《信息网络协议基础》课程教学资源（学习资料）The ITU-T’s New G.fast Standard Brings DSL into the Gigabit Era（GFAST）

ACCEPTED FROM OPEN CALL The ITU-T's New G.fast Standard Brings DSL into the Gigabit Era Vladimir Oksman,Rainer Strobel,Xiang Wang,Dong Wei,Rami Verbin,Richard Goodson,and Massimo Sorbara The standardized G.fast ABSTRACT unit (DPU)connected to the central office transmission method (CO)by fiber(PON or point-to-point fiber). and advanced crosstalk This article explores the recently issued ITU-T DPUs are installed close to the customer cancellation techniques Recommendations specifying "G.fast"(G.9701 premises,typically in mini-cabinets mounted [1]and G.9700 [2])that bring user bit rates up in basements of multi-dwelling units,on elec- are presented by the to 1 Gb/s over twisted pairs from the distribution trical poles,in curb boxes,or in manholes [3. authors,with specific point to customer premises.The overview and 6,and connected to customer premises equip- performance projec- some key research challenges of G.fast are dis- ment(CPE)via copper pairs.A DPU typically tions and measure- cussed in [3].The standardized G.fast transmis- serves 4-20 lines,but bigger DPUs are expected sion method and advanced crosstalk cancellation in the future.A single-line DPU may serve as ment results achieved techniques are presented here with specific per- a fiber-to-the-home (FTTH)copper extension. during the first demon- formance projections and measurement results DPUs can be powered locally,remotely,or by strations and trials, achieved during the first demonstrations and subscribers from the customer premises using showing bit rates of trials,showing bit rates of 500 Mb/s over 250 m reverse power feeding (RPF)[7];the latter is 500 Mb/s over 250 m and available reach up to 400 m.A description very convenient for small DPUs.The achiev- and available reach up of standardized tools for dynamic performance able bit rate over a particular line depends on maintenance,resource allocation,and power sav- its length and wire type.The maximum reach to 400 m. ing enhancing G.fast applications concludes this is 400 m,but the majority of installations are article. expected to be within 100 m. An example of a typical G.fast installation INTRODUCTION using RPF is shown in Fig.1.The G.fast trans- Modern life depends on the Internet,and thus ceivers in the DPU(FTU-O)and in the CPE the demand for high-speed Internet access is (FTU-R)are connected via a copper pair;the rapidly growing.Digital subscriber line (DSL) FTU-R resides in the network termination unit technology,accordingly,keeps up with both cus- (NTU).The broadband services delivered to tomer demand and the progress in competing the DPU via a PON feeder are conveyed to the access technologies,such as DOCSIS,WiMAX/ NTU and further distributed to various broad- Long Term Evolution (LTE),and gigabit pas band applications via a high-speed in-premises sive optical networking (G-PON).In 2010,the network (e.g.,WiFi).The RPF power sourcing International Telecommunication Union Tele- equipment(PSE)in the NTU generates suf- communication Standardization Sector (ITU-T) ficient power to supply the associated FTU-O developed Recommendation G.993.5 [41,which and common functions of the DPU via the cop- set a 100 Mb/s benchmark in DSL services [5]. per pair.The DPU power supply unit(PSU) The new G.fast Recommendations [1,2 specify gathers the power sourced by PSEs of all active 1 Gb/s access over copper lines through corresponding power extractors To reach such high bit rates,G.fast uses only (PEs).In other installations,the PSE may be the last leg of the existing copper access network separate from the NTU and also feed the NTU. and in-premises wiring.These wires are usually The PSE can work during power outages using unshielded,non-conditioned twisted pairs,flat a backup battery.More RPF details can be pairs,or quads (four twisted wires)and known found in (3,7]. for very strong crosstalk,especially inside quads Analog phones are connected through adapt- [3].Reaching high bit rates over such low-qual- ers because RPF uses DC:the Foreign Exchange ity copper is a difficult task that requires a sub- Office (FXO)adapter receives plain old tele- stantially new approach.The main challenges phone service (POTS)signaling derived from encountered by engineers and the potential tech- voice over IP (VolP)service by the analog tele- nical choices are discussed in [3],while this arti- phone adapter(ATA)and generates alternative cle describes the adopted technical solutions signaling capable of running over in-premises The G.fast-based access network uses the wiring with RPF;the Foreign Exchange Sub- fiber-to-the-distribution-point (FTTdp)archi- scriber (FXS)adapter further recovers the orig- tecture [6],which comprises a distribution point inal POTS signaling.Both the FXO and FXS Madimir Oksman and Rainer Strobel are with Lantig,an Intel Company;Xiang Wang and Dong Wei are with Huawei Technologies;Rami Verbin is with Sckipio Technologies;Richard Goodson is with AADTRAN Inc,Massimo Sorbara is with Qualcomm. 118 0163-6804/16/$25.00©2016IEEE IEEE Communications Magazine.March 2016

118 0163-6804/16/$25.00 © 2016 IEEE IEEE Communications Magazine • March 2016 Abstract This article explores the recently issued ITU-T Recommendations specifying “G.fast” (G.9701 [1] and G.9700 [2]) that bring user bit rates up to 1 Gb/s over twisted pairs from the distribution point to customer premises. The overview and some key research challenges of G.fast are dis￾cussed in [3]. The standardized G.fast transmis￾sion method and advanced crosstalk cancellation techniques are presented here with specific per￾formance projections and measurement results achieved during the first demonstrations and trials, showing bit rates of 500 Mb/s over 250 m and available reach up to 400 m. A description of standardized tools for dynamic performance maintenance, resource allocation, and power sav￾ing enhancing G.fast applications concludes this article. Introduction Modern life depends on the Internet, and thus the demand for high-speed Internet access is rapidly growing. Digital subscriber line (DSL) technology, accordingly, keeps up with both cus￾tomer demand and the progress in competing access technologies, such as DOCSIS, WiMAX/ Long Term Evolution (LTE), and gigabit pas￾sive optical networking (G-PON). In 2010, the International Telecommunication Union Tele￾communication Standardization Sector (ITU-T) developed Recommendation G.993.5 [4], which set a 100 Mb/s benchmark in DSL services [5]. The new G.fast Recommendations [1, 2] specify 1 Gb/s access over copper. To reach such high bit rates, G.fast uses only the last leg of the existing copper access network and in-premises wiring. These wires are usually unshielded, non-conditioned twisted pairs, flat pairs, or quads (four twisted wires) and known for very strong crosstalk, especially inside quads [3]. Reaching high bit rates over such low-qual￾ity copper is a difficult task that requires a sub￾stantially new approach. The main challenges encountered by engineers and the potential tech￾nical choices are discussed in [3], while this arti￾cle describes the adopted technical solutions. The G.fast-based access network uses the fiber-to-the-distribution-point (FTTdp) archi￾tecture [6], which comprises a distribution point unit (DPU) connected to the central office (CO) by fiber (PON or point-to-point fiber). DPUs are installed close to the customer premises, typically in mini-cabinets mounted in basements of multi-dwelling units, on elec￾trical poles, in curb boxes, or in manholes [3, 6], and connected to customer premises equip￾ment (CPE) via copper pairs. A DPU typically serves 4–20 lines, but bigger DPUs are expected in the future. A single-line DPU may serve as a fiber-to-the-home (FTTH) copper extension. DPUs can be powered locally, remotely, or by subscribers from the customer premises using reverse power feeding (RPF) [7]; the latter is very convenient for small DPUs. The achiev￾able bit rate over a particular line depends on its length and wire type. The maximum reach is 400 m, but the majority of installations are expected to be within 100 m. An example of a typical G.fast installation using RPF is shown in Fig. 1. The G.fast trans￾ceivers in the DPU (FTU-O) and in the CPE (FTU-R) are connected via a copper pair; the FTU-R resides in the network termination unit (NTU). The broadband services delivered to the DPU via a PON feeder are conveyed to the NTU and further distributed to various broad￾band applications via a high-speed in-premises network (e.g., WiFi). The RPF power sourcing equipment (PSE) in the NTU generates suf￾ficient power to supply the associated FTU-O and common functions of the DPU via the cop￾per pair. The DPU power supply unit (PSU) gathers the power sourced by PSEs of all active lines through corresponding power extractors (PEs). In other installations, the PSE may be separate from the NTU and also feed the NTU. The PSE can work during power outages using a backup battery. More RPF details can be found in [3, 7]. Analog phones are connected through adapt￾ers because RPF uses DC: the Foreign Exchange Office (FXO) adapter receives plain old tele￾phone service (POTS) signaling derived from voice over IP (VoIP) service by the analog tele￾phone adapter (ATA) and generates alternative signaling capable of running over in-premises wiring with RPF; the Foreign Exchange Sub￾scriber (FXS) adapter further recovers the orig￾inal POTS signaling. Both the FXO and FXS The ITU-T’s New G.fast Standard Brings DSL into the Gigabit Era Vladimir Oksman, Rainer Strobel, Xiang Wang, Dong Wei, Rami Verbin, Richard Goodson, and Massimo Sorbara Accepted from Open Call The standardized G.fast transmission method and advanced crosstalk cancellation techniques are presented by the authors, with specific performance projec￾tions and measure￾ment results achieved during the first demon￾strations and trials, showing bit rates of 500 Mb/s over 250 m and available reach up to 400 m. Vladimir Oksman and Rainer Strobel are with Lantiq, an Intel Company; Xiang Wang and Dong Wei are with Huawei Technologies; Rami Verbin is with Sckipio Technologies; Richard Goodson is with AADTRAN Inc.; Massimo Sorbara is with Qualcomm

CPE G.fast is a state-of-the- art copper access DPU CPE Copper pairs technology that CPE Customer premises provides fiber-grade DPU NTU transmission speed L2+ L2+ over existing copper, processing processing In-premises minimizes energy 0 YR BB network G.9701 DRA G9701 consumption,reduce色 PSE FTU-R TCE TCE FTU-R maintenance cost, VCE FXO adapter ATA and provides great U-R- U-0 robustness and flexibility Copper pair for the customers. FX5 In-premises adapter winng To the CO Access network FXS adapter Figure 1.Example of G.fast deployment using RPF and derived POTS are fed by the PSE.To avoid G.fast performance packets from upper layers(L2+)are mapped loss,no other devices should be connected to the into data transmission units (DTUs)that are in-premises wiring conveyed transparently over the line.Reed-Solo- In other installations,the ATA may reside mon forward error correction [8]improves noise in the NTU,while derived voice service can be immunity:each DTU is assembled from multi- distributed throughout the premises via cordless ple Reed-Solomon codewords,and DTU bytes phone technology or by using smartphone con- are interleaved.The number of codewords and nection to in-premises WiFi;no particular option their size are configured to fit the throughput of for voice distribution is implied by G.fast. the line.Noise is further mitigated by retrans- G.9701 uses the frequency spectrum from mission of DTUs received in error:the number 2.2 to 106 MHz with full crosstalk cancellation of retransmissions for a DTU is limited by the between the lines sourced by a DPU:near-end latency bound.For retransmission and latency crosstalk (NEXT)is avoided by using synchro- control,each DTU contains a sequence number nized time-division duplexing (STDD),and and associated timestamp. far-end crosstalk(FEXT)is cancelled using vec- Discrete multi-tone (DMT)modulation [8] toring.No alien crosstalk cancellation is defined. is used for passing DTUs and management data Other G.fast innovations include dynamic allo- over the line.The advantages of DMT are well cation of resources between DPU sourced lines, known,especially its capability to operate on lines efficient energy-saving techniques,and dynam- with multiple bridged taps such as in-premises ic performance maintenance.Pair bonding is wiring.The specified tone spacing of 51.75 kHz defined to allow multiplication of customers'bit is 12 times that of very-high rate DSL 2(VDSL2) rate. [8]because G.fast loops are much shorter.Thus, G.fast customer installations vary by signal 2048-tone DMT is sufficient to cover the cur- attenuation and noise,and may be influenced by rent G.fast frequency spectrum,simplifying the other technologies.For reliable self-installation design.Each DMT symbol is cyclically extended provisioning,which reduces the operator's cost, using both prefix and suffix.The prefix mitigates G.fast defines a flexible and robust transmission inter-symbol interference and is configurable to protocol,online reconfiguration,and dynamic address a wide range of loop lengths.The suf- adaptation of bit rate,all maintained via robust fix is applied for transmit spectrum shaping and management channels.Zero-touch management overlaps with the following symbol to improve further reduces the operator's cost by avoiding efficiency:suffix size always fits the size of the truck rolls for future equipment upgrades and windowing [8].The default cyclic extension yields adding new subscribers. a symbol duration of 20.83 s.Up to 14 bits can G.fast is a state-of-the-art copper access tech- be loaded per tone.To further increase bit rates nology that offers fiber-grade transmission speed future versions of G.fast may extend the frequen- over existing copper,minimizes energy consump- cy spectrum to 211.968 MHz. tion,reduces maintenance cost,and provides The advantages of the G.fast duplex- great robustness and flexibility for customers. ing scheme (STDD)compared to FDD are FUNDAMENTALS OF G.9701 TECHNOLOGY described in [3].With STDD,the upstream and downstream sets of DMT tones can be selected TRANSMISSION METHOD independently,making STDD flexible in the fre G.9701 specifies the functionality of G.fast trans- quency domain.A particular selection depends ceivers (FTU-O and FTU-R)that establish a on the deployment scenario,and involves chan- high-speed transmission path between y-O and nel characteristics and spectrum compatibility Y-R reference points(Fig.1).The user's data issues (see below). IEEE Communications Magazine.March 2016 119

IEEE Communications Magazine • March 2016 119 are fed by the PSE. To avoid G.fast performance loss, no other devices should be connected to the in-premises wiring. In other installations, the ATA may reside in the NTU, while derived voice service can be distributed throughout the premises via cordless phone technology or by using smartphone con￾nection to in-premises WiFi; no particular option for voice distribution is implied by G.fast. G.9701 uses the frequency spectrum from 2.2 to 106 MHz with full crosstalk cancellation between the lines sourced by a DPU: near-end crosstalk (NEXT) is avoided by using synchro￾nized time-division duplexing (STDD), and far-end crosstalk (FEXT) is cancelled using vec￾toring. No alien crosstalk cancellation is defi ned. Other G.fast innovations include dynamic allo￾cation of resources between DPU sourced lines, efficient energy-saving techniques, and dynam￾ic performance maintenance. Pair bonding is defi ned to allow multiplication of customers’ bit rate. G.fast customer installations vary by signal attenuation and noise, and may be infl uenced by other technologies. For reliable self-installation provisioning, which reduces the operator’s cost, G.fast defi nes a fl exible and robust transmission protocol, online reconfiguration, and dynamic adaptation of bit rate, all maintained via robust management channels. Zero-touch management further reduces the operator’s cost by avoiding truck rolls for future equipment upgrades and adding new subscribers. G.fast is a state-of-the-art copper access tech￾nology that offers fi ber-grade transmission speed over existing copper, minimizes energy consump￾tion, reduces maintenance cost, and provides great robustness and fl exibility for customers. fundAmentAls of G.9701 technoloGy trAnsmIssIon method G.9701 specifi es the functionality of G.fast trans￾ceivers (FTU-O and FTU-R) that establish a high-speed transmission path between -O and -R reference points (Fig. 1). The user’s data packets from upper layers (L2+) are mapped into data transmission units (DTUs) that are conveyed transparently over the line. Reed-Solo￾mon forward error correction [8] improves noise immunity: each DTU is assembled from multi￾ple Reed-Solomon codewords, and DTU bytes are interleaved. The number of codewords and their size are confi gured to fi t the throughput of the line. Noise is further mitigated by retrans￾mission of DTUs received in error; the number of retransmissions for a DTU is limited by the latency bound. For retransmission and latency control, each DTU contains a sequence number and associated timestamp. Discrete multi-tone (DMT) modulation [8] is used for passing DTUs and management data over the line. The advantages of DMT are well known, especially its capability to operate on lines with multiple bridged taps such as in-premises wiring. The specifi ed tone spacing of 51.75 kHz is 12 times that of very-high rate DSL 2 (VDSL2) [8] because G.fast loops are much shorter. Thus, 2048-tone DMT is sufficient to cover the cur￾rent G.fast frequency spectrum, simplifying the design. Each DMT symbol is cyclically extended using both prefi x and suffi x. The prefi x mitigates inter-symbol interference and is confi gurable to address a wide range of loop lengths. The suf￾fi x is applied for transmit spectrum shaping and overlaps with the following symbol to improve efficiency; suffix size always fits the size of the windowing [8]. The default cyclic extension yields a symbol duration of 20.83 s. Up to 14 bits can be loaded per tone. To further increase bit rates, future versions of G.fast may extend the frequen￾cy spectrum to 211.968 MHz. The advantages of the G.fast duplex￾ing scheme (STDD) compared to FDD are described in [3]. With STDD, the upstream and downstream sets of DMT tones can be selected independently, making STDD fl exible in the fre￾quency domain. A particular selection depends on the deployment scenario, and involves chan￾nel characteristics and spectrum compatibility issues (see below). G.fast is a state-of-the￾art copper access technology that provides fi ber-grade transmission speed over existing copper, minimizes energy consumption, reduces maintenance cost, and provides great robustness and fl exibility for the customers. Figure 1. Example of G.fast deployment using RPF and derived POTS. NTU CPE Copper pairs CPE •••••• ••• CPE DPU To the CO PON feeder Customer premises In-premises BB network In-premises wiring FXO U-R adapter U-O Copper pair Access network ATA FXS adapter L2+ processing γ-O γ-R TCE PE DRA TCE VCE PSU G.9701 FTU-R L2+ processing G.9701 FTU-R PSEHN-PHY PON-PHY FXS adapter DPU

STDD superframe (TsE=MsF x Tp) To minimize the per- Tp=Mpx Tsymb formance loss,down TDD sync frame TDD frame TDD frame stream transmit PSDs Logical frame are optimized across all DS TX US TX DSX■US TX网 DS TXDS TX the lines with precoder updates,induding trans- Sync symbol RMC symbol mit power reduction FTU-O DS TX US RX of tones causing high crosstalk. FTU-R DS RX US TX -line propagation delay Tg=T22×Td Figure 2.G.fast transmission format. The G.fast transmission format compris- a DPU (vectored group).The alignment is by es superframes,each composed of Msp TDD symbol boundaries,and only a small deviation frames (Fig.2).Each TDD frame contains Mr is tolerable to avoid NEXT and facilitate FEXT symbol periods(Tsymb).One set of contiguous cancellation,discontinuous operation,and fast symbol periods is assigned for downstream trans- reconfiguration. mission and another one for upstream transmis- sion.The sum of guard times between upstream FEXT CANCELLATION and downstream transmissions(Ts+T2)is one FEXT cancellation is imperative for reach- symbol period. ing high bit rates.Similar to G.993.5,G.fast Superframes follow each other with no gaps; performs FEXT cancellation at the DPU:the their boundaries are identified by downstream downstream transmit signals are precoded by sync symbols.Both downstream and upstream adding FEXT pre-compensation signals,and a sync symbols reside in a TDD sync frame,and post-processor subtracts FEXT components from carry probe sequences used for channel estima- the received upstream signal [3,5].The vector- tion and other purposes (described below). ing control entity(VCE)at the DPU performs The maximum duration of a TDD frame is channel estimation,and computes precoder and bounded by the propagation delay limit to Mg post-processor matrices for all connected lines. 36 symbol periods.A setting of Mg 23 reduces The particular methods of channel estimation, round-trip delay.A superframe contains 8 and matrix computation,and FEXT cancellation are 12 TDD frames,respectively,so its duration is vendor discretionary.For downstream channel always about 6 ms,which allows the superframe estimation,the VCE may assign the same or dif- period to be used as a time base for initialization ferent precoder matrices for sync symbols and and management procedures. data symbols (including the use of non-precoded The transmission path is maintained by the ync symbols)】 embedded operations channel (eoc)and robust Like G.993.5,G.fast uses linear precoding. management channel(RMC).The eoc is multi- However,the FEXT behavior in G.fast is fun plexed into DTUs;it has a flexible bit rate that damentally different,especially for quad-twist- can support high-volume management data,but ed cables,due to G.fast's much wider frequency its robustness is about the same as user data.The spectrum.Figure 3 shows a typical G.fast FEXT RMC,in contrast,is defined to carry short mes- channel:the in-quad crosstalk can be stronger sages and is much more robust due to high-re- than the direct channel at high frequencies dundancy Reed-Solomon coding.tone selection, This complicates precoding because the added and conservative bit loading.Multiple repetitions pre-compensation signals can substantially are applied for critical RMC commands. increase the transmit PSD.Thus,with a given One RMC symbol per direction is sent in PSD limit,pre-compensation signals associated each TDD frame (Fig.2).It carries the RMC on with a line generating high crosstalk can suppress dedicated tones and DTU bytes on other tones. the power of the direct signal in other lines of The positions of RMC symbols in a TDD frame the vectored group,causing substantial perfor- and the sets of RMC tones are configured at mance loss [31. initialization.Symbol positions from one to the To minimize this performance loss,down- next RMC symbol of the same direction repre- stream transmit PSDs are optimized across sent a logical frame.The RMC carries acknowl- all the lines with precoder updates,includ- edgments of received DTUs,supporting DTU ing transmit power reduction of tones caus- retransmission,and conveys management com- ing high crosstalk.A transmitter-initiated mands facilitating logical frame configuration, gain adjustment (TIGA)is used to accom- online reconfiguration (OLR),and transitions modate the change of precoder gain in the into and out of low-power states. peer FTU-R receiver.Figure 3 shows that With STDD,the time positions of super- PSD optimization substantially improves the frames,TDD frames,sync symbols,and RMC achievable signal-to-noise ratio (SNR).It symbols are aligned across all lines sourced by may even exceed the single-line SNR due to 120 IEEE Communications Magazine.March 2016

120 IEEE Communications Magazine • March 2016 The G.fast transmission format compris￾es superframes, each composed of MSF TDD frames (Fig. 2). Each TDD frame contains MF symbol periods (Tsymb). One set of contiguous symbol periods is assigned for downstream trans￾mission and another one for upstream transmis￾sion. The sum of guard times between upstream and downstream transmissions (Tg1 + Tg2 ) is one symbol period. Superframes follow each other with no gaps; their boundaries are identified by downstream sync symbols. Both downstream and upstream sync symbols reside in a TDD sync frame, and carry probe sequences used for channel estima￾tion and other purposes (described below). The maximum duration of a TDD frame is bounded by the propagation delay limit to MF = 36 symbol periods. A setting of MF = 23 reduces round-trip delay. A superframe contains 8 and 12 TDD frames, respectively, so its duration is always about 6 ms, which allows the superframe period to be used as a time base for initialization and management procedures. The transmission path is maintained by the embedded operations channel (eoc) and robust management channel (RMC). The eoc is multi￾plexed into DTUs; it has a flexible bit rate that can support high-volume management data, but its robustness is about the same as user data. The RMC, in contrast, is defi ned to carry short mes￾sages and is much more robust due to high-re￾dundancy Reed-Solomon coding, tone selection, and conservative bit loading. Multiple repetitions are applied for critical RMC commands. One RMC symbol per direction is sent in each TDD frame (Fig. 2). It carries the RMC on dedicated tones and DTU bytes on other tones. The positions of RMC symbols in a TDD frame and the sets of RMC tones are configured at initialization. Symbol positions from one to the next RMC symbol of the same direction repre￾sent a logical frame. The RMC carries acknowl￾edgments of received DTUs, supporting DTU retransmission, and conveys management com￾mands facilitating logical frame configuration, online reconfiguration (OLR), and transitions into and out of low-power states. With STDD, the time positions of super￾frames, TDD frames, sync symbols, and RMC symbols are aligned across all lines sourced by a DPU (vectored group). The alignment is by symbol boundaries, and only a small deviation is tolerable to avoid NEXT and facilitate FEXT cancellation, discontinuous operation, and fast reconfi guration. feXt cAncellAtIon FEXT cancellation is imperative for reach￾ing high bit rates. Similar to G.993.5, G.fast performs FEXT cancellation at the DPU: the downstream transmit signals are precoded by adding FEXT pre-compensation signals, and a post-processor subtracts FEXT components from the received upstream signal [3, 5]. The vector￾ing control entity (VCE) at the DPU performs channel estimation, and computes precoder and post-processor matrices for all connected lines. The particular methods of channel estimation, matrix computation, and FEXT cancellation are vendor discretionary. For downstream channel estimation, the VCE may assign the same or dif￾ferent precoder matrices for sync symbols and data symbols (including the use of non-precoded sync symbols). Like G.993.5, G.fast uses linear precoding. However, the FEXT behavior in G.fast is fun￾damentally different, especially for quad-twist￾ed cables, due to G.fast’s much wider frequency spectrum. Figure 3 shows a typical G.fast FEXT channel: the in-quad crosstalk can be stronger than the direct channel at high frequencies. This complicates precoding because the added pre-compensation signals can substantially increase the transmit PSD. Thus, with a given PSD limit, pre-compensation signals associated with a line generating high crosstalk can suppress the power of the direct signal in other lines of the vectored group, causing substantial perfor￾mance loss [3]. To minimize this performance loss, down￾stream transmit PSDs are optimized across all the lines with precoder updates, includ￾ing transmit power reduction of tones caus￾ing high crosstalk. A transmitter-initiated gain adjustment (TIGA) is used to accom￾modate the change of precoder gain in the peer FTU-R receiver. Figure 3 shows that PSD optimization substantially improves the achievable signal-to-noise ratio (SNR). It may even exceed the single-line SNR due to Figure 2. G.fast transmission format. t TDD frame TDD frame STDD superframe (TSF = MSF × TF) ••• DS TX DS TX FTU-O DS TX US RX DS TX US TX DS TX US TX TDD sync frame Logical frame TF = MF × Tsymb Tsymb Tg2 Tpd — line propagation delay Tg1’ = Tg2 - 2 × Tpd Tg1 Sync symbol RMC symbol FTU-R DS RX US TX Tg1’ Tg2’ To minimize the per￾formance loss, down￾stream transmit PSDs are optimized across all the lines with precoder updates, including trans￾mit power reduction of tones causing high crosstalk

-Direct channel -quad crosstalk Vectorizing,optimized PSD Out-of-duad crosstal vectonng,no PSD optimization No vedoring Single line W 50 60 100120140160180 200 f/MHz (/MHz Figure 3.Left:direct and crosstalk channels of PE 0.5 mm quad cable (Germany);right:achievable downstream SNR additional direct signal propagation via FEXT of the feedback in time and frequency are used, channels.The latter also has a negative effect: spreading the transmission over several probe turning off a line in a vectored group requires sequence cycles. a precoder update and likely changes perfor- mance of the other lines. OPERATION OF A VECTORED GROUP Vectored group operation comprises three CHANNEL ESTIMATION phases:tracking,joining,and leaving.During Channel estimation is necessary to compute the tracking,all lines of the group are in showtime: FEXT channel matrix,and to derive precoder lines neither join nor leave the group,and each and post-processor matrices.Similar to G.993.5 line tracks channel variations caused mainly by probe sequences are carried by upstream and temperature changes and discontinuous opera- downstream sync symbols (Fig.2).The values tion.The latter may require frequent precoder of the received probe signals are delivered to updates facilitated by OLR procedures (TIGA. the VCE,which computes the channel matrix. SRA.see below). The VCE assigns appropriate probe sequences In the joining phase,new lines are added to (orthogonal,like Walsh-Hadamard sequences,or the vectored group.During initialization of new pseudo-orthogonal)to the lines of the vectored lines,channel matrices,precoders,and transmit group using its own channel estimation strate- PSDs of new and showtime lines are jointly opti- gy.Probe sequences are repeated periodically, mized,improving overall performance.This also allowing efficient averaging for noise mitigation. requires multiple OLR procedures and high-vol- Unlike G.993.5,probe sequences may include ume transfers of vectoring feedback in showtime 0-elements (no transmission of a sync symbol). lines;both are supported by the eoc. By using 0-elements,the vectored group can be Lines can leave the group in an orderly or virtually divided into sub-groups;this reduces disorderly manner.Orderly leaving first ter- the aggregate FEXT inside each sub-group and minates transmission in both directions,then speeds up channel estimation. updates the channel matrices of the remain- The assigned probe sequences are commu- ing lines,allowing the FTU-R to then safely nicated to the FTU-R at initialization and can disconnect.A disorderly-disconnected FTU-R be updated via eoc during showtime (the state (e.g.,unplugging)usually disturbs other lines when user data is communicated).The VCE may substantially due to associated changes in their request of the FTU-R to report DFT-output direct channel and increase of residual FEXT, samples that represent the received probe signal which is obviously undesirable.Fast rate adap- in the frequency domain,or error samples that tation (FRA)and retransmission help mitigate represent a normalized error vector between this error bursts until channel matrices are updat- received probe signal and the associated refer- ed;the performance is restored by associated ence constellation point.The FTU-R uses the OLR procedures. communicated downstream probe sequence to Details of the joining procedure are shown identify this constellation point,since the error in Fig.4.After a G.994.1 handshake [9],during may be comparable or even exceed the received which the two sides exchange capabilities,agree signal due to strong FEXT. on a common operational mode,and set neces The FTU-R report (vectoring feedback)is sary parameters to facilitate STDD,the FTU-O sent to the VCE via the eoc during showtime and starts training by transmitting superframes via the special operations channel(SOC)during containing only sync symbols modulated by a initialization.If available upstream capacity is probe sequence (during O-VECTOR-1).This insufficient,tone interpolation and decimation allows the VCE to learn and cancel downstream IEEE Communications Magazine.March 2016 121

IEEE Communications Magazine • March 2016 121 additional direct signal propagation via FEXT channels. The latter also has a negative effect: turning off a line in a vectored group requires a precoder update and likely changes perfor￾mance of the other lines. Channel Estimation Channel estimation is necessary to compute the FEXT channel matrix, and to derive precoder and post-processor matrices. Similar to G.993.5, probe sequences are carried by upstream and downstream sync symbols (Fig. 2). The values of the received probe signals are delivered to the VCE, which computes the channel matrix. The VCE assigns appropriate probe sequences (orthogonal, like Walsh-Hadamard sequences, or pseudo-orthogonal) to the lines of the vectored group using its own channel estimation strate￾gy. Probe sequences are repeated periodically, allowing efficient averaging for noise mitigation. Unlike G.993.5, probe sequences may include 0-elements (no transmission of a sync symbol). By using 0-elements, the vectored group can be virtually divided into sub-groups; this reduces the aggregate FEXT inside each sub-group and speeds up channel estimation. The assigned probe sequences are commu￾nicated to the FTU-R at initialization and can be updated via eoc during showtime (the state when user data is communicated). The VCE may request of the FTU-R to report DFT-output samples that represent the received probe signal in the frequency domain, or error samples that represent a normalized error vector between this received probe signal and the associated refer￾ence constellation point. The FTU-R uses the communicated downstream probe sequence to identify this constellation point, since the error may be comparable or even exceed the received signal due to strong FEXT. The FTU-R report (vectoring feedback) is sent to the VCE via the eoc during showtime and via the special operations channel (SOC) during initialization. If available upstream capacity is insufficient, tone interpolation and decimation of the feedback in time and frequency are used, spreading the transmission over several probe sequence cycles. Operation of a Vectored Group Vectored group operation comprises three phases: tracking, joining, and leaving. During tracking, all lines of the group are in showtime: lines neither join nor leave the group, and each line tracks channel variations caused mainly by temperature changes and discontinuous opera￾tion. The latter may require frequent precoder updates facilitated by OLR procedures (TIGA, SRA, see below). In the joining phase, new lines are added to the vectored group. During initialization of new lines, channel matrices, precoders, and transmit PSDs of new and showtime lines are jointly opti￾mized, improving overall performance. This also requires multiple OLR procedures and high-vol￾ume transfers of vectoring feedback in showtime lines; both are supported by the eoc. Lines can leave the group in an orderly or disorderly manner. Orderly leaving first ter￾minates transmission in both directions, then updates the channel matrices of the remain￾ing lines, allowing the FTU-R to then safely disconnect. A disorderly-disconnected FTU-R (e.g., unplugging) usually disturbs other lines substantially due to associated changes in their direct channel and increase of residual FEXT, which is obviously undesirable. Fast rate adap￾tation (FRA) and retransmission help mitigate error bursts until channel matrices are updat￾ed; the performance is restored by associated OLR procedures. Details of the joining procedure are shown in Fig. 4. After a G.994.1 handshake [9], during which the two sides exchange capabilities, agree on a common operational mode, and set neces￾sary parameters to facilitate STDD, the FTU-O starts training by transmitting superframes containing only sync symbols modulated by a probe sequence (during O-VECTOR-1). This allows the VCE to learn and cancel downstream Figure 3. Left: direct and crosstalk channels of PE 0.5 mm quad cable (Germany); right: achievable downstream SNR. f/MHz 0 20 -90 TF/dB -80 -70 -60 -50 -40 -30 -20 -10 40 60 80 100 120 140 160 180 200 f/MHz 10 10 SNR/dB 20 30 40 50 60 20 30 40 50 60 70 80 90 100 Direct channel In-quad crosstalk Out-of-quad crosstalk Vectorizing, optimized PSD Vectoring, no PSD optimization No vectoring Single line

Dunng the channel G.9941 G.9701 channel discovery phase analysis and exchange G9701 handshake onase O-VECTOR-I R-VECTOR-1 O-VECTOR-2 O/R-PRM-UPDATE CA&E phase phase,FTUs establish their desired showtime o settings,such as bit loading,DTU size,and RMC tone sets.After CA&E,lines transition Start of Joining lines initialgzation 0.5 transition to into showtime.The 60 showtime expected joining time 50 of a single line is sig- nificantly less than in me VDSL2 due to the 0.5 15 shorter probe Time seconds sequence cyce. Figure 4.Joining timeline and examples of precoder and post-processor conversion in a 16-line DPU for tone #1160(60 MHz). crosstalk from joining lines into showtime lines tion of the CPE.Power-saving mechanisms are without disturbing showtime lines.After the pre- discontinuous operation (DO)and low-power coders of showtime lines are updated and the states. downstream crosstalk from joining lines is can- celled,the FTU-O turns on the downstream SOC Discontinuous Operation:DO scales a trans- and sends to the FTU-R the necessary upstream ceiver's power consumption with actual data initialization data.Since crosstalk between join throughput by transmitting only when data is ing lines is not cancelled,data transmitted over available.During the remaining (quiet)symbol the SOC is scrambled using a unique scrambling slots,essential analog and digital processing may seed in each joining line to avoid reception from be turned off,bringing substantial power savings. a non-peer FTU-O.Furthermore,to improve In a vectored group with high crosstalk,turn- robustness,the data is transmitted using repeti- ing off one line may change the direct channel tions and is modulated by an ID-sequence,which of other lines,causing performance degrada- is orthogonal relative to the ID-sequences of tion.Therefore,strict coordination of turning other joining lines. slots quiet is applied across all vectored lines. During R-VECTOR 1,the FTU-R transmits Specifically,each logical frame is divided into a sync symbols modulated by a probe sequence, normal operation interval (NOI,the first TTR and the VCE estimates the upstream channel. slots)and a discontinuous operation interval After upstream crosstalk between all joining and (DOI,the remaining slots).During NOI,no showtime lines is mutually cancelled,a high- quiet slots are allowed,while in DOI the first speed upstream SOC is established to convey TA slots are quiet,and the remaining slots may vectoring feedback for precoder training(during be quiet if no user data is available or active O-VECTOR-2).After O-VECTOR-2,the down- otherwise.In some NOI and DOI slots with stream crosstalk between all joining and show- no data available,an FTU may transmit only time lines is also cancelled. pre-compensation signals (idle slots).The DO During PRM-UPDATE,both FTU-O and parameters TTR,TA,and total number of FTU-R optimize their transmit PSDs in condi- active slots (TBUDGET,Fig.5)are determined tions when crosstalk is cancelled.One goal of by the DRA and VCE based on DPU dynamic optimization is to reduce the transmit power resource allocation (see below).They are con- on tones with extra SNR margin.Another is to figured per logical frame and coordinated across suppress tones generating very high crosstalk(to the vectored group via the RMC.Bit loadings avoid performance loss in showtime lines).Other and transmit PSDs during NOI and DOI are criteria,such as total power reduction,may also updated independently using OLR procedures. be applied [10]. Figure 5 shows an example of downstream DO During the channel analysis and exchange in a four-line DPU.The NOI includes five slots (CA&E)phase,FTUs establish their desired (TTR =5),and values of TA are set so that showtime settings,such as bit loading,DTU size, during DOI only one line is transmitting at a and RMC tone sets.After CA&E,lines transi- time,facilitating crosstalk avoidance;no trans- tion into showtime.The expected joining time of mission in other lines and reduced vector pro- a single line is significantly less than in VDSL2 cessing saves power. due to the shorter probe sequence cycle Power Saving States:During low-power states, POWER SAVING FTUs save power by transmitting only sync sym- Low power consumption is vital for G.fast,driv- bols and RMC symbols in a few assigned TDD en by limited heat dissipation,remote/reverse frames;all other slots are quiet.Furthermore, powering of the DPU,and battery-fed opera- the transmit PSD and number of active tones in 122 IEEE Communications Magazine.March 2016

122 IEEE Communications Magazine • March 2016 crosstalk from joining lines into showtime lines without disturbing showtime lines. After the pre￾coders of showtime lines are updated and the downstream crosstalk from joining lines is can￾celled, the FTU-O turns on the downstream SOC and sends to the FTU-R the necessary upstream initialization data. Since crosstalk between join￾ing lines is not cancelled, data transmitted over the SOC is scrambled using a unique scrambling seed in each joining line to avoid reception from a non-peer FTU-O. Furthermore, to improve robustness, the data is transmitted using repeti￾tions and is modulated by an ID-sequence, which is orthogonal relative to the ID-sequences of other joining lines. During R-VECTOR 1, the FTU-R transmits sync symbols modulated by a probe sequence, and the VCE estimates the upstream channel. After upstream crosstalk between all joining and showtime lines is mutually cancelled, a high￾speed upstream SOC is established to convey vectoring feedback for precoder training (during O-VECTOR-2). After O-VECTOR-2, the down￾stream crosstalk between all joining and show￾time lines is also cancelled. During PRM-UPDATE, both FTU-O and FTU-R optimize their transmit PSDs in condi￾tions when crosstalk is cancelled. One goal of optimization is to reduce the transmit power on tones with extra SNR margin. Another is to suppress tones generating very high crosstalk (to avoid performance loss in showtime lines). Other criteria, such as total power reduction, may also be applied [10]. During the channel analysis and exchange (CA&E) phase, FTUs establish their desired showtime settings, such as bit loading, DTU size, and RMC tone sets. After CA&E, lines transi￾tion into showtime. The expected joining time of a single line is significantly less than in VDSL2 due to the shorter probe sequence cycle. power sAVInG Low power consumption is vital for G.fast, driv￾en by limited heat dissipation, remote/reverse powering of the DPU, and battery-fed opera￾tion of the CPE. Power-saving mechanisms are discontinuous operation (DO) and low-power states. Discontinuous Operation: DO scales a trans￾ceiver’s power consumption with actual data throughput by transmitting only when data is available. During the remaining (quiet) symbol slots, essential analog and digital processing may be turned off, bringing substantial power savings. In a vectored group with high crosstalk, turn￾ing off one line may change the direct channel of other lines, causing performance degrada￾tion. Therefore, strict coordination of turning slots quiet is applied across all vectored lines. Specifi cally, each logical frame is divided into a normal operation interval (NOI, the fi rst TTR slots) and a discontinuous operation interval (DOI, the remaining slots). During NOI, no quiet slots are allowed, while in DOI the first TA slots are quiet, and the remaining slots may be quiet if no user data is available or active otherwise. In some NOI and DOI slots with no data available, an FTU may transmit only pre-compensation signals (idle slots). The DO parameters TTR, TA, and total number of active slots (TBUDGET, Fig. 5) are determined by the DRA and VCE based on DPU dynamic resource allocation (see below). They are con￾fi gured per logical frame and coordinated across the vectored group via the RMC. Bit loadings and transmit PSDs during NOI and DOI are updated independently using OLR procedures. Figure 5 shows an example of downstream DO in a four-line DPU. The NOI includes fi ve slots (TTR = 5), and values of TA are set so that during DOI only one line is transmitting at a time, facilitating crosstalk avoidance; no trans￾mission in other lines and reduced vector pro￾cessing saves power. Power Saving States: During low-power states, FTUs save power by transmitting only sync sym￾bols and RMC symbols in a few assigned TDD frames; all other slots are quiet. Furthermore, the transmit PSD and number of active tones in Figure 4. Joining timeline and examples of precoder and post-processor conversion in a 16-line DPU for tone #1160 (60 MHz). Time, seconds G.994.1 handshake phase G.9701 CA&E phase Joining lines transition to showtime O-VECTOR-1 R-VECTOR-1 O-VECTOR-2 O/R-PRM-UPDATE G.9701 channel discovery phase 0.5 10 Upstream SNR 20 30 40 50 60 1 1.5 2 2.5 3 0.5 Downstream SN 0 R 20 40 60 1 1.5 2 2.5 3 Start of initialization Showtime Joining During the channel analysis and exchange phase, FTUs establish their desired showtime settings, such as bit loading, DTU size, and RMC tone sets. After CA&E, lines transition into showtime. The expected joining time of a single line is sig￾nifi cantly less than in VDSL2 due to the shorter probe sequence cycle