The G.DMT and G.Lite Recommendations, Part 1
The International Telecommunications Union (ITU) has issued recommendations for asymmetrical digital subscriber line (ADSL) equipment based on discrete multitone (DMT) technology. Here’s a look at the two key recommendations: G.DMT and G.Lite.
By Jacques Issa and Roy Bieda
The promise of broadband services to the residential consumer is long overdue. Despite major advancements in silicon and network infrastructure, the lack of
standardization and interoperable equipment continues to hinder mass deployment of digital subscriber line (DSL) technology. In an effort to address this problem, the International Telecommunications Union (ITU) finalized a family of DSL recommendations for transmitting a range of bit rates over the existing copper local network. Each of these documents include mandatory requirements, recommendations, and options:
- G.992.1: Asymmetrical Digital Subscriber Line (ADSL) Transceivers, G.DMT
- G.992.2: Splitterless ADSL Transceivers, G.Lite
- G.994.1: Handshake Procedure for DSL Transceivers, G.HS
- G.996.1: Test Procedures for DSL Transceivers
- G.997.1: Physical Layer Management for DSL Transceivers.
Of the five documents listed above, only G.992.1 and G.992.2 specify physical layer transceivers. The other three documents provide standards for handshaking, testing, and physical layer management.
G.992.1 specifies the characteristics of the ADSL interface to
metallic loops. An ADSL transmission unit can simultaneously convey all of the following: downstream simplex and duplex bearers; a baseband analog duplex channel; and ADSL line overhead for framing, error control, operations, and maintenance. G.992.1 supports a minimum 6.144-Mbps downstream and 640-kbps upstream net data rate. It will be referred to as G.DMT.
G.992.2 specifies the physical layer characteristics of splitterless ADSL. The transmission unit can simultaneously convey a downstream and upstream
simplex bearer; a baseband analog duplex channel; and an ADSL line overhead for framing, error control, operations, and maintenance. G.992.2 supports a maximum 1.536-Mbps downstream and 512-kbps upstream net data rate. It will be referred to as G.Lite.
In some respects, G.DMT and G.Lite are closely related. Both G.DMT and G.Lite use the same discrete multitone (DMT) line code and its associated parameters (shown in
Figure 1
). G.Lite has been developed as a
technology that can be easily installed by end users, with considerations for possible interoperability with G.DMT. G.Lite, based on a modified G.992.1, meets the key objectives of lower equipment complexity (cost), lower power consumption, and splitterless operation (as shown in
Figure 1
).
Operating in a splitterless mode presents new technical challenges. The second part of this article will provide an overview of digital data transmission over analog media, and
will illustrate how different signal processing techniques can be used to minimize bit errors, and increase data rates and loop reach. Before discussing these processing techniques, however, we will examine the similarities between G.DMT and G.Lite, and discuss the issues that might affect interoperability.
The splitterless ADSL challenge
One of the driving factors behind the creation of G.Lite was the lure of a “no truck roll” installation. This means that a local exchange
carrier (LEC) can offer ADSL lite service without dispatching a service technician to install a splitter (see
Figure 2
).
The lack of a splitter in G.Lite is a major departure from the G.DMT specification. In G.DMT the functions of the splitter are:
- To protect the ADSL signal from plain old telephone service (POTS) nonlinearity and impedance changes, on/off hook, dial tone, ring-trips, home wiring, and electromagnetic interference (EMI)
- To
protect POTS from ADSL tones and intermodulation of ADSL tones into the POTS audible range
- To present known termination impedance to the ADSL transceiver unit (ATU).
With the G.Lite version of ADSL service, the modem and the POTS operate together on the same internal home wiring. Sharing the same in-house wiring creates the greatest potential for difficulty with G.Lite. By eliminating the POTS splitter, the termination impedance will vary from home to home, based on the variable home
wiring configurations and the number of phones. The impedance of a telephone going off the hook may be so low that it essentially shunts the strength of the ADSL signal. Nonlinearity of POTS handsets can convert voice signals to a modulated high-band signal that results in interference on the modem.
Splitterless operation presents some obstacles (summarized in
Table 1
) that can be overcome with signal processing techniques. In order to achieve a useable link, a bit error rate (BER) of
10
-7
is required. Techniques such as interleaving, Reed-Solomon forward error correction (FEC), and trellis encoding are employed (these techniques will be discussed further in the second part of this article).
A distributed splitter approach is another option for obtaining a usable link. This option involves shipping the ADSL modem with a number of small, in-line POTS splitters, which the customer plugs into each telephone in the house. This approach provides protection, and addresses a
number of the challenges associated with splitterless ADSL. However, this method introduces several new problems, including difficult installation of wall-mounted phones, the reduction of ADSL performance (due to having a number of filters in parallel), and tough manufacturer decisions regarding how many filters to ship with each modem. Despite these difficulties, the distributed splitter approach represents a compromise, offering a solution that delivers high-quality POTS and excellent ADSL
performance/reach, while avoiding professional ADSL installation.
If any possible solution to the splitterless environment appears to be most prominent, it is the need for fast retrain (as shown in
Table 1
). Fast retraining allows G.Lite to compensate for electrical changes on an active link without losing the connection (although some connection interruption will still be evident). Fast retrain is also necessary to wake up from a low-power state.
ATU-C and ATU-R handshaking
The
initialization timeline needed to establish an active ADSL link is outlined in
Table 2
.
Before any valid exchange of data can occur, the ADSL link must be activated. In most situations, the remote terminal ATU (ATU-R) will initiate the activation. However, it is possible for the central office ATU (ATU-C) to initiate activation. All messages in G.994.1 use differential phase-shift keying (DPSK) modulation.
- An ATU-R initiated startup begins by the unit transmitting
signals (TonesR) on one or both of its signal families, with a phase reversal every 16 ms.
- The ATU-C will respond by transmitting tones on one or both of its signal families (TonesC).
- The ATU-R detects the TonesC and enters into a quiet state, where it makes no transmissions for 50 ms to 500 ms.
- After this time period elapses, the ATU-R will transmit a tone (ToneR) on only one signal family.
- Once the ATU-C detects the ToneR, it will transmit GALF characters (0x81) on its
modulated carriers.
- The ATU-R receives these GALF characters and will respond by transmitting a series of flag characters (0x7e) on its modulated carriers.
- The ATU-C will then transmit flags, allowing the ATU-R to start its first transaction.
Each one of these transactions contains a message field coding. The message information field coding consists of an identification field, a standard information field, and an optional nonstandard information field (shown in
Figure 3
). The messages are delimited by flags, and are protected against errors using a 16-bit frame check sequence. The information carried by these messages allows each end of the link to construct a capability parameter tree. This tree is built via the passage of parameters that contain no subparameters (NPars) and parameters which have subparameters associated with them (SPars). Both ends then negotiate to a baseline of capabilities.
The allowable message types during a
transaction are:
-
CL:
capabilities list, indicating the capabilities of the transmitting station
-
CLR:
capabilities list and request, with the additional request of the remote station’s capabilities
-
MS:
mode select, which requests the initiation of a particular mode of operation
-
MR:
mode request, which asks the remote station to transmit an MS message
-
ACK(type) and NAK(type):
there are several types, two and seven, respectively, of
possible acknowledgement and negative acknowledgement messages.
The G.994.1 handshaking transactions are:
-
Transaction A:
ATU-R MS
*
ATU-C:
The ATU-R issues an MS message, requesting a mode of operation. The ATU-C may then either accept the request, request the ATU-R to initiate a transaction B, or request the ATU-R to initiate a transaction C.
-
Transaction B: ATU-R MR
*
ATU-C:
The ATU-R begins this transaction by requesting that the ATU-C
choose an MR of operation. The ATU-C may then respond by issuing an MS request. If the ATU-R acknowledges the MS, then both stations will transition to the selected mode. However, the ATU-C may respond to the initial MR by either requesting that the ATU-R initiate a transaction A or a transaction C.
-
Transaction C:
During a transaction C, both stations exchange their capabilities and negotiate. Once a common mode of operation is negotiated, a transaction A or a transaction B is initiated to
select this common mode.
The next step in the timeline is to initiate a DMT modulation session. This involves training and channel analysis of the ADSL link to determine the bit capacity and power levels for each DMT subcarrier, as well as the FEC parameters and interleaver depth. Then, both transceivers exchange the transmission settings to prepare the link for user data.
G.994.1 sets the stage for interoperability between the ATU-C and the ATU-R. However, to allow for future compatibility
with subsequent releases of this specification, it will be necessary for receivers to parse all information fields, ignoring those that are not understood.
Comparing G.DMT and G.Lite
To compare G.DMT and G.Lite, we can summarize the differences between the specifications in a Venn diagram (shown in
Figure 4
). Some of the major differences are:
-
Asynchronous transfer mode (ATM)-only transport.
G.DMT has the option to
allow for data transmission using bit serial synchronous transfer mode (STM) as an alternative to ATM cell transport. G.Lite specifies the sole use of ATM cell transport. This should not represent any loss in flexibility or functionality to the user, since ATM is a well-defined transport protocol. ATM, and its associated ATM adaptation layers (AAL0, AAL1, AAL2, AAL3/4, and AAL5), allow all types of traffic to be carried. The different AAL mechanisms permit the delivery of real-time and nonreal-time video, and
the transport of Internet Protocol (IP) traffic, using the point-to-point protocol (PPP) over ATM over ADSL protocol stacks. Designating ATM as the transport mechanism for both G.Lite and G.DMT ensures the robustness of ADSL.
-
Fast retrain.
Fast retrain is based on the concept of multiple stored profiles. Fast retrain is a necessity for G.Lite, since line conditions will vary, due to G.Lite’s splitterless implementation. Just as a G.DMT activation and training sequence allows for the
generation and storage of one profile at each ATU of a given ADSL link, G.Lite allows for the storage of up to sixteen profiles. A G.Lite ATU must be capable of storing at least two profiles. Each profile, at a minimum, contains bit allocation and gain tables, FEC parameters R and S, and interleaver depth D. G.DMT has no provision for fast retraining.
-
Power savings.
Power management has three distinct states: full rate and full power; sleep, which maintains an active link and achieves reduced
power consumption; and idle, which consumes little or no power. Power savings should encourage mobile, battery-powered laptop, and green computer users to rapidly adopt G.Lite. However, since G.DMT has no provision for power management, a potential incompatibility exists.
-
ATU-R loop timing.
ATU-R transmission of timing, synchronized to the ATU-C, is the classic method of timing a network. ATU-R loop timing is supported by both G.DMT and G.Lite.
-
Reduced overhead framing mode only.
G.DMT supports four framing structures. G.Lite only supports framing structure #3, using the interleave data buffer definition. The ATU-C negotiates which framing structure to use with the ATU-R during activation, and sets itself to the reported framing structure accordingly. A G.Lite ATU-R will request framing structure #3. There should not be any interoperability issues with the framing modes.
There are several other differences between G.Lite and G.DMT. Although the number of 64-kbps
carriers supported by G.DMT and G.Lite differs, the carriers are negotiated during initialization and training, and will default to the 128 tones supported by G.Lite. The lack of trellis coding and echo cancellation in the G.Lite specification (which will be discussed in the second part of this article) reduces both its robustness and capacity, but does not preclude its interoperability with G.DMT.
Which service to deploy?
For years now, the telcos have had the capability to install POTS-only
line cards, ISDN line cards, or multiservice line cards that support services such as ISDN and POTS. Yet only a handful of multiservice cards are deployed each year, compared to the large amount of new POTS-only line card installations. This difference is due to the fact that POTS-only cards are more cost-effective than multiservice cards.
As previously mentioned, G.Lite and its splitterless installation (and the less expensive “no truck roll” installation) are very appealing to a telco. Does a
telco install a G.DMT/POTS line card and support both G.DMT and G.Lite or a G.Lite/POTS line card? Given the history of the industry, multiservice cards are less probable.
The question then becomes whether to deploy G.DMT (and throttle back to support G.Lite) or G.Lite line cards. The ideal situation would be to use a silicon implementation that allows for both services at a G.Lite price. Silicon vendors are working towards this goal.
Silicon implementation options
With most new
technologies, there are multiple ways to implement a solution. G.DMT and G.Lite are not exceptions. Each solution will present both pros and cons. Whenever a design is to be implemented in silicon, the manufacturer must choose a design methodology. Often, this means choosing a fully custom-designed circuit, a semicustom-designed circuit, or a fully-programmable circuit. The pros and cons between these options are covered in
Table 3
.
In order to choose a methodology, one must consider the
application and its intended position within the marketplace. ADSL has the potential to be a large-volume consumer item, given the size of the PC and modem marketplaces. The most important factors under consideration are price and time-to-market.
A fully-custom implementation offers the best price advantage, but given the immaturity of the ADSL specification and marketplace, yields the highest risk should an evolving or misinterpreted specification necessitate a respin of the silicon.
A
fully-programmable solution, utilizing an off-the-shelf CPU or DSP, will yield the greatest flexibility. The program code may be provided by the silicon manufacturer, the modem manufacturer, or a combination of the two. However, the DMT modulation scheme is inherently complex. This gives rise to the need for a high-performance (high-MIPS) CPU or DSP, and can result in both higher cost and higher power dissipation.
A semicustom approach takes its cues from both the fully-custom and fully-programmable camps, utilizing
full-custom circuitry for the MIPS-intensive portion of the design, and a general purpose CPU or DSP for the areas of the specification that are likely to change or evolve. Often, this will yield the best balance of price, time-to-market, and flexibility.
Next steps
A lot of progress was made in 1998 to finalize and publish the DSL recommendations. The next step is to achieve interoperability between physical layer transceivers from different vendors. Since the handshaking mechanism is
common to both specs, it is possible for a G.Lite transceiver to activate with a G.DMT device, even though some of the features wouldn’t be active and performance would be affected. As the standards keep evolving, however, one option might be to add fast retrain and power management to the G.DMT standard.
Jacques Issa is a market development manager for the Motorola semiconductor products sector in Dallas, TX. He received his BS and MS degrees in electrical engineering from Purdue University.
He can be reached at
ryma40@email.sps.mot.com.
Roy Bieda is a field applications engineer with the Motorola semiconductor products sector in Dallas, TX. He received his bachelor’s degree in electrical engineering from Ryerson Polytechnical University in Toronto, Canada. He can be reached at
scn101@email.sps.mot.com.
|
TABLE 1: Splitterless challenges and potential solutions.
|
|
Issue
|
System Impact
|
Possible solutions
|
|
POTS on/off hook transitions cause dramatic and nonlinear impedance changes.
|
Retrain required to update filter coefficients
Decreased performance in off hook state.
|
Fast
retrain
Hook switch transition detection algorithm.
|
|
In-band POTS signaling (such as pulse dialing and dual tone multifrequency [DTMF]).
|
Bit errors
Potential connection loss.
|
Interleaver coding techniques (FEC)
Retrain if needed.
|
|
POTS band services spectral leakage interference into ADSL (such as fax).
|
Bit errors
Potential connection loss.
|
Interleaver coding techniques (FEC)
Retrain if needed.
|
|
ADSL interference into POTS during POTS off hook state.
|
Audible noise during phone conversation
Decreased fax/modem performance.
|
Power cutback
Peak-to-average ratio (PAR) reduction techniques
Retrain if needed.
|
|
Table 2: ADSL link initialization timeline.
|
Activation and acknowledgement
as per G.994.1.
|
Transceiver training
|
Channel analysis
|
Exchange
|
| Time ->
|
|
TABLE 3: Silicon implementation choices.
|
|
Design methodology
|
PROS
|
CONS
|
|
Full custom (nonprogrammable).
|
Smallest die size
Lowest cost
Often lowest power consumption
|
Longest design cycle time
Least able to accommodate bug and spec changes.
|
|
Semicustom (custom circuitry tightly coupled to either a CPU or DSP on one monolithic piece of silicon).
|
Small die size
Low power consumption
Ability to implement bug and spec changes quickly
Custom circuitry can reduce the MIPS requirement of the CPU or DSP
Mixed mode (analog/digital) reduces chip count.
|
Bug and spec changes could require a respin of the silicon.
|
|
Fully-programmable (general-
purpose CPU or DSP).
|
Full flexibility to implement
bug and spec changes
Time-to-market
|
Largest die size
Highest cost
Often highest power consumption
Analog functionality requires additional ICs.
|
Puzzled by a network processing design issue?
