Voice-over-broadband SOCs: a smarter way to slice the pie

A growing number of businesses and households already enjoy high-speed Internet connections using cable or DSL (digital-subscriber-line) modems. To meet the demand for such services, semiconductor companies have been quick to provide broadband SOC (system-on-chip) devices at an increasing level of integration. This approach has been somewhat effective in driving down modem cost and size and in increasing connectivity and feature options. Most broadband modems, for example, now offer routing, bridging, and network-security capabilities through a variety of connection choices, including Ethernet, USB, WLAN (wireless LAN), and HPNA (Home Phoneline Network Alliance). Broadband-modem chip sets have digested all of this technology despite chip counts that have decreased from an average of more than four ICs to as few as two.

High-speed broadband-to-home capabilities are enabling MSOs (multiple-system operators) to bundle together value-added services, such as video on demand, interactive TV, and telephony. By increasing customer retention and profit per user, telephone service via VOIP (Voice Over Internet Protocol) stands to benefit MSOs the most. With VOIP in the residential-broadband-services mix, VOB (voice-over-broadband) telephony becomes a candidate for integration into broadband SOC devices. In anticipation of rapid infrastructure investment and market acceptance, semiconductor companies have been quick to integrate voice capabilities into broadband SOC chip sets, and widespread acceptance is becoming a reality in some regions. For example, Japan has led the world in VOB deployment with about 2.5 million VOB subscribers signing up for the low-cost service. According to projections, more than 6.5 million homes, or 10% of Japan's population, will use VOB by 2005 (Reference 1).

Most of the world, however, has been slower to embrace VOB. Many observers argue that VoB is still in its infancy and that standards for VOIP implementation are both too numerous and still evolving. In light of the communications industry's continuing economic trauma, many wonder when VOB will gain general acceptance. Although broadband-SOC manufacturers recognize the potential advantages of integrating VoIP into their ICs in anticipation of market growth, they wonder whether the risk is worth taking. A VOB-SOC manufacturer must cope with this uncertainty either by making quick revisions to VOB devices in reaction to changing requirements or by creating a superset device that can handle any set of VOB requirements and standards. Both of these strategies can work, but neither is cost-effective. The addition of VoIP to broadband SOCs raises the question of whether integration has gone too far, too fast.

The impact of this uncertainty is painful for system designers, who must design competitive VOB platforms amid the chaos. These designers must consider whether it is better to migrate now to a fully integrated VOB architecture or to add VoIP as a module to data-only broadband products. VoIP hardware-partitioning strategies require analysis.

The first step in discussing VOB system partitioning is to define the operational blocks that make up the typical customer-premise VOB system and determine which sub-blocks are ideal for SOC integration. Figure 1 illustrates these blocks and their respective general circuit technologies.

The broadband block receives, demodulates, and decodes the incoming broadband signals to produce a digital data stream and encodes, modulates, and transmits the outgoing digital data stream in the broadband spectrum. Most broadband ICs today integrate the AFE (analog front end), MAC (media-access control), and PHY (physical-layer) subcircuits. A cable modem's tuner and a DSL modem's line driver consist of magnetic, passive, and high-voltage active analog circuits, which usually remain outside the SOC.

A residential gateway's LAN block can contain many data-networking physical interfaces: Ethernet, USB, WLAN, and HPNA, for example. AFEs tend to remain external to the SOC, but most MACs and PHYs are conducive to SOC integration.

The communications-processor block requires a significant amount of processing power and memory to perform all of the functions that residential gateways demand. Networking tasks, such as routing, bridging, and security, require hundreds of MIPS (millions of instructions per second) and tens of megabytes of memory. Typically, a small amount of cache and local RAM reside on the chip; you access most program and data memory through a memory-controller interface.

The VOIP block usually dedicates a DSP core and memory to the customized tasks of VOIP processing. Residential gateways usually connect directly to one or more analog telephones through a SLIC (subscriber-line interface circuit) and a voice codec. The "battery"—a legacy term from the telephony world—supplies high voltages for powering and ringing the telephone. VOIP tasks include any number and combination of voice-coder algorithms, fax relay, voice-echo cancellation, DTMF (dual-tone-multifrequency) decoding, impedance synthesis, tone generation, and telephony-control and -monitoring functions. The voice DSP also does a fair amount of voice-packet processing and buffering. The high-voltage, SLIC (subscriber-line interface circuit), and battery sub-blocks usually remain external; the other sub-blocks integrate into one or more ICs.

The preceding analysis of the sub-blocks that make up a VOB residential gateway defines which operational sub-blocks are compatible with current SOC process technologies and are therefore candidates for integration. Figure 1 suggests that this theoretical maximum-integration level constitutes a "realm of SOC integration." You should infer not that such a level of integration is commercially practical—only that it is technologically feasible. Of course, in some exceptions, suppliers venture beyond the standard boundaries of integration. For example, on the broadband side, some SOC designs integrate the DSL line driver. On the LAN side, some integrate Ethernet, HPNA, or USB AFEs. On the VOIP side, some SOC devices integrate audio codecs and many functional aspects of the SLIC and battery dc/dc-converter control. Vendors choose to assign functions differently among their chip sets' ICs; rarely do two chip sets use exactly the same partitioning.

The following sections investigate three scenarios of VOIP integration and consider the strengths and weaknesses of each. Key aspects are cost, size, design flexibility and reuse, and VOIP-channel scalability. Table 1 summarizes the three scenarios' key features.

The first scenario fully integrates VOIP into the VOB SOC with the exception of the SLIC and battery sub-blocks (Figure 2). This arrangement has several advantages over keeping the VOIP block external. First, it eliminates from one to three ICs and reduces board space by 2 to 4 sq in. depending on the number of voice channels on the board. Second, it promotes circuit efficiency within the SOC device. Some communications processors contain DSP instructions, permitting one high-performance core and an external memory IC to perform VOIP and network-communications processing. Space-constrained gateway designs can therefore benefit most from this scenario.

However, adopting such a highly integrated approach presents many disadvantages. First, because few vendors currently offer this level of integration, the approach greatly limits the user's product choices and buying power. Granted, as standards solidify and voice integration becomes widely accepted in gateways, more vendors will announce SOCs with fully integrated VOIP. Until then, however, the choices are few.

Additionally, the high level of integration greatly limits VOIP design flexibility and scalability. By placing the VOIP block in the VOB SOC, device suppliers can restrict VOIP software options to what they bundle in their own proprietary software packages, eroding buying power on the software side as well as on the hardware side. Furthermore, the voice-channel count and the scalability of the voice-DSP capacity are constrained to what is on the chip.

The few fully integrated SOCs available today offer two or four codecs. The problem of channel-count scalability becomes evident for designers who use a fully integrated two-channel VOIP SOC but now need to offer a single-channel version. In this scenario, half of the VOIP processing power and one codec is wasted. A different problem exists in expanding channel count beyond the capabilities of the VOIP SOC. For example, an eight-channel implementation would require an external codec IC, which is unlikely to be compatible with the SOC's internal codecs. The SOC vendor could also introduce an eight-channel version of the device to address higher channel counts, but adding more codecs eventually becomes too costly, because audio codecs use a fair amount of analog circuitry, which benefits little from the small-geometry CMOS processes that SOCs require. Therefore, the codec circuits are far more expensive to produce in a 0.13-micron SOC process than in a standard 0.5-micron process.

So, if a communications-equipment supplier offers both voice-enabled and data-only versions of a residential gateway, what VOIP-partition strategy makes sense? This question is valid, because nearly all companies that currently ship VOB residential gateways also ship the same products without the VOIP capability. Unless it's acceptable to develop and maintain two broadband products—one that uses the fully integrated VOB SOC and another that uses a data-only SOC—migrating to a fully integrated VOB approach loses its appeal. An SOC vendor could offer two versions of the same SOC—one with integrated voice capabilities and one without—to provide compatibility to such a customer, but few SOC vendors choose to burden their high volume, selling data-only broadband SOCs with voice codecs that are enabled only in VOB versions of the chip.

The second scenario integrates only the digital VLSI portions of the VOIP block—DSP and memory—into the VOB SOC (Figure 3). This arrangement maintains the advantages of the first scenario, but the voice codecs remain external and connect using a PCM interface and an SPI. Still, depending on the number of voice channels on the board, this approach eliminates one or two ICs and reduces board area by 1 to 2 sq in. Also, three times as many vendors currently offer VOB chip sets with this partitioning strategy than offer integrated codecs, increasing designers' choices and buying power.

Consider the increase in design flexibility and VOIP-channel scalability when the chip set uses external codecs. The SOC's contents no longer limit the channel count. Also, the industry-standard digital-PCM and SPIs can add any number of VOIP channels. The ability to connect any number of channels to a common digital interface substantially lowers the SOC's analog pin count. For example, a four-channel SOC having integrated codecs needs 16 analog connections—two differential pairs to each SLIC. Using PCM and SPIs requires only eight digital pins and simplifies the board layout, because you can move the SLICs away from the SOC without increasing the length of analog-signal traces.

However, because the voice DSP remains integrated, the DSP's maximum processing bandwidth ultimately limits scalability. In addition, VOIP software choices are not open to the entire market of DSPs. Finally, though still not an optimal strategy, the first scenario enhances an SOC vendor's ability to address both data-only and voice-enabled gateway markets with an optimal IC portfolio.

Although the third scenario scarcely reduces board space or component count, it maximizes hardware and software choices and offers complete flexibility in selecting and scaling the VOIP block and the number of VOIP channels (Figure 4). You can add a decoupled VOIP block to a data-only broadband gateway, whether it is ADSL, cable, or another type of upstream block. The flexibility to quickly port an established VOIP block to a different broadband SOC is particularly vital given the rapid evolution of broadband technologies. Companies that offer many broadband products can therefore benefit most from this scenario.

Decoupling VOIP from the broadband function may offer the most flexibility and scalability, but it strays from the benefits of a highly integrated approach. Consider now the impact of integrating many of the VOIP sub-blocks into only a few chips. A sufficient level of integration can regain the cost and size advantages of fully integrated VOB without sacrificing the benefits of decoupled VOIP.

So, what is the best approach to integrating the VOIP function into as few chips as possible? For VOB systems that have many VOIP channels, you can integrate common features into single ICs. For example, you could implement an eight-channel voice system by using one DSP, two quad codecs, a central battery supply, and four dual SLICs. You might describe this approach of integrating like functions together as "horizontal integration." Unfortunately, horizontal integration is relatively ineffective in one- to two-channel residential gateways, in which price competition is fiercest and potential volumes are greatest. For low-channel-count applications, you need to achieve integration by combining the codec, SLIC, and battery-control functions into one device—that is, by "vertical integration." Many semiconductor companies now offer vertically integrated devices for use in VOIP systems.

Vertically integrating the VOIP channel has an enabling effect in the global VOIP market. Specifically, integrating the SLIC and codec functions in the low-voltage-CMOS domain permits increased software control of SLIC-line-feed parameters. Traditional SLIC- control interfaces use three or four parallel input lines to directly control the line state. Line-feed attributes, such as ringing amplitude, loop current, ring-trip thresholds, and loop-closure thresholds, are either not programmable or programmable using external resistors and capacitors. Combining the SLIC and the codec into one chip overcomes the constraints of traditional SLIC interfaces and provides the SLIC with control and software programmability comparable with the codec's. When you can program SLIC parameters in software, you can optimize a single hardware implementation to meet any country's telephony requirements. Furthermore, if the VOB gateway is field-programmable via firmware downloads, an MSO can reconfigure the SLIC's operation to correct interoperability issues, upgrade to new VOIP standards, or even customize performance for a single user.

Another benefit of integrating SLIC and codec functions is ease in implementing remote subscriber-line diagnostics. For LECs (local-exchange carriers), subscriber-loop testing has always been important in reducing the cost of unnecessary service-truck rolls. In traditional carrier-access networks, each central-office hardware rack contains a telephony-test card that can connect via mechanical relays to any single subscriber loop in the rack. From the test results, the LEC can determine the source of the potential problem and dispatch the appropriate repair service only if necessary. With VOB residential gateways, the entire local loop resides on the customer premise, so the VOB gateway must contain the subscriber-loop test-and-diagnostic capabilities. An integrated SLIC-and-codec device's increased control of line-feed parameters allows for comprehensive subscriber-loop testing using existing SLIC circuits and thus obviates the need for external test circuits and relays.

Each company must decide on a system-partitioning strategy that implements its product road map amid the VOIP market's turbulence. It seems reasonable that the benefits of separating the VOIP function from the central SOC might outweigh those of a single SOC that integrates VOIP capabilities. An integrated broadband-gateway SOC together with a decoupled, though highly integrated, VOIP function can produce an economical yet flexible system.