Capacity Analysis of G.fast Systems ViaTime-domain Simulations

Igor Almeida and Aldebaro KlautauSignal Processing Lab. - Federal University of Para

Av. Perimetral S/N, Belem, PA, BrazilEmail: {igoralmeida,aldebaro}@ufpa.br

Chenguang LuEricsson Research

Farogatan 6, 164 80 Kista, SwedenEmail: chenguang.lu@ericsson.com

AbstractThe evolving broadband access systems using coppernetworks are currently deployed in a frequency band thatgoes up to 30 MHz, as specied in VDSL2. As hybrid ber-copper architectures become more important in the industryand academia, using shorter loop lengths (i.e. up to 250 meters)from the last distribution point to users enables adopting evenhigher frequencies to achieve very high data rates of 500 Mbpsand beyond, as is the case with the G.fast standard underdevelopment by ITU-T. In this work, a time-domain simulator hasbeen developed to evaluate G.fast system performance. Systemcapacity is evaluated with different cyclic extension lengthsand different reference loop topologies specied by ITU-T. Thesimulation results show that G.fast systems are robust to bridge-taps and capable of providing very high data rates for allsimulated loop topologies to support next generation ultra highspeed broadband services.

I. INTRODUCTION

Recently, an FTTdp (ber-to-the-last-distribution-point)approach has attracted a lot of attention from both operatorsand system vendors for its high cost efciency in deployingnext generation ultra-fast broadband access (e.g. capable ofproviding more than 500 Mbps services). The FTTdp approachreuses the copper infrastructure from the last distribution pointto the users, which normally consists of up to 250-metertwisted-pair telephony cables. In [1], FTTdp is shown to save80% in investment with respect to the FTTH (ber-to-the-home) approach that requires deploying ber all the way tousers. To support FTTdp, the ITU-T has been developingthe G.fast standard which is basically a TDD (time-divisionduplexing) system with DMT (discrete multitone) modulation.To support very high bit rates, the system will use a 100 MHzbandwidth in the rst release and extend it to 200 MHz inlater versions.

One potential issue that may affect the cost-effectiveness ofG.fast systems is the existence of bridge-taps in home net-works. Bridge-taps can signicantly reduce the line capacitywith increased line attenuation and channel dispersion. Thismay prevent self-install deployment and require truck rollsto remove the bridge-taps in order to achieve the target bitrate, which could signicantly increase the deployment costs.Therefore, it is important to know the impacts of bridge-tapson system capacity. In this work, we focus on analyzing thecapacity of different bridge-tap scenarios specied by ITU-T. As G.fast will support vectoring techniques [2] that can

cancel out crosstalk, single-line capacity is evaluated when nocrosstalk is considered.

To evaluate capacity more accurately, a time-domain simu-lator comprising a DMT transceiver chain has been imple-mented. In G.fast, a very short cyclic prex should be usedto minimize the overhead since the DMT symbol period issmall (about 20 s with a tone spacing of about 50 kHz [3]).Therefore, G.fast can be limited by intersymbol interference(ISI) as the cyclic prex may not be long enough to cover thechannel dispersion. To evaluate the ISI caused by insufcientcyclic prex length, a time-domain simulation is required. Inthis study, capacity results are simulated with different cyclicprex lengths and different loop topologies via this time-domain simulator, improving upon the theoretical derivationsin [4], [5]. The required cyclic prex lengths are also evaluatedfor different loop topologies from a capacity maximizationpoint of view. For example, the simulation results show thatabout 600 Mbps at 100 m can be achieved even with the worstloop topology, which has several bridge-taps. They also showa big spread (0.8 - 2.0 s) in optimal cyclic prex lengths fordifferent bridge-tap scenarios.

The rest of the paper is organized as follows: Section IIpresents the implemented time-domain simulator and describesin detail its key functionalities. In Section III, bit error rate(BER) simulation results with a measured 200-meter cable arepresented to verify the validity of the simulator. Then, SectionIV presents the capacity simulation results with differentcyclic extension lengths and different ITU-T reference looptopologies. The required cyclic extension lengths are given.Finally, Section V nalizes the work with conclusions.

II. G.FAST PHY TIME-DOMAIN SIMULATOR

A time-domain simulator with a DMT transceiver chain hasbeen implemented as shown in Fig. 1. At the transmitter, thebits are modulated to different tones in the frequency domainwith QAM modulation according to a bitloading table. Then,gain scaling is used to ne-tune the transmit power of eachtone and also control the power spectral density (PSD) in thesimulator. After IFFT, the frequency-domain DMT symbolsare translated to a time-domain signal. It is then cyclicallyextended (i.e. cyclic prex and cyclic sufx) to combat ISI. Atthe receiver, a preamble detector has been implemented with atraining sequence as a preamble to acquire the beginning of the

978-1-4673-3122-7/13/$31.00 2013 IEEE

IEEE ICC 2013 - Selected Areas in Communications Symposium

4008

Fig. 1. Block diagram of the simulator.

received signal. After preamble detection, the cyclic extensionis removed. The time-domain signal is translated back to DMTsymbols via FFT, and a frequency-domain equalizer (FEQ) isused to compensate for the attenuation and phase shift causedby the channel. In the end, the bits on different tones aredemodulated by a QAM demapper on each loaded tone.

To generate a real-valued time-domain signal, a DMTsymbol with N tones is rst extended to 2N tones with Ntones in negative frequency by Hermitian symmetry operationbefore the IFFT. Mathematically, the n-th sample of the timedomain signal for the i-th DMT symbol can be expressed as

x(i)R (n) =

2N1k=0

g(k) X(i)R (k) ej2kn/2N (1)

where X(i)R (k) denotes the QAM symbol loaded on tone kand g(k) is the gain scaling factor. Note: X(i)R (N) = 0,X

(i)R (2N k) = X(i)R (k) and g(2N k) = g(k) where k =

1, 2, . . . , N 1 for Hermitian symmetry extension.In this work, both cyclic prex and cyclic sufx are used.

The cyclic sufx is normally used to synchronize downstreamand upstream symbols to minimize the interferences betweendownstream and upstream bands in frequency-division duplex-ing (FDD) systems [6]. Because G.fast is a TDD system, thisshould not be an issue. However, the channel models speciedin the ITU-T reference loops for G.fast are not fully causal. Tofacilitate preamble detection, in this work, a very short cyclicsufx is used to cover the non-causal part of the channels.

For a time-domain simulation, channel impulse responsesare needed instead of the channel transfer function. In thiswork, the channel impulse responses are generated by IFFToperations on the channel transfer function. The receivedsignal at the receiver can thus be modeled as

y(n) = x(n) h(n) + w(n). (2)where x(n) denotes the transmitted signal, h(n) denotes thechannel impulse response, w(n) denotes the noise, and denotes convolution.

In the sequel, the implemented algorithms for timing acqui-sition, FEQ, SNR estimation and bitloading are described.

A. Timing acquisition: preamble detection

Before transmitting data symbols, a sequence of trainingsymbols as a preamble is transmitted to assist the receiver innding the beginning of the received signal. The implementedpreamble detector does a simple threshold-based detection viaa sliding inner product with a known preamble sequence [7]

that consists of a rst DMT symbol whose even tones areloaded with a 4-QAM modulated pseudorandom bit sequenceobtained from a linear feedback shift register (LFSR) with 13bits and a second DMT symbol which has all of its tonesloaded with a similar random sequence.

More specically, assume that r(m) is the received sampleat instant m, p(i) is the i-th sample of the preamble sequence,P is its total number of samples and T is the detectionthreshold. The detector calculates the inner product S(d)between the last received P samples and the known preamblesequence as:

S(d) =

P1i=0

r(d+ i P + 1) p(i). (3)

The cyclic prex of the rst DMT symbol is said to beginat instant d0 if

S(d0) T and S(d) < T, d < d0. (4)B. Frequency-domain equalization

The FEQ is basically used to compensate the channel effects(i.e. phase shift and attenuation) by inverting the channelcoefcients. After the FEQ, the received signal R(k) at tonek can be expressed as

R(k) = F (k) (g(k)T (k)H(k) +W (k) + I(k)) (5)

where F (k) is the FEQ coefcient at tone k, T (k) is thetransmitted signal, H(k) is the channel coefcient, W (k)denotes the noise component due to background noise andI(k) denotes the ISI component due to insufcient cyclicprex [8]. Ideally, F (k) = 1/g(k)/H(k).

To calculate F (k), a normalized least mean squares (NLMS)algorithm [9] has been implemented as

F (i+1)(k) = F (i)(k) + e(i)k R(i)(k)

|R(i)(k)|2 (6)

where e(i)(k) = T (i)(k)F (i)(k) R(i)(k) denotes the errorin tone k between the i-th transmitted and received trainingsymbol, () denotes complex conjugate and is the LMS stepsize. After a relatively long sequence of a thousand trainingsymbols, the NLMS algorithm converges closely to optimalcoefcients.

C. SNR Estimation

After the FEQ is trained, the SNR estimation processstarts. A different 4-QAM modulated pseudorandom trainingsequence obtained from an LFSR with primitive polynomial

4009

x13 + x12 + x11 + x8 + 1 is sent for L DMT symbol periodswith g(k) = 1. The SNR per tone is estimated after FEQ usingthe modulation error ratio (MER) [10]:

SNR(k) =P (k)

2(k)(7)

where 2(k) = 1LL1

i=0 |R(i)(k) T (i)(k)|2 is the noisepower estimate and P (k) = 1 is the power of the received4-QAM training symbols.

D. Bitloading

Once the SNR estimation subphase is nished, the Levin-Campello discrete bit loading algorithm described in [11] isexecuted to obtain a solution to the bit rate maximizationproblem [12]: given a total energy constraint ET , the algorithmoutputs an N -tuple b ZN+ that optimally distributes the bitsamong N tones.

The algorithm uses the concept of required energy ek(b) totransmit b bits of information in tone k. Because it is equallyvalid to use SNR relations whenever energy is stated in thealgorithm, we set

ek(b) =(b)

SNR(k) mc

(8)

where (b) is the pre-computed required SNR to achieve thetarget error rate for 2b-QAM [13] in AWGN, m is the SNRmargin and c is the total coding gain in the system [14].

To nd the optimal discrete solution, the following inequa-lities must hold:

ek(bk) ej(bj + ) k, j = 0, 1, 2, . . . , N 1, (9)

0 ET N

k=1

ek(bk) < minj

ej(bj + ), (10)

where is the information granularity and

ek(b) =

{ek(b) ek(b ) b ek(b) ek(0) b <

(11)

is the incremental energy required to transmit b bits in tone kwith respect to the energy required to transmit b bits inthe same tone.

Iteratively applying Equation (9) to an N -tuple, one obtainsa bit distribution b(e)k that is efcient, i.e. no movement ofbits from one tone to another reduces the symbol energy. Thecorresponding gain scaling vector g(e)(k) is obtained by

g(e)(k) =

ek(b

(e)k ). (12)

From b(e)k and g(e)(k), iteratively removing bits from the

most energy-consumptive tones until Equation (10) is satisedrenders a bit distribution b(t)k that is E-tight, i.e. no additionalunit of information can be carried without violation of thetotal energy constraint ET . The gain scaling vector g(t)(k) iscalculated in the same way as g(e)(k).

The E-tight pair b(t)k and g(t)(k) are used as the number

of bits and the gain scaling factor on tone k when the Levin-Campello algorithm converges.

III. BER EVALUATION ON MEASURED 200 M CABLETo verify the simulator, BER is evaluated with a measured

200-meter 0.5 mm cable over 200 MHz bandwidth. Theevaluation is done by comparing the target BER and thesimulated BER. Fig. 2 shows the impulse response of themeasured channel. It can be observed that the samples beforethe peak are very small but not zeros, which corresponds to thenon-causal part of the channel an artifact due to the imperfectmeasurement conditions. To illustrate preamble detection, thepreamble detection point is also marked in Fig. 2. It can beseen the detection point does not cover all non-causal samples.However, this can be easily solved with a very short cyclicsufx in this case 25 samples are sufcient.

Fig. 2. Impulse response of measured 200 m cable.

The simulation was set to have 2048 loaded tones andtone spacing of 48.84 kHz (thus having 100 MHz of systembandwidth), a minimum of 1 and a maximum of 12 bits pertone, no forward error coding mechanisms, an SNR marginof 0 dB and a target bit error rate of 107 for calculating thebitloading table. Furthermore, the background noise was setto -140 dBm/Hz and the system used a at transmission PSDat -76 dBm/Hz with an energy constraint of 0 dB per tonerelative to the PSD.

Fig. 3 compares the bitloading tables of two extreme caseswhen using a very long cyclic extension (5.09 s) and avery short extension (0.39 s). The long extension coversmore than 99.99% of the channel energy and therefore ISI isnegligible. As expected, due to the ISI caused by insufcientcyclic extension, the bit loading up to 4 MHz in Fig. 3b issignicantly reduced with respect to that using the long cyclicextension.

The achieved BER of the system with long cyclic extensionis shown in Fig. 4a, reaching an average of 1.138 107, whilethe same metric with short cyclic extension is shown in Fig. 4bhaving an average of 7.911108. Both results are very close tothe target BER, thus showing that the implemented simulatorcorrectly achieves the expected BER performance with bothshort and long cyclic extensions. Note: BER at several tonesare zero and thereby not visible, and the slight increase inBER in higher frequencies could be caused by the imperfectFEQ training due to lower SNR in these tones. Overall, thiscan be ignored for the purpose of this study.

IV. CAPACITY SIMULATION RESULTSSystem capacity is a key performance indicator for a com-

munication system. For G.fast, it is important to evaluate how

4010

(a) Bitloading table with long cyclic extension (5.09 s).

(b) Bitloading table with short cyclic extension (0.39 s).

Fig. 3. Results for bitloading.

(a) BER with long cyclic extension (5.09 s).

(b) BER with short cyclic extension (0.39 s).

Fig. 4. Calculated bit error rate per tone.

much the system capacity is impacted by different loop topolo-gies, especially bridge-taps in home wiring. As mentioned,the bridge-taps can increase both the line attenuation and thechannel dispersion, which may signicantly reduce the systemcapacity. In this section, the system capacity is studied withthe reference loops specied in ITU-T for G.fast. The mainfocus is on evaluating the impact of different cyclic extensionsin the system capacity.

A. Approach

The basic approach is to use the time-domain simulator tocalculate the bitloading table with assumed system parameterssuch as SNR margin, cyclic extension and coding schemes(i. e. coding gain and overhead). As the simulator was alreadyveried via the BER evaluation described in the previoussection, no time-consuming BER simulation is needed forcapacity evaluation. This saves a considerable amount of

simulation time and therefore speeds up the study on systemcapacity.

After the bitloading table is obtained, the system capacityR can be calculated as

R =RcTsym

N1k=0

bk (13)

where Tsym = Ts(LCE + 2N) and Ts are the DMT symboland sampling periods, respectively, Rc is the total coding rateand LCE is the cyclic extension length in samples. If Reed-Solomon and trellis coding are used as in VDSL2, the totalcoding rate can be calculated as follows:

Rc = RRS RTC (14)where RRS and RTC are the coding rate of the Reed-Solomoncode and the trellis coding, respectively. The two rates can becalculated as RRS = KRS/NRS, where NRS and KRS are theReed-Solomon codeword and data lengths, respectively, and

RTC = 1 1NL

N1k=0

tovh(k) (15)

tovh(k) =

{0.5/bk bk > 0

0 otherwise(16)

where NL is the number of loaded tones, i. e. with non-zerobitloading, and tovh(k) accounts for the overhead of Weis 4-dimensional 16-state trellis code used in VDSL2.

B. Simulated ITU-T reference loops

To facilitate the standardization work, ITU-T has speciedreference loops for G.fast evaluation [15]. Fig. 5 shows thereference topologies. The loop lengths are up to 250 meters.Topologies 1 and 2 are based on the CAD55 cable modelfor a typical drop cable in the UK [16]. Topology 2 has ashort bridge-tap at the home wiring, while Topology 1 is astraight line. Topologies 3 and 4 are based on the ANSI TP1cable model of 26 AWG copper cables normally used in USnetworks as drop cables [17]. Topology 4 represents the worstchannel with 5 bridge-taps connected at the same point.

Fig. 5. Topologies simulated in the Contribution presented to ITU-T.

It should be noted that, because both the CAD55 and theANSI TP1 model are based on the BT0 model, which is non-causal, timing detection had to cover the non-causal precur-sors. This potentially increases the required cyclic extension,

4011

TABLE ISIMULATION PARAMETERS.

Channel bandwidth 200 MHzTone spacing 48.84 kHz

Data bandwidth 100 MHzNumber of loaded tones 2048

FFT size 8192Signal PSD -76 dBm/Hz

Background noise PSD -140 dBm/HzEnergy constraint 0 dB per tone relative to PSD

Coding Reed-Solomon + Trellis codingReed-Solomon N/R = 255/16Trellis coding overhead of 0.5 bit per tone

Total coding gain 5 dBSNR gap 9.75 dB

SNR margin 6 dBBit loading 1-12 bits

Cyclic extension length 0.4, 0.8, 1.2, 1.6, 2.0, 2.4 sLoop length (DP to terminal) 50, 100, 150, 200, 250 m

which in real cables can be made even shorter since there isno non-causality.

C. Simulation setup

The system parameters used in the simulation are listed inTable I. The system bandwidth is assumed to be 100 MHz,in-line with the G.fast specication. When generating chan-nel impulse responses from channel transfer functions infrequency-domain, it is important to reduce the oscillation inthe time domain (especially for short loops) due to the implicitrectangular window in the frequency domain. To mitigate this,the system is oversampled twice up to 200 MHz and thesimulated channel impulse responses are generated from thechannel transfer functions with a 200 MHz bandwidth and arethen ltered using a 53-tap Kaiser lter with transition bandbetween 100 and 120 MHz, which also emulates the effects ofthe transmit and receive lters in practice. Fig. 6 shows howoscillation effects are mitigated by the ltering.

Fig. 6. A 50-meter CAD55 channel impulse response generated by 100 MHzchannel bandwidth without additional ltering compared to a ltered channelimpulse response using a 200 MHz bandwidth. Note: the coefcients arenormalized and shifted for comparison purpose.

D. Optimal cyclic extension lengths

The capacity results of Topology 1 and 4 are shown inFig. 7, where f is the start frequency of G.fast, to achievespectral compatibility with legacy ADSL2/2+ (f=2.2 MHz)and VDSL2 (f=17 MHz, f=30 MHz) systems. Fig. 7a showsthat the capacity decreases as the cyclic extension increases.Topologies 2 and 3 also show the same trend due to limitedspace, their results are not shown. This indicates that thechannel dispersion of straight lines (Topologies 1 and 3) andof a topology with one short bridge-tap (Topology 2) is smalland therefore can be well handled by a very short cyclicextension (i.e. 0.8 s). However, as shown in Fig. 7b, Topology4 requires much longer cyclic extensions (e.g. 2.0 s) dueto longer channel dispersion caused by the worst bridge-tapcondition of ve bridge-taps.

(a) Topology 1 capacity.

(b) Topology 4 capacity.

Fig. 7. Cyclic extension length impact on capacity for Topologies 1 and 4.

TABLE IIBEST CYCLIC EXTENSION LENGTHS (IN MICROSECONDS).

Start freq. 2.2 MHz 17 MHz 30 MHzTopology 1 0.8 0.8 0.8Topology 2 0.8 0.8 0.8Topology 3 0.8 0.8 0.8Topology 4 2.0 1.6 1.2

Table II summarizes the optimal cyclic extension lengths fordifferent start frequencies and different topologies. It shows

4012

the spread ranges from 0.8 to 2.0 s, corresponding to 3.76%and 8.9% overhead, respectively. For Topology 4, the optimalcyclic extension is decreased as the start frequency increases.This is because the ISI concentrates in lower frequencies.Therefore, higher frequencies are much less affected, as shownalso by the bit loading results in Section III.

E. Rate-reach results

Fig. 8 shows the rate-reach results (bit rate versus looplength) with the optimal cyclic prex lengths for all simulatedtopologies and with start frequencies of 2.2 and 17 MHz. Itshows the ability of G.fast systems to achieve very high bitrate even for the topologies with bridge-taps. Compared toTopologies 1 and 3, the bridge-tap(s) in Topologies 2 and4 reduce the rate signicantly. For example, at 100 m with2.2 MHz starting frequency, more than 300 Mbps bit ratereduction can be seen for Topology 4 and more than 100 Mbpsreduction can be seen for Topology 2. However, the systemcan still achieve about 350 Mbps at 150 m for Topology 4(the worst topology). For both straight line topologies withoutbridge-taps (Topologies 1 and 3), the system can even achieveabout 400 Mbps at 200 m and about 900 Mbps at 100 m.

As shown in Fig. 8, if 17 MHz legacy VDSL2 systemsare present in the same cable bundle, backing off the startingfrequency to 17 MHz reduces the bit rate. The system canstill maintain quite high bit rates, though: for Topologies 1-3,more than 400 Mbps can be achieved at 150 m. For the worstTopology 4, more than 250 Mbps can be achieved.

(a) Capacity decrease in Topologies 1 and 2.

(b) Capacity decrease in Topologies 3 and 4.

Fig. 8. Loop length impact on capacity.

V. CONCLUSIONSIn this work, capacity results of a 100 MHz G.fast system

are obtained by a time-domain simulator. The implementedalgorithms in the simulator are rst veried by BER evaluationusing a measured 200-meter cable. Then, capacity results arepresented with respect to different cyclic prex lengths anddifferent reference loop topologies specied by ITU-T.

It was shown that the optimal cyclic extension length isbetween 0.8 and 2.0 s for different loop topologies. Themost dispersive topology with several bridge-taps requires2.0 s, while 0.8 s is good enough for straight line andsingle bridge-tap scenarios. The results also show that G.fastis capable of providing very high bit rates to support nextgeneration ultra high speed broadband services. In the normalcases, for straight line topology and one bridge-tap topology,more than 500 Mbps can be achieved at 150 m. Even in theworst case, for the most difcult topology, the system can stillachieve more than 350 Mbps at 150 m. This indicates that thesystem is robust to bridge-taps, which will help reduce G.fastdeployment cost signicantly by avoiding truck rolls.

ACKNOWLEDGMENTThis work was partially supported by the Innovation Center,

Ericsson Telecomunicacoes S.A., Brazil.

REFERENCES[1] P. Odling, T. Magesacher, S. Host, P. O. Borjesson, M. Berg, and

E. Areizaga, The fourth generation broadband concept, Communica-tions Magazine, IEEE, vol. 47, no. 1, pp. 62 69, january 2009.

[2] G. Ginis and J. Ciof, Vectored transmission for digital subscriber linesystems, IEEE Journal on Selected Areas in Communications, vol. 20,no. 5, pp. 10851104, june 2002.

[3] Editor for G.fast, G.fast: Updated Issues List for G.fast, ITU-T Q4/15TD 758R2, Tech. Rep., Sep 2012.

[4] Lantiq, G.fast: On the G.fast symbol duration and guard interval, ITU-T Q4/15 Contribution 4A-050, Tech. Rep., Feb 2012.

[5] J. Maes, M. Guenach, K. Hooghe, and M. Timmers, Pushing thelimits of copper: Paving the road to FTTH, in Communications, IEEEInternational Conference on, june 2012, pp. 3149 3153.

[6] P. Golden, H. Dedieu, and K. Jacobsen, Fundamentals of DSL Technol-ogy. Auerbach Publications, 2006.

[7] T. Schmidl and D. Cox, Robust frequency and timing synchronizationfor OFDM, Communications, IEEE Transactions on, vol. 45, no. 12,pp. 1613 1621, dec 1997.

[8] W. Henkel, G. Taubock, P. Odling, P. Borjesson, and N. Petersson, Thecyclic prex of OFDM/DMT - an analysis, in Broadband Communica-tions, International Zurich Seminar on, 2002, pp. 221 223.

[9] S. Haykin, Adaptive Filter Theory, 3rd ed. Prentice Hall, 1996.[10] Y. Wang, Z. Chen, and K. Gong, MER performance analysis of M-

QAM-OFDM with Wiener phase noise, in ICMMT07, april 2007, pp.1 4.

[11] J. Campello, Practical bit loading for DMT, IEEE ICC99 Proceed-ings, pp. 801805, 1999.

[12] T. Starr, M. Sorbara, J. M. Ciof, and P. J. Silverman, DSL Advances.Prentice-Hall, 2003.

[13] ITU-T Rec. G.9960-2010, Unied high-speed wire-line based homenetworking transceivers - System architecture and Physical Layer Spec-ication, 2010.

[14] T. Starr, J. M. Ciof, and P. J. Silverman, Understanding DigitalSubscriber Line Technology. Prentice-Hall, 1999.

[15] Editor for G.fast, G.fast: Wiring topologies and reference loops, ITU-TQ4/15 Contribution 11GS3-100, Tech. Rep., Sep 2011.

[16] BT plc, G.fast: Cable models, ITU-T Q4/15 Contribution 11RV-026,Tech. Rep., Nov 2011.

[17] ANSI Standard T1.417-2003, Spectrum management for loop transmis-sion systems, 2003.

4013

/ColorImageDict > /JPEG2000ColorACSImageDict > /JPEG2000ColorImageDict > /AntiAliasGrayImages false /CropGrayImages true /GrayImageMinResolution 200 /GrayImageMinResolutionPolicy /OK /DownsampleGrayImages true /GrayImageDownsampleType /Bicubic /GrayImageResolution 300 /GrayImageDepth -1 /GrayImageMinDownsampleDepth 2 /GrayImageDownsampleThreshold 1.50000 /EncodeGrayImages true /GrayImageFilter /DCTEncode /AutoFilterGrayImages false /GrayImageAutoFilterStrategy /JPEG /GrayACSImageDict > /GrayImageDict > /JPEG2000GrayACSImageDict > /JPEG2000GrayImageDict > /AntiAliasMonoImages false /CropMonoImages true /MonoImageMinResolution 400 /MonoImageMinResolutionPolicy /OK /DownsampleMonoImages true /MonoImageDownsampleType /Bicubic /MonoImageResolution 600 /MonoImageDepth -1 /MonoImageDownsampleThreshold 1.50000 /EncodeMonoImages true /MonoImageFilter /CCITTFaxEncode /MonoImageDict > /AllowPSXObjects false /CheckCompliance [ /None ] /PDFX1aCheck false /PDFX3Check false /PDFXCompliantPDFOnly false /PDFXNoTrimBoxError true /PDFXTrimBoxToMediaBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXSetBleedBoxToMediaBox true /PDFXBleedBoxToTrimBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXOutputIntentProfile (None) /PDFXOutputConditionIdentifier () /PDFXOutputCondition () /PDFXRegistryName () /PDFXTrapped /False

/CreateJDFFile false /Description >>> setdistillerparams> setpagedevice