DSL Phantom Mode Transmission: Cable Measurements and Performance Evaluation

Transcription

1 DSL Phantom Mode Transmission: Measurements and Performance Evaluation D. A. Gomes, G. Guedes, A. Klautau and E. Pelaes Electrical Engineering Department UFPA Institute of Technology Belém, Pará - Brazil s: {diagomes,gustavo,aldebaro,pelaes}@ufpa.br Chenguang Lu Ericsson Research Kista, Sweden chenguang.lu@ericsson.com Abstract Phantom mode (PM) transmission aims at increasing the total rate of DSL when multiple copper pairs are available. This paper brings significant insights on PM mode usage and performance. It discusses procedures for performing PM measurements over frequencies from 1 khz to 3 MHz. Novel results are presented concerning direct channel, far-end crosstalk and mode conversion for a 5-meter Cat5e cable. Among other facts, the measurements show that mode conversion increases up to approximately 3 MHz and becomes almost flat over frequency afterwards. They also show that unrolling the cable significantly improves the PM measurements. The capacity of PM transmission is also estimated, indicating that the bit rate can be increased by 6% when PM is used with vectoring. I. INTRODUCTION Telephone lines were designed for voice transmission over copper pairs using frequencies up to only 4 khz. However, these lines are currently used for Internet access specially via digital subscriber line (DSL), the world s most popular broadband access technology. The most recently standardized DSL technologies is the very high bitrate DSL second generation (VDSL2), which provides up to 1 Mbps using a total bandwidth (including up and downstream) of 3 MHz [1]. And the used bandwidth can be extended to 2 MHz with the conclusion of the G.fast standard, which has been studied under the auspices of the International Telecommunication Union (ITU). This corresponds to an increase of four orders of magnitude in bandwidth when compared to the one used for the plain old telephone system and poses challenges with respect to modeling and even measuring the copper pair as transmission channels. VDSL2 and previous DSL standards were designed for differential mode (DM) transmission [2], where one copper pair is the transmission channel. However, when N pairs are available, combinations of wires and the use or not of the ground as a return path for the transmission signals provide new transmission modes such that a maximum of 2N 1 (theoretically independent) channels can be established [3] [5]. For example, one alternative is to reserve a single wire to be the return path of all remaining 2N 1 wires. However, this and some other alternatives would require the design of new transceivers, given that the current DSL standards adopt DM. This work investigates an alternative that allows using off-theshelf DSL transceivers by combining the common modes of two distinct pairs to create a phantom mode (PM). The PM construction allows to create N 1 PM channels and N DM channels. To achieve the maximum of 2N 1 channels, a first-layer of PM channels is used to composed a second-layer and so on. Assuming N is even, the first-layer is composed by N/2 PM channels with the signals properly combined via transformers, for example (see Fig. 1). A secondlayer PM channel uses four pairs together to transmit and it requires two more transformers on top of the first-layer PM channels. For example, N = 4 for a Cat5e cable, which allows four DM channels, two first-layer PM channels and one second-layer PM channel. In this work, only first-layer PM channels were considered. The PM construction allows the adoption of standard transceivers, which is an advantage when compared to the split-pairs discussed in [3]. Some significant work is discussed in the sequel. In [3], the theoretical limits of common mode transmission is discussed and a scheme that splits the wires of twisted pairs is evaluated. In [5] a channel model for phantom mode was derived, considering it as an eigenmode of the quad cable. The presented results were based on a maximum frequency of 25 MHz. In [6], impedance measurements were discussed, whereas the current work addresses measurements of direct transfer function, FEXT and mode conversion. In [7] a measurement setup for differential-mode and common-mode channels is discussed, assuming the VDSL2 frequency band. Similar to bonding [2], the PM transmission mode has the potential of increasing the total data rate when multiple copper pairs are available. However, its widespread usage depends on the development of techniques that circumvent the increased levels of crosstalk due to the fact that signals are transmitted in copper wires that are not paired and twisted accordingly. This work is organized as follows. Section II presents a brief theory on PM. Section III describes our measurement methodology for PM direct transfer function, MC and crosstalk. Section IV discusses the obtained results. Section V presents channel capacity estimations considering two scenarios with and without vectoring. Section VI concludes the paper. II. PHANTOM MODE CHANNELS Signal transmission over telephone lines is generally performed in DM, where the amplitudes of the signals transmitted

2 in each wire of a twisted pair are anti-symmetric (e. g., x(t) and x(t)). Recently, new transmission modes have been developed, such as PM [3] [5]. It is an interesting scheme that has also been used in applications other than DSL (e. g., [8]). Td1 Tp Td2 Fig. 1. Td1- + Tp+ Tp+ Tp- Td1+ + Tp+ Td2+ + Tp- Td2- + Tp- Rd1 Rp Rd2 Example of obtaining one PM channel from two copper pairs. In DSL, PM is a signaling mode that allows to convey a differential signal using the common mode signals of two distinct twisted pairs. Conceptually, this creates a new phantom pair, as illustrated in Fig. 1. The transmitter that uses the PM channel sends the signal T p represented differentially by T p + = T p/2 and T p = T p/2, where T p + and T p are imposed via transformers as CM signals to the (traditional) DM channels. Given that T d1 + = T d1/2 and T d1 = T d1/2 and assuming the transformers and lines are perfectly balanced, the receiver can recover T d1 by eliminating the CM signal T p + via Rd1 = (T d1 + + T p + ) (T d1 + T p + ) = T d1. Similarly, one can obtain Rd2 = T d2 and Rp = T p at the corresponding receivers. In practice, transformers and twisted pairs are not perfectly balanced and there is significant crosstalk among the channels. Hence, it is important to measure and model these channels in order to design efficient PM-enabled systems. This work presents results for the direct transfer function (DTF) and two kinds of crosstalk. The channel is assumed to be linear and time-invariant, such that the DTF or frequency response is H(f) = Y (f)/x(f), where X(f) and Y (f) are the Fourier transforms of the transmitted and received signals. The DTF magnitude results will be presented in db, i. e., 2 log 1 H(f). For PM transmission, crosstalk is of major importance. This is also true for FEXT in traditional DSL DM transmission, while the near end crosstalk (NEXT) is typically negligible due to the adoption of frequency division duplex (FDD) in most DSL systems. Another kind of crosstalk is the mode conversion (MC), also called leakage [7], which occurs when part of the signal in a mode (the common mode, for example) leaks to another mode (the differential mode, for example), and occurs mainly due to unbalance of lines and transceivers circuitry. MC generally is neglected for DM transmissions since the common mode is not used. However, for mixed PM+DM transmissions, it must be considered as an important limiting factor to the transmission performance. The next section discusses the adopted DTF, FEXT and MC measurement procedures. III. MEASUREMENT SETUP One contribution of this work is to present results that include PM channels for frequencies up to 3 MHz. The relatively high bandwidth demands a careful procedure. Based on the acquired experience, the most important aspects are discussed in the sequel. The results are presented for a 5- meters Cat5e cable. Similar measurements were conducted for a Quad cable but are omitted due to the lack of space. A. PM construction over a Cat5e cable The cross section of a Cat5e cable is depicted in Fig. 2. Dotted lines indicate options to compose a PM pair from two DM pairs. bl br A Fig. 2. or gr bl br B Cross section of a Cat5e cable. For example, the blue pair (bl) can be combined with the green pair (gr) to form the PM pair blgr (Fig. 2-A). For Cat5e, two arrangements of PM pairs are possible: edge, which is illustrated as the connection between DM pairs in Fig. 2-B, and diagonal, which is the connection between DM pairs as in Fig. 2-A. This terminology will be used throughout this document. Recall that, in this work, only first-layer PM channels were considered. Hence, the DTF measurements will be presented for all six PM pairs (blgr, blor, blbr, gror, grbr, orbr) but a given copper pair should not be used in distinct PM channels. Hence, there are three options to obtain two PM channels on a Cat5e cable: blgr-orbr, blor-grbr or blbr-gror. B. Methodology The following equipments were used for the measurements: - Agilent Network Analyzer (NA) 4395A; - Transmission/Reflection Test Set (TRTS) 87512A; - North Hills 5-1 Ω, 1 KHz - 3 MHz; - PM connectors to perform calibration and measurements; - MULTILAN Cat5e U/UTP 24AWGx4P cable. After calibration, the measurements were performed using standard procedures (see, e. g. [9], [1]), differing only in that each DM pair was treated as a single wire when measuring PM. The 5-meter cable was measured at frequencies ranging from 1 khz up to 3 MHz. or gr

3 Given the results in [11], which indicate that unrolling cables can decrease artifacts in the frequency domain and lead to a smoother impulse response, the measurements were conducted with both the Cat5e cable rolled and unrolled. Only DM was used in [11] and the current work shows that the need for unrolling the cable is even more evident in PM than DM measurements. C. Direct transfer function measurements in phantom mode The adopted procedure for PM was based in [6], which considers each DM pair as a single wire. Fig. 3 shows a scheme of PM DTF measurement using the NA, with the DTF being the ratio of the signal read in B by the signal sent in A (B/A), also called S 21 measurement. Note that each balun connector was plugged to a pair of wires of the cable, i.e, in each pair the signal travels in common mode. The baluns were shielded to reduce electromagnetic coupling. Preliminary tests without shielded baluns showed strong reflections, which were drastically reduced after their shielding. other pair, due to CM leakage. Note that the free end of PM was terminated with 1 Ω load ( T component) to reduce reflections. In the ideal case, no signal should be received in the unbalanced side of the balun since a common mode signal is sent and the balun is reading it in differential mode. Fig. 5. A NA B T MC measurement from PM to one of its composing DM pairs. Table I lists the pairs used in the FEXT and MC measurements (MC was not measured from DM to PM). TABLE I MEASURED FEXT AND MC. NA R A B DM to DM DM to PM PM to DM PM to PM MC bl-gr bl-gror gror-bl blgr-orbr blgr-bl bl-or bl-grbr grbr-bl blor-grbr blgr-gr bl-br bl-orbr orbr-bl blbr-gror blor-bl gr-or gr-blor blor-gr blor-or gr-br gr-blbr blbr-gr blbr-bl or-br gr-orbr orbr-gr blbr-br or-blgr blgr-or gror-gr or-blbr blbr-or gror-or or-grbr grbr-or grbr-gr br-blgr blgr-br grbr-br br-blor blor-br orbr-or br-gror gror-br orbr-br Fig. 3. Phantom mode DTF measurement setup. D. Crosstalk (FEXT and MC) in phantom mode Fig. 4 depicts the FEXT measurement setup and illustrates how the FEXT signal travels from one channel (in this case, a PM pair) to another (a DM pair). This work measured FEXT from DM to DM, DM to PM, PM to DM and PM to PM. A NA B IV. CABLE MEASUREMENT RESULTS This section presents the obtained measurement results. A. Direct transfer functions The DTFs where measured with both the cable rolled and unrolled. Figs. 6 and 7 show the plots of the DM and PM DTFs obtained with the cable rolled and unrolled, respectively Magnitude of DTF of Cat-5 cable for both DM and PM - 5 m (rolled) Fig. 4. T T Illustration of FEXT measurement from a PM pair to a DM pair. Magnitude (db) bl (DM) -25 br (DM) gr (DM) or (DM) blbr (PM) blgr (PM) blor (PM) grbr (PM) gror (PM) orbr (PM) The importance of using 1 Ω terminations (T) should be noted. They were connected to the channel ends for impedance matching and to better approximate actual systems. Differences of approximately 1 db were found when comparing measurements with and without terminations. Fig. 5 depicts how MC was measured and indicates that MC is composed by two parcels: from the same pair and from the Fig. 6. DTF of Cat5 in both DM and PM using rolled cable. Comparing Fig. 6 and 7, it can be noticed a significant reduction of artifacts, reduction of resonances and shift up in the PM DTF magnitude curves (dashed lines), mainly for edge PM. When comparing the two figures, a significant discrepancy can be observed around 1 MHz. For the DM

4 Magnitude of Direct Channel of DM and PM - 5 m -5-1 Magnitude (db) bl (DM) br (DM) gr (DM) or (DM) blbr (edge PM) blgr (diag PM) blor (edge PM) grbr (edge PM) gror (edge PM) orbr (diag PM) Fig. 7. Cat5e DTF in both DM and PM with cable unrolled. curves, unrolling the cable led to smoother curves. It can be also seen that PM has a larger DTF than DM. Also, Fig. 7 indicates that for frequencies larger than 1 MHz, the DTF of diagonal PM is slightly larger than for the edge PM, which deserves further investigation. Fig. 8 shows a comparison between the impulse responses corresponding to rolled and unrolled cable, and were obtained via the inverse discrete Fourier transform (IDFT). It can be seen that the use of unrolled cable led to a smoother impulse response, slight reduction of the amount of non causal energy (energy before the main peak, highlighted by a blue circle in the figure) and significant decrease of the reflection at approximately.7 microseconds. amplitude Impulse response of Edge PM on Cat-5 rolled unrolled delay ( s) Fig. 8. cable. Comparison of the impulse responses of the rolled and unrolled B. Crosstalk (FEXT) and MC Figs. 9 to 12 depict plots of FEXT from DM to DM, DM to PM, PM to DM and PM to PM, respectively. It should be noted in Fig. 9 that the FEXT between pairs bl-gr and or-br, corresponding to diagonal PMs, presented larger values than the FEXT between the edge PMs. This behavior was not expected given that the edge-located pairs TABLE II TWISTING RATE OF PAIRS OF CAT5E CABLE. Pair [cm] per turn Turns per [meter] bl gr or br are adjacent and, therefore, closer to each other than the diagonal ones. However, Table II indicates that there is a better impedance matching between the diagonal PMs due to their closer twisting-rate, which can be the reason for the obtained result giving that both twisting-rate and distance between pairs influence the channel electromagnetic characteristics [12]. Coincidently, the FEXT of bl-gr is the largest, as their twistingrates are the most similar ones. Fig. 12 indicates that the FEXT from one diagonal PM to the other appears weaker than from one edge PM to the other. This may be due to the quad-like configuration that is formed by the Cat5e DM pairs when they are used as single conductors to create phantom modes. Observing Fig. 2, it can be seen that the distances between wires involved in a diagonal-to-diagonal PM FEXT exhibit much greater symmetry than the ones of an edge-to-edge PM FEXT. So, the diagonal PM configuration in Cat5e cable is much probable to behave like the diagonal DM configuration in a quad cable, in which one diagonal DM receives almost no crosstalk from the other diagonal DM because of the canceling effects in the DM-to-DM coupling. Fig. 1 and Fig. 11 do not show significant differences between the FEXT from DM to PM and from PM to DM, which was expected. They appear to be slightly stronger than the FEXT from DM to DM. This result indirectly indicates that PM is more sensitive to ingress / egress. Fig. 13 presents MC in Cat5e. MC appears to increase up to 3 MHz and be almost flat over frequency afterwards, with MC from diagonal PM to DM being weaker than from

5 from bl to br from bl to gr from bl to or from gr to br from gr to or from or to br FEXT from DM to DM in Cat-5 unrolled cable - 5 m MC in Cat-5 cable - 5 m from edge PM to DM from diagonal PM to DM MC (db) Fig. 9. FEXT from DM to DM in 5m unrolled Cat5e. Fig. 13. MC in 5m, unrolled Cat5e FEXT from DM to PM in Cat-5 unrolled cable - 5 m Fig. 1. from bl to grbr from bl to gror from bl to orbr from br to blgr from br to blor from br to gror from gr to blbr from gr to blor from gr to orbr from or to blbr from or to blgr from or to grbr FEXT from DM to PM in 5m unrolled Cat5e. FEXT from PM to DM in Cat-5 unrolled cable - 5 m Fig. 11. FEXT from PM to DM in 5m unrolled Cat5e. FEXT from PM to PM in Cat-5 unrolled cable - 5 m from blbr to gr from blbr to or from blgr to br from blgr to or from blor to br from blor to gr from grbr to bl from grbr to or from gror to bl from gror to br from orbr to bl from orbr to gr from blbr (edge PM) to gror (edge PM) from blgr (diagonal PM) to orbr (diagonal PM) from blor (edge PM) to grbr (edge PM) Fig. 12. FEXT from PM to PM in 5m unrolled Cat5e. edge PM to DM. This may have occurred due to a stronger coupling between pairs in edge given their smaller distance when compared to pairs in diagonal. V. CHANNEL CAPACITY This section presents channel capacity [13] estimations adopting two different scenarios: (i) considering crosstalk and (ii) assuming that vectoring techniques completely eliminated FEXT and MC [3]. It is assumed a multicarrier transmission that splits the available bandwidth in K subbands with bandwidth f and using M modes to transmit, so that the overall channel capacity [14] is C = f M m=1 k=1 K log 2 (1 + S k,m ), (1) where S k,m is the SNR of subchannel (k, m) corresponding to mode m in subband k and is defined as S k,m = ( σ 2 p(k) + m H 2 k,m d G2 k,m,d + d L2 k,d ) Γ where H 2 k,m is the DTF squared magnitude (PM or DM) of subchannel (k, m), G 2 k,m,d is the FEXT squared magnitude from the d-th disturber into (k, m), L 2 k,d is the squared magnitude of MC from the d-th disturber, σ 2 is the background noise power and p(k) is the transmitted PSD mask, assumed to be the same for all transmitters. Note that the FEXT term is summed over m because a subchannel receives interference from disturbers in all considered modes. In Eq. 2, the SNR gap Γ is given in linear scale, but in db it can be conveniently written as Γ db = L mod + γ m γ c, (3) (2) where L mod is the modulation loss, γ m is the noise margin and γ c is the coding gain, all given in db [14]. Although this work exhibits measurement results up to 3 MHz, only frequencies up to 2 MHz (which is the maximum frequency of interest for G.fast) were considered for estimating capacity. It was also assumed f = khz, σ 2 = 14 dbm/hz, p(k) = 76 dbm/hz, k, L mod = 9.75 db, γ m = 6 db and γ c = 5 db. Estimates were obtained assuming a maximum number of of 12 and 15 bits per subchannel. Assuming the availability of a single Cat5e cable, two transmission scenarios were compared: 1) only DM and 2) DM and PM (edge or diagonal). The former scenario uses eight transceivers, two at each channel endpoint while the latter uses twelve transceivers. Fig. 14 summarizes the estimated capacities.

6 bit rate (Gb/s) DM (8 transceivers) DM + edge PM (12 transceivers) DM + diagonal PM (12 transceivers) No vectoring (12 bits) Vectoring (12 bits) No vectoring (15 bits) Vectoring (15 bits) Fig. 14. Estimated channel capacities in Gbps for the 5m Cat5e cable on a bandwidth of 2 MHz. Fig. 14 shows that the use of DM and PM together, considering all crosstalk sources, reduces the performance when compared to using only DM. This occurs because, although more transceivers are used, the use of DM and PM increases the crosstalk effects given that their simultaneous transmission violates the assumptions adopted when designing the cable and twisting its pairs. However, when considering the use of DM, PM and vectoring, the gain in performance is significant. When PM modes are using together with vectoring (or any other interference mitigation that can eliminate FEXT and MC), the bit rate per user is increased by 66% and 45% for 12 and 15 maximum bits, respectively, when compared to using only DM. A relatively small difference between the edge and diagonal configurations for PM was perceived, with the diagonal being slightly better than the edge PM. When a maximum of 12 bits per tone was used with vectoring, there was no difference on bit rates given that the relatively large SNR saturated the number of bits per subchannel. VI. CONCLUSIONS This work presented measurement results for DTF, FEXT and MC when using PM on a 5-meters Cat5e cable on a frequency range from 1 khz up to 3 MHz. Procedures to reduce reflections and other measurement artifacts were discussed. The channel capacity was estimated considering DM and mixed DM and PM scenarios, with and without the use of vectoring to eliminate the FEXT and MC interferences. It has been shown that unrolling the cable significantly improves the DTF measurements in PM, with a reduction of artifacts and of non-causal energy in impulse response. Cat5e DTF in PM presented larger magnitude than DTF of DM up to 2 MHz, with diagonal PM having a slightly larger DTF. Cases in which PM is used presented larger FEXT when compared to DM-only transmission. FEXT from PM to PM presented larger magnitude, with FEXT from edge PM to other edge PM representing the worst case. The measurements showed that MC caused by diagonal PM is weaker than the one caused by edge PM, with both increasing up to approximately 3 MHz and being almost flat over frequency afterwards. Estimation of channel capacity shows that, in systems that do not use any crosstalk mitigation technique (that is, when FEXT and MC are considered), it is not worth to consider PM transmission together with DM. However, when vectoring is applied, the performance gain was estimated as 6% and 45% (for scenarios with 12 and 15 bits per subchannel, respectively) when compared to DM-only for the 5m Cat5e cable considered in this work. As the quest for increasing bit rates continues, the use of PM transmission deserves investigation. The increasingly shorter copper cables allow to explore bandwidths of a couple of hundred MHz. Modeling and measuring PM (and DM) on this situation is challenging and this work presented results along this direction. REFERENCES [1] P. Ödling, T. Magesacher, S. Höst, P. O. Börjesson, and M. Berg, The fourth generation broadband concept, IEEE Communications Magazine, Jan. 29. [2] P. Golden, H. Dedieu, and K. Jacobsen, Implementation and Applications of DSL Technology. Auerbach Publications, Taylor & Francis Group, 27. [3] B. Lee, J. N. Cioffi, S. Jagannathan, K. Seong, Y. Kim, M. Mohseni, and M. H. Brady, Binder MIMO channels, IEEE Transactions on Communications, vol. 55, p. no. 8, Aug. 27. [4] V. L. Nir and M. Moonen, Study and optimisation of the common mode exploitation for xdsl application, Internal report 7-162, ESAT- SISTA, Katholieke Universiteit Leuven (KUL), Belgium, Tech. Rep., Jun. 27. [5] W. Foubert, C. Neus, L. V. Biesen, and Y. Rolain, Exploiting the phantom-mode signal in DSL applications, IEEE Transactions on Instrumentation and Measurement, vol. 61, pp , 212. [6] C. Neus, W. Foubert, L. V. Biesen, Y. Rolain, P. Boets, and J. Maes, Binder identification by means of phantom measurements, IEEE Transactions on Instrumentation and Measurement, vol. 66, p. no. 6, Jun [7] V. L. Nir and M. Moonen, Measurement setup for differential-mode and common-mode channels, Internal report 7-162, ESAT-SISTA, Katholieke Universiteit Leuven (KUL), Belgium, Tech. Rep., Jun. 27. [8] T. Gabara, Phantom mode signaling in VLSI systems, Conference on Advanced Research VLSI, pp. 88 1, 21. [9] ATMEL, Single-to-differential conversion in high-frequency applications, ATMEL, Tech. Rep., 24. [1] AGILENT, Agilent 4395a network/spectrum/impedance analyzer operation manual, Agilent technologies, Feb. 28. [11] ERICSSON, G.fast: Comparison between measurements of unrolled cables and cables winded on drums, ERICSSON, Tech. Rep., ITU-T Q4/15 Contribution 124A-24, may 212. [12] A. Lago, C. M. P. nalver, J. Marcos, J. Doval-Gandoy, A. N. M. endez, O. L. opez, F. Santiago, F. D. Freijedo, J. M. Vilas, and J. C. Lorenzo, Geometric analysis and manufacturing considerations for optimizing the characteristics of a twisted pair, IEEE Transactions on Electronics Packaging Manufacturing, vol. 32, pp , Jan. 29. [13] C. E. Shannon, A mathematical theory of communication, IEEE Transactions on Instrumentation and Measurement, vol. 27, pp , , [14] E. Karipidis, N. Sidiropoulos, A. Leshem, and L. Youming, Capacity statistics for short DSL loops from measured 3 MHz channel data, 25 IEEE 6th Workshop on Signal Processing Advances in Wireless Communications, pp , 25.