Pushing the limits of copper - Semantic Scholar

3 downloads 2529 Views 184KB Size Report
delivered data rates due to fierce competition with cable operators and due ... Telecommunications Union (ITU) has created a new study item called G.fast (Fast ...
IEEE ICC 2012 - Selected Areas in Communications Symposium

Pushing the limits of copper Paving the road to FTTH

Jochen Maes, Mamoun Guenach, Koen Hooghe, and Michael Timmers Bell Labs Alcatel-Lucent Antwerp, Belgium {jochen.maes,guenach}@ieee.org, {koen.hooghe,michael.timmers}@alcatel-lucent.com

Abstract—The ever growing demand for higher data rate is pushing existing electrical broadband communication systems, using coax or twisted pair cabling, to the limit. While optical access systems inherently have a much higher transmission capacity, upgrading the access network entirely to fiber remains a costly and time-consuming effort. Therefore, many operators deploy fiber gradually closer to the end user, while bridging the remaining distance through copper technologies such as VDSL2. In the next step of this evolutionary path, fiber is brought to distribution points that are in the range of no more than 200 m from the end user, with a copper access technology called G.fast delivering data rates of 500 Mb/s to 1 Gb/s over the remaining copper stretch. In this paper, we give an overview of the G.fast technology that is currently being standardized in ITU. We show the feasibility of its main technical requirements, namely i) powering of the access equipment through the customer premises modem, ii) high transmission efficiency for orthogonal frequency division multiplexing over short copper loops and iii) offering high net data rates to end users. Keywords- Subscriber loops; DSL; Power transmission; OFDM modulation; access protocols

I.

INTRODUCTION

Operators that use the copper telephony network for offering broadband services are incented to increase the delivered data rates due to fierce competition with cable operators and due to government broadband plans. By gradually deploying fiber deeper into their network, the access nodes (ANs) are brought closer to the end-users, hence shortening the copper loop to lengths typically below one km and increasing their capacity. Provisioning vectored Very high speed Digital Subscriber Line (VDSL2) technology on these shortened loops, allows to realistically achieve aggregate data rates of 50 (on relatively longer loops) to 200 Mb/s (on the shorter loops) [1]-[2]. In comparison, other access technologies like the latest coaxial transmission technology, Data Over Cable Service Interface Specification (DOCSIS) can achieve data rates in excess of 200 Mb/s.

longer time period [3]. Recently, the International Telecommunications Union (ITU) has created a new study item called G.fast (Fast Access to Subscriber Terminals) with the goal to define a standard that addresses the operators’ needs by the end of 2012. While stakeholders are converging on a number of system characteristics and high level requirements, technology options for the physical layer scheme are still open. Since G.fast is a project in process, requirements may be altered, removed, added or become optional in the future, and may become substantially different from what is addressed in this paper. Performance analysis (see further in this paper) shows that, when using orthogonal frequency division multiplexing (OFDM), aggregate data rates of 200 Mb/s to 1 Gb/s are within reach on copper loops of 30 to 200 m. From deployment point of view, the active equipment will be located in a node at one of the distribution points nearest to the endusers, which can be a pole, an underground utility vault (manhole), a small cabinet or inside a building. Each of these nodes will be serving a limited number of end-customers, about 10 or 20. Hence, these nodes will need to be abundantly present to cover the entire access network. For this reason, and due to space limitations at the last distribution point, the targeted system architecture is distributed, with a central access node (CO – central office) serving multiple remote equipment (RE) for line termination as indicated in the example in Fig. 1. Further, the RE’s are reverse powered from the customer premises equipment in order to avoid the need for dedicated

While fiber is clearly an optimal solution in terms of throughput, its deployment is slow, due to the high capital investment and the massive amount of construction work involved. Operators are therefore seeking short-to-medium term solutions to enable fiber-like speeds on the copper network and allow them to spread fiber investments over a

This work is supported by IWT through project PHANTER.

978-1-4577-2053-6/12/$31.00 ©2012 IEEE

3149

Passive splitter

RE

CPE

CO

RE G.fast transceiver

CPE Power extraction RE power supply unit

60 Vdc

Power insertion

G.fast transceiver

Power source

Figure 1. In the distributed architecture an access node at the central office serves multiple remote equipments (REs). The inset shows the reverse powering of the REs by the customer premises equipment.

mains power at every remote location (Fig. 1, inset).

20

On the system level, one novel aspect of G.fast is the desire to power the RE by the CPE over the copper line, called reverse power feeding. In Section II we evaluate the feasibility of reverse power feeding. On the physical layer level, one novel aspect of G.fast is the exploitation of higher frequencies. In Section III the feasibility of exploiting higher frequencies and the implications on OFDM modulation parameters are discussed. Finally, in Section IV we provide a performance evaluation based on realistic assumptions and the results from Section III.

18 16

Load power (W)

14 12 10 8 6

II.

POWER TRANSFER CAPABILITIES

Reverse power feeding of the access equipment is a new requirement not present in previous DSL technologies. The power is transferred from the CPE to the RE in dc while the broadband signal is present (Fig. 1). The mechanism proposed for power transfer is a 60 V Remote Feeding Telecommunication circuit (RFT-V [4]) with a maximum current of 0.3 A [5]. This allows the source to provide up to 18 W. The 60 V source is the maximum allowed in safety extra-low voltage (SELV) systems, and avoids the need for additional safety precautions. The low voltage is also desired to avoid flash-over in older paper insulated wire types that have low dielectric strength. The power is transferred in dc over the two wires of the twisted pair by applying the voltage only on one wire, while grounding the other wire. The voltage is kept negative to avoid electrolysis at places where the insulation is weak or damaged. Due to the resistance of the loop, the voltage seen by the load (the RE) and hence the available power at the load, is lower than provided by the source. We evaluate the load power for various typical and worst case loop conditions. At temperature T = T0 + 'T and wire diameter d, the resistance in Ohms of a twisted pair is:

Rline

2L

U (1  D'T ) d S ( )2 2

.

(1)

Here, U = 1.68 10-8 m is the resistivity of copper at T0 = 20°C and D = 0.0039 K-1 is the temperature coefficient. The factor 2L comes from the fact that for a twisted pair of length L the length of wire in the electrical chain is 2L. The wire gauge typically ranges from 0.4 to 0.9 mm. We consider a typical condition to be 0.5 mm copper wiring at a temperature of 20°C. The largest transfer of power into the load with resistance Rload occurs when Pload = I2Rload is maximized, where I is the current. For a given line resistance Rline and given source voltage Vsource, the power delivered into the load is:

Pload

§ Vsource ¨¨ © Rline  Rload

2

· ¸¸ Rload . ¹

0.9 mm 0.5 mm 0.4 mm 0.4 mm 0.4 mm

4

(2)

2 0

0

50

Cu at 20°C Cu at 20°C Cu at 20°C Cu at 65°C Al at 65°C

100

150 200 250 Loop length (m)

300

350

400

Figure 2. The G.fast RE needs to work on a budget of 7.2 W for a realistic scenario.

The available power at the load varies significantly with loop and ambient conditions (Fig. 2). While a budget in excess of 10 W is available for loops thicker than 0.5 mm and shorter than 400 m, the power budget drops to 7.2 W per line for a 400 m by 0.4 mm copper loop at elevated ambient temperature. A G.fast RE must therefore be designed to operate below this per-line budget for any number of active lines. Hence, it should consume less than this budget, when only a single line is active, which will strongly impact the hardware architecture of the REs. To achieve this, the power consumed by common functionalities such as the optical uplink to the access node, must be sufficiently low and may need to scale with the number of active lines. This is feasible by moving the interworking function to the AN, or making it scalable and integrate it in the digital signal processor (DSP). The bottom curve in Fig. 2 shows the budget available on aluminum twisted wiring, which has been deployed by some operators during eras where copper prices were high. In general, multiple copper lines will simultaneously power the RE. On the one hand, the power transfer needs to be efficient as to minimize Ohmic losses. On the other hand, fairness amongst end-users is desired to ensure that the power provided by any given end-user is proportional to the power needed for that line’s functionalities in the RE. Three strategies for balancing the power delivered by K end-users are assessed. In the first, each end user provides the same power at the CPE source (‘Equal Psource’). This implies that the load power contribution from longer lines will be less than that of shorter lines due to the higher resistance of the longer lines. In the second alternative, each end user provides the same power at the RE load (‘Equal Pload’). While each end user provides a fraction 1/K of the total required power at the load, the longer lines will have a significantly higher Psource than the shorter lines. In the third strategy (‘Efficient’), the power is drawn from the multiple lines in the most efficient way, by adapting the source power to the loop conditions. While short lines will provide relatively more power than long

3150

schemes. The digital complexity increase associated with utilizing a wider bandwidth has shown to be feasible based on Moore’s law scaling [7]. As we show here, the transmission efficiency of OFDM remains high when the cyclic extension (CE) is adapted to the loop conditions. The CE reduces intersymbol interference (ISI) of consecutive OFDM symbols, allows digital duplexing of down and upstream symbols and reduces out of band energy through the addition of a windowing operation within the CE. One of the factors that define the required CE length is the power delay profile (PDP) of the signal, which is related to the channel impulse response h(t) by:

0.4 mm Al at 65°C

0.4 mm Cu at 65°C

0.4 mm Cu at 20°C

0.5 mm Cu at 20°C Equal Psource Equal Pload

0.9 mm Cu at 20°C

Efficient 0

0.2

PDP(t ) 0.4 0.6 Power transfer efficiency

0.8

1

h(t )

2

³ h( x) dx 2

.

(4)

Figure 3. Comparison of the power transfer efficiency under three strategies.

lines, the overall efficiency is optimum. The power transfer efficiency is defined as K

¦P

load

(k )

k 1 K

(3)

¦P

source

(k )

k 1

Here, the indek k indicates the source or load power for user k = 1 to K. The three strategies are compared in Fig. 3, assuming the RE requires Kx7.2 W and uniformly distributed loop lengths between 40 and 400 m. The number of lines used for simulation is K = 10, each requiring 7.2 W. Depending on loop conditions, the power transfer efficiency of the ‘Efficient’ strategy is 1 to 15 percentage point (pp) higher than that of the ‘Equal Psource’ strategy. The worst strategy is ‘Equal Pload’ as it forces all lines to deliver the same amount of power at the load irrespective of their loop lengths which is clearly very inefficient for longer lines. Moreover, this strategy fails for the worst case loop condition because the longest loops are not able to provide the required Psource.

The PDP contains information on the signal delay. The arrival delay is defined as the duration between the transmission of a signal and the arrival of the first significant component of the signal at the receiver. It is the earliest time instance at which the PDP exceeds a pre-defined threshold, often taken around -40 dB. The total delay is the last time instance at which the PDP remains above that same threshold. The excess delay is a measure of the duration of the impulse and is the difference between the total delay and the arrival delay. These three quantities are shown in Fig. 4 for a 24 awg loop. In VDSL2, frequency division duplexing (FDD) is used, implying that the CE must be at least in the order of the total delay to ensure digital duplexing. A time division duplexing (TDD) approach is selected in which multiple consecutive symbols are transmitted in the same communication direction (either upstream or downstream). TDD alleviates the need for digital duplexing. In that case, the CE must be at least in the order of the excess delay to avoid ISI. By moving from FDD to TDD, the CE overhead attributed to symbol alignment can 4 Arrival delay Excess delay Total delay

3.5 3

TRANSMISSION EFFICIENCY

One of the most specific data rate requirement for G.fast expressed up to now can be formulated as a net aggregate service rate of 500 Mb/s that must be sustained using frequencies above 17.7 MHz over 50 m of 0.5 mm cable [6]. The aggregate traffic capacity limit should not be less than 1 Gb/s. Achieving these rates on a copper loop requires scaling up modulation and coding as compared to current DSL technology. Taking into account limitations on aggregate transmit power, spectral compatibility and equipment noise floor, the targeted capacity can only be achieved by utilizing a wider spectrum that spans significantly above 30 MHz. OFDM is selected for the next generation physical layer due to its multiple benefits as compared to other modulation

2.5 Time (μs)

III.

2 1.5 1 0.5 0

0

50

100

150 200 250 Loop length (m)

300

350

400

Figure 4. The cyclic extension length must be optimized to mitigate the effects of signal dispersion.

3151

around 0.5 to 1.5% of windowing overhead. For larger E, the inefficiency is dominated by the window length, while for smaller beta, the inefficiency is dominated by an increased number of vacated carriers. The efficiency at the optimal CE depends on the modulation parameters and ranges from 96 to 99%. Combined with the CE overhead required to avoid ISI, OFDM provides a transmission efficiency of 90 to 95% for the evaluated modulation parameters.

1 17 kHz, 17 kHz, 34 kHz, 34 kHz,

0.99 0.98

140 MHz 70 MHz 140 MHz 70 MHz

Efficiency

0.97 0.96 0.95

IV.

0.94 0.93 0.92 0.91 0.9

0

0.01

0.02

0.03

0.04

0.05 beta

0.06

0.07

0.08

0.09

0.1

Figure 5. Due to the trade-off between efficient notching and the windowing overhead, an optimal  exists that maximizes the efficiency for a specific scenario.

thus be halved. The actual CE overhead depends on the symbol duration, and thus on the carrier spacing. For a symbol rate of 16 and 32 kHz, a 2 μs CE corresponds to an overhead of 3.2 % and 6.4% respectively. Note that a time domain gap to compensate the delay spread on the subscriber loops must be introduced when the transmission direction changes in a TDD based scheme. This serves to avoid impact from echo and near end crosstalk. Frequent changes in transmission direction are needed to ensure low latency, which could limit the benefits of TDD. Spectral confinement is required near band edges and near notches that may be introduced to avoid the impact of ingress from or egress to other communication systems operating at the notched frequencies, e.g. the FM band. Spectral confinement is ensured by applying a windowing function to the CE prior to transmission. Typically this function is a raised-cosine window in the time domain. The spectral confinement increases with the duration of the window in the time domain. As the symbol duration decreases with carrier frequency, the windowing overhead will increase with increasing carrier frequency for a given spectral roll-off. Due to the finite window length, the spectral confinement is limited and a number of carriers may need to be vacated near band edges or notches. The number of vacated carriers and hence the bandwidth loss depends on the windowing overhead. The windowing efficiency, including CE overhead and bandwidth loss due to vacated carriers, is given in Fig. 5 as a function of the window overhead Efor different carrier spacings and utilized bandwidths. We assume 10 notches are present over the entire frequency band. Depending on the windowing overhead, a number of carriers needs to be vacated near each notch edge to restrict the power spectral density (PSD) within the notch due to spectral leakage. The PSD restrictions within the notches are taken from [8]. The efficiency includes the windowing overhead as well as the capacity loss due to vacated carriers. A maximum is observed

PERFORMANCE EVALUATION

On loops up to 400 m and with a gauge of 0.5 mm the attainable data rate is evaluated under different bandwidth utilization. The lowest starting frequency considered is 138 kHz, corresponding to the start of the downstream band of ADSL. An alternative starting frequency is 17.7 MHz, to allow co-existence of G.fast with deployed VDSL2 17 MHz band plans. Two upper frequencies are considered: 100 MHz and 141 MHz. The 100 MHz upper frequency fmax is currently the highest frequency used by standardized in-home powerline or phone line communication systems [9]. The fmax = 141 MHz upper limit is chosen to be a power of two multiplied with the carrier frequency used in ADSL. In the numerical evaluation, the carrier spacing fc depends on fmax (see Table I). However, the obtained results still hold if a common carrier spacing is selected. To keep the aggregate transmit power low, the PSD is kept below -76 dBm/Hz. The PSD is notched on frequencies in use by FM radio (87.5 to 108 MHz), on AM bands identified in [10], on emergency channels and on AM bands proposed for fixed notching in [11]. These notches are applied to obtain realistic performance evaluation: without notch, the ingress into the copper line may prevent full exploitation of these frequencies. A gap to capacity of * = 10.75 dB is typical for today’s DSL deployments and includes a 6 dB noise margin. The efficiency of K = 78.5% includes the transmission efficiency defined in Section III and coding overhead. The maximum bits loaded per QAM constellation is conservatively taken at bmax = 12 bits due to possible limitations of high bandwidth ADC components. The different settings in the numerical evaluation are summarized in Table I.

3152

TABLE I.

SIMULATION PARAMETERS

Parameter

Value

Binder type

0.5 mm poly-ethylene insulated (PE05)

PSD limit

-76 dBm/Hz

Notches

AM, FM and Emergency channels

Carrier fc

Spacing

48.828 kHz if fmax = 100 MHz 17.250 kHz if fmax = 141 MHz

Noise floor 02

-135 dBm/Hz

XdB

-6dB (below 99% worst case model)

*

10.75 dB

K

78.5%

bmax

12 bits

The net data rate R, aggregated over upstream and downstream, is derived as N 1

2

R Kf c ¦ min(log 2 (1  n 0

H (n) PSD(n) *(V 02  I (n))

), bmax ) .

Here, |H(n)|2 is the direct channel response evaluated at frequency nfc. The direct channel is provided by the analytical PE05 transmission line model of [12] describing the primary parameters of a type of 0.5 mm gauge twisted pairs generally used in performance testing. The crosstalk interference I(n) is derived from the 99% worst case model, reduced by XdB = -6 dB [13]. This offset allows evaluating the main rate requirement in [6], which stipulates that 500 Mb/s should be achieved on a 50 m loop using frequencies above

1200 0 to 100 MHz no XT 0 to 141 MHz no XT 18 to 100 MHz no XT 18 to 141 MHz no XT 18 to 100 MHz with XT 18 to 141 MHz with XT Vectored VDSL2

Aggregate data rate (Mb/s)

1000

800

simulation conditions, but with a PSD of -60 dBm/Hz and bmax = 15 bits, the curve labeled ‘Vectored VDSL2’ in Fig. 6 shows that a G.fast technology brings data rates well above VDSL2 capabilities within reach. At loop lengths typical for Fiber-To-The-Node deployments (400 to 1000 m), vectored VDSL2 remains the preferred technology. V.

REFERENCES [1]

600

[2] 400

[3] [4]

200

0

0

50

100

150 200 250 Loop length (m)

300

350

CONCLUSIONS

G.fast is an access technology under definition in ITU-T. It aims at providing multiple hundreds of Mb/s to endusers over the legacy copper access network. In this paper we show, through numerical evaluation, the feasibility of reverse powering of access equipment, with an available per-line power budget of 7.2 Watt. We illustrate that a transmission efficiency of cyclically extended OFDM of 90 to 95% is achievable depending on modulation parameters. We show a realistic data rate performance evaluation under scenarios with different use of bandwidth and different crosstalk conditions. Gigabit data rates are obtainable, but somewhat more modest rates can be achieved in case spectral compatibility with legacy systems must be ensured or in case crosstalk is not mitigated.

400

[5]

Figure 6. Gigabit speeds are within reach. Muliple hundred Mb/s can still be obtained with increased starting frequency and with crosstalk interference.

[6]

17.7 MHz and with crosstalk 6 dB below the 99% worst case crosstalk level. The 99% worst case crosstalk level is derived from the direct channel response |H(n)|2 using the phenomenological model described in [13]. The resulting net aggregate data rate as a function of loop length is provided in Fig. 6. It shows the feasibility of obtaining up to 1 Gb/s aggregate over a single short loop on the legacy copper network when the full bandwidth down to 138 kHz is exploited. When G.fast is deployed in co-existence with VDSL2 17 MHz lines, the net data rate reduces significantly, but rates well above 500 Mb/s remain obtainable. The curves indicated by ‘no XT’ correspond to scenarios where no crosstalk is present, or when the crosstalk is removed through vectoring [1], [2]. When crosstalk is present, even if it is 6 dB below the 99% worst case crosstalk level, the net data rate reduces significantly in the loop length region of interest (0 to 200 m). The impact of crosstalk is largest for loops between 50 and 100 m, where the rate reduction amounts up to 51%. With a limit of VDSL2 30a at 305 Mb/s under similar

[7]

[8]

[9]

[10]

[11]

[12] [13]

3153

V. Oksman et al., ‘The ITU-T’s new G.vector standard proliferates 100 Mb/s DSL’, IEEE Communications Magazine, October 2010. M. Guenach, J. Maes, M. Timmers, O. Lamparter, J.-C. Bischoff, M. Peeters, ‘Vectoring in DSL systems: Practices and Challenges’, proc. IEEE Globecom 2011, Houston, TX, dec. 2011. BBF OD-263, ‘FAST Digital Access over Short Subscriber Loops’. ITU-T Recommendation K.50, ‘Safe limits of operating voltages and currents for telecommunication systems powered over the network’; ITU-T Recommendation K.51, ‘Safety criteria for telecommunication equipment’. ETSI Technical Report 102 629, ‘Access, Terminals, Transmission and Multiplexing; Reverse Power Feed for Remote Nodes’. British Telecom, ‘G.fast: BT requirements for G.fast’, ITU-T SG15/Q4 contribution 11BM-088, August, 2011. M. Timmers, K. Hooghe, M. Guenach, and J. Maes, ‘Digital Complexity in DSL: an Extrapolated Historical Overview’, proceedings of Access 2011, Luxembourg, June, 2011. ITU-T Recommendation G.993.2, ‘Very high speed digital subscriber lines 2 (VDSL2)’, International Telecommunication Union recommendation, 2006. V. Oksman, S. Galli, ‘G.hn: The New ITU-T Home Networking Standard’, IEEE Communications Magazine, pp. 138-145, October 2009. British Telecom, ‘G.hn: Extension of dynamic notching above 30 MHz’, ITU-T SG14/Q4 contribution 10GS5-20, Geneva, Switzerland, October, 2010. ITU-T Recommendation G.9960, Corrigendum 2, ‘Unified high-speed wire-line based home networking trransceivers – system architecture and PHY layer specification’, International Telecommunication Union recommendation, 2011. ETSI Technical Specification 101 388 v1.4.1, August, 2007. J. Maes, M. Guenach, M. Peeters, ‘Statistical Channel Model for Gain Quantification of DSL Crosstalk Mitigation Techniques’, in Proc. IEEE Int’l Conf. Commun. (ICC09), Dresden, 2009.