Electronic Design

GSPx Shows Multiprocessors To Meet Video's Challenges

It’s not the everyday conference keynote that walks the audience through algebraic equations, but when you’re at a signal processing event, algorithms really are at the heart of it all.

The Global Signal Processing Exposition (GSPx) last month in Santa Clara provided a great opportunity to catch up on the latest challenges and solutions in this fast-moving world. To sum up the trends in a single word: multiprocessors. Intel and AMD have made headlines with the dual-core approach. But GSPx made it clear that the multiprocessor trend is far-reaching and rapidly becoming an essential approach to solving today's signal-processing challenges.

For the uninitiated, "dual-core" might connote two equal, parallel-processors that split up computational tasks. The dominant application for multiprocessors, however, is the intelligent shifting of specialized tasks to a processor best suited to that particular duty. In leading cell-phone systems-on-a-chip (like the TI OMAP functions shown in the graphic), there are multiple custom designed processors. Each processor performs its own function: graphics, versus voice processing, versus security, versus radio communications. In many other cases, the coprocessors may be standard DSPs or FPGAs, which were portrayed at the conference as either competitive or compatible co-processing solutions, depending on the conference panelist.

Just doubling the general-purpose processing power isn't usually the most efficient way to solve signal-processing challenges. Typically, the "bottleneck" of a given processing problem will only occupy 10% to 20% of the MAC capabilities of a given processor. So adding another parallel chip may add as much as 80% unneeded overhead. It's better to marry the main processor to a coprocessor that can uncork that specific bottleneck by offloading that function in the most efficient manner.

Still, there are powerful algebraic advantages to parallel processors, explained keynoter Dan Bouvier of Freescale. Most algorithms can be broken down into parallel computational methods to execute the algorithm in far fewer steps than if solved linearly. These alternate ways of parsing the algorithms have the biggest parallel processing impact.

VIDEO CHALLENGES Video processing was the "killer app" at GSPx, as seminars offered various ways to solve the challenges of processing video in cell phones, on portable devices, over networks, and for HDTV, DVDs, and set-top boxes. Beyond the challenges of processing higher-quality digital video in smaller or thinner packages, signal-processor vendors expect video to follow the path blazed by audio in which consumers will access video in multiple formats and "transcode" from one format to another.

In his keynote on 3G applications for cell phones, Qualcommcofounder-Andrew Viterbi outlined future scenarios for wide-area broadband wireless, where consumers may use the cell phone as a carrier in new distribution methods for video and other new media and content. In a session on Ultra-Wideband (UWB), Freescale's Jon Adams offered a similar vision, where UWB would enable cell phones to be the quick-loading portable carrier of choice to bring in media from outside the home or office. (Freescale recently demonstrated UWB coupled with Bluetooth, allowing Bluetooth devices to exchange data at 110 Mbits/s.)

VIDEO SURVEILLANCE Video surveillance and security systems pose another big challenge for video processing. Mike Harrison-Jones of London's CCTV Services Ltd. gave attendees a behind-the-scenes look at the video surveillance networks that allowed London police to quickly identify the July 7 bombers and to arrest perpetrators in the July 21 bombing attempt.

Most current crime prevention depends on human operators monitoring banks of video monitors in control centers. The capabilities of the systems are being greatly enhanced, however, by the move to digital image processing that includes image and pattern recognition. Harrison-Jones explained that pattern recognition can be coupled with the human monitoring to call out certain incidents that may be of interest, permitting far more efficient use of the operators' time. Images from a transportation hub are analyzed for behaviors such as running and loitering, as well as unattended packages or other suspicious items. In a garage, cameras can recognize pedestrian movement, then transmit only those frames involving pedestrians to the surveillance-center operators.

A processing challenge is capturing high enough resolution so that cameras can focus on large-scale activity patterns but still let operators zoom in to recognize facial features or even license plate numbers for identification purposes.

While issues of personal liberties in such detailed and constant surveillance are still being worked out in the British legal system, Harrison-Jones included plenty of statistics to show that the systems are successful, lowering crime rates and increasing Londoners' sense of security.

See associated figure

Hide comments

Comments

  • Allowed HTML tags: <em> <strong> <blockquote> <br> <p>

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.
Publish