XG-FAST: Towards 10 Gb/s Copper Access - IEEE Xplore

Globecom 2014 Workshop - Telecommunications Standards - From Research to Standards

XG-FAST: Towards 10 Gb/s Copper Access Werner Coomans, Rodrigo B. Moraes, Koen Hooghe, Alex Duque∗ , Joe Galaro∗ , Michael Timmers, Adriaan J. van Wijngaarden∗ , Mamoun Guenach, and Jochen Maes ∗ Bell

Bell Laboratories, Alcatel-Lucent, Antwerp, Belgium Abstract—Traditionally, copper network operators complement a Fiber-to-the-Home (FTTH) strategy with a hybrid fibercopper deployment in which fiber is gradually brought closer to the consumer, and Digital Subscriber Line (DSL) technology is used for the remaining copper network. In this paper, we propose the system concepts of XG-FAST, a technology capable of delivering 10 Gb/s connection speeds over short copper pairs. With a hardware proof-of-concept platform, it is shown that multi-gigabit speeds are achievable over typical drop lengths from front yard to customer premises equipment of up to 70 m. When an additional pair is available, 10 Gb/s service is shown to be possible up to 30 m by exploiting bonding and phantom mode. The XG-FAST technology will make Fiber-tothe-Door deployments feasible, which avoids many of the hurdles accompanying a traditional FTTH roll-out. Single subscriber XG-FAST devices would be an integral component of FTTH deployments, and as such help accelerate a worldwide roll-out of FTTH services. Standardization is required to specify the physical layer and the interface with the fiber network.

I. I NTRODUCTION About two decades ago, copper networks carried just a few kilobits per second. Today, digital subscriber line (DSL) technology offers rates exceeding 100 Mb/s over those same copper networks. Communication, computing and storage capacity continue to increase significantly every year, and fixed access network technologies must provide ever higher data rates to support a wide variety of high-speed applications, such as Video-on-Demand, fast downloads, support for cloud-based applications, and transport and backhaul capabilities for wireless networks. For copper networks, the achievable capacity is dominated by attenuation. Optical fiber technology has a much longer reach due to its inherent low attenuation, and is therefore ideal to transport high data rates over long distances. Current passive optical networks (PON) technologies provide data rates up to 10 Gb/s, and the next-generation PON systems will typically carry up to 40 Gb/s. However, a direct transition to full FTTH connectivity is hampered by enormous capital expenditures and the very long roll-out time needed to build these networks. At the same time, nearly every household or residential building in developed countries is connected to the copper-based telephone network. Reusing these copper cables as a physical transport medium for broadband access networks proves to be of great economical value. For this reason, the fiber connection has gradually been brought closer to the enduser, shortening the remaining copper loop to within a few hundred meters from the customer. This evolution has been accompanied by new generations of Digital Subscriber Line (DSL) technology that leverage the higher capacity of these shorter copper loops.

978-1-4799-7470-2/14/$31.00 ©2014 IEEE

Laboratories, Alcatel-Lucent, Murray Hill, NJ

For short and medium length copper loops, crosstalk across adjacent twisted pair cables is the limiting factor to improve data rates. Vectored VDSL2 [1], the most recently deployed DSL technology, effectively suppresses crosstalk across adjacent copper lines in a cable. Using this technology, single line performance can be attained, making the loop length the limiting factor once again. This has led to the development of a fourth generation DSL technology, called “fast access to subscriber terminals”, in short, G.fast [2], [3], [4]. A first variant of this technology has reached initial standardization consent in Dec. 2013 [4]. This technology targets copper loops with a maximum length of 250 meters, over which it aims to achieve maximum data rates exceeding 1 Gb/s. These high rates are attained by increasing the bandwidth used for signaling on these short copper loops to 106 MHz, and an extension to 212 MHz is considered. Naturally, G.fast will also include vectoring. The target deployment scenarios for G.fast include multi-home deployments, such as Fiber-to-theDistribution-point (FTTDp) and Fiber-to-the-Building. In these cases a fiber network, such as an passive optical network (PON), is terminated by an optical network terminal (ONT), and the remaining distance to the customer is bridged by a short copper-based network. This paper presents concepts for a fifth generation technology, referred to as XG-FAST [5]. It aims to provide data rates up to 10 Gb/s to the end user over very short existing copper-based lines. One of the targeted deployment scenarios for XG-FAST is an ONT-outside-the-home model or Fiberto-the-Door. The business case for this kind of deployment is primarily driven by the logistic difficulties of bringing the fiber into the home, caused by the need for digging and drilling on customer premises. Due to the very short loop lengths that are targeted with XG-FAST, a natural (additional) deployment scenario includes in-home networks. This would enable multigigabit access speeds over residential copper networks that are already in place today. The latest Wi-Fi 802.11ad standard aims for maximum data rates of almost 7 Gb/s [6]. XG-FAST would be very suited as an in-home wireline technology to provide the necessary capacity on the line feeding this WiFi access point. Although XG-FAST is presented here as a technology for twisted pair telephone cables, it can essentially be used with any cable type, as long as it is a point-to-point connection with sufficient capacity. Existing in-home Ethernet cabling (CAT3, CAT5) is a logical alternative for telephone cabling that, moreover, provides four twisted pairs per cable that can be exploited by XG-FAST. Another alternative is coaxial cabling. Since the attenuation on coaxial cables is typically much lower than that of twisted pairs, using coax

630


as a transmission medium for XG-FAST can extend its range. This paper is organized as follows. In Section II, the channel capacity is used to identify opportunities for high speed communication over short copper loops. Section III introduces the XG-FAST system concept, presents techniques to increase performance, and discusses implementation and deployment aspects. In Section IV, the viability of the XG-FAST technology is demonstrated with measurements performed on several short copper cables. Standardization aspects are discussed in Section V, and Section VI summarizes our results, compares them with competing technologies and their next-generation road maps, and proposes the next steps to be taken to turn the XG-FAST technology concept into practice. II. C OPPER N ETWORKS A first indication of the full potential of communicating over copper twisted-pair cables is provided by determining the available capacity for short copper loops of different lengths. Simulations are performed using a CAD55 model (0.5 mm gauge) for the copper loop. This model is based on measurements up to 300 MHz, and for our analysis it is extrapolated to higher bandwidths. Since the channel characteristics depend on the frequency, f , the channel is sampled using a tone spacing ∆f of 100 kHz. The dynamic range α is defined as the ratio between the signal level at the transmitter and the noise level at the receiver. The (protocol-independent) channel capacity C as a function of the copper loop length L, the bandwidth B and the dynamic range α is given by C = ∆f

N X

2 log2 1 + α |Hi | ,

(1)

i=1

where Hi is the channel at frequency i×∆f , and N = B/∆f . The (protocol-independent) capacity of a CAD55 twisted-pair cable with a 60 dB dynamic range is shown as a function of the loop length in Figure 1. A dynamic range of 60 dB is typical for VDSL2 and G.fast systems. For a bandwidth B = 17.7 MHz, the capacity is slightly above 300 Mb/s, while for B = 100 MHz and B = 200 MHz, the capacity on the shortest loops reaches 2 Gb/s and 4 Gb/s, respectively. For a bandwidth B = 800 MHz, a 10 Gb/s capacity can be achieved on a single twisted pair of up to 30 meters. The large bandwidth requirements are expected to reduce the achievable dynamic range due to hardware limitations. Exploiting the presence of two copper pairs relaxes the requirements to achieve a 10 Gb/s capacity. Using bonding and vectoring, the required rate drops to 5 Gb/s per twisted pair to achieve a total capacity of 10 Gb/s. The two pairs can also be exploited to create a third “phantom” pair to obtain three parallel channels over the two twisted pairs [9] (see Section III-A). Ideally, one can obtain a threefold increase of the capacity achieved over a single pair, reducing the required capacity over each of the three pairs to 3.33 Gb/s and still deliver a 10 Gb/s capacity to the end user. Using these techniques, a 10 Gb/s link capacity can ideally be achieved for short copper loops. The simulation results

Figure 1. Capacity versus the loop length for a CAD55 (0.5 mm gauge) twisted-pair cable with a 60 dB dynamic range (defined as the ratio between the signal level at the transmitter and the noise level at the receiver), and varying signal bandwidth. The benefit of increasing the signal bandwidth rapidly declines with the copper loop length.

also show that the benefits of increasing the bandwidth rapidly disappear for increasing loop lengths. Because of the Fiber-tothe-Door and in-home network deployment scenarios, a good guideline for the XG-FAST range is 50 m. However, this range specification is not carved in stone since this will depend on the cable quality and might even be substantially extended for a physical transmission medium with higher capacity, such as coaxial cables and Ethernet cables. In conclusion, copper loops shorter than 50 m still allow a large increase in capacity that is worthwhile to explore. In the next sections, we will introduce a system concept for XGFAST, taking into account different design requirements. III. XG-FAST S YSTEM C ONCEPTS All current DSL technologies use discrete multi-tone (DMT) modulation, which divides the frequency spectrum into equally spaced parallel channels or “tones” that are independent due to orthogonality. With DMT, the bit-loading can be configured separately for each tone. This is of great importance since the attenuation of the copper line is frequency-dependent. For a comprehensive review on DMT modulation, see [10] and references therein. The ability to use a different bit-loading on every tone allows the full potential of every copper loop to be exploited. Therefore, the use of DMT modulation for XG-FAST is a logical choice. In the Fiber-to-the-Door deployment scenario, XG-FAST devices will need to be powered by the customer premises equipment (CPE) due to the lack of locally available power circuitry. Compared to reverse powering in G.fast it is advantageous that the XG-FAST device hardware will be tailored to a single active user, and will not have the stringent requirements of being able to power large common hardware parts with a single user power supply. Moreover, the shorter cable lengths

631


also lead to a reduced loss of the reverse power provided over that cable. In this paper, we introduce some concepts that can be used in XG-FAST to leverage its bandwidth even further. We first explain how to exploit the presence of multiple twisted pairs using bonding and phantom modes, ideally yielding a linear increase in data rate with the number of available modes. Next, we present a vectoring technique that exploits twosided signal coordination for a single-user XG-FAST, which increases the performance compared to legacy (VDSL2 and G.fast) vectoring techniques. A Proof-of-Concept platform will be used to demonstrate that the increased bandwidth, combined with the proposed system concepts, significantly increase the achievable data rate. Although not covered below, another performance-boosting concept to be considered for XG-FAST is Transmitter Controlled Adaptive Modulation (TCAM) [7]. This is a modulation technique inspired by hierarchical modulation that automatically adapts to time-varying channel conditions. The technique allows to operate at significantly reduced SNR margins (which are now typically set at 6 dB to increase robustness), yielding a corresponding gain in data rate. Regarding forward error correction, TCAM is not compatible with currently used Trellis Coded Modulation. So although a shift from Trellis to Low Density Parity Check (LDPC) coding on merits of coding gain alone may not be worthwhile [8], the use of TCAM would be synergetic with LDPC coding as a forward error correction scheme. A. Bonding and Phantom Mode In many copper networks around the world, customer sites have been provided with two copper pairs, of which only one is currently being used for data transmission. This can be exploited by XG-FAST, which has the benefit of being a single subscriber device and allows each customer to be served with an optimized technology. This is more difficult for FTTNode deployments, as the second pair may not be connected through towards the “node”. Also in other deployment scenarios than Fiber-to-the-Door, such as home or business networks, several copper pairs can be available to the XG-FAST system (e.g., existing Ethernet cabling provides four copper pairs). The availability of multiple copper pairs provides opportunities to significantly increase the capacity by using bonding and phantom mode transmission [9]. Bonding means offering one big data pipe to the user, although underneath the data is communicated over separate physical channels. To obtain bonding, the data needs to be multiplexed over the multiple channels. The achieved data rate gain is proportional to the number of channels when vectoring is used. Without vectoring, crosstalk between the channels will lead to a sublinear increase in data rate. Two available twisted pairs can also be used as two “virtual” wires to create a third “virtual” pair, called the phantom mode (see Figure 2). This phantom circuit concept is well known and was already considered at the end of the 19th century [11]. The differential signal of the phantom mode is the difference

Figure 2. The phantom mode signal is the difference between the common mode signals of twisted pair 1 and twisted pair 2.

between the common mode signals on both twisted pairs, which does not impact the differential signals already present on the twisted pairs. The channel characteristics of the phantom mode are, however, not as good as those of the physical pairs because its two “wires” are two different twisted pairs. For example, a capacitive unbalance between the two physical pairs can increase the attenuation of the phantom mode signal. The phantom channel often experiences more ingress noise, more crosstalk, and causes more radiation (egress), because the twisting of the twisted pairs around each other is typically much worse than for a single twisted pair. Due to the possibly large crosstalk generated by phantom channels, it is essential to use vectoring to yield a data rate gain that is worth the hardware effort. The combination of bonding and phantom mode on two twisted pairs could potentially provide a threefold increase in data rate when compared to the data rate achievable over a single twisted pair. In practice, the increase is lower due to inferior channel characteristics of the phantom channel. However, the total data rate gain of using bonding, vectoring and phantom mode transmission on two twisted pairs can be close to a threefold increase [9]. In Section IV, we will demonstrate the feasibility of using phantom mode transmission at the high frequencies that we aim to use for XG-FAST. B. Two-Sided Signal Coordination Crosstalk cancellation techniques remain essential in XGFAST when multiple pairs are being used by a single user. In VDSL2 and G.fast, crosstalk cancellation is achieved by signal coordination solely at the access node, which in downstream is effectively a Multiple-Input-Single-Output (MISO) scheme. Since XG-FAST is primarily a single-user technology, it is now possible to achieve signal coordination in both the transmitter and the receiver, creating opportunities for new precoder and equalization schemes. It allows to use a Multiple-Input-Multiple-Output (MIMO) scheme that has a better performance than the legacy MISO schemes, which was not possible in legacy DSL. Consider a bonded vectoring group comprising K channels. When two copper pairs are available for a Fiber-to-the-Door deployment, one can use two direct channels and a phantom channel (K = 3). The system model is given by y = F (HPx + n) ,

(2)

where the K × 1 vectors y, x, n represent the received signal, the transmitted signal and the noise, respectively, and the K × K matrices H, F and P denote the channel matrix, the receiver equalizer, and the precoder, respectively. In a bonded

632


group the vector x comprises data derived from a single data stream. In G.fast, due to a lack of diagonal dominance, a linear gain-scaled precoder has been introduced and nonlinear precoders are being studied [3]. With two-sided coordination, the power penalty due to linear or non-linear precoding can be removed by exploiting the eigenmodes of the channel through singular value decomposition (SVD). The channel matrix can be decomposed into unitary matrices U and V, and a diagonal matrix S as follows H = USVH .

Parameter Sampling frequency Bandwidth Transmit power per mode Cyclic extension Forward error correction

Value in PoC 1 GHz 0–500 MHz -2 dBm 0.6 µs RS and TCM

(3)

Two-sided coordination is best explained by the zero-forcing precoder P = V and equalizer F = S−1 UH . This leads to y = x + S−1 UH n.

TABLE I PARAMETER VALUES U SED FOR THE P ROOF - OF -C ONCEPT D EMONSTRATION P LATFORM .

(4)

Because both V and U are unitary, the precoder P = V rotates the signal vector x without increasing the transmit power, and at the receiver UH rotates the received signal vector without noise enhancement. These operations diagonalize the channel, decomposing it into K separate virtual sub-channels (the channel eigenmodes). Each of these eigenmodes has an attenuation given by the corresponding diagonal element of S. We can use these eigenmodes to simultaneously transmit K interference-free symbols. The SVD transmission scheme can be interpreted as the coherent combination of the signals so as to achieve improved SNR. Variants for the precoder and equalizer matrices can be obtained based on singular value decomposition that take a Minimum Mean-Square Error criterion, rather than ZeroForcing, or that take noise covariance into account. Although XG-FAST is primarily a single-user technology, it does not preclude coordination at the central side among multiple bonded vectoring groups. In that case, zero-forcing may be applied across groups, while maintaining the above pre- and postcoder structure [15].

FEC overhead is 6.27 %. All parameters are summarized in Table I. The channel measurements of the cables on which XGFAST were tested are shown in Figure 3. The performance was measured on CAT5e cables (AWG-24 or 0.51 mm diameter) of lengths 30 m, 40 m and 50 m and on telephone cables provided by a European operator, which have a 0.6 mm diameter. For this second cable type, the performance was measured for a 30 m, a 60 m and a 70 m cable. We used a transmit power spectral density of -89 dBm/Hz, which corresponds to a modest ATP of -2 dBm per channel. This transmit power was determined by the upper limit of the receiver ADC on the short cables. It is possible to further increase this transmit power on the longer cables, but the transmit powers were kept equal for the sake of comparison. Figure 4 shows the net data rates that were achieved over these cables. The blue curves (triangles) correspond to the CAT5e cables, and the red curves (circles) correspond to the operator cables. The solid lines show the net data rate over two pairs using bonding and phantom mode transmission, the dashed lines show the rate over two pairs using only bonding, and the dotted lines show the rate achieved over a single pair. For a 30 m CAT5e loop, a combined data rate of 11.1 Gb/s is achieved over two twisted pairs using two differential and one phantom mode with SVD-based vectoring. Longer loops of

IV. P ROOF OF C ONCEPT In this section, we show the potential of XG-FAST with a proof-of-concept (PoC) platform that uses two twisted pair cables. It incorporates the system concepts mentioned above: vectoring, exploiting two-sided coordination, and phantom mode transmission (effectively increasing the number of available channels to three). The purpose of this demonstration is to show the raw potential of XG-FAST. It is not meant to be a fully optimized platform. For the proof-of-concept platform, a 500 MHz baseband spectrum and a sampling frequency of 1 GHz is used. Forward error correction (FEC) is provided using trellis coded modulation (TCM) with a Reed-Solomon outer code, as in G.fast and VDSL2. As a duplexing method, we use time division duplexing (TDD) as in G.fast. The length of the cyclic extension (CE) was set to 0.6 µs in all measurements, unless mentioned otherwise. The use of such a short cyclic extension is possible due to the short length of the copper loops. The framing overhead due to the TDD scheme is 6.25 %, the RS

Figure 3. Channel measurements of the different cables. The longer operator cables are not able to exploit the full 500 MHz bandwidth.

633


a gain between 2.6 and 2.8 for all cables, except for, again, the 0.6 mm operator cable of 70 m, which yields a combined gain as high as 3.2. V. S TANDARDIZATION

Figure 4. Achieved aggregate net data rates over a single pair in the proofof-concept platform for different cable types and cable lengths for a fixed transmit power. See Table I for parameter values. A 10 Gb/s rate is achieved on a 30 m operator cable using bonding and phantom mode, and a 2 Gb/s (or 1 Gb/s symmetric) rate is achieved on a 70 m operator cable using a single pair.

40 m and 50 m yield a data rate of 10.95 Gb/s and 10.25 Gb/s, respectively. This demonstrates that it is possible to achieve 10 Gb/s over CAT5e cables of at least 50 m, using only two pairs. This already brings the performance very close to that of 10GBASE-T, which delivers an aggregate data rate of 20 Gb/s over four pairs of a higher quality CAT6 cable up to 55 m long. The measurements on the 0.6 mm operator cable show that it is possible to achieve 10 Gb/s on a 30 m traditional copper telephone line. Even at 70 m, we still achieve a net data rate of almost 7 Gb/s using two pairs. Note that on this longer cable we set the cyclic extension to 1.5 µs, increasing the overhead. The single line performance (dotted lines in Figure 4) show that aggregate rates up to 4 Gb/s can be achieved using only a single pair. At 70 m, an aggregate rate of 2 Gb/s is achieved, thus enabling symmetric 1 Gb/s FTTH-like services, which is an important milestone for data communication over a single telephone line. When using SVD vectoring, the channels used for transmission are mutually orthogonal “modes” of the transmission medium. Hence, we cannot refer separately to the data rate of a physical pair. A comparison of the dashed and dotted lines provides an indication of the gain achieved by bonding. The bonding gain varies between 1.95 and 2.05 for all measurements, except for the 0.6 mm operator cable of 70 m, which has a bonding gain as high as 2.35 due to the MIMO benefits of vectoring with two-sided coordination. The solid and dashed lines give a clear indication of the gain achieved by adding phantom mode transmission, which on average yields an extra 36 % on top of the bonded rate for all cables (as compared to the theoretically optimal 50 %). Compared to single line rate, bonding and phantom mode combined provide

Standardization of XG-FAST is beneficial, even though it is a single user technology. The standardization of the physical layer helps to grow the market for operators, leading to more cost-effective chipsets, due to economies of scale. The physical layer of XG-FAST can leverage the consented ITU-T G.fast standard, and it would require an updated bandplan covering frequencies up to 500 MHz. Furthermore, new SVD-based vectoring methodologies need to be standardized. In G.fast, the transmit PSD has been limited to avoid interference with, e.g., FM radio and airport control signaling. The International Telecommunications Union (ITU) Terrestrial and Radio groups have specified transmit PSD limitations for a variety of wireline systems. G.fast also facilitates coexistence by providing a highly configurable PSD mask, which can be used to notch out frequencies that could potentially harm any of the co-existing services. Similar efforts will be needed to allow the introduction of XG-FAST technology, in particular when larger bandwidths are being used for signaling. The broadband forum (BBF) has also launched its own survey on FTTDp with the aim to start a G.fast certification program to avoid inter-operability issues. A similar interoperability scheme is needed to allow the operator to select the distribution point and CPE independently. At their Q2/2014 meeting, the BBF agreed to a management architecture based on the use of a Persistent Management Agent (PMA) and a single management interface to a DPU. A Fiber-to-the-Door deployment using XG-FAST is supposed to use the same architecture as FTTDp. VI. D ISCUSSION In this paper, we introduced the basic concepts of XGFAST, a novel multi-gigabit transmission technology capable of transmitting more than 10 Gb/s over two twisted pairs. One of the possible deployment scenarios for XG-FAST is Fiberto-the-Door (ONT-outside-the-home), in which XG-FAST is used as the next-generation DSL technology succeeding G.fast. The performance will depend on the cable quality and will gracefully degrade with increasing copper loop length. We have experimentally demonstrated a net data rate exceeding 2 Gb/s over a single pair in a 70 m operator cable (0.6 mm) using a bandwidth of 500 MHz. When a second twisted pair is available, as is the case in many DSL networks around the world, the XG-FAST data rate can be boosted up to 10 Gb/s using bonding and phantom mode transmission for loops up to 30 m. A second deployment scenario for XG-FAST are home networks, as XG-FAST could equally well be operated over in-home telephone lines, Ethernet or coaxial cables. Other ultra-broadband access technologies that are already or almost available include XG-PON (optical), G.fast, G.hn, DOCSIS 3.1, 10GBASE-T (copper), next-generation Wi-Fi, and LTE (wireless). G.fast provides an alternative over copper

634


twisted pairs up to 1 Gb/s, and the home network standard G.hn specifies data transport up to 1 Gb/s over several wiring types including twisted pair, power line and coaxial cable [12]. DOCSIS 3.1 provides a data rate of 10 Gb/s downstream and 1 Gb/s upstream over coaxial cable [13]. These data rates are to be shared among several users, since the coax-network topology typically provides point-to-multipoint (P2MP) instead of point-to-point (P2P) links. 10G Ethernet (10GBASE-T), on the other hand, is a P2P technology that provides a dedicated single user data rate of 10 Gb/s symmetric. The 10GBASE-T technology requires the availability of four pairs in a cable of at least CAT6 quality. With CAT6 cabling, the specified range is limited to 55 meters [14]. If the quality of the cable is insufficient, the 10G Ethernet protocol fails to handshake and falls back to 1G Ethernet. There is no intermediate operating point. The data rate for XG-FAST, on the other hand, will be matched to the cable quality. As such, a particular cable may be able to support 9 Gb/s using XG-FAST, while 10GBASE-T would default to 1 Gb/s. This flexibility is a crucial advantage in brown field deployments that reuse an existing copper infrastructure. The XG-FAST technology would be complementary to XGPON, and next-generation Wi-Fi as a further evolution of the G.fast and G.hn technologies. Fiber-to-the-Door (ONToutside-the-home) deployments will allow the existing copper plant to be exploited once more to deliver rates of 10 Gb/s over this last copper drop. These deployments will go hand in hand with XG-PON, and are a logical evolution after G.fast. XG-FAST is also a perfect candidate for backhauling nextgeneration mobile and Wi-Fi technologies that are specified to operate at multi-gigabit data rates. Most importantly, it can do so by reusing cables that are already deployed today. The experiments and concepts presented in this paper only illustrate the raw potential of XG-FAST. It is necessary to initiate a standardization effort for this new ultra-broadband wireline technology to enable a successful market adoption. Further study is desired in the framework of this standardization, such as for instance the targeted loop lengths and RF bandwidth. The very first G.fast products are announced for 2015–2016. Standardization of XG-FAST should ideally be finished by the time G.fast reaches large-scale deployment.

[7] M. Timmers, D. Vanderhaegen, D. Van Bruyssel, M. Guenach, and J. Maes, "Transmitter-Controlled Adaptive Modulation: Concepts and Results," Bell Labs Tech. J., vol. 18, no. 1, pp. 153-169, 2013. [8] J. Neckebroek, M. Moeneclaey, M. Guenach, M. Timmers, and J. Maes, "Comparison of error-control schemes for high-rate communication over short DSL loops affected by impulsive noise," In proceedings IEEE ICC 2013, pp. 4014-4019. [9] M. Peeters and S. Vanhastel, "The Copper Phantom," OSP Mag., 2011. [10] T. Starr, J. M. Cioffi and P. Silverman, Understanding digital subscriber line technology, Prentice Hall, 1998. [11] D. G. Tucker, "A technical history of phantom circuits," Proc. IEE, vol. 126, no. 9, pp. 893–900, Sep. 1979. [12] Unified high-speed wire-line based home networking transceivers, ITUT recommendations G.9960 to G.9964, 2011. [13] DOCSIS 3.1 Specification, Cable Television Laboratories (CableLabs), 2013. [14] 802.3 - 2012 - Standard for Ethernet, IEEE Standards Association, 2012. [15] R.B. Moraes, P. Tsiaflakis, J. Maes, and M. Moonen, "DMT MIMO IC rate maximization in DSL with combined signal and spectrum coordination, " IEEE Trans. Signal Proc., vol. 61, no. 7, pp. 1756-1769, July 2013.

R EFERENCES [1] Self-FEXT cancellation (vectoring) for use with VDSL2 transceivers, ITU-T recommendation G.993.5, Apr. 2010. [2] "P. Ödling, T. Magesacher, S. Höst, P. O. Börjesson, M. Berg, and E. Areizaga, “The fourth generation broadband concept,” IEEE Commun. Mag., vol. 47, no. 1, pp. 63–69, Jan. 2009. [3] M. Timmers, M. Guenach, C. Nuzman, and J. Maes, “G.fast: Evolving the copper access network,” IEEE Commun. Mag., vol. 51, no. 8, pp. 74–79, Aug. 2013. [4] Fast access to subscriber terminals (FAST) – physical layer specification, ITU-T recommendation G.9701 (draft), Dec. 2013. [5] J. Maes, P. Spruyt, and S. Vanhastel, “G.fast breaks through the Gigabit barrier," Alcatel-Lucent Techzine, July 2014. [6] Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications – Amendment 3: Enhancements for Very High Throughput in the 60 GHz Band, IEEE Std. 802.11ad, Dec. 2012.

635