The G.DMT and G.Lite Recommendations
Part 2
This second in a two-part article on ADSL technologies focuses on the physical-layer challenges involved in sending digital in
formation over a copper medium originally intended to support analog communications.
By Jacques Issa and Roy Bieda
Part 1 of this article gave a detailed review of the differences between the International Telecommunications Union’s (ITU’s) recommendations for G.DMT and G.Lite. Part 2 will take a step back to the basics, since both
recommendations are physical layer interface specifications for digital transmission over metallic (twisted pair copper) physical plants.
Part 2 will provide an overview of digital data transmission over analog media (wiring that was originally intended to support voice-band analog signals), and show how different signal processing techniques can be used to minimize bit errors, increase data rates, and extend loop reach. The article will also point out common areas between G.DMT and G.Lite.
Metallic copper
loops
Since digital subscriber line (DSL) technology uses the existing copper infrastructure, a brief review of the installation and nature of the copper-wire subscriber loops is in order. Traditional voice-grade distribution cables use twisted pair copper wire (19 to 26 gauge). The media is run overhead (aerial) or buried underground. The limiting criteria for loop design are attenuation, noise, and crosstalk.
- Attenuation
. Attenuation (or loss) is the dissipation
of energy encountered by an electrical signal as it progresses down a transmission line. Original loop deployments targeted analog voice services in the 4-kHz region of the spectrum, ignoring future utilization of higher-frequency bands. To overcome loss and extend reach, phone companies opted to reduce the series resistance of the line by using larger gauge wire. They also increased the series inductance of the line with loading coils, and used analog electronic amplifiers to provide compensating gain to
the transmission line.
- Noise
. The most common sources of extraneous electrical signals are the adjacent electrical power lines and equipment. Devices such as spark plugs, electrical appliances, and motors can generate noise, termed electromagnetic interference (EMI). Noise can also be caused by broadcast TV, radio stations, cellular radio systems, and microwave radio systems. This is called radio frequency interference (RFI).
- Crosstalk
. Crosstalk occurs when energy
is coupled between two adjacent cable pairs. Crosstalk introduces bit errors and distortion on digital signals. This phenomenon occurs because practical transmission lines are laid in binder groups such as paired cables. When part of the energy in one transmission line is coupled with an adjacent line appearing as noise, crosstalk can also happen.
There are additional issues to consider. While deploying the copper plant, the original goal of local exchange carriers (LECs) was to economically
provide a voice quality loop (4 khz) to their service areas without exceeding loop loss (8 dB) standards. This was done by using a combination of cable gauges (usually 24 to 26 guage), creating impedance mismatches and reflections. For rural subscriber loops, most LECs used analog amplification devices (line extenders) at the central office (CO) to boost signal levels. To achieve longer loop reach, LECs also used inductors (loading coils) to compensate for the cable pair capacitances.
To better
understand the difficult environment the DSL transceivers must operate in, an overview of some digital transmission techniques is covered next.
Digital-data transmission methodology
Transmitting a digital stream of data from point A to point B requires some processing. At the very least, the transmitted signal power must be sufficient for an adequate signal-to-noise ratio (SNR) at the receiving end. This is accomplished using amplifiers with acceptable linearity and minimal distortion. In the digital
domain, however, there are many other ways for the signal to be processed before transmission (see
Figure 1
):
- Channel coding
. Concerned with maintaining the integrity of the conveyed data sequence (frame), channel coding may incorporate error correction techniques such as Reed-Solomon forward error correction (RS FEC) and interleaving.
- Line coding
. This type of coding involves converting the sequences of binary digits into
patterns suitable for transmission as pulses (modulation). Line coding also maintains regular timing of the bit sequence, and may use error-detection techniques such as trellis or Viterbi.
- Pulse generation
. Concerned with transmitting and detecting individual pulses, pulse generation can be modeled as a filter with an impulse response equal to the desired pulse shape. This layer is closely tied to the analog output because it is concerned with handling the voltage waveforms on the
line as well as maximizing the SNR.
RS FEC
Transmission errors such as noise, interference, or distortion will occur due to problems with the signal. Once an error is detected, the most common correction scheme is to request a retransmission.
Another correction method adds some bits to the transmitted data, allowing the data to be corrected at the receiving end. This is known as FEC. One method of FEC is RS coding, which uses block codes. RS block codes are organized on the basis of
groups of bits. These groups of bits are referred to as symbols (see
Figure 2
). Consider an example where a symbol = 6 bits and a code word = sixty-three symbols (47 payload/16 FEC). These parameters would allow the hardware on the receiving end to correct payload errors that are up to 12 bits in length.
How is a burst of noise handled? That’s where interleaving comes in. The interleaver stores symbols from different frames by row and reads them out by column (as
shown in
Figure 3
).
So, by applying RS FEC and interleaving, the receiving unit can overcome and recover from bit errors induced by noise. While both G.DMT and G.Lite use FEC and interleaving, G.Lite uses a smaller parameter set for the FEC coding and interleaving.
DMT
Dividing the available bandwidth into a set of independent, orthogonal subchannels is the key to DMT performance. By measuring the SNR of each subchannel and then assigning a number of bits
based on its quality, DMT transmits data on subcarriers with good SNRs and avoids regions of the frequency spectrum that are too noisy or severely attenuated. The underlying modulation technique is based on quadrature amplitude modulation (QAM). Each subchannel is 4.3125 kHz wide and is capable of carrying up to 15 bits. The subchannels are assigned according to the following parameters:
- Downstream: 26 to 1104 kHz, offering 249 subchannels for G.DMT; 26 to 578 kHz, offering 127
subchannels for G.Lite
- Upstream: 26 to 138 kHz, offering 25 upstream subchannels.
DMT can be used in either an echo-cancellation (EC) or frequency division multiplexing (FDM) mode (see
Figure 4
). Because the attenuation of any copper transmission line increases with frequency, it is desirable to transmit data using a frequency band that is as low as possible. Even though EC is harder to implement, it reuses a higher-quality frequency spectrum in the
26- to 138-kHz region. EC is optional for G.DMT.
Another optional feature in G.DMT is trellis coded modulation (TCM), which can provide additional coding gain and therefore could increase the SNR and bit allocation per DMT subchannel.
Trellis and Viterbi coding
TCM is a combination of convolutional coding and QAM. The redundancy of convolutional coding requires more states in the TCM QAM constellation than with QAM alone. TCM uses an encoding scheme similar to the one used for QAM, but
adds extra bits for its error-correction work. TCM operates on the input data continuously, so that the encoding of any input bits depends upon the encoding that has gone before, and is said to have “memory.”
That memory allows the receiving modem to determine, based on the value of the preceding signal, whether or not a given signal element is received in error. TCM has a set of allowable transitions based on the previous data. The trellis encoder and the Viterbi decoder use what amounts to
a data state machine. The state machine uses the current data and the previous data to determine the state of the bits that will be transferred. Allowing only certain transitions of the data (from one point on the trellis to another) gives TCM an inherent error correction capability.
If trellis coding had been selected on the transmit side, the data would pass through the Viterbi decoder on the receive side. The Viterbi algorithm for decoding uses the structure of the trellis (the allowable
transitions) and the input data to determine the most likely path through the trellis. A conceptual example of this is shown in
Figure 5
.
Trellis and Viterbi are optional features in G.DMT. When these features are enabled, the throughput, noise immunity, and reach of the asymmetric DSL (ADSL) link is improved.
ADSL over ISDN
Almost a decade ago, ISDN promised simultaneous voice and data communications to replace the present POTS using the same copper twisted pair.
And ISDN has delivered, albeit mainly in Europe and Japan, with over ten million ISDN lines installed (and that number is growing). Although available, ISDN has not been as widely deployed in North America.
ISDN provides the user with two 64-Kbps channels (referred to as B or bearer channels) as well as a lower-speed signaling channel (D channel) that is often used for X.25 data packet services. Each of the two 64-kbps channels can support the simultaneous transport of voice and data. On the voice
side, ISDN offers enhanced calling features such as digital voice quality, speed dialing, call return, caller ID, call forwarding, and local number portability. On the data side, both bearer channels can be bonded, yielding a 128-Kbps bidirectional data connection. ADSL over ISDN (AOI) promises even higher speeds over the same link while preserving the features of ISDN.
AOI’s main hurdle is the overlap of the ISDN and ADSL frequency spectrums (see
Figure 4
). ISDN occupies
the frequency spectrum of the twisted pair up to 80 kHz (2B1Q ISDN line coding) or 120 kHz (4B3T ISDN line coding). ADSL occupies the frequency spectrum from 26 kHz to 1.1 MHz (256 bins, each 4 kHz wide). A guard band from 4 to 26 kHz separating POTS and ADSL services is left to allow for splitter (filter) rolloff.
To eliminate the spectrum overlap, one approach would be for ADSL to occupy its normal frequency spectrum and transmit ISDN voice and data bit streams inband (within the ADSL payload).
Both G.Lite and G.DMT would be free to activate and use the twisted pair unencumbered. A splitter would no longer be necessary for either G.DMT or G.Lite and this approach would remain ITU compliant. However, the routing of voice-ISDN over ADSL means that all traffic must pass through the ADSL modem, creating two major issues. The first problem is that the power requirements of an ADSL modem make it unlikely that the modem can be line powered. A local power failure could interrupt lifeline voice services.
Another issue is that the ADSL processing of the ISDN information would introduce latency exceeding 2 ms, violating the ISDN standard.
An alternative to in-band transmission would be to move the entire ADSL signal above the frequency spectrum utilized by ISDN. This is called up-banding. The ADSL activation procedure would no longer be able to use the spectrum normally allotted to it for startup, as this spectrum would fall directly into the ISDN spectrum. Thus, both ADSL modems would need to be
up-banded. This approach is under consideration by the G.DMT specification as Annex B.
G.DMT Annex C addresses the issues of AOI in Japan. Japanese loops carry more wires, use different insulation techniques, utilize half-duplex ISDN, and often carry power within the same conduit.
Up-banding will allow ADSL and ISDN services to coexist, preserving lifeline communications and adhering to the ISDN standard.
Jacques Issa is an engineering 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
.
Puzzled by a network processing design issue?
