Estimation of the achievable xdsl service

Transcription

1 Estimation of the achievable xdsl service C. Neus, P. Boets, L. Van Biesen Vrije Universiteit Brussel Pleinlaan 2 B-1050 Brussels, Belgium Abstract- As a general trend, optical fibers are penetrating farther and farther into traditional telephony networks, thus shortening the length of the copper loop. Nevertheless, primarily for economic reasons, it is certain that in the shortto-medium term a significant percentage of the telephony network in most countries will remain as twisted pair cable between the subscriber and the exchange. However, since the telephone network was primarily not intended for highspeed data transfer, several problems are encountered which limit the maximal achievable bit rate. This paper gives an overview of the limiting factors and describes how the maximal data rate of a xdsl line can be estimated. Keywords: DSL, loop qualification, Single-Ended Line Testing I. INTRODUCTION Surely Fiber-to-the-Home would satisfy the bandwidth requirements of the most demanding user. Unfortunately, this scenario is unlikely to be achieved for some considerable time to come because of the cost involved in the installation of a complete fiber network. Particularly labor and other non-equipment costs are extremely difficult for the service provider to recover in a reasonable time frame. In the meantime, the service providers shorten the copper telephone loop by renewing the common backhauls with optical fibers. Next to fiber, there are several other media that can be used to provide broadband access to residential and business subscribers, for example wireless communication, television coax lines or power lines. But in general none of them has the ubiquity and the level of maturity of development of the telephone network [1]. The present revival of the copper network has been made possible by xdsl modems, such as HDSL, ADSL, ADSL2, ADSL2+, VDSL and VDSL2. In this paper we will not focus on any specific xdsl technology, but we will consider xdsl services in general, where the x denotes any of the possible DSL technologies. xdsl modems make use of higher frequencies than voice-band modems. As a consequence higher data rates are achievable. The copper cables are now used up to 1.1 MHz for ADSL and up to 12 MHz for VDSL. However, communication in such broad frequency bands encounters several problems because the telephone network was designed to function optimally in the voice-band ( Hz). This original service is referred to as Plain Old Telephony Service (POTS). Not having been designed for high-speed digital communications at these high frequencies, the existing telephone lines are prone to attenuation, distortion, noise and interference from external sources. When customers request a service, request a change or drop in service, the appropriate connections must be made, rearranged or taken away. Obviously, a telecom operator must understand his network and know where the connections and the flexibility points such as junction boxes are. Records of the loop make-up have been kept on paper and more recently, have been manually entered into computer databases. Whether on paper or stored in a computer, some of the records are inaccurate, others are not up-to-date and some are simply missing. One way to improve the existing records is to examine and update them manually. This method is expensive and time consuming. Furthermore, new technologies such as xdsl require additional information that was not previously kept for voice services, so new information must be added to the existing records. In any case, when a customer asks for a certain xdsl service, the first thing the operator needs to do, is verifying whether this telephone line can support the requested service. This is called loop qualification and is different for each customer since it depends on the cabling between the customer s house and the central office. Inaccurate loop qualification may lead to an overestimation of the achievable bit rate, resulting in customer dissatisfaction, or to an underestimation of the achievable bit rate, resulting in loss of revenues. Reliable loop qualification is thus very important and the best way of qualifying a loop for xdsl service is to test it. The maximum transmission rate over a given channel is referred to as the channel capacity. The channel capacity is limited by the transfer function of the subscriber loop and the corresponding noise environment. In this paper we will focus on the first item, namely the estimation of the transfer function. The transfer function describes how a signal is attenuated and distorted when traveling along a line. The noise estimation will not be described in detail but as one can imagine, the noise will certainly limit the achievable bit rate [2, 3]. The cabling between the customer and the central office is called the subscriber loop and it determines the transfer function. Many loop make-ups are possible, but typically a subscriber loop consists of several cable sections, possibly with a different diameter, spliced to each other, connecting the customers equipment to the central office. Fig. 1 gives an overview of the most common topologies. The remainder of this paper is structured as follows. Section II will illustrate the main problems encountered when working at high frequencies, as is the case with xdsl services. Section III will explain the possible measurement techniques for the evaluation of the transfer function, while Section IV will describe one specific measurement technique (SELT) in detail. Section V will

2 L LL LTL LTTL Central Office () C1 C2 LLL Fig. 1 : Examples of topologies = central office, = customer premise LTLTL FEXT NEXT Fig. 2 : Near-end Crosstalk (NEXT) and Far-end Crosstalk (FEXT) illustrated for two customers (C1 and C2) discuss the possible applications and advantages of this technique and Section VI will summarize the most important points. II. DIFFICULTIES With xdsl systems running over existing telephone lines, environmental conditions are not easily controlled. Changing weather, varying loop quality and age, inconsistent indoor wiring, radio interference and xdsl services running on adjacent wire pairs, to name a few, can all cause errors with different characteristics. Depending on the application running over the xdsl service, errors can cause varying degrees of disruption. For data applications, for example Internet, information is typically delivered in packets. If there are non-correctable errors anywhere in a particular packet, the entire packet is discarded and a retransmission of the packet is requested. However, for video applications where there typically is not enough time or memory to retransmit packets, errors can show up as an annoying picture distortion. This section describes the most important channel impairments encountered by xdsl-technologies due to the use of a broader frequency band than with voicecommunication [1, 4 ]. As a consequence the channel capacity will be limited. A. Noise Every electrical transmission system suffers from noise. Amongst others radio interference, thermal noise, impulse noise, etc. are always present. Therefore we will not discuss them is this paper. Besides these well-known noise sources, there is also an important disturbance typical for xdsl services, called crosstalk. Several twisted pairs are bundled together in cable binders. Due to the physical proximity of these twisted pairs, there is leakage of signal power from one channel into another. Crosstalk is precisely this coupling between pairs in the same cable. It increases with frequency and is worst between adjacent pairs. Two kinds of crosstalk can be distinguished. Near-end crosstalk (NEXT) is the coupling between transmitters and receivers at the same end of a cable. NEXT typically imposes the limit to xdsl system performance when the co-located transmitters and receivers use the same frequencies. Far-end crosstalk (FEXT) is the coupling between transmitters and receivers at opposite ends of a cable. Fig. 2 illustrates the difference between NEXT and FEXT. All these noise sources and especially crosstalk limit the maximum throughput of xdsl systems [2, 3]. B. Line length and attenuation Every electrical signal propagating along a transmission line is attenuated. The greater the distance between the customer and the central office, the higher the signal attenuation. In addition, in a cable high frequencies are more attenuated than low frequencies. Thus it is possible that a line of a certain length is perfectly suitable for voice communication, while being unable to support ADSL at full capacity. This problem is even more pronounced for VDSL, as here even higher frequencies are used. As a consequence, VDSL will only be applicable for very short lines (less than 1.5 km). C. Gauge change The wire diameter is an important characteristic when assessing the ability of a loop to act as a medium for data transmission. As a general rule one can state that the ability of a channel (of a given length) to carry information will increase with diameter. In North America twisted pairs are characterized by the American Wire Gauge designation (AWG), which is indicative of wire diameter. Typical twisted pair gauges are #19, #22, #24 and #26. In most markets outside of North America, e.g. in Europe, wires are classified according to their diameter in mm. Table 1 indicates the correspondence between AWG, wire diameter and DC resistance (in Ohms per kilometer) [1]. Loop resistance is an important parameter, since it fixes the maximal length of the loop. Thinner wires have a higher resistance but are cheaper and were thus used for short loops. Due to their lower resistance, thicker cables were used for longer loops. For manageability, long pairs tend to share cables with the thinner pairs for some distance, so the build of a long pair typically starts thin from the central office and gets progressively thicker. It is common to deploy 0.4 mm twisted pairs (AWG 26) along the first few kilometers from the central office to some primary or secondary connection point. Beyond this, successively heavier gauge can be used in order to avoid excessive attenuation. Thus, along its path from central office to customer, a loop can consist of several sections having different diameters (see left column of Fig. 1). TABLE 1 MMON TWISTED PAIR GAUGES CHARACTERISTICS AWG Wire diameter (mm) Loop resistance (Ω/km) at 20ºC

3 At these connection points, a part of the transmitted signal is reflected. These reflections have a negative impact on the overall system capacity. D. Taps In some networks, it can be quite common for a given twisted pair line to have another section of twisted pair connected at some point along its line, the final end of this unused twisted pair being an open circuit. This is known as a bridged tap: a section of wire pair connected to a loop on one end and not terminated at the other end. Fig. 3 shows a loop of length 2 km that connects a central office and a customer with a bridged tap of 500 m at the loop midpoint. Sometimes even several bridged taps can exist along the same loop (see right column of Fig. 1). Taps were often placed earlier by operators, typically when cabling a new neighborhood, in order to leave flexibility for future connections. Although bridged taps don t have any discernible effect on POTS performance, this is not true for xdsl transmission. Since taps have an open line end, the signals arriving at the end of the tap are completely reflected. This means in download, the customer receives the signal from the central twice: once through a direct path and once after reflection through the tap. The customer thus receives the sum of these two reflections, which causes interference. Due to the destructive interference, the transfer function will look approximately like depicted in Fig. 4. Certain frequencies will suffer severe loss and will therefore become unusable [5]. Taps are mainly present in the USA and Japan. They are not used as such in Europe, but customers typically make branches at home to connect several telephones and computers at different places in the house. This in-house wiring is also a kind of tap and has the same adverse effects on the channel capacity. E. Load Coils The technique of loading a loop was invented in 1900 by Michael Pupin. In essence, it involves placing a series of physical inductors called load coils at equally spaced intervals along the loop. A typical value of 88 mh placed at 1.8 km intervals has been used in the past for long line deployment. The benefit of this technique is that it improves the POTS transmission at the expense of greater attenuation at frequencies above the POTS band. The load-coil enhanced loop exhibits minimized attenuation in the POTS band but one highly unfortunate consequence of this is that xdsl transmission cannot take place on these loops because the rejection of frequencies in the xdsl band is magnified. On the other hand load coils are rarely found on loops shorter than 5 km. If any load coils exist on lines intended for xdsl, they must be located and removed in order for xdsl to be successfully deployed. III. SELT VS DELT From Section II, it is now clear that estimating the loop capacity is not an easy task, as it depends on many factors. It is therefore difficult for the operators to make an Central Office 1 km 0.5 mm Open line end 500 m 0.5 mm 1 km 0.5 mm Fig. 3: Example of a bridged tap customer educated guess about the achievable xdsl service, without a measurement. Commercial available instrumentation for the measurement of the transfer function already exists but it is based on double-ended line testing (DELT). It requires a technician at both line extremities in order to quantify the loop. Sending a technician at the customer s home for each subscription makes the determination of the achievable xdsl service expensive and in addition it is disturbing for the customer. Therefore, it is desirable to have a technique that could identify and qualify all the subscriber loops in an automated and highly accurate manner without the intervention of staff at the subscriber s location. This explains the recent shift of focus to single-ended line testing (SELT). The idea is to perform measurements only at the Central Office and using advanced processing techniques to come to a reasonable estimate of the channel capacity in bits/s of the subscriber line. This eliminates the dispatching of a technician but is of course much more complex than with DELT. With DELT the transfer function can be measured directly, which is not possible with SELT because connecting a measurement device at the customers house is not allowed. With SELT, the loop make-up (see Fig. 1) has to be estimated first and from this, the transfer function and the channel capacity can be calculated. At the moment no operational SELT-system exists which can predict the subscriber loop capacity accurately in all cases. Consequently research in this field is still ongoing [4, 6-12]. IV. LOOP CAPACITY ESTIMATION WITH SELT Discovering information about the loop make-up through single-ended line tests is possible thanks to Time Domain Reflectometry. Transfer Function no tap tap Fig. 4: Transfer function with and without tap Frequency

4 A. Time Domain Reflectometry The basic principle of Time Domain Reflectometry (TDR) is to inject an excitation signal in the transmission line under test and to collect and analyze the reflections in order to extract the desired information. For SELT, this means an excitation signal is injected in the subscriber line at the central office. The signal propagates along the line and if a discontinuity is present, a part of the signal will be reflected and thus returns to the central office. This is schematically illustrated in Fig. 5. Several kinds of discontinuities can occur in a copper access network. As discussed in Section II.C, a subscriber loop typically consists of several sections having different wire diameters. Each gauge change is a discontinuity and causes a reflection. The line end at the customer side is also a discontinuity because the line ends there. Possible taps and load coils are also discontinuities. All of these cause reflections, which can be collected at the central office. By analyzing these reflections, the make-up of the subscriber loop can be discovered. The amount of reflection is given by (1) where Z 1 is the characteristic impedance of the line before the discontinuity and Z 2 is the characteristic impedance of the line after the discontinuity. Z Z 2 1 ρ = (1) Z 2 + Z1 Because cables with different diameters have different characteristic impedances (Z 1 Z 2 ), each junction will cause a reflection. As explained in Section II.C, in most European networks the changes are from thinner to thicker cables and the thicker the cable, the lower the characteristic impedance. This means at a gauge change the reflection will be negative (Z 2 < Z 1 ). If the line end is left open or has a sufficiently high impedance, as is often the case in practice, the reflection will be positive (Z 2 = ). Following a similar reasoning, a shorted line end will have a negative reflection (Z 2 = 0). Fig. 6 gives an example of such reflections. By determining the start time of a peak (t start ) it is possible to discover the line length l with (2) if the velocity of propagation v is known a priori. v. t l = (2) 2 with t = t start t ref. The factor 2 comes from the fact that the signal travels forwards until it reaches the discontinuity and then travels backwards to the central office. In Fig. 6 the first reflections starts approximately 5 µs after the injection of the signal (t ref = 0). Since the propagation speed in a twisted pair is about meters per second, this means the first cable has an approximate length of 500 m. The second reflection starts 15 µs after the start of the first reflection, thus leading to an approximate length of 1500 m for the second line segment (t ref = t start1 = 5 µs). B. Loop estimation with SELT The first systems that exploited TDR used a device that presented a graphical representation of the reflectogram (voltage vs. time) to the user. The user had to be experienced to decipher the information that was shown central office voltage excitation response: reflectogram subscriber loop Z 1 Z 2 Fig. 5: Time Domain Reflectometry customer on the screen. The expert user then emitted a judgment about the topology of the loop. However, this technique lacks the precision to detect small but important reflections. Furthermore, the use of this kind of system leads to delays and inaccuracies in the loop identification process. For this reason, research is conducted to investigate the possibility to create a fully automated system for the measurement and interpretation of reflectograms of the local access network of telephone companies [4, 6-9]. The system should automatically interface with the subscriber loops in order to obtain the TDR measurements. The resulting signal should then be cleaned-up by filtering, followed by signal processing algorithms to allow extracting its most important features (e.g. the start of a reflection). Once the features are defined, a reasoning system deduces the most probable topology and determines the line length of each section. This produces the estimated subscriber loop topology [9]. The system in this way will not require human intervention for its normal operation, avoiding as such delays and human errors. If the noise characteristics are known, using a direct measurement or a model-based estimation, then a bit rate prediction is possible [8]. Several approaches for the reasoning system are possible. For example, neural networks or models with s11 [dimensionless] x valid maximum start of first reflection start of second reflection valid minimum Physical loop make-up 500m 1500m time [s] time reflectogram maximum minimum inflection point Fig. 6: A reflectogram with maxima, minima and inflection points indicated for a 500 m segment (0.4 mm diameter) in series with a 1500 m segment (0.6 mm diameter) x 10-4

5 lots of parameters are possible but lack physical meaning (black box approach). Another approach, used at our department, is based on physical cable models (white box approach) [7, 8, 9]. Several other research groups are also working on a white box aproach [10, 11, 12, 13]. C. Difficulties First of all, some pre-processing is needed to clean up the measured signal. The measured signal inevitably contains noise, but also artifacts introduced by the imperfect measurement [7]. Secondly, the feature extraction analyzes the reflectogram in order to extract the features, being the number of peaks visible in the signal and their attributes. These peaks correspond to reflected pulses travelling back to the central office due to impedance discontinuities. Important attributes are the start and the end of each observed reflection, the position of the extremum and the type of extremum (maximum or minimum). Lets discuss Fig. 6 to show that the extraction of the features is not straightforward. The first reflection is negative and is due to the splice of the two line segments. The second reflection is due to the line end after a total of 2 km and is positive. It can be seen that the second reflection is superimposed on the tail of the first reflection because the second reflection arrives while the tail of the first reflection is not yet completely extinct. These slow decaying tails are inherent to the copper transmission line and are complicating the feature extraction. Fig. 6 is a very simple example, but once the topology gets more complicated or when one deals with longer line lengths, the feature extraction becomes very difficult [6]. Consequently, the feature extraction s task can be reformulated as follows. It should decompose the reflectogram, which represents the sum of all the reflections, to a series of non-interfering reflections. This is necessary because each k-th peak p k is superimposed on the tail of the preceding peaks p k-1, p k-2,, p 1, therefore distorting the isolated shape of the individual peaks. Finally, a reasoning system has to analyze these features in order to identify the topology of the loop. For this, expert knowledge has to be translated into coding rules to be programmed in the reasoning system. This is a very difficult task because an important part of this expert knowledge is intuitive or is hard to describe in the form of if-then rules [4, 7]. V. APPLICATIONS Determining the transfer function from single-ended measurements is thus possible, but far from easy. Due to the fact that all the measurements can be done from the central office, SELT is often preferred by the telecom operators. However, in contrast to DELT, the loop transfer function cannot be measured directly from SELT data. As explained in Section IV, first the loop make-up must be identified from the TDR-measurements by advanced signal processing techniques. Once the loop make-up is known, the transfer function can be calculated and the achievable xdsl service can be estimated. At the moment no operational SELT-system exists which can predict the subscriber loop accurately in all cases. But if it would be feasible to perform loop identification via single-ended measurements with sufficient accuracy and in an automated way, then operators would benefit substantially. The following subsections give some examples of possible applications. Being able to perform all the measurements and identification from the central office however opens many more possibilities, most of which probably cannot even be imagined at the moment. A. Estimation of the achievable xdsl service As described in this paper, SELT is mainly used for the estimation of the make-up of a subscriber loop. An accurate knowledge of their network allows the telecom operators to estimate the achievable xdsl service for a particular customer. Especially when deploying VDSL, the knowledge of the network will be a valuable asset. Even more, this knowledge might induce completely new commercial applications. Instead of offering a number of packages the customers can subscribe to, in the future a completely personalized approach would be possible. When a new customer wishes to subscribe the operator just needs to perform a measurement. This will allow the estimation of the maximum capacity of this particular customer line and a personalized offer can then be made. B. Updating records As explained in the introduction, loop topology records are often missing or incomplete. If it is feasible to perform loop identification via single-ended measurements in an automated way, all the lines of the network can be tested on a regular base. Using this automated system, the information of the local loop can be updated in an accurate and detailed way. Ideally, this will lead to complete and up-to-date records. With this information the telephone company will be able to deploy xdsl services in a more cost-effective way. Moreover, the complete knowledge of the network strongly facilitates its maintenance and administration, resulting in a better service for the customer. For example, assume the complete network make-up is known and a customer calls because his connection is not working anymore, or working at reduced speed. Performing a new measurement and comparing it with the known loop topology, allows easy troubleshooting. If the cable is broken for example, the point where the two measurements differ indicates the location of the fault. C. Quality control tool With the liberalization of the European telecommunication markets, it has become common for several operators to share a same network infrastructure. The service providers leasing a line often don t have access to the whole network, but are only allowed to some parts. In this context, SELT measurements can be very useful as a quality control tool. Because the service providers don t have physical access to the lines except at the racks in the central office, they have no guarantee about the quality of the leased lines. With SELT measurements, the loop make-up can be identified, which opens much more possibilities for marketing purposes.

6 VI. NCLUSIONS AND FUTURE WORK One of the key reasons for the wide deployment of xdsl technologies is that it runs over the existing telephone lines. Therefore a further enhancement of the copper telephone network as a medium for carrying data information is to be expected. Because the telephone lines were initially not designed for high-speed digital communications, they encounter several problems when using broadband signals. Nowadays, telecom operators base their xdsl-tariffs mainly on the offered bit rate. However, the maximal achievable bit rate is difficult to predict as it is different for each customer and depends on the cabling between the central office and the subscriber. In order to identify whether a subscriber loop is suitable for xdsl service, the transfer function of the loop has to be estimated. Several measurement techniques exist, however Single Ended Line Testing (SELT) is gaining much attention lately. This method allows only measuring at the central office side and yields a reflectogram. From this reflectogram, with the necessary expert knowledge and signal processing techniques, the loop make-up can be identified. From this, the transfer function can be calculated and with some knowledge about the noise environment, the maximum achievable bit rate of the line under consideration can be estimated. When it is feasible to perform loop topology identification via single-ended measurements with sufficient accuracy and in an automated way, then several promising applications open up for the operators. Besides automatic qualification of a loop for xdsl service, this capability will allow the updating of telephone company loop make-up records. These records can in turn be accessed to facilitate administration, provisioning and maintenance operations. REFERENCES [1] P. Golden, H. Dedieu and K. Jacobsen, Fundamentals of DSL Technology, Auerbach Publications, [2] T. Bostoen, M. La Fauci, M. Luise and P. Boets, Disturber Identification for Single-Ended Line Testing (SELT), IASTED International Conference on Communications, Internet and Information Technology (CIIT 2003), Scottsdale, AZ, USA, November, 2003 [3] S. Galli, C. Valenti, K. Kerpez, A Frequency-Domain Approach to Crosstalk Identification in DSL Systems, IEEE Journal on Selected Areas in Communications, Special Issue on Multiuser Detection Techniques with Application to Wired and Wireless Communications Systems (Part I), vol.19, no.8, August 2001 [4] X. Ochoa Chehab, Expert System for Automated Time Domain Reflectogram Interpretation using a Double Level Generate-Test- Debug Approach, Master Thesis, Dept. ELEC, Vrije Universiteit Brussel, 2002 [5] A. Wang, J. Werner, S. Kallel, Effect of bridged taps on channel capacity at VDSL frequencies, IEEE International Conference on Communications, Vol. 1, 6-10 June 1999 pp [6] C. Neus, Feature Extraction of One Port Scattering Parameters for Single Ended Line Testing, Master Thesis, Dept. ELEC, Vrije Universiteit Brussel, 2004 [7] T. Vermeiren, T. Bostoen, F. Louage, P. Boets, and X. O. Chebab, Subscriber Loop Topology Classification by means of Time Domain Reflectometry, IEEE International Conference on Communications, Anchorage USA, May, 2003 [8] T. Bostoen, P. Boets, M. Zekri, L. Van Biesen, T. Pollet, and D. Rabijns, Estimation of the Transfer Function of the Access Network by means of One-Port Scattering Parameter Measurements at the Central Office, IEEE Journal of Selected Areas in Communication-Twisted Pair Transmission, Vol. 20, No. 5, june 2002, pp [9] P. Boets, Tom Bostoen, L. Van Biesen, D. Gardan Single-Ended Line Testing - A Whitebox Approach, 4th IASTED International Multi-Conference Wireless and Optical Communications, Banff, Canada, 8-10 July, 2004 [10] S. Galli and D.L. Waring, Loop Makeup Identification Via Single Ended Testing: Beyond Mere Loop Qualification, IEEE Journal of Selected Areas in Communication-Twisted Pair Transmission, Vol. 20, No. 5, June 2002, pp [11] S. Galli and K.J. Kerpez, Single-Ended Loop Make-Up Identification -Part I: A Method of Analyzing TDR Measurements, IEEE Transactions on Instrumentation and Measurement, Vol. 55, No. 2, April 2006, pp [12] K.J. Kerpez and S. Galli, Single-Ended Loop Make-Up Identification -Part II: Improved Algorithms and Performance Results, IEEE Transactions on Instrumentation and Measurement, Vol. 55, No. 2, April 2006, pp [13] J. Yoho, Physically-Based Realizable Modeling and Network Synthesis of Subscriber Loops Utilized in DSL Technology, PhD Thesis, Virginia Polytechnic Institute and State University, 2001