DSL: Technology and Implementation
The popularity of Internet access, combined with demand for speed accelerated by the use of multimedia content, has highlighted the bottleneck of analog modem technology in the last mile -- the connection between the subscribers and telephone companies (telcos). Digital subscriber line, or DSL, is a technology developed by the telcos to solve the problem. This article examines the underlying technology, and explores some of the implementation issues faced by administrators.
Because DSL can use existing phone lines, no new wiring is required. In addition to providing a high-speed data pipe, DSL provides voice service, even under a power failure that brings down the DSL equipment. DSL is an always on technology, so no connection setup is needed, as is the case for dialup connections using analog and ISDN (Integrated Services Digital Network) modems. Another big incentive behind the DSL push is its competitive pricing compared to other high-speed access technologies. However, DSL comes with many restrictions and challenges -- its deployment may require careful planning, research, and evaluation. This article will first help you build a solid foundation on various DSL technologies, then discuss the implementation issues in DSL deployment for your networks. A four-step implementation guideline and an example of using DSL technologies in enhancing network connectivity are provided.
The x of xDSL
All DSL technologies, collectively called xDSL, use advanced modulation schemes to carry high-speed data traffic in existing telephone wires. The differences among these flavors mainly lie in their maximum line rate, reach (operating distance), and supported applications. The following discussion will concentrate on common DSL flavors. It is important to understand where they differ so that you can use the ones that most appropriately address your network needs. The following DSL flavors are discussed in this section -- Asymmetric DSL (ADSL), High-bit-rate DSL (HDSL), Symmetric DSL (SDSL), and ISDN DSL (IDSL).
ADSL, as the name implies, is an asymmetric DSL, which limits its use primarily for home users or small offices at high speeds, although symmetric services are provided at lower speeds. Its downstream speed is significantly higher than its upstream speed. The actual line rate depends on the distance between the subscriber and CO (Central Office), wire gauge, modulation techniques, and line quality. ADSL is suitable for applications that primarily download heavily downstream to the users, such as Web surfing, ftp file download, and video on demand.
Within ADSL, there are two competing technologies that use different modulation schemes. Developed by Globespan Technologies when it was the Paradyne unit of AT&T, CAP (Carrierless Amplitude and Phase modulation) became a popular ADSL technology because its coding scheme is well understood, and it is less expensive to implement. CAP uses a single carrier modulation, and is therefore more susceptible to narrow band interference. DMT (Discrete MultiTone), the competing technology, uses a much more complex modulation scheme. It splits the bandwidth to 256 discrete sub-bands of 4 kHz each and modulates each sub-band individually. To increase performance, more data resides on lower frequency bands and less on higher frequency bands. DMT is a multicarrier technique and much more complex to implement. DMT consumes more power but can achieve a higher line rate than CAP. The biggest advantage of DMT over CAP is that DMT is an ANSI (American National Standards Institute) standard. The current DMT standard, Issue 2, is specified in the ANSI T1.413 document.
HDSL is the most extensively deployed and mature DSL. HDSL is commonly provisioned by telcos to provide T1 (T-carrier digital signaling level 1) service. HDSL uses a modulation scheme called 2B1Q (2 Binary, 1 Quaternary), used extensively in ISDN. 2B1Q improves coding efficiency by sending two bits of information with each cycle of analog waveform or baud. In traditional T1 provisioning, repeaters need to be installed within 2,000-3,000 feet of the endpoints and not more than 3,000-6,000 feet between repeaters, depending on the wire gauge used. For example, the maximum distance between two repeaters is less than 6,000 feet if 22 AWG (American Wire Gauge) wire is used. Traditional T1 also requires that all bridge taps be removed. The loop engineering work that is required tremendously increases the cost of traditional T1 service. By using 2B1Q and splitting the service into two pairs of wires (four wires), HDSL achieves T1 speed up to 12,000 feet without using repeaters. HDSL can also tolerate some bridge taps.
SDSL is HDSL on a single loop (two wires). Using only two wires reduces the cost of SDSL deployment, although the reach is also reduced compared to HDSL. SDSL can be a low-cost replacement for leased-line service, with proper wire gauges and within a shorter distance. Both HDSL and SDSL provide symmetric line rates. Because HDSL and SDSL use the voice band frequency, analog voice service is not provided. Typical applications for HDSL and SDSL are data services that currently use leased line, Frame Relay, and ISDN.
IDSL provides DSL service using ISDN circuits. The same 2B1Q line coding is used for IDSL. Line rate is generally the same as ISDN BRI, 128 kbps, but can be as high as 144 kbps when the D channel is also used for data. IDSL provides the longest reach of all DSL flavors. Unlike ISDN, IDSL does not connect through telco's voice switch and does not provide voice service. Existing ISDN customers can easily switch to IDSL if the connection is only for data service. A summary table for the various DSL technologies is provided in Table 1.
The local loop, the twisted-pair connection between a phone subscriber and the telco's CO, was designed to provide POTS (Plain Old Telephone Service), which occupies frequencies from 0 to 3.4 kHz. This narrow band has been used to provide analog voice service and analog modem connections.
To provide high-speed line rate, DSL eliminates the POTS frequency limit. For example, ADSL uses frequencies up to 1 MHz. However, there are two problems when using high frequencies, both of which reduce the reach and distort the signals -- attenuation and crosstalk increase as frequency gets higher. It is important to note that the effect of the two is uneven. In some situations, the attenuation may be the primary factor; in other situations, the crosstalk may be the primary problem. There are two types of crosstalk -- NEXT (Near End Crosstalk) and FEXT (Far End Crosstalk), with NEXT being the most significant in terms of its impact on performance.
To reduce attenuation, DSL uses an advanced modulation scheme to send more information bits into each frequency cycle, thus raising the information bandwidth per frequency spectrum. In the case of HDSL, the frequency spectrum required is reduced by splitting the service into two pairs of wires. Equalization is another technique used to compensate for attenuation and phase error.
DSL uses two techniques to reduce crosstalk. Crosstalk is created when there is frequency overlap. DSL uses FDM (Frequency Division Multiplexing) to separate upstream frequency from downstream frequency, thus eliminating NEXT. This technique is used in ADSL, where upstream frequency is between 25 to 138 kHz and downstream frequency is between 142 and 1104 kHz. The POTS occupies the lower bandwidth. The reason that upstream spectrum is lower than that of downstream is that more crosstalk is introduced when transmitting toward the CO (upstream) than away from the CO (downstream). The downside of FDM is that the total spectrum used for data is reduced.
Another technique that DSL uses to reduce the crosstalk is echo cancellation. Upstream and downstream signals are allowed to overlap. Echo cancellation is used to separate the far-end signal from the near-end echo in the overlapped band. Echo cancellation is an optional feature in ADSL DMT. Because the far-end signal is attenuated, FEXT normally does not present a big problem. Echo cancellation increases available bandwidth and is effective when there is echo from one source. Bridge taps can complicate the echo. The downside of the echo cancellation system is increased cost in circuitry. Echo cancellation must be supported on both ends for it to work. A side effect of echo cancellation is self NEXT, a form of NEXT caused by the neighboring echo cancellation loops in the same cable binder. Therefore, standard performance testing, such as defined in ANSI T1.413 for DMT, includes various types of crosstalk and different numbers of interferers. This should be kept in mind when evaluating DSL performance claims.
This section focuses on the complete picture of how DSL is connected. Different DSL flavors use a slightly different connection, and I have selected ADSL as an example because of its additional POTS service. The overall architecture, however, is the same for DSL implementations.
There are several major components in a typical ADSL configuration (Figure 1). Starting from the subscriber side, a POTS splitter splits DSL (upper band) from voice (lower band). The CPE (Customer Premise Equipment), called ATU-R (ADSL Transmission Unit-Remote) for ADSL, provides the data connection, usually via an Ethernet port. The POTS splitter on the central office side passes the POTS to a voice switch, and the DSL data connection is terminated in a DSLAM (DSL Access Multiplexer), which houses many DSL modem cards called ATU-C (ADSL Transmission Unit-CO). From DSLAM on, the network is cell-based (ATM, or Asynchronous Transfer Mode) or frame-based (Frame Relay). The access network generally consists of ATM or Frame Relay switches and routers. The physical location and ownership of DSLAM and access network depend on who is providing the service. In the case of ILEC (Incumbent Local Exchange Carrier), both pieces of equipment would be at the central office and belong to ILEC. In the case of CLEC (Competitive Local Exchange Carrier), DSLAM could be located in an ILEC's central office in a collocation cage but the access network would be in a different location controlled by the CLEC. ISPs (Internet Service Providers) and corporate networks must provide the connections to the access network.
DSL providers generally oversubscribe the DSL modem banks to increase the modem utilization (i.e., there is no one-to-one relationship between ATU-R and ATU-C in the case of ADSL). The ratio of oversubscription determines how many subscribers are competing for one DSL modem. Assuming that not all subscribers are online at the same time, this generally may not present a problem. It is also important to note that all subscribers from the same DSLAM compete for the uplink, usually DS3 (45 Mbps) or OC3 (155 Mbps). To reduce the cost and complexity of installation, proposals of using splitterless (no POTS splitter) or microfilters (replacing regular POTS splitters) are being considered. Voice over IP is another option to carry POTS over DSL.
Like all emerging technologies, DSL implementation presents a lot of challenges in terms of technologies, service availability, ease of use, setup, cost, and more. I'll cover many of these issues in this section, so you'll know what to expect when deploying DSL service.
Local Loop Conditions
To increase the coverage for voice services, telcos over the years have deployed various techniques such as installing loading coils, bridge taps, and remote terminals. Loading coils are not compatible with DSL and must be removed if DSL service is to be provided on that loop. Although a limited number of bridge taps are tolerable to DSL, they reduce the overall performance and reach. Remote terminals are also not compatible with DSL; if they are used, DSL must be terminated at the remote terminal and be converted to a signal compatible with the remote terminal.
Many types of loops exist today, depending on wire gauge, length of each gauge, and bridge taps. Before a loop can carry DSL, the telco has to determine the quality of the loop. For a loop to be DSL qualified, all loading coils and remote terminals must be removed. Additionally, there is a distance limitation, as indicated in Table 1. The distance is further modified by the type of wire gauge used and the number and length of bridge taps. For example, themaximum distance for ADSL between the CO and a subscriber is 15,000 feet for 26 AWG copper and 18,000 feet for 24 AWG copper. Some specific limitations for bridge taps are that all bridge taps must be counted toward total loop length and that the maximum length of any bridge tap should be less than 2,000 feet.
Currently DSL services are only available in major metropolitan cities. As customers are more receptive and telcos become more confident, services are expected to expand rapidly to other areas. The type and speed of DSL may vary from different providers. Other considerations of the service availability issue are how high the modem oversubscription is, how many subscribers compete for the uplink, and the speed of the uplink. Resource sharing can lead to problems of service availability.
To use DSL for the Internet access, ISPs must provide connections from their POPs (Point Of Presence) to DSL providers. However, there are many ISPs that still do not have such connections. On the positive side, many DSL telcos provide Internet access -- such national telcos as RBOCs (Regional Bell Operating Companies), GTE, and SNET. A few CLECs, such as Covad, Northpoint, and Rhythms, also provide DSL services. The partnership between DSL service providers and ISPs helps smooth the installation and service.
Different DSLs may provide different selections of services, depending on applications, speed requirement, and loop conditions. Businesses tend to select symmetric DSL, such as HDSL, SDSL, or IDSL because the nature of data transfer is symmetric. ADSL can also be symmetric at lower speeds. ADSL is the choice if higher download speeds are required. Achievable speed is dependent on loop conditions. Longer loops may be forced to use lower speeds.
DSL pricing is generally dependent on the speed, the type of DSL, and the type of service. Each DSL flavor comes with many speeds selectable by the customer. The customer generally pays for installation, inside wiring if required, CPE, and a flat monthly charge. The type of service may include voice over IP, quality of service, the number of IP addresses, etc. If the same company does not provide DSL and Internet service, pricing may be more complex. Pricing and availability information can be found at http://digiquote.telcoexchange.com/.
There are two competing ADSL flavors -- DMT and CAP. While CAP has an early lead, DMT is the ANSI standard. It appears that telcos are slowly migrating from CAP to DMT. From a user's point of view, there is no big difference if staying with either. The problem is migrating from one to another, which increases the uncertainty. Staying with the standard may prove beneficial down the line, as there will be fewer equipment interoperability problems.
The following general guidelines can be used when setting up DSL connections:
1. Determine the network needs. If the purpose of DSL is for telecommuters to access existing corporate LANs, then ADSL and IDSL are probably good choices. For inter-office connections, HDSL and SDSL are good candidates. SDSL is a low-cost replacement for existing T1-leased lines. The final result of this step should be a network diagram.
2. Once the needs are clearly defined, find appropriate service providers. When selecting a DSL provider, run through the issues discussed previously. If the purpose of the DSL line is for Internet access, it is probably best to find a DSL provider that is also an ISP. If the DSL is to provide the inter-office connection, both sites can have DSL links, or one end can have other types of WAN (Wide Area Network) links. One important aspect for this step is to decide whether to outsource the support for the customer equipment. If delay-sensitive traffic, such as packetized voice, is to be passed on the network, it is a good idea to obtain a quality of service guarantee.
3. Purchase the right equipment based on requirements set in the previous two steps. The type of DSL equipment must match the type used by the providers (e.g., DMT vs. DMT, SDSL vs. SDSL). If both the CO and subscriber equipment fully support the standard, there should not be an interoperability issue. It is best to purchase the CPE from the vendor recommended by the DSL provider to avoid the potential problem from the start.
4. Setup and testing. The provider would provision the line based on the agreement. There are several configuration models, but users are generally required to use the model supported by the provider. There should not be a compatibility issue if previous steps are executed correctly. Depending on the flavor of DSL, the CPE can generally act as a router or a bridge. In the router mode, the CPE provides a lot of layer 3 functions, as discussed in the next section. In the bridge mode, the CPE acts as a transparent device, thus address assignment and negotiation and user authentication must reside on a PC or a conventional router. The number of layer 3 addresses (e.g., IP addresses) may be limited in a bridge mode. In either configuration, it is a good idea to conduct extensive setup tests before full-scale rollout. Once the connection is up and running, it is also important to validate the DSL performance against the provider's claims and monitor it regularly.
An Implementation Example
This section provides an example that offers a more detailed look on how DSL can be used to provide network connectivity. I will discuss ADSL and, for simplicity, I will focus only on data connections.
The major piece of user equipment is the ADSL modem (ATU-R) that is capable of supporting the type of ADSL (CAP or DMT Issue 1 or Issue 2) that must match the type of ATU-C on the DSLAM. Depending on the connection model used by the DSL telco, there are several variations. The ATR-R is generally able to act as a layer 2 device (a bridge) or a layer 3 device (a router). ATU-R may come as a form of NIC (network interface card), if only one PC needs to be connected. If the CPE acts as a bridge, the configuration complexity shifts to the PC. If it acts as a router, the PC configuration would be the same as if it were connected to a LAN. The advantage of ATU-R as a router is that it can handle a lot of layer 3 functions, such as address negotiation and assignment, firewall, address translation, and terminating layer 2 broadcast traffic. Multiple PCs can be connected to the hub and then to the Ethernet port of the ATU-R. Wiring is straightforward, as shown in Figure 2.
A common configuration is shown in Figure 2. In this configuration, ATU-R acts as a router. ADSL connection is terminated in both ATU-R and ATU-C. ATM is the transport running on the telco's network. ATU-R uses PVCs (Permanent Virtual Circuit) to send and receive traffic from DSLAM, which maps them to telco's ATM network, completing the end-to-end layer 2 connection that terminates at the telco's router. PPP (Point-to-Point Protocol) runs on top of ATM, providing services to layer 3 (IP). PPP is a preferred choice because of its support for authentication, layer 3 address autoconfiguration, layer 3 transparency, and support for RADIUS (Remote Authentication Dial-In User Service).
On startup, ATU-R and ATU-C will go through an ADSL training process to negotiate a proper speed based on line provisioning and line quality. Various DSL capabilities are exchanged and synchronized at this step. Once it is successfully completed (trained), ATU-R would normally indicate a steady WAN link LED. The PPP session will then be initiated between the ATU-R and the router. Centralized RADIUS servers can perform authentication, and ATU-R would negotiate a dynamic IP address from the RADIUS server. ATU-R can also act as a DHCP (Dynamic Host Configuration Protocol) server or proxy, providing dynamic addresses to PCs while translating addresses using the dynamically obtained address for itself. Most of these tasks on ATU-R can be accomplished with little or no configuration on the part of the user. In many cases, the default configuration coming with the ATU-R is enough to get the user connected immediately. Once PPP sessions are successfully established, user traffic can flow from the PC, through the DSL and ATM, past the router, to other networks, such as the ISP and the Internet. User traffic can be prioritized through the ATM network, providing quality of service support.
As an emerging technology, DSL evolves rapidly. DSL brings with it a high-speed broadband network that promises to change the whole network access landscape. This network would finally enable newer and bandwidth-demanding contents and applications, such as video on demand and multimedia teleconferencing. This article has provided extensive discussion on technical and implementation-related issues and guidelines for taking advantage of the DSL technology. Further information and development in DSL can be found on the Internet at www.adsl.com, www.telcoexchange.com, www.cisco.com, www.alcatel.com, www.coppermountain.com, www.3com.com, and www.compaq.com.
About the Author
Randy Zhang, Ph.D., is a software engineer at Cisco Systems DSL Business Unit. He can be reached at: email@example.com.