preamble or remains silent, while a non-participating node always remains silent duringthe back-off stage.When there are no nodes transmitting a preamble in a back-off time slot, all nodes detect thisback-off time slot to be idle and nodes participating in the back-off stage decrease their back-offcounters (BCs) by one. A participating network node transmits a preamble in a back-off timeslot if and only if its BC = 0 during that slot. The nodes transmitting a preamble gain access tothe channel and then transmit an FC message of type start-of-frame (SOF), followed by the datapayload. The other nodes detect the preamble transmission and freeze their BCs until the start ofthe next back-off stage they participate in.Apart from BC, the back-off stage also involves three other counters: (a) back-off procedureevent counter (BPC), (b) deferral counter (DC), and (c) transmit counter (TC). The TC associateseach MAC frame with a life-time. Whenever a network node attempts to transmit an MAC frame,the associated TC will be decreased by one. When TC is decreased to zero, the network nodediscards the MAC frame. When a network node first enters the back-off stage, it will reset BPCto zero. Each time the BPC is changed or reset, positive integer values of DC and contentionwindow (CW) are set depending on the current BPC value, and BC is randomly set to an integervalue uniformly distributed between 0 and CW. Every time a collision is encountered, the BPCis increased by one, and every time a node detects the transmission of MAC frames of the samepriority level as its, it decreases DC by one. When DC reaches zero, the network node increasesBPC by one.14Collision in the NetworkDespite the precautions taken to avoid collisions, CSMA/CA does not guarantee collision-freetransmissions, especially with increased number of contending nodes. A collision in the networkoccurs when multiple nodes simultaneously transmit the preamble signal in a back-off time slot togain access to the channel. Every node uses a timer called the virtual carrier sense (VCS) timerto determine a collision. At the end of the back-off stage, the timer is set to tEIFS, the time intervalof the extended inter-frame space (EIFS), at each network node. A collision is determined at anetwork node if the node does not receive a selective acknowledgment (SACK) frame before thetimer expires. Transmitted by the destination node, an SACK frame is either an ACK frame or annegative acknowledgment (NACK) frame, depending on whether the data payload was successfullydecoded. As shown in Fig. 2.2, upon receiving the data payload, the destination node waits fortRIFS, the time interval of the response inter-frame space (RIFS), before transmitting an SACKframe. Note that the VCS timer is also used by the non-transmitting stations to recover from acollision together with all the transmitting stations when the VCS timer expires.Collision RecoveryNodes recover from a collision immediately after the EIFS timer expires and start to back-offregardless of the priority level of the transmitting MAC frames. This operation is illustrated inFig. 2.3. The EIFS has a duration oftEIFS = 2tp+2tFC+MaxFL+ tRIFS+ tCIFS, (2.1)where tp, tFC and MaxFL, tCIFS are the time intervals of the preamble, the FC, the maximum timeinterval of the data payload and the contention inter-frame space (CIFS), respectively. EIFS istherefore lengthy, and renders a collision very costly to recover from.152.4.2 MAC Throughput and EfficiencyAs shown in Fig. 2.2, a data payload is only transmitted during the ‘Data Payload’ time interval.All other time intervals essentially are overheads that impede the ability of the MAC layer toefficiently translate PHY data rate into MAC throughput. We define the MAC throughput as therate of successful data packet transmissions, i.e., data rate of packets that are acknowledged by thedestination nodes with an ACK.For Npkt packets successfully transmitted over a time duration T , we define the MAC through-put asS, 1TNpkt∑i=1riti, (2.2)where ri and ti indicate the PHY transmission rate and data payload interval of the ith packet,respectively. For this scenario, the average PHY transmission rate can in turn be expressed asP=Npkt∑i=1ritiNpkt∑i=1ti. (2.3)Thus, we define the ratio of the achieved data rate in the MAC layer to the average performance ofPHY layer as the MAC efficiencyη , SP=1TNpkt∑i=1ti. (2.4)Therefore, to improve S we would typically want to increase the product ηP. However, P isdependent on the nature of the communication channel and advancement of the PLC technologyin the PHY layer. In this thesis, we focus on improving η .The maximum MAC efficiency under CSMA operation can be computed under the conditionthat a network node continuously transmits frames of maximum length without incurring a trans-mission error, collision, or backoff2. Under such conditions, the maximum MAC efficiency can be2For the sake of simplicity, we ignore the bursting and inverse bursting procedure specified in the HPAV protocolfor this computation, without any adverse effects on our proposed solution. Results obtained in this thesis can be easilyextended to cases with bursting.16expressed asηmax =MaxFLEIFS+2tSLOT, (2.5)where 2tSLOT is the time interval for two PRSs. Using the parameters specified in the HPAVstandard [6], we obtain ηmax = 78.24%.2.4.3 Possible Improvement of MAC EfficiencyIn order to enhance practical values of η as close as possible to ηmax, we need to reduce the MACoverheads.• A collision costs the HPAV network a time interval of EIFS to recover from, which is verylengthy and costly as shown in Fig. 2.3. If we could know beforehand that a collision wouldoccur and avoid it, the MAC efficiency could be considerably improved. We realize thisthrough MPD.• When a node is the only node transmitting, there will be no contention or collision. Therandom back-off stage in Fig. 2.2 is superfluous. If we can avoid this superfluous randomback-off stage, the MAC efficiency can be further improved. We realize this through CFP.17Chapter 3Queuing Enhancements for In-VehicleTime-Sensitive StreamsIn this chapter, we propose two queuing enhancements [37] to address two of the problems weencounter when we apply BB-PLC in IVNs. One one hand, the latency performance of the orig-inal HPAV protocol is not satisfactory. In this regard, we propose VC as an enhancement to therandom back-off procedure specified in the HPGP protocol. VC improves queuing fairness andthus reduces the overall network traffic latency. On the other hand, a new TXSA is needed, whichbetter deals with the relationship between different priority level messages. In this aspect, we com-bine the AVB CBS TXSA with the strict priority TXSA in order to guarantee the time constraintof real-time control signals and to simultaneously meet the latency requirements of other networktraffic.3.1 Virtual CollisionIn the following, we first rewrite the random-back off procedure introduced in Section 2.4.1 in theform of an algorithm. Then we identify its inherent deficiency in the queuing fairness guaranteeand introduce the VC mechanism to combat this problem.183.1.1 Backoff Procedure in HPGPThe random back-off procedure in CSMA specified by the HPGP/HPAV protocol can be dividedinto three steps, as listed in Algorithm 1. Note that Algorithm 1 is exactly the same procedure asis described in Section 2.4.1. Here we only rewrite it in the form of an algorithm in order to betterpresent our proposed modifications.Each time the network finishes an MAC frame transmission or the network recovers from acollision, step 1 (back-off status update) is executed. Upon winning the PRP, a network node ini-tializes itself if it has not done so previously. Then in step 1, if BC is zero (in case of initializationor collision) or DC is zero (maximum number of packet deferral reached), the network node resetsits BC with an increased BPC. Otherwise DC and BC are decreased since the on-going MAC frametransmission is deferred. The function setCounters() sets DC according to the current BPC andresets BC to be uniformly distributed between zero and the current CW. Then, network nodes win-ning priority resolution continuously listens to the channel in each back-off time slot and decreaseBC by one for each time slot that the communication channel is sensed idle, until BC=0, in whichcase the node sends out the preamble followed by the SOF FC message and the data payload. Thedata transmission sets the channel state to busy and the algorithm goes to step 3. In step 3, when anACK or NACK is received, the transmitting node is reset, or when the timer expires and the ACKor NACK is still not received, the network decides that there is a collision.3.1.2 Problem IdentificationIf there is no collision and there is only a single type of network traffic, in order to reduce themaximum latency of packet transmissions, obviously the optimal packet to transmit is the packetthat has arrived first and been in queue for the longest time; i.e., FIFO.In practice, there are multiple types of network traffic interacting with each other. However, thecontentions between different types of network traffic are resolved according to the TXSA. Accessopportunity allocated by the TXSA to a particular type of network traffic is not affected by thescheduling of packets belong to this traffic type. Thus, for packets belonging to the same traffic19Algorithm 1 Random Backoff ProcedureStep 1 - Backoff Status Update.Wait f or the end o f the priority resolution;if (wonPriorityCheck) thenif (!Initialized) thenBC := 0, BPC := 0 and DC := 0;Initialized := true;end ifif (BC== 0||DC== 0) thensetCounters(BPC);BPC++;elseBC−−,DC−−;end ifListen to the channel;end ifStep 2 - Random Backoff.At the beginning of subsequent backoff slot,if (ChannelBusy) thenGoto Step 1;else if (BC> 0) thenBC−−;elseSend out message;end ifStep 3: End of FrameBefore the timer expires,if (ACKReceived or NACKReceived) thenInitialized := false;Goto Step 1;end ifWhen the timer expires,Goto Step 1;type, we should still aim for the FIFO queuing. Moreover, retransmissions of collided packets canbe considered as another type of network traffic for this discussion.Without centralized control, Algorithm 1 conforms well to the FIFO queuing principle in mostcases. Despite the randomness of the back-off procedure, an early arriving packet on average hasundergone more random back-off slots, which means that an early arriving packet has a higherchance to be transmitted. However, Algorithm 1 violates the FIFO queuing principle in the follow-20ing situation.Assume that there is a collision in the network. For network nodes with non-empty buffers,those involved in the collision will go to a higher contention stage as a penalty and those notinvolved in the collision will decrease their DC, which is also a penalty with respect to contention.Now, when a data packet is originated at a node with a previously empty buffer, it will initializethe random back-off procedure at the zero contention stage according to Algorithm 1, which yieldsa smaller contention window size and higher probability of gaining access to the communicationmedium through contention compared to nodes with non-empty buffers. In the above scenario,the random back-off procedure favours late arrivals of nodes with previously empty buffers, whichcontradicts the FIFO principle.3.1.3 Proposed SolutionWe propose the VC mechanism, in which Step 1 of the aforementioned random back-off procedureis replaced by that shown in Table 3.1, which differs from the original Step 1 in two respects.Firstly, the buffer of a network node that does not participate in a particular priority level in thePRP is empty with respect to the corresponding type of data packet; BPC is increased each timethe random back-off procedure undergoes Step 1 in these nodes. Secondly, the previous BPC valueis kept instead of being initialized to zero.For nodes with non-empty buffers, they follow the exact procedure as specified in Algorithm 1.For nodes with empty buffers, each time a collision occurs, they are considered to be virtuallyinvolved in the collision (hence VC) and BPC is increased. As a result, when a data packet isoriginated, these nodes are penalized to the same degree as network nodes actually involved in thecollision, which results in better fairness of buffer queuing with a better conformity to the FIFOqueuing principle. Note that the network node with empty buffers will never be over penalizedbecause at each successful transmission, these nodes also assume a virtual successful transmissionand reset BPC. Such an improvement in buffer queuing fairness naturally results in a better latencybehaviour for the whole network.21Table 3.1: Modified Backoff Status UpdateStep 1 - Backoff Status Update.Wait f or the end o f the priorityresolution;if (wonPriorityCheck) thenif (!Initialized) thenBC := 0 and DC := 0;Initialized := true;end ifif (BC== 0||DC== 0) thensetCounters(BPC);BPC++;elseBC−−,DC−−;end ifListen to the channel;elseBPC++;end if3.2 Incorporating AVB Credit Based ShapingAt the beginning of each MAC frame transmission, the network selects the type of data packetto transmit according to the TXSA. We propose to modify the original strict priority TXSA bycombining it with the AVB credit-based TXSA. With the proposed new TXSA, the time constraintof real-time control signal can be met and the latency requirements of all other network traffic canalso be satisfied.3.2.1 Existing Transmission Selection AlgorithmsIEEE Standard 802.1Q describes the strict priority TXSA [19], whereby lower priority packets arenot transmitted until all higher priority packets have been transmitted. During long bursts of highpriority traffic, the transmission of lower priority traffic is completely blocked, which significantlyincreases the latency of lower priority traffic [38]. The AVB credit-based TXSA in IEEE Standard802.1Qav [39] applies CBS to space out high priority traffic, thus preventing long bursts of highpriority traffic.223.2.2 Proposed Transmission Selection AlgorithmThe strict priority TXSA works well to guarantee latency of high priority network traffic. WhenAVB credit-based TXSA is applied, the latency requirements of all AVB streams can be satisfiedunder admission control. A combination of these two TXSAs is proposed in this thesis. As a result,the advantages of these two TXSAs can be both realized.Real-time control signals should be delivered within strict time constraints. Therefore, wegive the real-time control signals the highest priority level and whenever there is real-time controlsignal, other network traffic should yield. Since the real-time control signal is small in packet sizeand low in data rate requirement, it is not bursty and will not severely block the transmission oflower priority data packets. On the other hand, best effort control signals should yield wheneverthere is any other network traffic because it is not necessary to guarantee the latency of best effortnetwork traffic. We assign the critical application network traffic as AVB class A stream and theexcellent effort network traffic as AVB class B stream. In such a setting, when there is no real-timecontrol signal, AVB class A stream will give way to AVB class B stream if and only if the credit ofclass A stream is negative and the credit of class B stream is non-negative.3.2.3 Realization of CBS in HomePlug GPThe CBS is realized through the AVB credit updating procedure. When a class A stream is beingsent, the credit of class A stream is decreased with SentSlopeA. Otherwise, the credit of class Astream is increased with IdleSlopeA. The same applies for AVB class B stream with SentSlopeBand IdleSlopeB. IdleSlopeA and IdleSlopeB are exactly the reserved bandwidths for AVB classA stream and class B stream, respectively. SentSlope is the network bandwidth minus the corre-sponding IdleSlope. For example, we have 10 Mb/s in network bandwidth. We reserve 45 percentnetwork bandwidth for AVB Class A stream; thus IdleSlopeA is 4.5 Mb/s and SentSlopeA is 5.5Mb/s. In this thesis, we assume an equal bandwidth allocation; i.e., 50% of the network bandwidthare allocated to each type of AVB streams, which also means that we reserve no bandwidth for nonAVB streams.23Table 3.2: Simulation Parameters for the IVNParameter ValueSimulation time 30 sNumber of network nodes 10 (including CCo)Real-time control traffic 50 packets/s/nodeOther types traffic 60 packets/s/nodeReal-time control packet size 64 byteOther packet size 128 byteData transmission rate 9.8 MbpsNumber of Priority Bits 3PRS and Backoff Time Slot 5 µs3.3 Performance EvaluationsWe set up a simulation model using OMNet++ to verify the effectiveness of our proposed enhance-ments. We first specify the simulation parameters and then present and analyze the simulation.3.3.1 Simulation ConfigurationThe simulation parameters are presented in Table 3.2. Other parameters are as specified in theHPGP specifications [8]. Different types of data packets arrive from the upper layer of the sourcenode independently according to the Poisson process. All the network nodes are assumed identicaland each type of network traffic is originated at each network node independently with the samerate as listed in Table 3.2. The simulation topology is shown in Fig. 3.1. Ten ECUs including theCCo are interconnected with each other through the PLC bus.3.3.2 Simulation Results and AnalysisThe simulation results are presented in Table 3.3, Fig. 3.2 and Fig. 3.3. Table 3.3 shows themaximum latency of all the network nodes. The figures show the maximum latency at variousnetwork nodes. Note that the significance of the results is guaranteed by the sufficient simulationtime (30s continuous simulation) and multiple independent samples (samples collected from 10independent and identical network nodes). Admittedly, the maximum latency results become largeras the number of network nodes or the simulation time increases. However, we can apply admission24CCoECU1ECU5ECU6ECU2ECU4 ECU3ECU7 ECU8 ECU9PLC bus lineFigure 3.1: Simulation Topologycontrol to confine the collision rate as the number of network nodes increases. Meanwhile, we canresort to back-up communication systems, should the latency fail to meet the requirements over alonger period of time.In the results, HPGP corresponds to the latency results obtained by the original HPGP protocol,VC adds the VC mechanism to HPGP, VC+CBS further adds AVB CBS to VC, and IC is the idealcentralized benchmark. In IC, a perfect control channel is assumed over which each network nodeis able to instantly inform the CCo of its buffer queuing condition, and in each MAC frame accessopportunity the CCo selects the network node to transmit according to the FIFO principle.The results show that when VC is applied, the maximum latency behaviours of all other typesof network traffic except best effort are considerably improved, since the improved fairness ofbuffer queuing causes the network to behave more like a FIFO buffer. From Fig. 3.2 and Fig. 3.3,not only the peak of VC is lower than the peak of HPGP among all the nodes, but also the value ofVC for each node is generally lower than the corresponding value of HPGP.In HPGP and VC, only strict priority TXSA is applied, which is good at securing the latency25Figure 3.2: Real-Time Control Maximum Latency over 30s Simulationsof higher priority packets, but lower priority packets may be blocked during long bursts of higherpriority packets. When AVB CBS is applied, after a burst of AVB stream A, its credit will benegative thus preventing its transmission while the credit of AVB stream B is non-negative. Asshown in Table 3.3, when CBS is applied over VC, the maximum latencies of critical applicationnetwork traffic and excellent effort network traffic are both within 5 ms. Moreover, CBS TXSAis combined with the strict priority TXSA so that the latency of real-time control traffic can beguaranteed.The results show that the two proposed enhancements, i.e., VC+CBS, effectively lower thelatency of real-time control traffic so that it is able to meet more stringent latency requirements, andalso keep the latency of critical application and excellent priority traffic within acceptable limits of5 ms. While the latency of the respective traffic classes is not as low as IC, this is achieved withoutthe overhead of centralized control.26Figure 3.3: Excellent Effort Maximum Latency over 30s SimulationsTable 3.3: Maximum latency of transmitted packets over 30s simulationsHPGP VC VC+CBS CentralizedReal-Time Control 1.80 ms 1.53 ms 1.37 ms 0.92 msCritical Application 3.80 ms 3.07 ms 4.56 ms 2.13 msExcellent Effort 7.78 ms 5.94 ms 4.87 ms 3.88 msBest Effort 21.09 ms 22.01 ms 22.04 ms 10.86 ms3.3.3 Discussion on Metrics Used in EvaluationsIn this chapter, we want to improve the real-time performance through the adoption of our proposedschemes. Thus, we evaluate the network performance using latency.Another network metrics to our interest is the throughput. However, we accommodate the net-work traffic incurred by new applications of high data rate requirements using Ethernet backbone.27The remaining network traffic will not pose much pressure on the PLC network with regard todata throughput. Thus, we do not present network throughput results in our IVN performanceevaluations.3.4 SummaryIn this chapter, we have proposed some efficient access control schemes for IVN. We have pro-posed two enhancements to data packet queuing: the VC mechanism improves the queuing fairnessand thus reduces the overall network latency; combining AVB credit-based TXSA with straight pri-ority TXSA minimizes the latency of real-time traffic while satisfying the latency requirements ofdifferent traffic classes. We have presented simulation results to demonstrate the effectiveness ofour proposal.28Chapter 4Efficient Access Control Schemes for HomeArea NetworksIn this chapter, we develop some efficient access control schemes to address two problems thatarise when BB-PLC are employed in HANs. On one hand, the PHY data rate is not efficientlytranslated into throughput in the MAC layer. In this regard, we propose two new techniques thattake advantage of IBFD to improve the MAC efficiency. Specifically, we propose CFP to eliminatesuperfluous random back-off stages and we propose MPD to avoid lengthy collision recovery. Onthe other hand, an interface to accommodate heterogeneous network traffic generated by variousapplications running over PLC HAN is not yet available. In this aspect, we propose an interfacethat uses network traffic prioritization to accommodate their different QoS requirements and trafficshaping to realize centralized bandwidth management and admission control.4.1 Improving the MAC EfficiencyThe format of an HPAV MAC frame transmission in CSMA/CA mode is shown in Fig. 2.2. All theother time intervals except the ‘Data Payload’ are MAC overheads impeding the MAC efficiency.Note that in case of a collision, the transmitted data payloads corrupt each other and thus all thetime intervals shown in Fig. 2.3 are also MAC overheads. With the goal of bringing practical29values of η as close as possible to ηmax, we attempt to reduce the total time duration of these MACoverheads as much as possible. Specifically, we propose CFP to eliminate the redundant back-offstage and MPD to avoid the lengthy collision recovery.4.1.1 Contention Free Pre-sensingIn this subsection, we propose our first scheme called CFP to detect a CFC during the PRP.Network Operation with CFPA CFC is identified by a network node when it does not detect any other network node transmittingwith the same or higher priorities during the PRP. To detect a CFC, we equip network nodes withIBFD capability, and allow them to detect the PRS transmitted by other nodes while transmittinga PRS themselves. If a node does not detect any other PRS signal during its PRS transmission, itidentifies the channel to be contention free, i.e., detects the presence of a CFC. In case a CFC isdetected, as shown in Fig. 4.1, the source node skips the random back-off stage that is traditionallyfollowing the PRP, and transmits a preamble signal to gain access to the power line medium imme-diately after the PRP. Due to this, we observe from Figs. 2.2 and 4.1 that we save a time durationof up to CWmax · tSLOT, which is otherwise wasted for a redundant back-off, where CWmax is themaximum contention window size.Figure 4.1: MAC frame transmission with CFP when a CFC successfully detected.Detecting CFC using IBFD CFPConsider a network of K nodes, where N of those nodes contend to transmit a frame, with prioritylevels pn associated with each of the n = 1,2, ...,N nodes. A CFC occurs when only one of theN nodes transmits a message of the maximum priority level of all transmitting nodes, max(pn),with 0 < max(pn)≤ pstd, where pstd = 2m−1 (m ∈ Z+) is the highest supported priority level of30a message in the operating standard. For example, in the IEEE 1901 and HPAV standards, m= 2.In order to provide upward compatibility for future standards that may decide to support a greaternumber of priority levels to effectively serve traffic of varied nature, we present an analysis of ourproposed CFP procedure for general m.We designate a px-CFC to arise when only one network node transmits a message with max(pn)=px. Our CFP scheme is aimed to successfully detect such px-CFCs, for all 0 < px ≤ pstd. Everypriority level pn can be expressed aspn =m−1∑i=02m−1−iχi, (4.1)where χi ∈ {0,1} is the binary value of the ith PRS, PRSi, i.e., χi = 1 when a PRSi is transmittedby the node and χi = 0 otherwise. The priority levels are resolved bit-by-bit through the m prioritybits from PRS0 to PRSm−1, with the most significant bit, χ0, transmitted first as PRS0. During thePRP, a node of priority pn transmits the PRS signal in PRSi (0≤ i≤ m−1) if and only if χi = 1.Every pn is associated with a slot position j¯n, for which χ j¯n = 1 while χ jn = 0, ∀ jn > j¯n. Thatis, χ j¯n is the least significant ‘bit 1’ in the binary notation of pn. With an intention to preserve thelegacy PRP and introduce CFP as an add-on feature, we compel the nth network node with prioritypn to perform CFP only at PRS j¯n , when it has won all previous PRSs. If the node loses the PRPbefore PRS j¯n , it resigns from the PRP contention as per the legacy PRP, and therefore does notproceed to perform CFP.When the node has won in all the previous priority resolution slots, it transmits a PRS atPRS j¯n , and so will any other node with the same or a higher priority level. Therefore, if the nthnode detects another PRS transmitted at PRS j¯n , it deduces the presence of other node(s) of eitherthe same or a higher priority level. In either case, the nth node deduces a non-CFC. However,if it does not detect any PRS in PRS j¯n , it deduces the absence of any other node with the sameor a higher priority level. In such a CFC, the node skips the following back-off procedure asdescribed in Section 4.1.1. However, it still continues to listen to the medium in the subsequent31PRSs to complete the PRP, to inform other nodes of the on-going transmitting message’s prioritylevel in a robust way. By doing so, we also ensure that the CFP procedure does not interfere withthe conventional PRP, and is only a supplementary feature introduced to eliminate the redundantback-off time slots under CFC.The successful detection of a CFC is dependent on the extent of SI cancellation achieved bythe IBFD solution. A non-ideal SI cancellation in IBFD could subject CFP to detection failure andfalse alarms. In the following, we analytically compute the probabilities of detection errors andfalse alarms using realistic SI cancellation gain values reported in [33].Detection Error and False Alarm Rates of the CFPWe denote the false alarm and detection error rates of the CFP at a network node as PFA and PDE,respectively. To aid our derivations, we define the following three events at a given network node.• E0: The node transmits a PRS.• E1: The node detects the presence of at least one PRS signal transmitted by another node inthe network.• E2: At least one node other than the considered node actually transmits a PRS.We can now represent PFA = P(E0∩(E1|E¯2)), and PDE = P(E0∩(E¯1|E2)), where E¯n represents thenon-occurrence of the event En.In order to calculate PFA and PDE, we consider a network with two nodes A and B, with nodeA continuously transmitting PRSs, while node B either transmits a PRS or remains silent. Todetermine the detection error and the false alarm rate at node A, we view this scenario as an on-offkeying (OOK) transmission, with node B transmitting ‘bit 1’ when it transmits a PRS, and ‘bit 0’when it does not. Here, ‘bit 1’ corresponds to the PRS signal s1, whose samples are given by [6]s1[`] =103/20√L ∑c∈Ccos(2pi · c · `L−ψ(c)),`= 0,1, ...,L−1, (4.2)32where L is the total number of time samples transmitted in the PRS signal, C is the set of orthogonalfrequency division multiplexing (OFDM) sub-carriers used for PRS transmission, and ψ(c) is asub-carrier specific phase angle [6]. Thus, PFA represents the probability of node A detecting a ‘1’when a ‘0’ is transmitted by node B, and PDE represents the probability of detecting a ‘0’ whena ‘1’ is transmitted. Considering that the signal is subject to possible phase distortions along theline, we apply non-coherent detection at node A to get [40, Ch. 7]PFA = exp(−b202), (4.3)PDE = 1−Q(√2γ,b0), (4.4)where Q(·, ·) is the first-order Marcum-Q function, b0 is the decision threshold of non-coherentOOK normalized to the root-mean-square noise value, and γ is the signal-to-noise ratio (SNR).The latter is given asγ =EbN0+ΨRSI, (4.5)where Eb represents the received energy per-bit, N0 is the average power spectral density (PSD) ofthe cumulative noise at the receiver of node A, and ΨRSI is the average residual self-interference(RSI) PSD after non-ideal SI cancellation. For brevity, we define N0,FD = N0 +ΨRSI as the neweffective ‘noise floor’ under IBFD operation.To determine realistic values of PDE and PFA in a HAN, we derive an expression for γ in termsof known transmission parameters and channel conditions. The received bit-energy can be writtenas Eb = ΦRtpd, where tpd is the time interval in which CFP is performed, and ΦR is the power ofthe received signal, which can in turn be written asΦR =f2∫f1ΨR( f )d f , (4.6)with ΨR( f ) being the PSD of the received signal at a frequency f , and f1 and f2 are the lowerand upper frequency limits of the transmission band, respectively. We further express ΨR( f ) in33terms of the known transmit PSD, ΨT( f ), as ΨR( f ) = ΨT( f ) · |H( f )|2, where H is the powerline channel frequency response between nodes A and B. The maximum transmit PSD is typicallyregulated to limit the electromagnetic interference caused by BB-PLC [25]. For our analysis,we consider the devices to always transmit signals with maximum PSD ΨT,max, although newerdevices support variable transmit PSDs [14]. Further, it is safe to assume the channel gain to beflat within each sub-carrier, since the HPAV PRS sub-carrier spacing is smaller than the observedchannel coherence bandwidth in typical in-home BB-PLC networks [6, 41]. We can thereforere-write (4.6) asΦR =ΨT,max ∑c∈C|H( fc)|2 ·∆ f , (4.7)where fc is the center frequency of the cth OFDM sub-carrier, and ∆ f is the sub-carrier spacing.Since N0,FD = 1|C | ∑c∈CN0,FD( fc), where N0,FD( fc) is the effective noise floor of the cth OFDMsub-carrier under IBFD operation, we express (4.5) asγ =ΨT,maxtpd∆ f |C |∑c∈C|H( fc)|2∑c∈CN0,FD( fc). (4.8)We now determine the optimal value of the threshold b0 to be used in (4.3) and (4.4). Denotingthe overall network node error rate asPe = P(E2|E0)PDE+P(E¯2|E0)PFA, (4.9)we define the optimal threshold asbopt , argminb0Pe. (4.10)In the appendix, we derive the following approximate solution to (4.10),bopt ≈√(γ+ lnτ)2+4(γ+ lnτ)2γ, (4.11)34Table 4.1: PRS SNR under Varying Minimum Channel Gains|Hmin|2 SNRHD SCINR N0,FD γmin(dB) (dB) (dB) (dBm/Hz) (dB)-5 65 32 -87 60-10 60 30 -90 58-20 50 27 -97 56-30 40 27 -107 56-40 30 21 -111 50-50 20 12 -112 40-60 10 2 -112 30where τ = P(E¯2|E0)P(E2|E0) . The value of τ depends on the network condition (between idle and heavilyloaded).Since Pe is a monotonically decreasing function with respect to γ for optimal threshold bopt [40,Ch. 7], and since |H( fc)|2N0,FD( fc)decreases with |H( fc)|2 [12], we use a flat minimum channel gain,|Hmin|2, to obtain an upper bound for the total error, Ptot, asPtot = PFA+ τPDE (4.12)≤ 1−Q(√2γmin,bopt)+τ exp(−b2opt2), (4.13)where γmin =ΨT,max|C ||Hmin|2∆ f tpdN0,FD,and N0,FD is now the effective IBFD noise floor for a channel gain of |Hmin|2.Using (4.13), we calculate practical upper-bound values of Ptot that we expect to encounterwhen CFP is deployed in in-home networks. In accordance with the HPAV specifications, we setΨT,max = −50 dBm/Hz, ∆ f = 195 kHz, |C | = 153, and tpd = 25.92 µs [6]. We use an ambientpower line noise PSD of N0 =−120 dBm/Hz as the noise floor at the receiver in half-duplex mode.Note that the effective noise floor under IBFD operation is related to N0 asN0,FD = N0 · SNRHDSCINR, (4.14)35where SNRHD is the SNR of the signal-of-interest (SOI) in half-duplex mode, and SCINR is thesignal-to-canceled-interference-plus-noise-ratio after the SI cancellation [12]. We then use the SIcancellation gain values reported in [12, Table II] to obtain γmin for different minimum channelgains in Table 4.1. From (4.11), we can further calculate bopt for each obtained γmin.As an example, we consider the case where τ = 1 to evaluate PDE and PFA. With τ = 1, (4.11)simplifies tobopt ≈√γ2+2. (4.15)With the values computed in Table 4.1 and (4.15), we calculate Ptot to be near-zero (< 10−100)under various minimum channel gain conditions down to |Hmin|2 =−60 dB. From (4.12), we canalso use Ptot as the individual upper-bound for both PFA and PDE, since 0≤ PFA,PDE ≤ 1. For otherτ values, we have similar results and Ptot, PFA and PDE are all vanishingly small. This assures usthat the deployment of CFP in practical in-home BB-PLC network environments results in virtuallyno detection errors or false alarms.4.1.2 Mutual Preamble DetectionWe now introduce our second access control enhamcement called MPD, where we use the medium-aware transmission ability provided by the IBFD capability to avoid lengthy collision recovery bypredicting a future frame collision. Through MPD, we essentially propose a practical scheme torealize CSMA/CD in BB-PLC networks by detecting the overlapping preamble signals.Network Operation with MPDWe consider all nodes in the network are equipped with the ability to transmit a preamble andsimultaneously sense the power line medium for other possible preamble signal transmissions. Inthis way, when two or more network nodes transmit a preamble signal at the same back-off timeslot and gain access to the power line channel simultaneously, they each predict a future datapayload collision by detecting a preamble other than their own. Under such circumstances, wecompel these conflicting nodes to transmit another preamble signal subsequently. This acts as a36jamming signal to ensure that all network nodes are made aware of a potential collision. We thenlet the nodes follow the standard HPAV collision recovery procedure in a guard interval of tCIFS.This operation is illustrated in Fig. 4.2.Figure 4.2: Activity on the medium in case of a collision with the deployment of MPD.We notice that the time interval between the two back-off stages (the back-off stage of thecolliding MAC frame and the back-off stage after the collision recovery) is reduced to 2tp+ tCIFSwith our MPD scheme, while it is of duration tEIFS in the original HPAV MAC protocol, as shownin Fig. 2.3.IBFD Preamble DetectionTo determine the success of an IBFD preamble detection, consider a BB-PLC network with twonodes, A and B, with node A continuously transmitting preambles slot-by-slot, while in each timeslot, node B either transmits a preamble or not. Similar to Section 4.1.1, we view the behaviorof node B as a source continuously transmitting information bits using OOK, with the preamblesignal being the transmission pulse. In every preamble time slot, node B transmits a bit ‘1’ to senda preamble to node A, and a ‘0’ when it has nothing to transmit. An IBFD enabled node A is ableto continuously detect the information bit sent by node B in each time slot. This scenario is similarto the one considered in Section 4.1.1, albeit, the transmission pulse is now a preamble signal, s2,wheres2[`] =103/20√L ∑c∈Ccos(2pi · c · `L+ψ(c)),`= 0,1, ...,L−1, (4.16)37with the value of L being the same as that for the PRS signal. The preambles also use the same setof OFDM sub-carriers as the PRSs, with the phase shift of each corresponding sub-carrier being thesame in magnitude but opposite in sign. Therefore, the PFA and PDE formulations in Section 4.1.1also apply to the above scenario. This also implies that all the computations in (4.3) to (4.13), aswell as the data reported in Table 4.1, are valid for detecting preambles as well. Thus, the detectionerror rate and the false alarm rate for MPD are also practically zero.4.1.3 Implementation of CFP and MPD in a BB-PLC DeviceHardware Implementation CostsThe elementary requirement for implementing CFP and MPD in power line networks is to in-corporate BB-PLC modems with IBFD capability. Recent works have shown that adding IBFDcapability to a BB-PLC device requires minimal changes to the modem chipsets, with only an ad-ditional power consumption of about 0.1 W for the active hybrid circuit that is used at the powerline-modem interface [33].InteroperabilityOur proposed CFP and MPD schemes are completely inter-operable with half-duplex (HD) de-vices. For CFP, an IBFD-enabled node can detect a CFC when it is the only node transmitting thehighest priority message regardless of whether the other nodes are IBFD-enabled. However, anHD node is unable to detect a CFC. For MPD, when only a part of the network nodes are IBFD-enabled, MPD still offers improvements in η , but with reduced effect compared to the case whenall the network nodes are IBFD-enabled. As long as the conflicting nodes are IBFD-enabled, adata payload collision can be successfully predicted and avoided using MPD. However, when anHD node is involved as a conflicting node, the ensuing data payload collision is inevitable, and ittakes the network a time interval of EIFS to recover from it.384.2 An Interface to Accomodate HAN ApplicationsWe aim to provide a unified solution to support various applications running over PLC HANs,which include in-home multimedia applications as well as home automation applications. To ac-commodate the heterogenous network traffic generated by these applications, in this section, wepropose an interface with network traffic prioritization and traffic shaping.4.2.1 Network Traffic Generated by HAN ApplicationsWe first look into the network traffic generated by various HAN applications.1. Home Automation Applications: Home automation traffic is typically short in duration butfrequent in occurrence. It generally consists of data collected by various sensors, like tem-perature monitoring data, or control commands sent by control units, like turning on/off adevice. Due to the large number of sensors, controllers, and actuators in future in-home en-vironments [42], we expect the number of data packets to be also large. These data packetsrequire in-time robust delivery. However, there are also applications like video surveillancethat require greater bandwidth. In addition, there is always a constant background informa-tion network traffic that only requires a best-effort delivery.2. In-home Multimedia Applications: Multimedia data are generally AV streams that are, forexample, generated by video conferencing, in-home gaming, or high-definition AV stream-ing. These are bursty in nature and have high data throughput requirements [43]. Concur-rently, network traffic generated by in-home multimedia applications are also required tobe delivered robustly in time. However, multimedia traffic also consists of accompanyingcontrol messages that are to be handled similar to home automation traffic, and backgroundinformation data that only require best-effort delivery.3. PLC Network Management Function: HANs also contain critical network management traf-fic that requires in-time robust delivery.39Figure 4.3: Ilustration of Prioritization and Traffic Shaping4.2.2 Prioritizing HAN TrafficThe heterogeneous network traffic generated by these applications require HANs to provide QoSdifferentiation function. Since the HPAV protocol supports MAC frame transmissions with differ-ent priority levels, we propose a specific network traffic prioritization scheme for the HAN. Weassign the highest priority level to network management traffic, control messages, and real-timemonitoring data, which we classify as general control message. Next, we assign the next twolower priority levels to multimedia traffic, which mainly includes AV streaming data. Finally, weassign priority 0 to background information traffic flow and all other best-effort network traffic.An illustration of this network traffic prioritization scheme can be found in Fig. 4.3.Since we prioritize the HAN traffic, we briefly describe the characteristics and requirements ofthe associated MAC frames.1. Priority 3: General control messages typically require in-time robust delivery. Some of thesemessages are generated in a cyclic manner while some other are triggered event based. Weexpress the average arrival rates of priority 3 MAC frames at the nth network node as λn,3,40which depends on the average message generation rates of general control messages.2. Priority 2 and 1: We reserve these priorities to multimedia traffic. The multimedia contentis either stored locally at the source node or is retrieved from external sources (typicallythe Internet). In either case, the source node pre-fetches multimedia content into the bufferto ensure that there is always some AV streaming data ready to be transmitted beforehand.Such a pre-fetching (also referred to in the literature as buffering or caching) is required atthe source node for the smooth delivery of multimedia streams [44–46]. This avoids jittersin the AV playback that could significantly degrade user experience. In order for the efficientplayback of the buffered data, as well as to prevent lower priority starvation and to preventmultimedia traffic from using up all the network resource, in Section 4.2.3, we propose theuse of traffic shaping as is shown in Fig. 4.3.3. Priority 0: After serving the higher priority traffic, we tune the network nodes to utilize theremaining resources to transmit as many priority 0 MAC frames as possible. In order tofully utilize the network resources, we supply each network with a constant requirement totransmit priority 0 MAC frames.4.2.3 Traffic ShapingSeveral traffic shapers like Poisson traffic shaping (PTS), token-bucket shaping, and credit-basedshaping could be implemented in our network [39, 47, 48]. As an example, we incorporate thePTS as our traffic shaping scheme. A Poisson traffic shaper simply pushes the content stored in thebuffer into the MAC layer as a Poisson process [48].The PTS is realized through the CCo of the PLC logical network [7, Ch.]. The CCodynamically allocates λn,i to each of the network nodes through beacon payload messages, whereλn,i denotes the arrival rate of the nth network node for the ith priority MAC frame (i ∈ {1,2}).Once a node receives this allocation, it pushes the buffered packets of the ith priority level with arandom inter-arrival time drawn from an exponential distribution with a mean of 1λn,i .414.2.4 Admission ControlIn order to provide each priority level with an appropriate share of the network resource, we im-plement admission control for the data packets. We denote the average MAC frame intervals at thenth network node as µn,i for the ith priority MAC frame (i ∈ {1,2,3}). This can be seen in Fig. 2.2as the duration from the beginning of the PRS0 to the end of the CIFS.For a network with N active network nodes, admission control at the CCo ensures thatN∑n=1(µn,3λn,3+µn,2λn,2+µn,1λn,1) = κ, (4.17)where κ < 1. This guarantees the first three priority level messages to be delivered without accu-mulating at any node. We should also choose a parameter κ way smaller than 1 to account for afair portion of the network resource to be reserved for collisions, retransmissions as well as besteffort message transmissions.Under scenarios where the relationship of (4.17) cannot be satisfied, switching from CSMA-only mode to a CSMA-plus-TDMA mode is appropriate [13]. For example, when multiple nodesin the network are transmitting high-speed video streams of equal priority levels, it is more suitableto serve them in a round-robin fashion.4.3 Performance EvaluationsIn order to verify the effectiveness of our proposed schemes, we simulate the in-home PLC networkand evaluate their performance in three different network settings using a discrete event simulator,OMNeT++ [49].4.3.1 Simulation ConfigurationThe simulation network topology is shown in Fig. 4.4. Several network nodes, including the CCo,are interconnected to each other through the power line medium in a star configuration. Out ofthese, only a set of N nodes are active with data to transmit. In our simulations, we assume42Figure 4.4: Network Simulation TopologyTable 4.2: Simulation Parameters for the HANParameter ValueSimulation Time, TS 30 stCIFS 100 µsPRS and Back-off slot time, tSLOT 35.84 µstp 35.84 µstFC 133.92 µsMaxFL 2341.12 µstRIFS 140 µstEIFS 2920.64 µsan identical time interval of the data payload, tFL, regardless of its priority level. The simulationparameters are listed in Table 4.2, and are based on the HPAV specifications [6]. The significanceof the simulation results is guaranteed by the sufficient simulation time, TS, where the resultantMAC efficiency of each simulation run is the average performance of several thousands of MACframe transmissions.By denoting the total number of transmitted MAC frames with successful acknowledgments asnACK, we compute the MAC efficiency, η , at the end of our simulation runs asη =nACKtFLTS, (4.18)43where TS is the total simulation time (see Table 4.2).We assume that the physical layer uses a robust transmission mode (referred to as ROBO modein [25]) to transmit control sequences, while we consider the multimedia streams with inherentdata redundancy to be error tolerant. Therefore, for simplicity, we do not consider transmissionerrors as well as the associated retransmissions in our simulation [50, 51]. In such conditions,nACK is simply equal to the total number of frames transmitted without encountering collisions.4.3.2 Performance of CFP with Single Node FloodingFor our first result, we use the following network setting to test the effectiveness of the CFP scheme.We set |N | = 1 by letting the CCo be the only active network node that continuously transmitspriority-3 MAC frames to all the other network nodes without channel idling. Under such a sce-nario, our proposed MPD scheme has no effect as no contention or collision occurs with thissetting. The impact of varying tFL on the achieved MAC efficiency with and without CFP is shownin Fig. 4.5.Since the network experiences no collision or idle time intervals, the power line medium iskept busy by continuously transmitting MAC frames shown in Fig. 2.2. Under such conditions, theMAC efficiency can also be represented asη =tFL(tEIFS−MaxFL)+ tFL+(2+E[nBF])tSLOT , (4.19)where E[nBF] is the expected number of back-off time slots. Using the HPAV protocol withoutCFP, the contention remains at the base stage of 0 as no node other than the CCo transmits packets,and hence the transmission of these packets encounters no collisions or deferral [6]. As a result,every time the CCo transmits a MAC frame, the BC is randomized to an integer value uniformlydistributed between 0 and CWmin, where CWmin is the smallest contention window size (7, as spec-ified in the HPAV protocol). Therefore, E[nBF] = 72 = 3.5. However, with our CFP deployed, theCCo detects a CFC at every frame transmission, since there are no other contending nodes. Thus,the redundant back-offs are avoided at the CCo. This results in E[nBF] = 0, thereby increasing η .440 20 40 60 80 100tFL (% of MaxFL)ηwithout CFPwith CFPFigure 4.5: MAC efficiency as a function of tFL under single node flooding.Furthermore, when we set tFL = MaxFL for our CFP deployment, we observe η = 78.21%,which is close to the theoretical maximum of ηmax = 78.24% (from (2.5)). The slight differencecan be attributed to the bootstrapping process of our simulations, as well as the last frame in oursimulations not being able to finish its transmission within the total simulation time. When thesesimulation non-idealities are accounted for, the achieved MAC efficiency reaches the theoreticalmaximum with tFL =MaxFL.4.3.3 Performance Evaluation with Multiple Active NodesTo evaluate the network performance with multiple active nodes, we enable 1 < |N | ≤ 25. Weform two sub-settings where we fix |N | = 10 and vary tFL in the first case, while we fix tFL =MaxFL and vary |N | in the second.45Network Resource AllocationWe calculate the average MAC frame interval at the nth network node in our simulations asµn,i = (tEIFS−MaxFL)+ tFL+(2+E[nBF]n,i)tSLOT, (4.20)for the ith priority MAC frame, where E[nBF]n,i represents expected number of back-off time slotsat the nth network node for the ith priority data packet. Since the number of back-off time slots ishard to predict, we approximate µn,i asµn,i ≈ µ = (tEIFS−MaxFL)+ tFL+2tSLOT. (4.21)Thus, by considering the above approximation, the admission control in our simulations can beexpressed by re-writing (4.17) asµ|N |∑n=1(λn,3+λn,2+λn,1) = κ. (4.22)To account for the back-off time slots we ignored in our approximation of µn,i, we choose a smallκ = 0.65. By setting|N |∑n=1λn,i = κiµ , we allocate a certain portion of network resource to the ithpriority network traffic, which for simplicity, we further allot equally to all |N | network nodes.Thus, we have λn,i = 1|N |κiµ , ∀n ∈N and i ∈ {1,2}. For general control messages, we assume anequal event rate at each network node as λn,3 = 1|N |κ3µ , ∀n ∈N . In our simulations, we assumeκ3 = 0.25, and κ2 = κ1 = 0.2.Performance Evaluations with Varying tFLWe first simulate the network with varying tFL and fix |N | = 10. The variation of the MAC effi-ciency with varying tFL is shown in Fig. 4.6. The curves essentially resemble those in single nodeflooding, but with reduced η , because of contentions and collisions. However, all our proposedschemes improve the MAC efficiency compared to the standard HPAV protocol.460 20 40 60 80 100tFL (% of MaxFL)ηw/o CFP, w/o MPDw/ CFP, w/o MPDw/o CFP, w/ MPDw/ CFP, w/ MPDFigure 4.6: MAC efficiency as a function of tFL.We observe that simultaneous deployment of CFP and MPD yields an η = 76.80%, whichachieves 98.16% of the optimal MAC efficiency, ηmax = 78.24%. At the same time, we notice thata conventional HPAV protocol without CFP and MPD only manages to provide η = 69.43%.Performance Evaluations with Varying Number of Active NodesFor our final result, we simulate the network with varying |N | and a fixed tFL = MaxFL. Thesimulation results of this subsetting are shown in Fig. 4.7. We observe that without our MPDscheme, η decreases as the number of active network nodes increases due to the increased collisionrate as well as the lengthy collision recovery time. However, we achieve a stable η across differentnumber of nodes with MPD since our proposed MPD scheme virtually achieves CD. With our MPDscheme, the potential collisions can be successfully detected and avoided, which brings the cost of475 10 15 20 25Number of Active Network Nodes0.660.680.70.720.740.760.78ηw/o CFP, w/o MPDw/ CFP, w/o MPDw/o CFP, w/ MPDw/ CFP, w/ MPDFigure 4.7: MAC efficiency as a function of the number of active nodes.a potential collision to the minimum. Because of this, as the number of network nodes increases,although the collision rate is increased, we do not observe a degradation of MAC efficiency whenour MPD scheme is applied. We further observe that although the use of CFP without MPD doesnot provide a stable η across varying nodes, it still manages to improve η from that obtained withthe original HPAV protocol, as a result of redundant back-off times being eliminated under CFCs.We also observe that we obtain the best η , both in terms of absolute value and stability acrossincreasing nodes, using both our proposed schemes of CFP and MPD.4.3.4 Discussion on Metrics Used in EvaluationsIn this chapter, we aim to improve the MAC efficiency through the adoption of our proposedschemes. Thus, we evaluate the network performance under various conditions using MAC effi-48ciency.Another network metrics to our interest is the latency of network traffic with non-zero prioritylevel. However, we use admission control to ensure that messages of first three priority levels canbe emptied from time to time. The probability of a message with non-zero priority level stayingin the network goes exponentially small as the staying time increases. Thus, we do not presentlatency data in our HAN performance evaluations.4.4 SummaryIn this chapter, we have proposed some efficient access control schemes for BB-PLC in HAN. Inorder to efficiently translate PHY data rate onto the MAC throughput, we have leveraged IBFD topropose two novel schemes, CFP and MPD. CFP eliminates the redundant back-off stages whileMPD avoids the lengthy collision recovery. To accommodate heterogeneous network traffic gener-ated by various applications running over PLC HAN, we have proposed an interface with networktraffic prioritization and traffic shaping. We have presented simulation results showing under sin-gle network node flooding and maximal MAC frame size conditions, theoretical optimal MACefficiency can be achieved. We have shown that under various conditions, CFP works well withMPD to considerably increase the MAC efficiency close to optimal value while each scheme stand-ing alone can also provide MAC efficiency improvements over the original HPAV protocol.49Chapter 5Conclusions5.1 SummaryIn this thesis, we have proposed some efficient access control schemes for IVNs and HANs. Ineach of the two considered application scenarios, on one hand we have made some refinementsto the original MAC protocols to better support the QoS requirements of different network traffic,while on the other hand, we have developed some upper layer specifications to better interface theinvolved applications.In Chapter 3, we have explained the details of our proposed efficient access control schemesfor IVNs. Through network simulations we have shown that the introduced VC mechanism ef-fectively reduces the latency of different classes of network traffic involved. The proposed TXSAcombining strict priority TXSA and AVB CBS TXSA better deals with the relationship betweentransmissions of different classes of network traffic so that the latency requirements of differentclasses of network traffic are decently satisfied.In Chapter 4, we have explained the details of our proposed efficient access control schemesfor HANs. Through theoretical and numerical evaluations we have shown our proposed CFP andMPD work well together to considerably increase the MAC efficiency. The PHY data rate are bettertranslated into the throughput in MAC layer. We have also developed an interface to accommodateheterogenous network traffic generated by various applications running over PLC HAN.50We have adopted a comprehensive model in Chapter 3. While such a model is more implicativeof what can be achieved in a real scenario, it prevents us from a thorough theoretical analysis, whichis more viable under a simplistic model. As a result, the reasoning in this chapter is more heuristicthan analytical and the evaluations following is more empirical than theoretical. To improve this,we can first derive analytical results using a simplified model and then discuss implications of theseresults in a real scenario.It is noticed that the MAC protocol of a BB-PLC network much resembles that of a wirelessnetwork. In particular, CSMA is implemented as the main operation mode in both of the MACprotocols. Access control schemes developed for wireless networks can give us a lot of inspirationswhen we study related topics in BB-PLC networks. With some adaptions, some schemes developedfor wireless networks can be implemented in a BB-PLC network to further improve the networkperformance.5.2 Future DirectionsFor potential extensions of the current work, we suggest to reconsider our proposed schemes ina multi-hop network environment. This also brings about the hidden terminal/exposed terminalproblem in which case some enhanced schemes are required to reduce the lengthy collision re-covery time or to avoid the superfluous random back-off stage. For example, it is possible thattwo network nodes both transmit a preamble and each detects no preambles transmitted by othernetwork nodes because of hidden terminals. The original HPAV protocol implements request-to-send/clear-to-send (RTS/CTS) mechanism to overcome this issue. We may also use it in an FDBB-PLC network. However, we expect a more efficient realization using IBFD by overlapping(partially) the transmission of RTS and CTS.Furthermore, the MAC efficiency can be further improved by monitoring network statisticsand dynamically choosing optimum CW. For example, we can implement adaptive schemes witha threshold to vary the size of CW according to the estimated network load.51Bibliography[1] S. Tuohy, M. Glavin, C. Hughes, E. Jones, M. Trivedi, and L. Kilmartin, “Intra-vehiclenetworks: A review,” IEEE Transactions on Intelligent Transportation Systems, vol. 16,no. 2, pp. 534–545, 2015. → pages 1, 3, 7, 9[2] J.-P. Javaudin, M. Bellec, D. Varoutas, and V. Suraci, “OMEGA ICT project: Towardsconvergent gigabit home networks,” in IEEE International Symposium on Personal, Indoorand Mobile Radio Communications (PIMRC), Cannes, France, pp. 1–5, September 2008.→ pages 1[3] S. Galli, A. Scaglione, and Z. Wang, “For the grid and through the grid: The role of powerline communications in the smart grid,” Proceedings of the IEEE, vol. 99, no. 6,pp. 998–1027, 2011. → pages 1, 6[4] A. Majumder and J. Caffery, “Power line communications,” IEEE Potentials, vol. 23, no. 4,pp. 4–8, 2004. → pages 1, 6[5] L. Lampe, A. Tonello, and T. Swart, “Introduction,” in Power Line Communications:Principles, Standards and Applications from Multimedia to Smart Grid (L. Lampe,A. Tonello, and T. Swart, eds.), ch. 1, pp. 1 – 7, John Wiley and Sons Ltd, 2016. → pages 2,6, 10[6] “HomePlug AV specification,” HomePlug Powerline Alliance, 2007. → pages 2, 10, 11, 17,32, 33, 34, 35, 43, 4452[7] S. Galli, H. Latchman, V. Oksman, G. Prasad, and L. Yonge, “Multimedia PLC systems,” inPower Line Communications: Principles, Standards and Applications from Multimedia toSmart Grid (L. Lampe, A. Tonello, and T. Swart, eds.), ch. 8, pp. 475 – 511, John Wiley andSons Ltd, 2016. → pages 2, 10, 12, 41[8] H. A. Latchman, S. Katar, L. W. Yonge, and S. Gavette, “HomePlug Green PHY,”Homeplug AV and IEEE 1901: A Handbook for PLC Designers and Users, pp. 302–311,2013. → pages 2, 8, 24[9] N. Pavlidou, A. H. Vinck, J. Yazdani, and B. Honary, “Power line communications: state ofthe art and future trends,” IEEE Communications Magazine, vol. 41, no. 4, pp. 34–40, 2003.→ pages 2[10] R. P. Antonioli, M. Roff, Z. Sheng, J. Liu, and V. C. M. Leung, “A real-time MAC protocolfor in-vehicle power line communications based on HomePlug GP,” in IEEE VehicularTechnology Conference (VTC Spring), Glasgow, Scotland, pp. 1–5, May 2015. → pages 3,9, 10[11] H. Hrasnica, A. Haidine, and R. Lehnert, Broadband Powerline Communications: NetworkDesign. John Wiley and Sons Ltd, 2005. → pages 4, 11[12] G. Prasad, L. Lampe, and S. Shekhar, “Enhancing transmission efficiency of broadband PLCsystems with in-band full duplexing,” in IEEE International Symposium on Power LineCommunications and Its Applications (ISPLC), Bottrop, Germany, March 2016. → pages 4,11, 12, 35, 36[13] IEEE Standards Association, “IEEE standard for broadband over power line networks:Medium access control and physical layer specifications,” IEEE Std 1901, pp. 1–1586, 2010.→ pages 6, 12, 42[14] L. Yonge, J. Abad, K. Afkhamie, L. Guerrieri, S. Katar, H. Lioe, P. Pagani, R. Riva, D. M.53Schneider, and A. Schwager, “An overview of the homeplug AV2 technology,” Journal ofElectrical and Computer Engineering, vol. 2013, 2013. → pages 7, 34[15] S. Poledna, W. Ettlmayr, and M. Novak, “Communication bus for automotive applications,”in IEEE European Solid-State Circuits Conference (ESSCIRC), Villach, Austria,pp. 482–485, September 2001. → pages 7[16] G. Leen, D. Heffernan, and A. Dunne, “Digital networks in the automotive vehicle,”Computing and Control Engineering Journal, vol. 10, no. 6, pp. 257 – 66, 1999. → pages 7[17] N. Taherinejad, R. Rosales, L. Lampe, and S. Mirabbasi, “Channel characterization forpower line communication in a hybrid electric vehicle,” in IEEE International Symposiumon Power Line Communications and Its Applications (ISPLC), Beijing, China, pp. 328–333,March 2012. → pages 8[18] G. Alderisi, A. Caltabiano, G. Vasta, G. Iannizzotto, T. Steinbach, and L. L. Bello,“Simulative assessments of IEEE 802.1 Ethernet AVB and time-triggered Ethernet foradvanced driver assistance systems and in-car infotainment,” in IEEE Vehicular NetworkingConference (VNC), Seoul, Korea, pp. 187–194, November 2012. → pages 9[19] IEEE Standards Association, “IEEE standard for local and metropolitan area networksvirtual bridged local area networks,” IEEE Std 802.1Q-2005 (Incorporates IEEE Std802.1Q1998, IEEE Std 802.1u-2001, IEEE Std 802.1v-2001, and IEEE Std 802.1s-2002),pp. 1–285, 2006. → pages 9, 22[20] C. Cano and D. Malone, “Performance evaluation of the priority resolution scheme in PLCnetworks,” in IEEE International Symposium on Power Line Communications and itsApplications (ISPLC), Glasgow, Scotland, pp. 290–295, March 2014. → pages 10[21] C. Cano and D. Malone, “When priority resolution goes way too far: An experimentalevaluation in PLC networks,” in IEEE International Conference on Communications (ICC),London, U.K., pp. 952–957, June 2015. → pages 1054[22] Z. Sheng, A. Kenarsari-Anhari, N. Taherinejad, and V. C. M. Leung, “A multichannelmedium access control protocol for vehicular power line communication systems,” IEEETransactions on Vehicular Technology, vol. 65, no. 2, pp. 542–554, 2016. → pages 10[23] “HomePlug 1.0 specification,” HomePlug Powerline Alliance, 2001. → pages 10[24] S. Nowak, F.-M. Schaefer, M. Brzozowski, R. Kraemer, and R. Kays, “Towards aconvergent digital home network infrastructure,” IEEE Transactions on ConsumerElectronics, vol. 57, no. 4, pp. 1695–1703, 2011. → pages 10[25] H. A. Latchman, S. Katar, L. Yonge, and S. Gavette, Homeplug AV and IEEE 1901: AHandbook for PLC Designers and Users. Wiley-IEEE Press, 2013. → pages 11, 34, 44[26] A. Leon-Garcia and I. Widjaja, Communication Networks. McGraw-Hill Inc., 2003. →pages 11[27] S. Sen, R. R. Choudhury, and S. Nelakuditi, “CSMA/CN: Carrier sense multiple access withcollision notification,” IEEE/ACM Transactions on Networking, vol. 20, no. 2, pp. 544–556,2012. → pages 11, 12[28] T. Vermeulen, F. Rosas, M. Verhelst, and S. Pollin, “Performance analysis of in-band fullduplex collision and interference detection in dense networks,” in IEEE Annual ConsumerCommunications & Networking Conference (CCNC), Las Vegas, NV, pp. 595–601, January2016. → pages 12[29] Y.-S. Choi and H. Shirani-Mehr, “Simultaneous transmission and reception: Algorithm,design and system level performance,” IEEE Transactions on Wireless Communications,vol. 12, no. 12, pp. 5992–6010, 2013. → pages 12[30] T. Vermeulen and S. Pollin, “Energy-delay analysis of full duplex wireless communicationfor sensor networks,” in IEEE Global Communications Conference (GLOBECOM), Austin,TX, pp. 455–460, December 2014. → pages 1255[31] Y. Liao, K. Bian, L. Song, and Z. Han, “Full-duplex MAC protocol design and analysis,”IEEE Communications Letters, vol. 19, no. 7, pp. 1185–1188, 2015. → pages 12[32] L. Song, Y. Liao, K. Bian, L. Song, and Z. Han, “Cross-layer protocol design for CSMA/CDin full-duplex WiFi networks,” IEEE Communications Letters, vol. 20, no. 4, pp. 792–795,2016. → pages 12[33] G. Prasad, L. Lampe, and S. Shekhar, “In-band full duplex broadband power linecommunications,” IEEE Transactions on Communications, vol. 64, no. 9, pp. 3915 – 3931,2016. → pages 12, 32, 38[34] C. Cano, A. Pittolo, D. Malone, L. Lampe, A. M. Tonello, and A. Dabak, “State-of-the-art inpower line communications: From the applications to the medium,” IEEE Journal onSelected Areas in Communications, vol. 34, no. 7, pp. 1935 – 1952, 2016. → pages 12[35] Y.-J. Lin, H. A. Latchman, M. Lee, and S. Katar, “A power line communication networkinfrastructure for the smart home,” IEEE Wireless Communications, vol. 9, no. 6,pp. 104–111, 2002. → pages 12[36] G. Hallak and G. Bumiller, “PLC for home and industry automation,” in Power LineCommunications: Principles, Standards and Applications from Multimedia to Smart Grid(L. Lampe, A. Tonello, and T. Swart, eds.), ch. 7, pp. 449 – 472, John Wiley and Sons Ltd,2016. → pages 12[37] Y. Huo, Q. Zheng, Z. Sheng, and V. C. M. Leung, “Queuing enhancements for in-vehicletime-sensitive streams using power line communications,” in IEEE/CIC InternationalConference on Communications in China (ICCC), Shenzhen, China, pp. 1–6, November2015. → pages 18[38] M. D. Johas Teener, A. N. Fredette, C. Boiger, P. Klein, C. Gunther, D. Olsen, andK. Stanton, “Heterogeneous networks for audio and video: Using IEEE 802.1 audio videobridging,” Proceedings of the IEEE, vol. 101, no. 11, pp. 2339–2354, 2013. → pages 2256[39] IEEE Standards Association, “IEEE standard for local and metropolitan area networks -virtual bridged local area networks amendment 12: Forwarding and queuing enhancementsfor time-sensitive streams,” IEEE Std 802.1Qav-2009 (Amendment to IEEE Std802.1Q-2005), Jan. 2009. → pages 22, 41[40] M. Schwartz, W. R. Bennett, and S. Stein, Communication Systems and Techniques. JohnWiley & Sons, 1995. → pages 33, 35, 60[41] F. J. Canete, J. Corte´s, L. Dı´ez, and J. T. Entrambasaguas, “A channel model proposal forindoor power line communications,” IEEE Communications Magazine, vol. 49, no. 12,pp. 166–174, 2011. → pages 34[42] F. Xia, L. T. Yang, L. Wang, and A. Vinel, “Internet of things,” International Journal ofCommunication Systems, vol. 25, no. 9, p. 1101, 2012. → pages 39[43] Z. Sahinoglu and S. Tekinay, “On multimedia networks: Self-similar traffic and networkperformance,” IEEE Communications Magazine, vol. 37, no. 1, pp. 48–52, 1999. → pages39[44] L. Golab and M. T. O¨zsu, “Issues in data stream management,” ACM Sigmod Record,vol. 32, no. 2, pp. 5–14, 2003. → pages 41[45] S. A. Saberali, H. E. Saffar, L. Lampe, and I. Blake, “Adaptive delivery in cachingnetworks,” IEEE Communications Letters, vol. 20, pp. 1405 – 1408, July 2016. → pages[46] R. Rejaie, H. Yu, M. Handley, and D. Estrin, “Multimedia proxy caching mechanism forquality adaptive streaming applications in the Internet,” in Annual Joint Conference of theIEEE Computer and Communications Societies (INFOCOM), Tel Aviv, Israel, vol. 2,pp. 980–989, March 2000. → pages 41[47] F. Gomez-Cuba, R. Asorey-Cacheda, and F. J. Gonzalez-Castano, “Smart grid last-mile57communications model and its application to the study of leased broadband wired-access,”IEEE Transactions on Smart Grid, vol. 4, no. 1, pp. 5–12, 2013. → pages 41[48] F. P. Kelly, “Networks of queues,” Advances in Applied Probability, pp. 416–432, 1976. →pages 41[49] A. Varga, “OMNeT++,” in Modeling and Tools for Network Simulation, pp. 35–59,Springer, 2010. → pages 42[50] G. Bianchi, “Performance analysis of the IEEE 802.11 distributed coordination function,”IEEE Journal on selected areas in communications, vol. 18, no. 3, pp. 535–547, 2000. →pages 44[51] M. Y. Chung, M.-H. Jung, T.-J. Lee, and Y. Lee, “Performance analysis of HomePlug 1.0MAC with CSMA/CA,” IEEE Journal on Selected Areas in Communications, vol. 24, no. 7,pp. 1411–1420, 2006. → pages 4458AppendixOptimal Detection ThresholdIn this appendix, we provide a closed-form expression for the optimum detection threshold, bopt,by solving (4.10). We start with (4.9), and divide both sides by P(E2|E0) to getPτ =PeP(E2|E0) = PDE+ τPFA, (5.1)where τ = P(E¯2|E0)P(E2|E0) . Since P(E2|E0) is a constant for a given network operation, minimizing (5.1)also solves (4.10). Therefore, bopt is the b0 that solves∂Pτ∂b0= 0. (5.2)We first individually find the partial derivatives of both the components of Pτ using (4.3) and (4.4)to get∂PFA∂b0=−b0 exp(−b202)(5.3)∂PDE∂b0=∂∂b01− ∞∫b0xexp(−x2+2γ2)I0(x√2γ)dx= exp(−2γ+b202)I0(b0√2γ)b0, (5.4)59where In(·) is the nth-order modified Bessel function of the first kind. By using (5.1)-(5.4), we getexp(−γ)I0(bopt√2γ)= τ. (5.5)Next, we find a closed-form approximation for the transcendental equation of (5.5). To this end,we use the proven result that b=√2+ γ2 provides an excellent analytic approximation to thesolution of exp(−γ)I0 (b√2γ) = 1 [40, Eqns. 7-4-13, 7-4-14, Fig. 7-4-3]. Thus, we haveexp(γ)≈ I0(√γ(γ+4))=⇒ I0(bopt√2γ)≈ exp(√2b2optγ+4−2). (5.6)Using (5.6) in (5.5), and simplifying with some simple manipulations gives us the closed-formexpression for bopt asbopt ≈√(γ+ lnτ)2+4(γ+ lnτ)2γ. (5.7)60


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