Voice Vise

Although much of the hype surrounding 3G focuses on wireless data, the best-kept 3G secret might be that better voice service is just around the corner. Beyond faster pipes carrying more calls, greater voice capacity and improved quality is attainable through the use of vocoding, a chip-based technology with a 25-year history that recently has gone through significant innovation.

At a basic level, vocoders convert analog voice to digital data. New, lower-rate vocoders can squeeze more calls into a single channel — effectively doubling or, in some cases, tripling voice capacity. These vocoders also can dedicate a larger number of bits to the task of error correction within strained channels. The trade-off still must be made between capacity and quality, but either choice is better than it was a few years ago.

The GSM community has chosen adaptive multi-rate (AMR) as its next-generation vocoder; CDMA carriers will deploy selective mode vocoder (SMV) technology. SMV vocoders are being embedded on Qualcomm's (www.qualcomm.com) 6000 series mobile chipset and should make it onto the next generation of handsets shortly thereafter. Although the SMV standard should be approved by the end of the month, carriers and vendors interviewed for this article would not commit to a commercial release date for SMV-enabled service.

AMR vocoders were available early this year from Ensigma, a Welsh technology provider. Many handset and chip makers are using Ensigma's AMR vocoder, but the company would not identify its customers, citing confidentiality agreements (www.ensigma.com).

As with most new services, seeding the market with enough handsets to improve overall network capacity and quality will be a long process.

Until recently, CDMA networks used the QCELP-13 vocoder, which operates at 13kb/s. Rick Ross, Sprint PCS (www.sprintpcs.com) manager of air interface standards, explained that within the past year, new EVRC vocoders have enabled CDMA carriers to achieve an 8kb/s coding rate.

“Going from 13kb/s to 8kb/s for the coding rate saves you bandwidth,” Ross said. He said improvements in vocoder technology have two goals: To maintain, if not improve, voice quality and to reduce the average data rate, increasing the system's capacity.”

Even as EVRC was being deployed in networks last year, the new SMV vocoder was going through the standards process at 3GPP2. The standards work began in the TIA and was later moved to 3GPP2, explained Craig Greer, Nokia Research Center (www.nokia.com) manager and chair of the 3GPP2 TSG-C Voice Services Subworking Group, the group responsible for creating the SMV vocoder standard (www.3gpp2.org).

“The SMV program began with a requirements document that considered most, if not all, of the IMT-2000 codec-related requirements,” he said. “Additional requirements were imposed upon the SMV to ensure backward compatibility with the 2G IS-95 system.”

The SMV vocoder enables the network to control the average rate of the codec. The vocoder can toggle between four data rates, which it selects based on speech traffic.

“This allows the network to make trade-offs between voice quality and capacity,” Greer said.

The description of the codec algorithm will be published on the 3GPP2 Web site this month, the specifications for minimum performance standards in October.

Measured Gains

Sprint PCS estimates the SMV will bring a 30% gain in network capacity, said Dean Prochaska, Sprint PCS director of wireless industry standards.

The operational differences between the EVRC and SMV vocoders boil down to dead air.

“EVRC is an 8kb vocoder,” Ross said. “So when you're speaking, the data rate of your speech is 8kb/s. But there are pauses when we speak. The way EVRC works is when it senses that you're not speaking, it cuts that coding rate back to one eighth of the rate so you're not using as much capacity over the air (during silence). But if you're speaking at all, it goes back up to full rate, 8kb/s.”

Alternatively, the SMV vocoder capitalizes on the parts of speech that don't require a full 8kb/s.

“Certain sounds such as ‘s’ could be coded at a lower rate,” Ross said. “With EVRC, whenever I make the ‘s’ sound, it's going at full rate, but with the SMV vocoder, it's able to sense that sound and send the ‘s’ at 1-half or 1-quarter rate. Parts of speech that would have been in EVRC transmitted at 8kb/s now are transmitted at 2kb/s or 4kb/s.”

By choosing a mode — operating at between zero and 8kb/s — the SMV vocoder selects the most efficient rate for current conditions.

Mode 0 would run at the same rate as EVRC but with a bit higher quality than EVRC and/or 13kb/s. Mode 1 would operate at a lower average data rate, and then capacity would be more on par with EVRC. If you go to Mode 2, then carriers are operating at an even lower data rate, but they start to take a hit on quality. Depending on the time of day, carriers can make trade-offs between voice quality and capacity.

Prochaska said that although SMV mode 0 shows no capacity gain, tests of SMV mode 1 have yielded 30% capacity gains.

“In all of the tests that have been run, this (SMV) vocoder outperforms the EVRC vocoder,” he said. “And the EVRC vocoder is better than the QCELP-13 vocoder. Mode 1 SMV is going to outperform EVRC and is showing a 30% capacity gain on simulations run up to this time.”

Sprint PCS is in the process of running vocoder tests and has been working to drive SMV standards. The company has put its infrastructure vendors on notice that it will be looking at the technology carefully, and it wants to implement this as soon as possible, Prochaska said. SMV vocoders would be deployed well after Sprint's 3G network is operational. In the meantime, it will use EVRC vocoders.

One of the vendors likely to be involved is Qualcomm, which has committed to SMV for its next-generation chipsets, said Mark Frankel, Qualcomm senior director of product management.

“We have a next-generation chip in our 6000 series that we're targeting to be SMV-enabled,” Frankel said. “The chip easily has the computational power for this. The real issue is divining the schedule between us as a handset chip provider and the implementation of the vocoder by the network equipment suppliers. The network operator has to confirm that they still have head room in the DSP cards in their infrastructure so that those DSPs aren't maxed out and have enough processing power to handle the SMV algorithms. It would be somewhat minor in terms of an upgrade.”

Frankel said he expected vocoder gains to continue, with the next entry coming from the GSM community.

“There were tests done comparing SMV with the various AMR rates,” he said. “AMR is the standard vocoder in GSM today. As you would expect, SMV, being a newer vocoder, scored better. I don't know what's coming after AMR from the GSM community.”

Dual Path

Cingular (www.cingular.com), with its TDMA and GSM networks, is following vocoder developments for both air interfaces, said Keith Radousky, Cingular director of engineering.

With Cingular's TDMA lineage from BellSouth and SBC, the first vocoder it deployed in the early 1990s was the vectorial sum excited linear predictive (VSELP) vocoder, which achieved 7.95kb/s and mediocre voice quality. VSELP was used until 1996, when the company deployed an algebraic code excited linear predictive (ACELP) vocoder with a lower bit rate and enhanced error correction.

“The ACELP vocoder is serving us well in TDMA, and there aren't any plans to change it,” Radousky said.

On the GSM side, a 13kb/s vocoder was replaced in the early 1990s by enhanced full rate (EFR) vocoders. GSM then developed a half-rate vocoder. When it was tested, subscribers rejected the service because the voice quality was so bad.

“That vocoder went over like a lead balloon,” he said. “The quality just wasn't there to produce a quality vocoder only using 6.5kb/s. In its place, the GSM community decided to do an AMR vocoder.”

Ensigma Technologies released its AMR vocoder for GSM markets in January. Dr. Graham Wacey, Ensigma project manager, said the AMR vocoder creates capacity for carriers while achieving the best possible speech quality.

He said the AMR is better than the half-rate GSM vocoder, which was trialed by Vodafone. “It halved the number of bits to double the people on the network, but the deployment was quite poor. They thought the most important thing was capacity, but when they turned it on, they got tons of complaints about the low quality of speech. It's still used in emergency systems where the most important thing is for lots of people to get a channel. It was interesting to see that people preferred to lose calls rather than lose quality of speech. That's where AMR comes in with better quality, and it is a benefit to the network in terms of capacity.”

AMR vocoders find their sweet spots by varying the bit rate based on radio-link quality, Radousky said.

“If the radio link works perfectly, every time you put in a zero, a zero comes out the other end,” he said. As the radio link degrades, the AMR vocoder will put more and more bits in there to maintain consistent quality across the radio link.”

Cingular has AMR vocoders in its base stations and is waiting for AMR-enabled handsets to complete the upgrade, which it expects to have toward the end of this year.

“We're counting on those AMR phones for our network planning,” Radousky said. “We're expecting that vocoding technology to double our capacity.”

With 6.5kb/s as an average, carriers can adjust the AMR vocoder to match network conditions. From an operational standpoint, Cingular is conducting trials to optimize those parameters, Radousky said.

“We have the EFR vocoder deployed in both of our technologies, which is the 7.4kb/s ACELP in TDMA and the 12.2kb/s ACELP in our GSM markets,” he said. “We're planning to get our feet wet with AMR vocoding at the end of the year in our GSM markets, but TDMA is not changing. If you were fully deployed where all the mobiles were AMR on the network, you could realize at least a doubling of the capacity.

“However, you're going to have a mix of EFR and AMR. The sooner we begin deploying the AMR coder, the sooner we would enjoy the benefits of the capacity gains.”