PCM Modem Technology: Extending V.34

Pulse code modulation (PCM) modem technology and its applications, its network topology, its digital impairments, and the status of V.pcm are discussed.

By Les Brown and Marc Davidson

PCM modems, or so-called "56k" modems, take advantage of the digital connections that ISPs have to the public switched telephone network (PSTN). This digital connection allows PCM modems to avoid the principal impairment limiting V.34 modem performance - PCM quantizing noise - allowing higher downstream data rates to be achieved. In the first issue of the PCM modem recommendation, V.34 modulation will be used in the upstream direction, making this new analog modem asymmetric. This article discusses the technology behind these new modems.

As of the writing of this article, the Recommendation has not been ratified. Accordingly, the commentary expressed in this article reflects discussions and progress to date in the standards organization. There are still several open issues to resolve. Therefore, this article should not be viewed as a complete treatise on the subject.

A future issue of this modem Recommendation may add PCM in the upstream direction. This would enhance the viability of inherently symmetric applications, such as Internet telephony and Internet video conferencing, but will also bring on a new set of challenges to the modem industry. It is expected that the demand for analog modems will continue to drive activities in this direction.

The PCM modem concept

When Recommendation V.34 was completed in 1994 (and then slightly revised in 1996 to 33.6 kbps), it was believed by many that this would be the last modem Recommendation since the channel capacity had been virtually reached.1 However, in late 1996, several modem manufacturers announced a new generation, the 56k modem, claiming data rates approaching twice that offered by V.34 modems.

Figure 1 illustrates the transmission channel for a typical V.34 modem. Both modems are connected to analog local loops that terminate in central office line-cards, where the analog signals pass through a hybrid circuit H, and are then digitized and converted to G.711 codewords for transmission across the digital backbone network.2 The process of digitizing an analog signal and converting it to G.711 codewords (with the PCM encoder typically part of the A/D converter) generates quantizing noise, which limits the signal-to-noise ratio (SNR) to about 38 dB.

Figure 1

Figure 2 illustrates the transmission channel for a PCM modem. Only one modem is connected to an analog local loop. The other modem is connected directly to the digital backbone network. The digital end may be a bank of modems connected to either a T1 interface (E1 in Europe) or a Primary Rate ISDN interface, or it may be a single modem connected to a Basic Rate ISDN interface.

Figure 2

In the downstream direction, there is no PCM encoder, and, therefore, no quantizing noise is generated. This results in a higher SNR in this direction. Without the PCM encoder and its associated bandpass filter, a wider bandwidth is also available for PCM modems than is available for regular analog modems. In Figure 1, the usable bandwidth is 200 Hz to 3,700 Hz, whereas in Figure 2 it is approximately 70 Hz to 4,000 Hz. The significantly greater bandwidth, lower end-to-end loss, and higher SNR are the principal reasons that PCM modems can achieve higher data rates than V.34 modems.

In the upstream direction, the PCM encoder is still present, which makes it difficult to avoid generating quantizing noise and, therefore, difficult to achieve significantly higher data rates than V.34. PCM modems are inherently asymmetric.

Performance barriers

There are numerous network impairments and conditions that can limit the achievable downstream data rate of PCM modems. These include tandem encodings, adaptive differential (AD) PCM, loaded loops, code conversions, regulatory power requirements, D/A nonlinearities, frequency dependent nonlinearity, talker echo, and loop noise.

Figure 2 illustrates a completely digital connection between the digital modem and the line card in the central office that is connected to the analog modem. This is not always the case. There are often additional D/A and A/D conversions in the network. These additional conversions can occur when there are analog switches or analog trunk sections in the network. (Analog trunks, "N-carrier" in North America, are not very common today in the PSTN in North America, but they are still common in other parts of the world.) It can also occur with digital loop carrier (DLC) systems.3 PCM modems are expected to proceed with V.34 operation if any tandem encodings are detected.

DLC systems are used to replace bundles of analog local loops. The local loop signals are carried digitally between the central office and a location closer to the customer premises. DLCs have been and continue to be deployed in large numbers in the PSTN in North America. Some of these systems are integrated, which means there is a common digital interface between the DLC and the central office switch. Other DLC systems do not share a common digital interface with the central office switch and convert the signals to the analog domain in order to interconnect them. These are referred to as universal DLCs.

ADPCM, specified in Recommendation G.726, is a speech compression algorithm that is commonly used on overseas connections to reduce the bit rate required to carry a voice channel.4 Instead of the usual 64-kbps PCM channel, ADPCM normally operates at 32 kbps, although there are enhancements for data transmission that may be enabled upon detection of 2100 Hz modem answer tone. G.726 specifies a standard 40-kbps algorithm, but there is at least one commonly used proprietary solution that operates at 32 kbps with a sampling rate of 6.4 kHz, instead of 8 kHz. PCM modems are expected to proceed with V.34 mode when operating over ADPCM links. V.34 includes a means to detect the presence of the 6.4-kHz sample rate and limit its symbol rate accordingly.

Loading coils are placed in long analog local-loops at regular intervals to improve voice quality by flattening the frequency response up to about 3.2 kHz. Above this frequency, the channel rolls off very quickly and is unusable for modem signaling. PCM modems will likely proceed with V.34 mode when operating over loaded local loops.

Robbed bit signaling (RBS), also referred to as ABCD signaling, is commonly found in the North American PSTN.5 RBS steals the least significant bit (LSB) of every sixth PCM sample and uses it for network signaling. In most cases, the network sets this bit to binary 1 before delivering it to the customer. There is often more than one RBS link in a connection, resulting in more than one in every six samples being affected.

In order to optimize voice performance by reducing echo levels, loss pads are normally inserted in the network at the receiving end. These may be implemented either in the analog domain or the digital domain. Common loss values in North America are 3 dB and 6 dB. Other values are used in different countries. Digital implementations may be done in the linear signal space (equivalent to analog loss pad) or the PCM codeword space. The latter is normally performed by a table lookup, where each 8-bit PCM codeword maps to another PCM codeword of lesser value. This form of translation results in code duplication, where more than one input value maps to the same output value. There are no standards for these code translations, which means that different switch manufacturers may implement the translation differently, resulting in a potentially wide variety of code translations in the network. This should be kept in mind when learning digital impairments and selecting signal constellations.

The order of RBS links and digital loss pads varies throughout the network, which makes the selection of optimum signal constellations challenging. For example, a DLC system may use RBS, in which case there may be robbed bits in different phases before and after a digital loss pad.

There are two different PCM coding schemes in G.711, A-law and ý-law. In the US and Canada, ý-law is used. In many other parts of the world, A-law is used. When calls are placed between countries that use different coding laws, a code translation is performed in the network. This translation, which also generates code duplication, is completely specified in Tables 3 & 4 of G.711. Again, this must be kept in mind when selecting signal constellations.

In the US, there is an FCC regulatory requirement that limits the power of any digital signal that is to be converted to an analog signal by a central office line card. This limit is -12 dBm, and is measured at the digital modem output, not the line- card PCM decoder (which may be after a digital loss pad). An effort is underway within the standards bodies to have this constraint relaxed. The original reason for this restriction was to prevent overload induced crosstalk in analog carrier sections of the network. PCM modems will have the capability to determine whether there are analog trunk sections in the network (tandem PCM encodings) and to proceed with V.34 modulation if there are any. Therefore, it is argued, higher transmit levels should be allowed for PCM mode, because this mode would only be used when there are no analog trunk sections in the connection. It is also argued that, since PCM modems will have the capability to detect the presence of digital loss pads, the power level should be measured after the digital pad, at the line-card PCM decoder. It will be some time before all of this is sorted out with the FCC.

G.711 specifies the D/A conversion level for each codeword, and the actual conversion level may deviate from the specified values. This may result in a dc offset and/or nonlinear distortion. Dc offset is not a problem; however, other deviations of the levels from their proper values may need to be compensated for in the PCM modem receiver.

Central office line-cards that use transformers in their hybrids may generate nonlinear distortion, particularly when excited by high power levels at low frequencies. This may be avoided by applying spectral shaping to reduce low frequency content.

As for talker echo, with the network configuration of Figure 2, the digital modem will have only far-end talker echo and the analog modem will have only near-end talker echo. There will be no listener echo. However, there will always be some level of loop noise that may or may not be the impairment that limits modem performance. Uncancelled echo or residual intersymbol interference may be more significant performance-limiting factors in many cases.

PCM modem technologies

The G.711 codewords have a non-uniform spacing, with larger codewords being spaced further apart. PCM modems attempt to select a subset of these codewords that are spaced as uniformly as possible. Of course, this becomes more difficult as the data rate is increased and more codewords are needed. Not including any redundancy associated with the modem mapping function, 2n codewords would be needed to operate at n x 8 kbps. For 56 kbps then, 128 codewords would be needed. Many of the smallest codewords would not be used, since their spacing is too close to be able to be received properly in the presence of noise, and a few of the largest codewords would not be used, since they may result in a violation of the regulatory power constraint.

The mapping/framing function takes the user's data bits and translates them to G.711 codewords for transmission. A good mapping function should be able to generate a fractional number of data bits per symbol to allow for a finer granularity in the selection of data rates. Since RBS may occur for some symbol phases and not for others, a good mapping function should also be able to use different constellations for different symbol phases in order to minimize data rate loss. As of the writing of this article, the mapping algorithm for the PCM modem recommendation has not been selected. Two proposals are being considered, multiple modulus conversion and shell mapping (based on the V.34 algorithm).6, 7

Spectral shaping may be used to reduce the very low frequency energy content, possibly to minimize frequency-dependent nonlinearity. As of the writing of this article, the spectral shaping algorithm for the PCM modem recommendation has not been selected. Two proposals are being considered, convolutional spectral shaping (CSS) and maximum symbol inversion.6, 8 V.pcm is truly adaptive when it comes to the detection and mitigation of digital impairments. This has two advantages. First, it makes it easier to reach consensus in the committee since each manufacturer is free to use his favorite sequences and constellations. Second, as the network evolves, or our understanding of the network improves, the Recommendation does not need to be revised.

The detection and mitigation works as follows. The analog PCM modem sends a digital-impairment learning descriptor to the digital PCM modem. The digital PCM modem then constructs the digital-impairment learning sequence from this descriptor, and transmits it to the analog PCM modem. The analog PCM modem then analyzes the received signal to determine the phase(s) of RBS and the type of loss pad present. Based on this analysis, the analog PCM modem then selects a set of constellations for the digital PCM modem to use during data mode, and conveys this information to the digital PCM modem.

PCM modem training and interworking

The first phase of the startup procedure for PCM modems will use Recommendation V.8 to negotiate modulation capabilities, just as V.34 modems do. This will allow PCM modems and V.34 modems to interwork properly. A new category octet in V.8 has been defined for PCM modem use.

The second phase of the startup procedure will use the same structure as V.34. During this probing phase, a PCM modem will determine whether the channel is capable of PCM operation. If it is not, the modems will proceed into the third phase of training in V.34 mode.

The third phase of the startup procedure is half-duplex training. During this phase, adaptive elements of the receiver (e.g., equalizer, timing loop) and transmitter (e.g., echo canceller) are trained. After initial receiver training, the APCM modem will also receive and analyze the digital-impairment learning sequence in order to select the constellations for the downstream transmission.

The final phase of the startup procedure for PCM modems is full-duplex training, where the modem's adaptive elements are fine tuned and modulation parameters such as data signalling rate are exchanged.

Standards status

In September 1996, several modem manufacturers announced that they were working on 56k modem technology. This prompted the Telecommunications Industry Association (TIA) TR-30.1 to initiate a "fast track" standards project with the goal of completing a TIA interim standard during 1997. When the International Telecommunication Union (ITU-T SG16) met in March of this year, they also initiated a new project (Question 23) to complete an international recommendation for PCM modems. Continuing from where TR-30.1 had left off, the goal of this project was to complete the technical work on a first PCM modem recommendation, referred to as V.pcm-Issue 1, which was to be "determined" (the first stage of the approval process) at the SG16 Working Party (WP) 1 meeting in September 1997 and "decided" (the final stage of the approval process) at the SG16 meeting in January 1998. Another goal was to complete a second PCM modem recommendation, referred to as V.pcm-Issue 2, by late 1998. The purpose of V.pcm-Issue 1 is to address the immediate market need and minimize the impact of noninterworking proprietary solutions in the marketplace. The purpose of V.pcm-Issue 2 is to enhance the performance, possibly including PCM upstream.

The September 1997 WP1 meeting has come and gone, and the work on V.pcm-Issue 1 was not completed in time to initiate the approval process. Some key data mode issues (including the mapping/framing and spectral shaping algorithms) remained unresolved. The new goal is to determine the recommendation in January 1998 and "decide" it no later than September 1998.

To summarize, the following are some of the principal characteristics of V.pcm-Issue 1 that have been agreed to so far:

• PCM downstream (symbol rate of 8,000), V.34 modulation upstream (support for 3,000, 3,200, and 3,429 symbol rates only)
• Downstream user data signalling rates from 28 kbps to 56 kbps in 1 1/3 kbps increments
• Automatic determination of the capability of a connection to support PCM mode downstream. If the connection is unable to support PCM mode downstream, it is to proceed with V.34 mode.
• Adaptive selection of downstream signal constellations
• A startup procedure similar to that of V.34.

Enhancements such as an all digital mode and PCM upstream are being considered for the next revision of the recommendation. Trellis coding and support for an auxiliary channel, left out of V.pcm-Issue 1, may also be considered.

Normally, the digital modem (Figure 2) is the ISP's modem and the analog modem is the client's modem. Suppose the client is not connected to an analog local loop, but rather has Basic Rate ISDN service. In this case, both modems will be digitally connected and it is possible to operate with PCM in either or both directions. A different startup procedure is needed for this all digital mode of operation.

For the network connection of Figure 2, it is possible to use PCM upstream to achieve higher than V.34 data rates in the upstream direction as well. This is a more difficult problem than PCM downstream, since it is more difficult to avoid generating A/D quantizing noise in this case. To avoid generating quantizing noise, it is necessary to accurately lock the timing of the upstream transmitter to the network clock, which means running loopback timing in the analog PCM modem. Then, the analog channel must be modeled so that signals arrive at the network A/D converter at the sampling instances with voltage levels equal to the correct PCM encoding levels. One of the challenges that needs to be faced is how the digital modem will deal with the nonlinear echo resulting from the PCM encoding process in the echo path.

Les Brown is a Member of Technical Staff at Motorola ISG. He is also the Motorola ISG representative in modem-related standards committees, including Chairman of TIA TR-30.1 and Rapporteur for Question 23 (PCM modems) in ITU-T Study Group 16. He received his BASc in electrical engineering and MASc in electrical engineering, specializing in active filter design, from the University of Toronto. Brown can be reached at LLB005@email.mot.com.

Marc Davidson is an analog modem product manager at Motorola's Semiconductor Products Sector. He recieved his BASc in electrical engineering from the University of Ottawa.