End-to-End Simulation of 5G mmWave Networks

Please download to get full document.

View again

of 25
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Similar Documents
Information Report

Screenplays & Play


Views: 4 | Pages: 25

Extension: PDF | Download: 0

End-to-End Simulation of 5G mmWave Networks
  1 End-to-End Simulation of 5G mmWave Networks Marco Mezzavilla,  Member, IEEE  , Menglei Zhang, Michele Polese,  Student Member, IEEE  , Russell Ford,Sourjya Dutta,  Student Member, IEEE  , Sundeep Rangan,  Fellow, IEEE  , Michele Zorzi,  Fellow, IEEE   Abstract —Due to its potential for multi-gigabit and low latencywireless links, millimeter wave (mmWave) technology is expectedto play a central role in 5th generation (5G) cellular systems.While there has been considerable progress in understanding themmWave physical layer, innovations will be required at all layersof the protocol stack, in both the access and the core network.Discrete-event network simulation is essential for end-to-end,cross-layer research and development. This paper provides atutorial on a recently developed full-stack mmWave moduleintegrated into the widely used open-source ns–3 simulator. Themodule includes a number of detailed statistical channel modelsas well as the ability to incorporate real measurements or ray-tracing data. The Physical (PHY) and Medium Access Control(MAC) layers are modular and highly customizable, making iteasy to integrate algorithms or compare Orthogonal FrequencyDivision Multiplexing (OFDM) numerologies, for example. Themodule is interfaced with the core network of the ns–3 Long TermEvolution (LTE) module for full-stack simulations of end-to-endconnectivity, and advanced architectural features, such as dual-connectivity, are also available. To facilitate the understanding of the module, and verify its correct functioning, we provide severalexamples that show the performance of the custom mmWavestack as well as custom congestion control algorithms designedspecifically for efficient utilization of the mmWave channel. This work has been submitted to IEEE Communication Surveys and Tutorials for possible publication.Copyright may be transferred without notice, after which this version may no longer be accessible.  Index Terms —mmWave, 5G, Cellular, Channel, Propagation,PHY, MAC, multi-connectivity, handover, simulation, ns–3 I. I NTRODUCTION Millimeter Wave (mmWave) communications are emergingas a central technology in 5G cellular wireless systems due totheir potential to achieve the massive throughputs required byfuture networks [1]–[5]. In particular, mmWave has becomea key focus of the 3rd Generation Partnership Project (3GPP)New Radio (NR) effort currently under development [6].Due to the unique propagation characteristics of mmWavesignals and the need to transmit in beams with much greaterdirectionality than previously used in cellular systems, muchof the recent work in mmWave communications has focusedon channel modeling, beamforming and other physical layerprocedures. However, the design of End-to-End (E2E) cellularsystems that can fully exploit the high-throughput, low-latencycapabilities of mmWave links will require innovations notonly at the physical layer, but also across all layers of thecommunication protocol stack. For mmWave systems, E2Edesign and analysis are at a much earlier stage of research[7]–[9]. Marco Mezzavilla, Menglei Zhang, Russell Ford, Sourjya Dutta and Sun-deep Rangan are with NYU WIRELESS, NYU Tandon School of Engineer-ing, Brooklyn, NY, USA (email:  { mezzavilla, russell.ford, menglei, sdutta,srangan } @nyu.edu).Michele Polese and Michele Zorzi are with the Department of Informa-tion Engineering, University of Padova, Padova, Italy (email:  { polesemi,zorzi } @dei.unipd.it) Discrete-event network simulators are fundamental andwidely used tools for developing new protocols and analyz-ing complex networks. Importantly, most network simulatorsenable  full-stack simulation , meaning that they model alllayers of the protocol stack as well as applications runningover the network. This full-stack capability will play a crit-ical role in the development of 5G mmWave systems. Theunique characteristics of the underlying mmWave channelhave wide ranging effects throughout the protocol stack. Forexample, the use of highly directional beams increases thecomplexity of a number of basic MAC-layer procedures suchas synchronization, control signaling, cell search and initialaccess, which in turn affect delay and robustness [7]. MmWavesignals are also highly susceptible to blockage [1], [10]–[12], which results in high variability of the channel quality.This erratic behavior complicates the design of rate adapta-tion algorithms and signaling procedures, requiring advancedsolutions for multi-connectivity, fast handover and connectionre-establishment [13]–[16]. New transport layer mechanismsmay also be required in order to utilize the large capacity,when available, and to react promptly to rapid fading to avoidcongestion [9], [17]–[19]. The need for ultra-low latency ap-plications [1], [20], [21] may require solutions based on edgecomputing and distributed architectures that will determinea considerable departure from current cellular core network designs.To better capture these design challenges, this work presentsa comprehensive tutorial on the open-source mmWave simula-tion tool developed by New York University and the Universityof Padova for LTE-like 5G mmWave cellular networks, whichcan be used to evaluate cross-layer and end-to-end perfor-mance. This mmWave simulation tool is developed as a newmodule within the widely used ns–3 network simulator [22].ns–3 is an open-source platform, that currently implements awide range of protocols in C++, making it useful for cross-layer design and analysis. The new mmWave module presentedhere is based on the architecture and design patterns of the LTELENA module [23], [24] and implements all the necessaryService Access Points (SAPs) needed to leverage the robustsuite of LTE/Evolved Packet Core (EPC) protocols providedby LENA. The code (publicly available at GitHub [25], alongwith examples and test configurations) is highly modular andcustomizable to facilitate researchers to design and test novel5G protocols.The ns–3 mmWave module was first presented in [26],[27]. The 3GPP channel model implementation is introducedin [28], and the dual connectivity functionality is describedin [14], [29]. This paper extends these works by presenting thens–3 mmWave module from a single and organic point of view,and is intended as a tutorial for any researcher that plans touse the simulator. In addition to its comprehensive description   a  r   X   i  v  :   1   7   0   5 .   0   2   8   8   2  v   3   [  c  s .   N   I   ]   5   F  e   b   2   0   1   8  2 and discussion, we provide in Sec. X a brief guide on how toset up a simulation, followed by a number of representativeexamples.The rest of the paper is organized as follows. In Section II,we provide some background on mmWave cellular communi-cations and highlight some key problems at the higher protocollayers to motivate the need for a robust full-stack simulator.We also describe the main challenges related to the design of ammWave cellular networks simulator. Then, in Section III, weintroduce ns-3, the network simulator on which our mmWavemodule is developed, and in Section IV we present the overallarchitecture of the mmWave module. We then take a closerlook at each component, starting with the suite of MultipleInput, Multiple Output (MIMO) channel models in Section V.In addition to an implementation of the latest 3GPP “above6 GHz” model [30], several custom channel models are alsoprovided. Section VI discusses the features of the OFDM-based PHY layer, which has a customizable frame structurefor evaluating different numerologies and parameters. In Sec-tion VII, we provide a MAC-layer discussion that includesour proposed flexible/variable Transmission Time Interval(TTI) Time Division Multiple Access (TDMA) MAC scheme,which is supported by several scheduler implementations.Section VIII presents the enhancements that we introducedto the LTE Radio Link Control (RLC) layer. The dual-connectivity architecture is reported in Sec. IX. In SectionX, we show how the module can be used for cross-layerevaluation of multi-user cellular networks through a numberof representative examples, and provide pointers to a largeset of general results that have been obtained so far with thismodule. The integration of native Linux Transmission ControlProtocol (TCP) implementations, performed through the ns–3 Direct Code Execution (DCE) framework, is discussed inSection XI. In Section XII, we provide details on our futureplans for the simulator and suggest possible research topicsin which it could be used. Finally, we conclude this tutorialpaper in Section XIII.II. M ILLIMETER  W AVE  C ELLULAR  B ACKGROUND Millimeter wave communications is an advanced PHY layertechnology, which has recently come to the forefront of research interest and may be able to rise to the challenge of providing high-rate mobile broadband services, in addition tooffering opportunities for reducing over-the-air latency for 5GNew Radio.MmWave makes use of the radio frequency spectrum roughlybetween 30 and 300 GHz, even though the research challengesextend also to lower frequencies (i.e., above 6 GHz) whichare considered for 3GPP NR. Systems that can operate inthese bands are attractive because of the large quantities of available spectrum at these higher frequency ranges and thespatial degrees of freedom afforded by very high-dimensionalantenna arrays, which are possible thanks to the smallersize of antenna elements at higher frequencies. Most currentcommercial wireless systems operate below 6 GHz, wherelower frequencies allow for long-range propagation and lowpenetration loss (i.e., attenuation by walls and other obstacles),which makes them well-suited for radio communications. As aresult, the sub-6 GHz spectrum has become heavily congestedand individual bands are generally not available in contiguouschunks wider than 200 MHz. However, large swaths of spec-trum are available at the higher mmWave frequencies, whichoffer the possibility of very wide bandwidths of over 1 GHz,in some cases.Although the mmWave bands are already used by a vari-ety of commercial applications, such as satellite and point-to-point backhaul communications, until recently they wereconsidered impractical for mobile access networks due to thepoor isotropic propagation and the vulnerability to shadowingat these higher frequencies. However, it has now been shownthat the limitations of the mmWave channel can be overcomewith the help of high-gain, directional antennas so that thisvast region of spectrum can now be exploited to provide anorder of magnitude or more increase in throughput for mobiledevices [3], [31].Directional smart antennas are the major technology enablerthat will make it possible for mmWave devices to overcomethe poor propagation effects and unlock this high-frequencyspectrum. The theoretical free space path loss (as governed byFriis’ Equation) is proportional to the square of the frequency,resulting in the magnitude of received power for a mmWavesignal being over 30 dB (1000x) less than conventionalcellular systems at equivalent distances between transmitterand receiver [32]. Multi-element antenna arrays and MIMObeamforming techniques offer a means of compensating forthis high attenuation. With millimeter waves, the antenna sizeand spacing shrinks to be on the order of millimeters, makingit possible to pack hundreds of elements onto a small cell basestation and dozens onto a handheld device. Smaller antennasize also allows for multiple arrays to be integrated ontomobile devices to provide diversity and maintain connectivityeven if the signal from one array is blocked (for instance, bythe user’s hand) [3].It is clear that mmWave will be highly disruptive in the wire-less space thanks to the prospect of massive bandwidth andhigh-gain antennas. Nevertheless, before mmWave technologycan be effectively realized in 5G networks, there are numerouschallenges to be addressed, not only at the physical layer, butalso at higher layers of the radio stack, namely: •  Adaptive beamforming and beam tracking:  The require-ment of directionality introduces new challenges for sup-porting mobility in mmWave networks. The transmitterand receiver must continually track the channel as themobile user moves in order to align their antenna ar-rays to achieve the maximum directional gain. MmWavesignals are also known to be particularly susceptibleto shadowing and can be completely blocked by manymaterials such as brick, tinted glass and even the humanbody [33], [34]. Fortunately, recent field measurementshave demonstrated that reflected power can be sufficientfor Non Line of Sight (NLOS) communications to bepossible. A blocked link may therefore be able to recoverby steering the beam from the primary Line of Sight(LOS) path to an alternate NLOS path. The UE and BS  3 must then jointly initiate a procedure to search for andselect another path to reestablish the link. •  Directional synchronization and broadcast channels:  Di-rectionality also complicates the design of many controlchannels and procedures. The cell discovery and initialaccess procedures, where the UE must search for nearbybase stations to which it can attach, will require aninnovative approach to be handled efficiently. Traditionalcells periodically broadcast out synchronization signals(known as the Primary Synchronization Signal (PSS) inLTE systems) omnidirectionally, which are received byall devices within the cell’s coverage range and used toinitially connect to the cell. If a 5G mmWave evolvedNode Base (eNB) were to broadcast the PSS in anomnidirectional antenna pattern, the signal would notbenefit from the directional gain and might not haveadequate range to be detected by many UEs. Therefore,the eNB and User Equipment (UE) must perform anangular search in order for users to detect the PSS andhone in on the optimal Transmitter (TX)/Receiver (RX)beamforming angles [35]. A similar problem also arisesfor other control signals, such as the Downlink ControlInformation (DCI) assignments, which indicate the re-sources assigned to each user for Downlink (DL)/Uplink (UL) transmission in the subframe or slot. •  Issues for the MAC, Network and Transport Layers:  Therapid channel dynamics and vulnerability of mmWavelinks to shadowing will require frequent, near instanta-neous handovers between neighboring 5G or 4G cells.  Dual-connectivity , where mobiles are continuously con-nected to both the 5G and legacy 4G network, may there-fore be essential to recover from an abrupt failure of theprimary 5G link [14], [29]. Additionally, at the transportlayer, the congestion control and avoidance mechanismsprovided by TCP must be able to quickly adapt to suddenfluctuations in capacity to maximally utilize the link bandwidth while avoiding overwhelming the network bysending too many packets, resulting in congestion andaffecting other flows in the network. Current versionsof TCP may not be optimized for mmWave channeldynamics [9], [18], so new algorithms may be called forto provide high rates for E2E sessions [17], [36], [37]. Potentials and Challenges of System-level Simulations of mmWave Networks An End-to-End network simulator for mmWave cellularnetworks is an invaluable tool that can help address thesechallenges by allowing the evaluation of the impact of thechannel and of the PHY layer technology on the wholeprotocol stack. However, given the characteristics of mmWavecommunications described in the previous paragraphs, in orderto have accurate results it is of paramount importance to modelin detail the behavior of the different elements that interact in acellular system. In the following paragraphs we will introduceand discuss some of the most important elements that needto be considered when designing a mmWave cellular systemssimulation, and show how they depend on one another: •  The channel model is the fundamental component of every wireless simulation. Given the harsh propagationconditions at mmWaves, the channel is one the mainelements that affect the end-to-end network performance.Firstly, it has to account for the different LOS andNLOS states for the propagation loss and the fading [30].Moreover, beamforming should be applied on top of thechannel to accurately model directional transmissions,which have an impact on the link budget, the interferenceand the control procedures. Finally, the Doppler effectis particularly relevant at mmWave frequencies, espe-cially with high mobility [3]. An important considerationrelated to the channel model is the trade off betweenthe accuracy and the computational complexity: veryaccurate models that require the computation of thecomplete channel matrix are usually also computationallyintensive [28]. •  The users’ mobility and the network deployment havean important impact on the communication performance,intertwined with that of the channel model. Given thesmall range of the mmWave cells, the deployment willbe dense and will require frequent access point updates,which should be simulated for a realistic performanceassessment [38]. Moreover, mobility affects the perfor-mance of beam tracking algorithms [15]. Therefore, whensimulating a mmWave network it is important to userealistic deployments and mobility models. •  The level of detail when modeling the protocol stack of the mmWave links and of the end devices is another im-portant parameter for network simulations. A simplifiedmodel of the protocol stack can be accurate enough forstudies that do not involve complex interplay betweendifferent layers, but cannot capture the behaviors thatemerge from the interaction of the different layer, andtherefore could not generate realistic results for end-to-end performance evaluations. For example, at mmWavefrequencies, it has been shown that the channel behaviorhas an impact on the TCP performance [9], [18], thereforea model of the TCP/IP stack is needed when analyzingthe data rate that an application can reach in an end-to-end mmWave network.To the best of our knowledge, there are no open sourcesimulators capable of thoroughly modeling the mmWavechannel along with the cellular network protocol stack aswell as other protocols (e.g., the TCP/IP stack), realisticscenarios and mobility. There exists an ns–3-based simulatorfor IEEE 802.11ad in the 60 GHz band [39], [40], whichhowever cannot be used to simulate cellular and 3GPP-likescenarios. Other papers [41]–[44] report results from systemlevel simulations, with custom (often MATLAB-based and notpublicly available) simulators which are not able to capturethe complexity of the whole stack with a very high level of detail. This is what motivated us to develop an open sourcecellular mmWave module for the ns–3 simulator, which wewill describe in the following sections.  4 Figure 1: Class diagram of the end-to-end mmWave module. III.  NS –3The ns–3 discrete-event network simulator [22], [45] is avery powerful tool available to communication and networkingresearchers for developing new protocols and analyzing com-plex systems. It is the successor to ns–2, a tried and tested toolthat has been in use by the networking community for over adecade in the design and validation of network protocols. ns–3is open source, and can be downloaded from the website of theproject 1 . An active community of researchers from both indus-try and academia has enriched the basic core of the simulatorwith several modules, and ns–3 can be now used to simulatea wide variety of wireless and wired networks, protocols andalgorithms. There is a complete documentation 2 on the modelsin the ns–3 website, in terms of both the design of the modelsand what a user can do with the models. Moreover, a completetutorial on how to install ns–3, set up ns–3 scenarios andtopologies, handle the collection of statistics and log usefulmessages is provided in the documentation 3 . The tutorial is agood starting point for a researcher who approaches ns–3 forthe first time.The ns–3 simulator is organized into multiple folders. The src  folder provides a collection of C++ classes, whichimplement a wide range of modular simulation models andnetwork protocols. The different modules can be aggregatedand instantiated to build diverse simulated network scenarios,making ns-3 especially useful for cross-layer design and anal-ysis. The modularity and use of object-oriented design patternsalso allows for new algorithms to easily be incorporated intothe network stack and experimented with. Each module is 1 http://www.nsnam.org 2 https://www.nsnam.org/documentation/  3 https://www.nsnam.org/docs/tutorial/html/  itself organized into multiple subfolders, which contain thedocumentation and the source code of the model itself, thehelpers, the examples and the tests. The helpers associatedwith each model have a very important role. They are classeswhich hide to the final user the complexity involved insetting up a complete scenario, for example by automaticallyassigning IP addresses, or connecting the different classes of a protocol stack. The  build  folder contains the binaries of the simulator. Finally, the  scratch  folder is a special folderin which scripts with examples and scenarios can be built on-the-fly.Besides the core module, which provides the basic structureof the simulator, there are modules for networking protocols(e.g., the TCP/IP stack protocols [46]), wireless protocols(LTE [23], Wi-Fi [47], WiMAX [48]), routing algorithms [49],mobility, embedding obstacles and buildings in the simulationscenarios, and data collection. All the modules are listed inthe model library 4 .In the following sections, we will describe in detail themmWave module for ns–3, following the same approach whichis used for the other ns–3 modules. We will first describe themodel in terms of implementation of the different componentsof a mmWave cellular network and protocol stack, and thenthe examples and scenarios that can be simulated with it andhow they can be set up.IV.  MM W AVE  M ODULE  O VERVIEW The ns–3 mmWave module is designed to perform end-to-endsimulations of 3GPP-style cellular networks. The architecture 4 https://www.nsnam.org/docs/release/3.27/models/html/index.html for ns–3version 3.27 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  !0)+6*/ !A)2 015)2A!A)2 01-$A&!8+75#)C0A2%&$B6&':%#$N.2<= @+A&!8+?.$.A@.%&$B6&':%#$N.2<= @+A& !0)A3*?%!0)A3*"%$%&'()#7>D)0E)(4*) !A)2 01$.--(&!A)2 01.#5*$)!8+6.)&:%#$N.2<= ,++- $%&'()?)D)0E)(4*) !A)2 01A0CA!A)2 01.#5*$)!8+6.)&:%#$N.2<= ,++- !0)A3*.%!0)A3* !A)2 015)2A!A)2 01-$A&!8+75#)C0A2%&$B%&':%#$N.2<= @+A&!8+?.$.A@.%&':%#$N.2<= @+A&!8+Q+2A*D73TBB+52')A2D:;;;<= @+A& !0)?)A1* !6D)$I5+)AU.7+625+)H.C2>.--:<= @+A&!6.)&".#C'5.0.)2?.B+52:;;;<= @+A& !0)#7>A1* !>+)*AH'5.>.--:;;;<= @+A& Figure 2: Unified Modeling Language (UML) class diagram for the end-to-end mmWave module.
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!