IEEE Communications Magazine • March 2015108 0163-6804/15/$25.00 © 2015 IEEE

Belal Hamzeh is withCableLabs.

Mehmet Toy is with Com-cast USA.

Yunhui Fu is with CentralSouth University atChangsha.

James Martin is withClemson University.

1 http://www.leichtman-research.com/press/081514release.html.

INTRODUCTION

In the United States, cable companies serve about50 million broadband access subscribers, 40 per-cent more than DSL and fiber subscribers.1 Amajor reason for this trend is due to the contin-ued evolution of cable technology. Current cablenetworks are based on the Data-Over-Cable Sys-tem Interface Specification (DOCSIS) version 3.0(DOCSIS 3.0) [1]. Cable Television Laboratories,Inc. (CableLabs), the research and developmentorganization for multiple system operators (werefer to this simply as network operators), hasrecently released an update to DOCSIS, referredto as the DOCSIS 3.1 specifications [2, 3].

Figure 1 illustrates a modern broadbandcable network that applies to both DOCSIS 3.0and 3.1 systems. Hybrid fiber coaxial (HFC) sys-tems are designed to push content over fiberwith the fiber terminating as close to end usersas possible. A cable modem termination system(CMTS) interacts with cable modems (CMs)over the coaxial cable. The CMTS resides in thehead-end that houses network operator facilitieswhile the CM resides at the subscriber’s homeresidence or business. A subscriber is likely tohave multiple devices that consume traditionalbroadcast video such as televisions and digitalvideo recorders and that use broadband dataservice such as IP delivered video content andInternet applications. The DOCSIS standardspecifies the media access control (MAC) andthe physical (PHY) layer procedures.

The network operator’s Operational Support

System (OSS) contains, among other capabilities,various necessary services including DynamicHost Configuration Protocol (DHCP), a configu-ration file and software download server for CMs,a Network Time Protocol server, and a certificaterevocation server. The combined functionality ofthese servers allow for the proper connection andoperation of the CMs with the DOCSIS network.

DOCSIS 3.1 was developed to provide sub-scribers with service speeds up to 10 Gb/s down-stream and up to 1 Gb/s upstream. The mostsignificant area of change provided by DOCSIS3.1 involves spectrum access at the physicallayer. Motivating application directions forDOCSIS 3.1 might include:• Higher video quality such as 4K• Higher levels of on-demand video• Network operator provided digital video

recorder services;• Backhaul network service to support cellu-

lar operator WiFi offloading;• Emerging applications enabled by advances

in virtual environments, cloud computing,and machine-to-machine communications. The central focus of this tutorial article is to

explore how future cable systems based on DOC-SIS 3.1 specifications will scale to support gigabit-per-second network speeds. We introduce thebuilding blocks of modern cable systems providedby the current DOCSIS 3.0 specifications. Wehighlight DOCSIS 3.1 extensions and presentopen issues and challenges. An important objec-tive of this article is to identify research issuesrelated to emerging DOCSIS-based cable systems.

This tutorial article is organized as follows.The next section provides a brief overview of theDOCSIS 3.0. Next, we DOCSIS 3.1 enhance-ments, paying particular attention to modulationand coding profile management and to activequeue management (AQM). We highlight specificchallenges and identify several research questions.We conclude the article with a brief summary.

DOCSIS BACKGROUNDThe DOCSIS 3.1 specifications define the fifth gen-eration of high-speed data-over-cable systems and ispart of the DOCSIS family of specifications devel-oped by CableLabs. It builds upon the previous gen-erations of DOCSIS specifications commonlyreferred to as the DOCSIS 3.0 and earlier specifica-

ABSTRACT

The cable industry has recently released theDOCSIS® 3.1 specifications which are the fifthgeneration of the DOCSIS family of specifica-tions. In this article, we introduce the buildingblocks of DOCSIS 3.1 and highlight the specificcapabilities that will help broadband cable sys-tems scale to support gigabit-per- second net-work speeds. While this article is a tutorial onDOCSIS 3.1, an important objective is to engageand motivate the academic community to partici-pate in research related to emerging DOCSIS-based cable systems. We identify a set of openissues that represent challenges and opportuni-ties for academic researchers to explore.

NEXT GENERATION CABLE NETWORKS WITHDOCSIS® 3.1 TECHNOLOGY

Belal Hamzeh, Mehmet Toy, Yunhui Fu, and James Martin

DOCSIS 3.1: Scaling Broadband Cable toGigabit Speeds

MARTIN_LAYOUT_Author Layout 3/4/15 4:11 PM Page 108

IEEE Communications Magazine • March 2015 109

tions. In this section, we provide an overview of theDOCSIS 3.0 and earlier specifications.

The coaxial medium provides in excess of 3GHz of usable bandwidth (in ideal signal-to-noiseconditions). Frequency division multiplexing isused to isolate bandwidth for downstream (CMTSto CM) and upstream (CM to CMTS) communi-cations. In current DOCSIS 3.0 systems, down-stream communications assumes 6 MHz (8 MHzin Europe) channels which are located in the fre-quency range between 108 MHz and 870 MHz(and optionally 1002 MHz). Most of the spectrumis allocated for traditional broadcast video. Somedownstream channels are allocated to supportInternet access. Downstream channels can achieve,assuming 256 quadrature amplitude modulation(QAM), data rates of 42 Mb/s per 6 MHz chan-nel. The CMTS uses asynchronous time divisionmultiplexing to share downstream bandwidthbetween multiple end-stations. In DOCSIS 3.0 sys-tems, packets sent over the downstream channelare segmented based on a 188 byte MPEG frameformat that includes 4 bytes of header and a 184byte payload. The use of MPEG frames facilitatesmultiplexing the coaxial medium for use with bothdata and broadcast video. The CM rebuilds IPpackets that arrive over the downstream channel.

Upstream channels are 3.2 or 6.4 MHz wideand are located within the frequency rangebetween 5 MHz and 42 MHz (65 MHz in Europe,and optionally 85 MHz in North America).Upstream communications, assuming 64 QAMand a 6.4 MHz channel, can achieve data rates of30 Mb/s. Additional upstream physical layeroperating modes were added in the DOCSIS 2.0standard. Asynchronous Time Division MultipleAccess (ATDMA) and Synchronous Code Divi-sion Multiple Access (S-CDMA) to provide com-munications options that can improve robustnessand increase data rates over impaired channels.

In a DOCSIS network, downstream andupstream traffic is classified and managed by ser-vice flows. A service flow provides a unidirectionaltransport of packets over the cable network. Traf-fic is shaped, policed, and managed based on ser-vice quality parameters that were defined for theservice flow. A service flow typically includes multi-ple IP flows. By default, a subscriber is provisionedwith two service flows: a downstream service flowthat represents all downstream user traffic and anupstream service flow that represents all upstreamuser traffic. A further example involves a sub-scriber configuration that supports toll quality IPtelephone service and best effort data. This couldbe provisioned with four service flows: one eachfor the upstream and downstream VoIP signalingand one each for upstream and downstream aggre-gate consisting of all the best effort data traffic.

To support higher data rates, DOCSIS 3.0specifications introduce channel bonding whichrequires the CMTS to schedule downstream traf-fic from services flows over potentially morethan one channel. A bonding group is an abstrac-tion that represents the group of channels thatare available to a service flow. CMs, equippedwith multiple tuners and transmitters, areassigned a set of downstream and upstreamchannels. A common configuration involves 8bonded downstream channels and 4 bondedupstream channels which can support data rates

up to 320 Mbit/s downstream and 120 Mbit/supstream. The downstream scheduler, whichoperates at the CMTS, can allocate bandwidthto service flows from any channel that is in thebonding group assigned to the service flow.

Support for upstream bonding groups is morechallenging than downstream due to the request-grant mechanism. Prior to DOCSIS 3.0, a CMwas limited to a single outstanding request forbandwidth. This stop-and-wait behavior does notscale to higher speeds. Therefore, DOCSIS 3.0defined a new request-grant mechanism referredto as continuous concatenation and fragmenta-tion (CCF). CCF allows a CM to send multiplerequests for bandwidth concurrently. The CMTSstripes data grants over the set of upstream chan-nels in the service flow’s bonding group. A pro-cess referred to as segmentation allows the CMto stream data over the bonding group. The CMorganizes user data (i.e., packets) into segmentswhich include a sequence number allowing theCMTS to reorder the segments if necessary.

Since every CM tuned to an upstream channelis a potential sender, controlling access to theupstream channels is considerably more complexthan downstream operation. The upstream channelis time division multiplexed with transmission slotsreferred to as minislots. Permission to transmit datain a block of one or more minislots must be grant-ed to a CM by the upstream schedule which oper-ates at the CMTS. The CMTS grants mini-slotownership by periodically transmitting UpstreamBandwidth Allocation Map messages, which werefer to as MAP messages, on the downstreamchannel. In addition to ownership grants, the MAPalso typically identifies some minislots as contentionslots in which CMs may bid for quantities of futureminislots. To minimize collisions in the contentionslots, a non-greedy back-off procedure is employed.When a CM has a backlog of upstream packets itmay also “piggyback” a request for minislots for thenext packet in the current packet.

QUALITY OF SERVICE MECHANISMSMany applications have distinctive network servicerequirements which in turn requires broadbandnetworks to provide quality of service (QoS) mech-anisms. VoIP, online gaming, and video streamingare just a few examples of such applications. Addi-tionally, the migration to an all IP video broadcast

Figure 1. Modern broadband cable network.

WiFi devices

Subscriber’s networkCoax splitter

Video-on-demandTraditional broadcastvideo

Network operator: servicesand applications

Switched digitalvideo

DVR 1DVR 2

TV1

Cablemodem

WiFiAP

OSS

Networkoperator’sback officenetwork

Fiber

Cable modemterminationsystem (CMTS)

Internet audio/video Web contentCloud servicesSocial networking

Internet:Services and applications

Fiber Coaxial cableOptical-to-coaxnode

The hybrid fiber coaxial(HFC) network

TV2

MARTIN_LAYOUT_Author Layout 3/4/15 4:11 PM Page 109

IEEE Communications Magazine • March 2015110

network makes it critical for the network to pro-vide service guarantees. Each service flow isassigned a scheduling service which is associatedwith a set of service attributes. For downstreamservice flows, attributes include traffic priority, traf-fic shaping parameters (maximum sustained trafficrate and maximum traffic burst), and minimumtraffic rate. Because upstream data rates are typi-cally lower than downstream, a set of specific net-work services have been defined to supportapplication service requirements. As an example,the Unsolicited Grant Service (UGS) is intendedto support upstream latency sensitive applicationsthat generate fixed size data packets on a periodicbasis, such as Voice over IP. UGS offers fixed sizegrants (in bytes) on a real-time periodic basis. Thedefault upstream service flow provisioned for eachsubscriber is treated as a best effort service. Thedetails of the upstream and downstream schedulersare not specified in the DOCSIS specifications andconsequently are left for vendor differentiation.Please refer to [1] for further information onDOCSIS services and configuration parameters.

DOCSIS 3.1 SYSTEMENHANCEMENTS

In this section, we provide an overview of themain enhancements provided by the DOCSIS3.1 specifications. We focus in particular on pro-file management and AQM.

DOCSIS 3.1 PHYSICAL LAYER

The DOCSIS 3.1 specifications adopt the use ofOrthogonal Frequency Division Multiplexing(OFDM) in the downstream channel and Orthog-onal Frequency Division Multiple Access(OFDMA) in the upstream channel. The DOC-SIS 3.1 specifications supports 25 kHz and 50 kHzsubcarrier spacing options on both channels; thedownstream supports channel bandwidths from24 MHz up to 192 MHz while the upstream chan-nels’ bandwidths range from 6.4 MHz to 96 MHz.

The DOCSIS 3.1 specifications extend therange of frequencies that can be used over thecoaxial medium. Prior to DOCSIS 3.1 specifica-tions, upstream channels operate in specificband ranges such as 5–42 MHz, 5–65 MHz or5–85 MHz. The DOCSIS 3.1 specifications intro-duce two additional bands at 5–117 MHz and5–204 MHz. Although the use of specific bandranges is typical, for operational reasons a net-work operator might support other upstreamband ranges such as 5-80 MHz by taking advan-tage of the highly granular channel bandwidthsenabled by OFDM signaling. On the down-stream, the lower band edge can be as low as108 MHz (assuming the upstream upper bandedge is less than 108 MHz), while the upperband edge can extend up to 1794 MHz.

To maximize the overall spectral efficiency ofthe system (i.e., to operate as close to the Shannonlimit as conditions allow) the DOCSIS 3.1 specifi-cations introduce higher order modulationschemes, with mandatory support up to 4096-QAM on the downstream and 1024-QAM on theupstream. The DOCSIS specifications 3.1 alsoallow for optional support of 8192-QAM and16384-QAM on the downstream and 2048-QAMand 4096-QAM on the upstream; this allows thenetwork operator to maximize network capacityand take advantage of signal quality improvements.

Table 1 and Table 2 summarize the down-stream and upstream channel parameters. Thetables indicates that 4096 or 8192 point FastFourier Transform (FFT) is used for down-stream and that 2048 or 4096 point FFT is usedfor upstream. To further enhance spectral effi-ciency, forward error correction based on lowdensity parity check (LDPC) coding is appliedalong with interleaving of code words in timeand in frequency prior to transmission in bothdownstream and upstream.

Proactive Network Maintenance — The DOC-SIS 3.1 specifications provide enhanced diagnostictools and statistics monitoring and reporting forproactive network maintenance (PNM). Thestatistics and measurements are collected at theCMs and the CMTSs distributed within the net-work. The goal for PNM is to rapidly and accu-rately characterize, maintain and troubleshoot theupstream and downstream cable plant. Informa-tion such as mean error rates per subcarrier, noisepower, and forward error correction statistics ana-lyzed over large timescales can be used by the net-work operator to proactively identify potentialperformance issues and respond accordingly.

The measurements for PNM are in-servicemeasurements, which mean they are transparentto the end-user without disruption of service.

Table 1. Downstream channel parameters.

Parameter Value

Lower Band Edge Mandatory: 254 MHzOptional: 108 MHz

Upper Band Edge Mandatory: 1218 MHzOptional: 1794 MHz

Signal Type OFDM

Subcarrier Spacing 50 kHz, 25 kHz

FFT Time Duration 20 ms (50 kHz subcarriers)40 ms (25 kHz subcarriers)

FFT Size

50 kHz: 4096 FFT; 3800 Maximum activesubcarriers25 kHz: 8192 FFT; 7600 Maximum activesubcarriers

Minimum OFDM ChannelBandwidth 24 MHz

Maximum OFDM ChannelBandwidth 192 MHz

Number of Independentlyconfigurable OFDM channels Minimum of 2

Modulation Type

Mandatory: BPSK, QPSK, 8-QAM, 16-QAM,32-QAM, 64-QAM, 128-QAM, 256-QAM,512-QAM, 1024-QAM, 2048-QAM, 4096-QAM Optional: 8192-QAM, 16384-QAM

MARTIN_LAYOUT_Author Layout 3/4/15 4:11 PM Page 110

IEEE Communications Magazine • March 2015 111

The PNM functionality provides the networkoperator with a centralized view of the completenetwork health and can help predict networkfaults before they cause a service disruption. Thecollected measurements and statistics enable thenetwork operator to have embedded functionali-ties including network analyzers, spectrum ana-lyzers, vector signal analyzers in addition tovarious functionalities provided by the collectedmetrics. A detailed description of the collectedmeasurements and statistics is provided in [3].

DOCSIS 3.1 MAC LAYERIn this section we summarize the main MAClayer functional extensions introduced withDOCSIS 3.1 specifications.

Profile Management — By usingOFDM/OFDMA, DOCSIS 3.1 systems can varythe modulation details over all subcarriers, effec-tively optimizing the transmission to account fordifferent channel conditions observed by differentCMs. In its simplest form, a profile defines a setof subcarriers and associated modulation ordersthat are assigned to a CM. On the downstream, aCM is required to support 4 profiles for live traf-fic and 1 test profile for each OFDM channel.For upstream, a CM is must support at least 2profiles for each OFDMA channel. Profiles areassigned on a per CM basis, thus CMs will beable to operate using the highest modulationorder possible according to the signal-to-noiseratio (SNR) of the channel (or profile-assignedsubcarriers). Profiles are channel dependent butindependent of service groups. For example, if aCM is operating with two downstream channels,the profiles for each channel can be set indepen-dently. This can potentially increase system effi-ciency. This is in contrast to previous versions ofthe DOCSIS specifications where CMs wererequired to use the same profile (or modulationorder) within the same service group.

Figure 2 illustrates the use of five downstreamchannels along with four profiles (referred to asProfiles A, B, C, and D). The downstream sched-uler must be “channel aware” in that it deter-mines which profile should be used for eachtransmission. Channel quality information is peri-odically sent by the CMs to the CMTS. TheCMTS selects the best burst profile for a CMbased on current channel conditions and on theset of profiles available to the CM. Over longertime scales, the CMTS can update the set of pro-files available at a CM. The specification indicatesthat Profile A will offer a robust configurationthrough the use of relatively low bit loading toensure all CMs can receive it. This profile will beused by the CMs for initialization and to carrybroadcast MAC management messages. A testprofile will be used to check whether a CM canreceive a certain profile before the profile isassigned to the CM and live traffic sent on it. Themethodology to select the profile configuration,which profiles to assign to a CM, and when toswitch between profiles is not specified in theDOCSIS 3.1 specification. As with scheduling,this functionality is meant to be vendor specific.

For upstream, the CMTS specifies the specif-ic channel as well as the specific burst profilethat a CM should use for a particular grant.

CMs use a profile denoted as IUC 13 for initial-ization and then can be assigned to a differentprofile with higher bit loading for live traffic. Asecond upstream profile may be assigned to theCM for test purposes or for live traffic.

Downstream multicast supports the use ofmultiple profiles. The CMTS should attempt toconserve bandwidth by selecting the highestbandwidth profile common to the CMs that aremembers of the multicast group. If this is notpossible, or if a CM requests to join a sessionbut does not support the current session profile,the CMTS can either move all CMs in the ses-sion to a lower bandwidth profile or it can repli-cate the multicast on multiple profiles.

Active Queue Management (AQM) —Bufferbloat in access networks has received sig-nificant attention recently [4]. Bufferbloat iscaused by network equipment that is provisionedand configured with large unmanaged bufferqueues. Its effects are persistently large queuelevels that in turn lead to large and sustainedpacket latency. Active Queue Management(AQM) has long been considered a solution tobufferbloat. Controlled Delay (CoDel) and Pro-portional Integral controller Enhanced (PIE) aretwo recently proposed AQM algorithms thataddress Random Early Detection (RED) issues[5, 6]. Both are delay-based as they proactivelydrop packets to maintain an average packetqueue delay that is less than a configured latency

Table 2. Upstream channel parameters.

Parameters Value

Frequency ranges

5–42 MHz5–65 MHz5–85 MHz5–117 MHz5–204 MHz

Signal Type OFDMA

Subcarrier Spacing 50 kHz, 25 kHz

FFT Time Duration 20 ms (50 kHz subcarriers)40 ms (25 kHz subcarriers)

FFT Size

50 kHz: 2048 FFT; 1900 Maximum activesubcarriers25 kHz: 4096 FFT; 3800 Maximum activesubcarriers

Minimum OFDMA OccupiedBandwidth

6 MHz for 25 kHz subcarrier spacing10 MHz for 50 kHz subcarrier spacing

Maximum OFDMA ChannelBandwidth 96 MHz

Number of Independently configurable OFDMA channels Minimum of 2

Modulation Type

Mandatory: BPSK, QPSK, 8-QAM, 16-QAM,32-QAM, 64-QAM, 128-QAM, 256-QAM,512-QAM, 1024-QAM Optional: 2048-QAM, 4096-QAM

MARTIN_LAYOUT_Author Layout 3/4/15 4:11 PM Page 111

IEEE Communications Magazine • March 2015112

target. The DOCSIS 3.1 specification requiresAQM turned on by default for both downstreamand upstream service flows. Further, the CMmust support PIE AQM on a per-upstream ser-vice flow basis. For downstream, the CMTS isrequired to support a published AQM. Presum-ably this will be PIE or CoDel.

The DOCSIS 3.1 specifications introduceother MAC enhancements which we brieflydescribe below. A complete description of theenhancements is available in [2]. HierarchicalQoS (HQoS) adds an aggregate service flowabstraction allowing an intermediary level ofscheduling located between aggregates of serviceflows and bonding groups. The concept is similarto class-based queueing where different groups offlows (perhaps organized by organization or typeof traffic) is allocated a percentage of availablebandwidth [7]. Scheduling within one class has noeffect on flows that are scheduled within a differ-ent class. An energy management mode specifi-cally targeting CMs uses OFDM downstreamchannels. The CM supports multiple levels of CMfunctionality and operation, with each level offer-ing reduced capabilities but with lower powerrequirements. A new DOCSIS Timing Protocolallows CMs to maintain time synchronization withthe CMTS within 3000 nanoseconds.

DEPLOYMENT DISCUSSIONThe DOCSIS 3.1 specification is backwards com-patible with DOCSIS 3.0 (or earlier) specifica-tions, where a DOCSIS 3.1 CMTS mustinteroperate seamlessly with DOCSIS 3.0, DOC-SIS 2.0 and DOCSIS 1.1 CMs and a DOCSIS 3.1CM must interoperate seamlessly with a DOCSIS3.0 CMTS. The CMTS is required to supporttime and frequency multiplexing of OFDMA andSingle Carrier QAM signals (DOCSIS 3.0 signal-ing) on the upstream. Depending on the schedul-

ing methodology used by the CMTS, DOCSIS3.1 and DOCSIS 3.0 signaling can be frequencydivision, time division or time and frequency divi-sion multiplexed which provides the highest effi-ciency in spectrum utilization.

The various generations of video technologydeployed in operators’ networks define therequired video transmission formats over the net-work. In the extreme case an operator may haveto broadcast traditional video channel in multipleforms such as analog, digitally using MPEG2 inboth standard and high definition formats, and inIP format for DOCSIS capable set-top boxes. Asnetwork operators migrate their access networkto all IP (including video), the amount of dupli-cate broadcasts that has to be maintained isgreatly reduced. For example, once an operatordecides to no longer support analog, there is nolonger the need to have commonly viewed con-tent broadcast in both analog and digital formats.Further, network operators might need to broad-cast the same content over a 6 MHz channel aswell as over an OFDM channel. The transitionfrom analog video, to digital video to IP video ishappening but at different rates depending onthe number of legacy devices deployed in thenetwork. Once beyond this migration period,emerging DOCSIS systems will greatly increasethe spectral efficiency of the coaxial medium.

EMERGING ISSUESThere are many areas related to the DOCSIS 3.1specifications that are fertile for innovativeresearch. We briefly identify several examples.

DOCSIS 3.1 Scheduling — The DOCSIS specifi-cations purposely do not address specific schedul-ing techniques for either the downstream or theupstream. We expect that some form of channelaware scheduling will be supported by a CMTS.The following simple downstream example ismeant to illustrate the concept. We consider twosubscribers that have the same service plan in aDOCSIS 3.1 network. Assume that one subscriberis poorly connected and is assigned a profile that isoptimized to mitigate low signal-to-noise level. Thesecond subscriber is well connected and is assigneda profile that can achieve high data rates over thechannel. If these are the only two flows in the net-work and all packets are the same size, simpleround robin downstream scheduling at the CMTSwill allocate bandwidth such that the throughputachieved by both flows will be the same (this isreferred to as max-min fair allocation [8]). Unfor-tunately, this underutilizes spectrum capacity. Analternative allocation strategy referred to as ‘maxthroughput’ would favor the better connected flowbut might starve the poorly connected subscriber.This scheduling complexity has been well studiedin wireless networks. A widely accepted compro-mise is proportional fair scheduling that applies aslight bias towards the better connected subscriberbut ensures the poorly connected subscriber alsoreceives a share of available bandwidth [9].

High Bandwidth Applications and Issues — Itis well known that TCP has issues in certain net-work environments. Well studied examples includeTCP over wireless and TCP over high speed net-

Figure 2. DOCSIS spectrum management.

Profiles A,B,C,Drepresent fourprofiles

Figure courtesy ofCableLabs

1212

t

f

B C D A

1020

A B

828

A B C D

636

A B C D A B C D

444

A B C D

252

200

Upstream

5

MARTIN_LAYOUT_Author Layout 3/4/15 4:11 PM Page 112

IEEE Communications Magazine • March 2015 113

works. TCP over cable also has been studied. Infact much of the early research in cable networksfocused on TCP issues [10, 11]. Better performingDOCSIS upstream support along with more robustTCP implementations have addressed these earlyconcerns. However, DOCSIS 3.1 systems thatoffer gigabit-per-second data rates over networksthat involve new delay-based AQM mechanismswill require further study to better understand andengineer future deployments.

Wireless Last Hop — Broadband access is like-ly to involve wireless as the last hop network. Anatural evolution for cable networks is to extendservices over wireless networks.

Ethernet over DOCSIS — Extending carriergrade Ethernet over DOCSIS is an open issue.The challenge is matching the precise timing andquality required by Ethernet to available DOC-SIS profiles and service qualities.

Autonomic Capabilities — Network faultcharacterization based on PNM measurementsand statistics. This can be viewed as a “big data”problem where software needs to analyze largeamounts of information searching for early signsof errors, faults, network performance issues,and subscriber quality of experience issues.

Innovative Economic Models — If currenttiered pricing models are scaled to support giga-bit per second data rate services, the top tiers(i.e. tiers that offer the highest service rates)might not be widely used due to the high cost.New economic models are required.

Virtualization and Software Defined Net-working — In the future, cable systems involvingvirtualizations of CM and CMTS functionality arelikely as this can simplify equipment and reduceequipment cost. Further simplification of equip-ment and automation of provisioning are alsoexpected with applications of Software DefinedNetwork (SDN) concept to DOCSIS systems.

CONCLUSIONSThe DOCSIS 3.1 specifications define a flexibleand highly extensible technology that is capableof meeting the growing capacity requirements ofbroadband networks. The main enhancementsinclude more flexible and efficient use of coaxialspectrum. These capabilities will allow broad-band cable to scale to gigabit speeds. The morechallenging issue might be related to economics.Innovative applications and network servicesmust synergistically evolve to ensure appropriateeconomic factors are in place so that gigabit-per-second broadband access services see wide usage.

REFERENCES[1] Data-Over-Cable Service Interface Specification, MAC

and Upper Layer Protocols Interface Specification DOC-SIS 3.0, CM-SP-MULPIv3.0-124-140403, Apr. 2014.

[2] Data-Over-Cable Service Interface Specification, MACand Upper Layer Protocols Interface Specification DOC-SIS 3.1, CM-SP-MULPIv3.1-104-141218, Dec. 2014.

[3] Data-Over-Cable Service Interface Specification, PhysicalLayer Specification DOCSIS 3.1, CM-SP-PHYv3.1-104-141218, Dec. 2014.

[4] J. Gettys, “Bufferbloat: Dark Buffers in the Internet,”IEEE Internet Computing, vol. 15, no. 3, 2011.

[5] K. Nichols and V. Jacobson, “Controlling Queue Delay,”ACM Queue, vol. 10, no. 5, 2012.

[6] R. Pan et al., “PIE: A Lightweight Control Scheme toAddress the Bufferbloat Problem,” Proc. IEEE HPSR, July2014.

[7] S. Floyd and V. Jacobson, “Link-Sharing and ResourceManagement Models for Packet Networks,” IEEE/ACMTrans. Net., vol. 3, no. 4, Aug. 1995, pp. 365–86.

[8] C. So-In, R. Jain, and A. Al-Tamimi, “Resource Allocationin IEEE 802.16 Mobile WiMAX,” Orthogonal FrequencyDivision Multiple Access (OFDMA),” Edited by T. Jiang,L. Song, Y. Zhang, Auerbach Publications, CRC Press,ISBN: 1420088246, Apr. 2010.

[9] L. Li, M. Pal, and Y. R. Yang, “Proportional Fairness inMulti-Rate Wireless LANs,” INFOCOM 2008. The 27thConf. Computer Commun., 13–18 Apr. 2008, pp.1004–12.

[10] R. Cohen, S. Ramanathan, “TCP for High Performancein Hybrid Fiber Coaxial Broad-band Access Networks,”IEEE/ACM Trans. Net., vol. 6, no. 1, Feb. 1998.

[11] O. Elloumi et al., “A Simulation-based Study of TCPDynamics over HFC Networks,” Computer Networks,vol. 32, no. 3, 2000, pp. 301–17.

BIOGRAPHIESBELAL HAMZEH is Director of Broadband Evolution withCableLabs driving technology development activities forwired and wireless technologies. Belal has extensive experi-ence in research and development activities for wired andwireless technologies and currently leads the DOCSIS 3.1specification development efforts and is the Principal Archi-tect for the RF and Physical layers. Prior to that, Belal ledresearch and development efforts for 3G/4G systems,including standardization, product development and net-work deployment. He holds an M.Sc. and Ph.D. degrees inElectrical Engineering from Penn State.

MEHMET TOY [SM] is a Distinguished Engineer at Comcast,USA, involved in network architectures and standards. Hereceived his B.S and M.S from Istanbul Technical University,Istanbul, Turkey, and Ph.D from Stevens Institute of Tech-nology, Hoboken, NJ. He held management and technicalpositions at ADVA Optical Networking, Intergenix, IntelCorporation, Verizon Wireless, Axiowave Networks, FujitsuNetwork Communications, AT&T Bell Labs and Lucent Tech-nologies. In addition, he served as a tenure-track andadjunct faculty at universities in the US and Turkey; andserved in IEEE Network Magazine Editorial Board, IEEE, andIEEE-USA. He has contributed to research, developmentand standardization of Cloud Services Architectures, DataCenter Network Virtualization Overlays, Self-Managed Net-works, Metro Ethernet, DOCSIS Provisioning of EPON(DPoE), IP Multimedia Systems (IMS), Asynchronous Trans-fer Mode (ATM), optical, IP/MPLS, and wireless. He has fourpatent applications; authored five books, a video tutorial,and numerous articles and standards contributions in theseareas and signal processing. He received various awardsfrom IEEE and the companies, including multiple InventorAwards from Comcast and Exceptional ContributionAwards from AT&T Bell Labs. He is a member of Cloud Eth-ernet Forum (CEF) Board, co-chair of CEF Architecture Com-mittee, and founder and chair of IEEE ComSoc CableNetworks and Services (CNS) sub-committee.

YUNHUI FU received a B.S. degree in electrical engineeringfrom the Central South University at Changsha, China, in2000, and an M.S. degree in Computer Science from theUniversity of Texas-Pan American in 2010. He is currently aPh.D. student in the School of Computing at Clemson Uni-versity. His research focuses on improving the reliabilityand scalability of multimedia communication over wiredand wireless packet networks.

JIM MARTIN is an associate professor in the School of Com-puting at Clemson University. His research interests includebroadband access, wireless networks, Internet protocols,and network performance analysis. Current research pro-jects include heterogeneous wireless systems and DOCSIS3.x cable access networks. He has received funding fromNSF, NASA, the Department of Justice, BMW, CableLabs,Cisco, Comcast, Cox, Huawei, and IBM. He received hisPh.D. from North Carolina State University. Prior to joiningClemson, he was a consultant for Gartner, and prior tothat, a software engineer for IBM.

The more challeng-

ing issue might be

related to economics.

Innovative applica-

tions and network

services must syner-

gistically evolve to

ensure appropriate

economic factors are

in place so that giga-

bit-per-second

broadband access

services see wide

usage.

MARTIN_LAYOUT_Author Layout 3/4/15 4:11 PM Page 113