|
 |
 dspZONE Products for the week of November 12, 2007
Octasic Says…
Most Competitive Multi-Core Media Gateway DSP
Vocallo’s unique licensing enables new business model; first Octasic product to use groundbreaking Opus DSP architecture
Octasic, Inc. has announced its next generation multi-core gateway DSP platform for voice, video and data over IP applications. Vocallo combines ultimate flexibility, with pristine voice and video quality, to deliver the most comprehensive solution for current and future media gateways.
The Vocallo expandable media gateway solution is the first DSP platform of its kind to enable a new flexible business model for OEMs designing media gateways. This business model enables an OEM to design a cost-reduced gateway to handle today’s Voice over IP (VoIP) needs with licensable performance to accommodate future applications. This allows OEMs to defer the costs of additional performance capacity until they deliver and charge for an application that requires it.
“Because requirements are constantly changing, there is a need for greater flexibility and expandability in media processing devices. Vocallo is the only solution today that allows OEMs to pick their features, add-in their own software and upgrade,” said Frederic Bourget, director of product management at Octasic. “Vocallo is powerful and flexible. It is a comprehensive solution for current and future applications.”
Vocallo is based on Opus, Octasic’s unique DSP core architecture, which offers the highest performance-to-power ratio in the industry with a traditional programming model. In high density applications Opus is the most efficient DSP core processor by two-fold. In addition, Opus provides a rich instruction set that goes beyond DSP functions allowing for expandability into new applications. (See Octasic’s release on the Opus core. http://www.octasic.com/en/about/news.php?pr=50)
Optimal Flexibility for Evolving Applications
With its flexible hardware and software platform, Vocallo allows media gateway developers to match an offering to their specific requirements. The features and benefits of the Vocallo DSP can then be scaled to create a wide range of cost-effective products from a single design. As an added benefit, Vocallo allows designers to add their own software to the Vocallo framework to further differentiate their products.
"As VoIP adoption grows, service providers will continue to rely on media gateways," said Jon Arnold, principal of independent analyst consultancy, J Arnold & Associates. "Their needs will become increasingly complex, and Vocallo provides maximum flexibility to quickly adjust to changing market conditions."
Pristine Voice Quality for Today’s Applications
Vocallo’s Echo Cancellation and Voice Quality Enhancement (VQE) features use Octasic’s carrier-approved algorithms. To maintain pristine voice quality in the latest standards, Vocallo is built on a packet-based software architecture designed from the ground-up to support wideband voice traffic and process media connections with minimum latency.
Delivering on Emerging Applications
Forty million voice lines in networks worldwide rely on Octasic’s carrier-recognized solutions. Building on this rich history, Vocallo supports emerging network architectures and applications, such as IMS and Fixed Mobile Convergence (FMC). In addition, Vocallo provides support for robust real-time video transcoding, transrating, and conferencing.
“Our experience in designing high density voice solutions for carrier equipment OEMs over the last ten years has led us to develop the Opus DSP core,” said Michel Laurence, Octasic’s president and CEO. “Through its use in Vocallo and follow-on products, Octasic is providing processing platforms with new models of cost and performance allowing our customers to be more flexible and more competitive.”
EN-Genius Says…
If I were Texas Instruments, I’d be looking in my rearview mirror about now because the technology used in the Octasic Vocallo processing chip could be a disrupting factor in at least one of their key markets. Octasic has been one of a handful of companies producing specialized signal processing products that managed to survive against much larger traditional DSP makers (like Brand T) by picking a specific market niche and delivering a carefully-crafted product that delivers better channel density and better price/power/performance ratio than their general-purpose cousins. In the past, Octasic's focus on echo cancellation, acoustic echo control, and noise reduction actually made their products a good choice to complement DSP-based VoIP solutions by using their dedicated cores to do much of the heavy lifting, freeing the programmable elements to handle more channels of coding and other tasks. Now, their introduction of the first commercial application of asynchronous logic I’ve seen has enough power to challenge TI, Freescale, and other DSP makers by taking on both processing and coding tasks.
 Vocallo is a three-chip set based on Octasic Opus asynchronous DSP architecture (see their October 16, 2007 release) which sandwiches the 15-core DSP element between an external (mobile DDR), and an I/O chip that harnesses the DSP to a GbE or UTOPIA line interface, and a TDM bus. Together, they can support up to 672 channels (a DS3 worth) of G.711 voice traffic, complete with tone detection, 128 ms of echo cancellation, auto fax relay switching, and other features. If you can drop some of the special features and trim the echo tail a bit, it can support even more channels.
Octasic claims that because the architecture is packet-based, transcoding functions are easy to implement -- using only two channels' worth of regular voice processing capability to implement a typical EnDEC function. This makes Vocallo especially good for 3G or VoIP session border controllers and other high-density applications that bridge voice coding formats. It’s also been designed specifically to accommodate the new wideband codecs (like AMR-WB) as well as variable bit rate video such as H263/264 for advanced services that are starting to pop up within 3G wireless and VoIP networks.
As I hinted earlier, Vocallo gets much of its punch from the asynchronous logic that drives each of its fifteen 32-bit compute elements. Vocallo home-grown architecture is the result of a four-year project work to reduce the power and cost involved with voice processing through the use of a smaller, simpler core that runs contrary to the parallelization techniques typically used to increase capacity and speed. Unlike a data flow architecture which pours data through a pipeline of instruction stages, the Opus architecture pours instructions through the processor which grabs data from memory and manipulates it as appropriate. In these small multi-stage cores, information cascades to the next stage as it is processed without the need for a clock.
Although each compute stage does not have to wait for the drumbeat of a master clock to do its work, there is logic that ensures that it does not go on to the next step until all other results are available -- at least most of the time. To take advantage of the core’s maximum speed, Octasic has developed a sneaky speculative complete feature that can use the results from a previous calculation that’s assumed to be stable many nanoseconds before the results are absolutely guaranteed stable. To cope with the occasional glitch, the processor is designed to re-compute the next step if one of its inputs does change. Octasic says that, if desired, you can lock out the speculative feature to save power. I’d also conjecture that disabling the speculative feature could be useful for certain classes of applications that generated enough re-computations to consume more time than it saved.
Each core runs independently -- incoming data is stored in memory and then accepted by a dedicated processor core that classifies it, a second core schedules it for processing by one of the 14 remaining cores. Although the processors use asynchronous logic internally, they are connected via a common bus and network interfaces using conventional timing and clock signals. The same bus is used to support resource sharing, messaging, and memory protect functions among the cores. Even with the synchronous outer layer, each core has the theoretical equivalent throughput of a conventional DSP core running at 1.5 GHz -- although Octasic cautions that data dependences in the timing of real-world applications will result in somewhat lower actual performance.
To keep developers and applications engineers from having a nervous breakdown over the complexities of its asynchronous DSP core, Octasic wisely chose to wrap it in traditional clocked logic that lets it present itself to the outside world like a traditional SIMD processor. Its conventional-looking instruction set could have been lifted from any other 32-bit DSP commonly used for signal processing operations. Vocallo’s single-threaded programming model allows each core to be separately programmed and is complemented by a C compiler tailored to the processor’s instruction set. This traditional programming and optimization model means that developers need only a little training to become proficient at coding or developing algorithms for this multi-core engine.
One could speculate that putting a friendly face on the core’s decidedly unconventional innards could compromise its ultimate performance, but I’d say that any penalty is more than repaid by the fact that you don’t need a PhD in computer architecture to program it. One nice example of the simplified programming approach at work here is the host controller software. The in-band signaling scheme uses specially tagged packets in the processor’s Ethernet line connection to exchange commands, status and error messages, without a separate control plane interface.
Vocallo’s high levels of channel density and low power-per-channel put it in some rarified company with only a handful of chips to compare it to. Off the top of my head, Centillium’s Entropia IV boasts a 3 mW/channel power budget, one of the few devices that comes close to Octasic 2.2 mW/channel. If I was comparison shopping, I’d probably also look at TI’s TNETV2030 high-density VoIP solution. The TNETV2030 six ripping-fast cores and deep library of application software could give it more flexibility in application mix, video capability, carrier-grade management features, although I think Vocalo may still beat it in absolute channel density and overall power consumption. Since Octasic was very tight-lipped on pricing it’s impossible to make any meaningful comparison on the basis of cost-per-channel.
Octasic did share one interesting tidbit regarding pricing, though. It turns out that they are using an on-chip key file system to ration out features and channel capacity as customers are willing to pay for them. Octasic says that the same systems will be used to allow special features (special packet headers, custom encryption, etc…) to be added via custom programming or 3rd-party vendors. In theory the customer will be able to design equipment that can be easily upgraded in the field, although I worry that locking down so much of the chip could also affect its flexibility.
This is not the first time a pay-per-feature scheme has been used in a chip. One notable example is the Broadcom BCM8603 SAS/SATA RAID-on-chip storage controller.
Vocallo devices will be available programmed with capacities ranging from around 16 channels to just under 1 k channels. The basic cost for a mid-sized application is anticipated at around $1/channel, although this will vary considerably according to the volume and application set. Despite any concerns I may have, it certainly makes sense for Octasic to be able to be able to sell products with different capacity ranges without the massive tooling costs required to make three or four different chips. Should sales volumes begin to become significant enough to warrant it, Octasic already has a roadmap for developing right-sized products with both higher and lower channel densities.
|
|
|
|
|
