Houston (October 15, 2001) - Texas Instruments Incorporated announced today a software library of 21 imaging functions that allow designers to rapidly develop power-efficient, digital imaging applications utilizing video hardware extensions designed within the TMS320C5510 and TMS320C5509 DSPs. A single 200 MHz C5510 DSP, available today, can easily perform MPEG4 video encoding and decoding at 30 frames per second in quarter common intermediate format (QCIF) resolution, while consuming only 50 milliwatts of power for applications such as PDAs, digital camcorders, Internet-enabled set-top boxes and other digital imaging and video appliances.

The new software library developed by Texas Instruments includes C-callable functions (ANSI-C language compatible) to utilize the hardware extensions for core components of MPEG-4, JPEG and MJPEG processing such as discrete cosine transform, motion estimation, pixel interpolation and a library of video operations such as 1-D/2-D pyramid wavelet decomposition and reconstruction. These components are suitable for inclusion in creating eXpressDSP-compliant algorithms, which will shorten development time in designing high-performance video encode and decode into portable applications. The software library and complete documentation are free and downloadable from the web today.

The TMS320C55x DSP core was created with an open architecture that allows the addition of application-specific hardware to boost performance on specific algorithms. The hardware extensions on the C5510 and C5509 DSPs strike the perfect balance of fixed function performance with programmable flexibility, while achieving low-power consumption, and cost that traditionally has been difficult to find in the video-processor market.

The extensions allow the C5510 and C5509 DSPs to deliver exceptional video codec performance with more than half its bandwidth available for performing additional functions such as color space conversion, user interface operations, security, TCP/IP, voice recognition, and text-to-speech conversion. As a result, a single C5510 or C5509 DSP can power most portable digital video applications with processing headroom to spare.

"Traditional programmable solutions like RISC processors provide more than enough flexibility, but lack the performance to do more than the task at hand," said Marc Guillaumet, product manager, A.T.E.M.E. "On the other end of the spectrum, an ASIC solution provides all of the performance defined for the application, but lacks flexibility. Other DSPs take the approach of using general purpose processing units such as arithmetic logic unit's (ALU) that require special multiple issue instructions and add complexity for the programmer. TI's approach is the best solution in regards to performance and ease of use."

Unlike accelerator units that are attached to a DSP core as a peripheral, the C55x DSP is designed so that hardware extensions can be integrated within the core's resources thus allowing them to become part of the core signal flow. An external accelerator requires hardware to be duplicated that already exists in the DSP core and communication overhead to handle the data transferred from the core accelerator. Both requirements can increase latency, degrade chip resource utilization, and increase programming complexity. The C55x DSP core has none of these drawbacks, stripping away roadblocks to optimum circuit design with exceptional imaging and video performance.

"The C5510 and C5509 DSPs incorporates hardware video extensions allowing designers to add imaging and video capabilities to their portable applications," said Christine Wu, TI C55x DSP product marketing manager. "They have been carefully tailored for the C55x DSP generation since we began development of the core. TI will continue developing extensions for other applications in the future."

For portable video and imaging applications that require higher resolution image and video processing, TI offers the TMS320C5000 DSP based TMS320DSC24 DSP solution. This system-on-a-chip packs the processing power needed to deliver encoding or decoding of real-time video in applications such as streaming video. The DSC24 combines on a single device the ultra-low power TMS320C5000 DSP to perform real-time media processing and an ARM7TDMI RISC processor for system control functions.

For high-performance video client and infrastructure solutions, the TI TMS320C6000 DSP platform, the industry's highest-performance and most flexible fully software programmable DSPs, includes special-purpose instructions tailored for video and imaging applications. It's flexible on-chip memory scheme and high-bandwidth I/O capability are designed to handle dense data streams that are characteristic of high-performance video and imaging applications including video head-end infrastructure equipment, wireless video gateways, digital media centers, streaming media servers and IP-based video phones & other surveillance/netcam type applications.

Proven Development Tools Improve Design Success

The C55x DSP generation is supported by the most complete, field-proven development platform in the industry - eXpressDSP real-time software technology. This easy-to-use suite of tools features the Code Composer Studio Integrated Development Environment; the DSP/BIOS real-time kernel and a large and growing range of application software compliant to the TMS320 DSP Algorithm Standard from TI's nearly 500-member third-party network.

Designed for the Power-Efficient Needs of Portable Applications

The C5510 and C5509 DSPs combine up to 400 MIPS of processing speed with the industry's lowest power of 0.25 mW/MIPS. The devices provide a seamless, code-compatible migration path from TI's programmable TMS320C54x DSP generation - the world's most popular DSP. TI's C55x DSP generation incorporates functions that would otherwise need to be enabled by external devices, thus reducing board space, parts count and overall product cost. This integration combined with the inherent power efficiency of the C55x DSP generation improves battery life up to 70 percent and allows the end product to offer more features in a smaller footprint.

Availability

Samples of the TMS320C5509 DSP at 144 MHz and the TMS320C5510 at 200 MHz are available today. Full production is scheduled to begin in the first half of 2002 at a planned price of $18 for the C5509 and $35 for the C5510 in 10,000 unit quantities. The C5509 and C5510 DSP evaluation modules (EVM) are available today for a special introductory price of $995. Complete documentation and technical training are available for free.

When TI introduced the TMS320C5509 and 5510 last spring, they didn't tell the whole story. They didn't disclose some features of the underlying device architecture, primarily because they didn't yet offer software to take advantage of those features. With this announcement, they have the software and are now starting to reveal details about these hardware features. Indeed, while the press release emphasizes the libraries, which are available at no charge, the story really revolves around the underlying hardware that TI is starting to talk about.

The applications they're addressing with these developments involve digital imaging using the latest compression algorithms. Certainly, designers can add an accelerator chip to a system to assist with these tasks, but that approach has several drawbacks. Among them are the fact that the two chips must duplicate some functions such as program sequencers and program address units. You'll also pay a penalty in communications overhead, adding to latency times and increasing program complexity. Then, of course, you're paying for two chips and must also supply the extra real estate and power.

Why not include these accelerator functions directly in the base DSP? That's what TI has done with the 55x, plus the 5509 and 5510 feature three "bolt-on" hardware extension engines that perform a DCT, motion estimation and pixel interpolation. Indeed, the 55x architecture supports eight such bolt-on engines, says Dennis Barrett, C5000 product marketing manager. He also explains that the engines are gated and don't draw power unless activated. Further, they don't take much die area because each of the three engines he is talking about now require in the order of 50,000 gates.

Note that you won't find these "bolt-ons" on every 55x series device; TI plans to add them only when it makes sense, in this case for embedded video applications. For instance, the firm has announced (but is not yet delivering) the 5502, which has limited internal memory and thus requires external memory for many applications. This situation means the device isn't well suited for handheld battery operated systems, a prime target for this new imaging technology now being announced. Also, while the firm has announced OMAP class products based on the C55 core in the wireless space, those devices won't have these add-ons, either. However, you can expect to see later OMAP devices and other future products based on the C55 core that have the ability to use these extensions as well as other extensions that might come along.

But isn't today's announcement really about software and a free library? Indeed, and TI is now making available a library of 21 imaging functions, functions required to implement popular imaging standards. As you might imagine, most of them not only take advantage but indeed require the C55 hardware extensions (although the same functions are available for other processors in code that doesn't require them). The library functions are delivered as a mix of source code and object code. For instance, some of the more basic functions such as a 1D or 2D pyramid decomposition and reconstruction are all in source code, while some higher level functions such as the DTC come all as object code. TI's Barrett acknowledges that some engineers want to examine the details of algorithms and says the delivery situation could change in the future. He also points out that for key customers TI can work out a separate agreement with regard to supplying source code.

Also note that TI previously made the details of these library functions available to third-party software suppliers such as Ingenient Technologies, ATEME, Power Image and ubvideo, so users can already purchase an entire imaging solution based on these algorithms.

So what benefits can you expect in terms of raw performance thanks to these on-chip accelerators? Let's take just one example from the TI media briefing, which compares the C55x to a RISC processor, specifically a 200-MHz C55x to a 233-MHz StrongArm SA110, both running MPEG-4/H.263 QCIF encoding at 15 frames/sec. The C55x needs 21M cycles/sec of its total processing power to perform the task, whereas the StrongArm needs 153M cycles/sec, a factor of more than 7.

And what about some of the other chips on the market with architectures intended to improve performance on video applications? Here, the media briefing compares the C55x with the Intel/Analog Devices FRIO, which is now available as the ADSP-21535 Blackfin. For an 8x8 DCT, the C55x requires 147 cycles, whereas the Blackfin reportedly needs 284 cycles, a factor more than 1.9 greater.

A final point that TI likes to make regarding these on-chip accelerators and libraries is how much horsepower is available for other system tasks. For instance, the firm points out, even when running CIF resolution at 30 frames/sec, the 200-MHz C55x still has almost half of its bandwidth available for auxiliary functions such as color-space conversion, the user interface, security features, TCP/IP, voice recognition and text-to-speech capabilities. Thus, most portable video applications should need only one C55x or OMAP device to perform all the required processing. Certainly, an ASIC solution can provide very strong performance on video encode and/or decode, but the system generally needs a second processor to handle auxiliary functions.