Use G.Lite DSL Modems For Broadband Residential Applications

Digital-subscriber-line modems are spearheading the race for high-connect speeds in the residential marketplace. There, modern web pages and e-commerce are creating a demand for greater bandwidth. Growing interest has given rise to a number of issues in the design and implementation of systems based on digital subscriber lines (DSLs). These include monthly connect fees, ease and cost of installation, interoperability with central-office (CO) equipment, and the ultimate issue: "How fast will it connect?"

One key distinction, in terms of use and equipment needs, is between consumer-broadband and commercial-broadband service. A high level of activity resides in the commercial segment, where service fees are estimated to be as high as $49 billion annually. Its opportunities appear so attractive that many residential-related issues have been neglected.

The primary differences between commercial and consumer service involve monthly service fees and guaranteed level of service between the two account types. Most commercial accounts are being offered monthly services in the $100 range, with minimum guaranteed download speeds of 768 kbits/s and higher. The typical residential account is being offered rates of $30 to $50 per month for minimum guaranteed download speeds of 384 kbits/s.

The differences aren't merely the result of marketing and positioning, but are derived directly from system issues. Commercial establishments tend to be located relatively near their CO, while the average household is roughly 12.5 kft. away. This 12.5-kft. median distance is significant in terms of available bandwidth. The only way to shorten it is to install more COs in the residential territories. While this would alleviate some of the distance-related bandwidth issues, however, it wouldn't solve the Internet delivery problems that exist in today's implementation.

Local Internet traffic may well be available at rates above 1 Mbit/s. But practical data rates for information traveling from one city to the next appear to be in the 1-Mbit/s range and below. The net result is that today's Internet should be able to consistently supply data in the 1-Mbit/s range. Meanwhile, rapid improvements in backbone capability are being consumed by the growing demand for Internet-related content. With the continued increase in new users, there won't likely be a major increase in the Internet's "practical capability" for residential users during the next few years.

Supporting widespread adoption of DSL technology requires easy and straightforward installation, with no surprises like unexpectedly low connect rates. The line between full-rate (8-Mbit/s maximum download) and G.Lite service (1.5-Mbit/s maximum download) is somewhat blurred. So many consumers will expect full-rate connect speeds when they bring a DSL device home for installation. Signal loss through the line, combined with noise sources at both the near and the far ends, tends to limit the connection's available bandwidth. Consumers located more than 10 kft. from the CO will see a significant drop in available bandwidth when compared with the promises seen in much of the promotional material.

The full-rate specification chip sets support a number of protocols. They're also expensive. A modem like the G.Lite-compliant one, which is specifically designed for the residential environment,will cost less and perform better.

The distance from the CO to the CPE site affects signal strength and frequency response. Plus, the current local-loop infrastructure has been installed over many decades. A typical local loop may have several gauges and types of wire between the CO and the CPE site. Because the real-world installed plant of twisted-pair copper isn't uniform, the ITU has specified a number of representative test loops that normalize testing and help ensure interoperability. Several of the ANSI T1.601 test-loop standards are shown in Figure 1.

Test loop #12 is composed of a 7500- ft. 26-gauge segment trailed by a 4500-ft. 24-gauge segment. That second segment is followed by a 1500-ft. 26-gauge one stretching from the CO to the CPE. The characteristics of a given segment can be modeled, as in Figure 2. Each of these sections has a fairly well-defined impedance characteristic that can be approximated using the standard tables for R/C/L/G components and transmission line theory.

Using a standard 2-port model in Figure 3, a section of twisted-pair line can be modeled in standard ABCD matrix form, as seen in equations 1 and 2:

V₁ = AV₂ + BI₂      (1)

I₁ = CV₂ + DI₂      (2)

By multiplying the resulting matrices from each segment of the local loop together, a single set of equations are generated. These closely approximate the actual loop's performance characteristics:

Rearranging Equations 1 and 2, it's possible to derive a transfer function for the entire loop characteristics:

In practice, the local loop from the CO to the CPE site may include various branches and stubs. At one time, these may have been used to service another location in the particular multi-pair bundle's service area. Known as "bridge taps," the bundles present unterminated loads in the middle of the local loop.

Load coils may have been installed on longer lines to improve the 0- to 4-kHz voice-band performance. Those coils are incompatible with broadband xDSL operation, so they'll have to be taken off the line. The procedures for removing load coils are well understood. Today, they're a standard aspect of ISDN installations. For the sake of simplicity, assume that the example loop doesn't have load coils and the test-loop case doesn't include bridge taps.

T1.601 test-loop case number 2, loop #7 has been selected and simulated using standard published characteristics for 26-gauge wire extending for 13.5 kft. The resulting attenuation and frequency response can be seen in Figure 4. By applying those line characteristics to the standard power-spectral-density (PSD) transmission template in Figure 5, it's possible to generate the CO output signal level seen at the CPE input in Figure 6. Clearly, significant attenuation of the transmitted signal is due to the inherent line impedance.

When analyzing a DSL line, several noise sources must be considered in addition to the local-loop attenuation. These include additional CO transmitters, more CPE transmitters operating with the same copper-pair bundle, and the local CPE transceiver noise sources. If the CO can handle one DSL line, it stands to reason that several others are in operation there as well. Secondary CO transmitter noise is coherent to the CO transmit signal on the DSL local loop. When the other local loops share the same bundle, it couples into it as a crosstalk effect.

The case for local CPE transmission disturbers is similar, but they're considered noncoherent noise sources. The CO transmitters all operate from the same clock. The CPE transceivers are locked to this clock. Since the CO transmitter signals travel in the same bundle, they arrive at the local CPE in sync and coherently.

In contrast to this, the near-end local CPE transmitters travel various distances to reach the CPE transceivers. So they're considered noncoherent noise sources. This noncoherent character generates significant sideband energy content.

The far-end crosstalk (FEXT) is modeled in Equations 5, 6, and 7. The line-attenuation effect calculated previously is applied to arrive at the level of FEXT noise present at the CPE transceiver input (Fig. 7). The FEXT disturbers are highly attenuated, because they must travel the entire length from the CO to the local CPE. From the ANSI T1.413-1995 specification, the FEXT is approximated as follows:

This equation gives the single-sided PSD. K_ADSL is the total transmitted power for the downstream ADSL transmitter before the application of the ATU-C maximum-transmission PSD template (Fig. 5, again). The sampling frequency is f_o,

is a fourth-order low-pass filter with a 3-dB point at 552 kHz, and

is a fourth-order high-pass filter with a 3-dB point at 20 kHz, separating the ADSL and POTS spectrums.

While approximating the near-end crosstalk (NEXT), note that the energy content is quite substantial as evidenced in Figure 7. This analysis assumes that the near-end transmitters adhere to the PSD of the ATU-R transmission template in Figure 8. That template defines the allowed CPE transmit power-spectral density. Using it permits reduction of the NEXT noise approximation, especially compared to the near-end transmitters that don't adhere to the stringent out-of-band attenuation limits.

The last external noise source to ponder is the inherent noise of the line. This noise is typically modeled at a −140-dBm level. But in practice, slightly less noise is usually seen. Typically, it's at a −144-dBm level.

The CPE transceiver ADC quantization-noise contribution also is indicated in Figure 7. It too is the result of noise shaping. Naturally, if converter technologies other than those assumed here are used, different DAC and ADC quantization-noise profiles would be expected.

When the effects of all noise sources are combined, it's clear that the NEXT contribution dominates the noise seen at the CPE receiver

The available signal strength for each carrier can be seen by comparing the CPE received signal level and the noise present at the receiver. The discrete-multitone (DMT) downstream signal is made up of many discrete carriers on a 4.3125-kHz spacing. Assessing the signal-to-noise ratio (SNR) of each carrier can determine the available data rate for downstream transmissions (Fig. 6 again).

There are several methods for approximating the maximum data rate. The most common involves establishing an SNR_GAP to which an SNR_MARGIN is added. For a QAM signal, the SNR_GAP needed to limit the maximum bit error rate (BER) to 10⁷ is understood to be 9.8 dB. The SNR_MARGIN is specified to be 6 dB for G.992.1. Combining the two gives a total of 15.8 dB.

The maximum amount of data that can be conveyed in a channel is related to the SNR as follows:

where the symbol rate is 4.058 kHz on a 4.3125-kHz carrier spacing, and Γ = 15.8 dB. One in 69 symbols is allocated to synchronization, leaving only 68 for actual data and protocol transport.

Using the 13.5-kft. loop #7, the highlighted carrier in Figure 6 has an SNR of 32 dB. This 32-dB available SNR, when factored into Equation 8, supports 5 bits/Hz. Integrate this over the 4 kHz of available bandwidth in the carrier and a total symbol rate of 20 kbits/s results. When adding the symbol rate for all downstream carriers, the maximum theoretical downstream symbol rate is 1500 kbits/s. Some of this symbol rate is allocated for ATM protocol overhead. In practice, the actual downstream data bandwidth is lower than the theoretical maximum symbol rate.

Consider a simple, single-wire-type local-loop example. Its distance-related bandwidth fall-off can be estimated. To make the analysis more realistic, assume that 49 disturbers operate within the bundle. This assumption would be quite reasonable during peak hours for residential users. Comparing the G.992.1 full rate and G.992.2 G.Lite bandwidth performance in Figure 9, it's clear that at the longer CO-to-CPE distances, the full-rate symbol rate is reduced to that of G.Lite.

The line-attenuation effects and local disturbers restrict the maximum theoretical symbol rate of both protocols at distances beyond 14 kft. The impact of bridge taps, unusual home-wiring topologies, and AM radio will reduce that rate even further. The AM radio issue is especially detrimental to the full-rate G.992.1 data rate, because the 525- to 1610-kHz AM band falls directly on top of the full-rate upper carriers. This only tangentially affects the G.Lite G.992.2 carriers.

Recent industry analysis¹ has indicated that the performance of the copper local loop may be significantly impaired as a result of moisture damage and oxidation. Local-loop degradation effects can move the convergence of G.992.1 and G.992.2 performance from 14 kft. down to as low as 8 kft. For many residential users, who find themselves on a long local-loop run back to the CO, the additional cost of a full-rate implementation isn't warranted.

Data from IEEE Communications shows that median loop length in the U.S. is 11.5 kft. (including commercial service). Roughly 25% of the population is at least 16.4 kft. from their CO, implying that G.Lite hardware is sufficient to handle the available bandwidth of this segment. If loop performance is degraded by moisture and oxidation, full-rate DSL performance and G.Lite performance issues will only matter to the 35% of the population that can detect a difference between the two standards. Those beyond the 8-kft. distance would experience nearly identical symbol rates, regardless the standard under which they operated.

Residential grade service is being marketed and sold uniformly to those with both short and long local loops, so it will be difficult to promote anything more than G.Lite download speeds. The current trend is toward 384-kbit/s minimum guaranteed download speeds in residential service offerings. Given that G.Lite hardware is less complicated and expensive than full-rate hardware, it's hard to see what value full-rate hardware brings to residential users.

Small- and medium-sized businesses tend to be located near their COs. Thus, the full-rate solutions will probably continue to dominate that segment. With the CO located only a short distance away, full-rate hardware does support and deliver higher bandwidths for these businesses.

Many CO line cards are being built with full-rate and multi-rate capabilities. So low-cost G.Lite residential modems will need to support the G.hs handshaking protocol that allows G.Lite modems to negotiate with full-rate line cards. The modems indicate to the line card that the upper carriers are inoperable. Once this initial negotiation is complete, the line card will only send and receive data on the carriers supported by the G.Lite CPE modem. This is possible because the G.Lite standard is very similar to the full rate standard.

By understanding the end-user needs and the distance-related line effects and disturbers, the most appropriate complexity and cost tradeoffs can be made.

References:

1. Peden, Mark; Markee, Joe; Hering, Uwe; and Aalaei, Faraj, "Splitterless DSL and the Myth of Multi-Mode" (white paper), 1999.

Sources:

1. ITU-T G.992.2-1999 Splitterless Asymmetric Digital Subscriber Line (ADSL) Transceivers.
2. ANSI T1.413-1995 Network and Customer Installation Interfaces—Asymmetric Digital Subscriber Line (ADSL) Metallic Interface.
3. ANSI T1.601-1998 Integrated Services Digital Network (ISDN)—Basic Access Interface for Use on Metallic Loops for Application on the Network Side of the NT (Layer 1 specification).
4. ITU-T G.996.1 Test Procedures for Digital Subscriber Line (DSL) Transceivers.
5. Pozar, David M., Microwave Engineering, Addison-Wesley, 1990.
6. Rauschmayer, Dennis J., ADSL/VDSL Principles, MacMillian Technical Publishing, 1998.
7. Starr, Thomas; Cioffi, John M.; and Silverman, Peter J., Understanding Digital Subscriber Line Technology, Prentice Hall PTR, 1999.