Digital audio processing is the key to "full employment" for engineers. That's according to Tomlinson Holman, the inventor of Lucasfilm Ltd.'s THX theater sound and the keynote speaker at DSP World Spring Conference this week (April 26-28) at the Santa Clara Convention Center, Santa Clara, Calif. Holman argues that Dolby digital-5.1 speaker surround sound-is really not powerful enough to convey the realism of a dramatic situation on film or in a concert hall. The 5.1 topology, with left front, center front, right front, left rear, right rear and directionless subwoofer, does not do justice to the "early reflections" from the wall of a theater or concert hall, he said.
What does it take to make picture and sound transparent to the viewer? Holman believes that the answer lies in "everything you can offer." But then, the home-theater enthusiast may not have the same sound-equipment budget as a Hollywood studio. Given a choice between sample rate, word length and number of channels, Holman believes the latter will be most "psycho-acoustically useful." Sampling data with a 24-bit converter, in principle, gives you sound information with a 144-dB dynamic range-low-amplitude signals that you wouldn't be able to hear and high-amplitude signals that could destroy your hearing. "That's silly," said Holman, "get on with more channels."
San Diego State's Fred Harris believes multirate signal prcoessing,
where a DSP-based system samples the analog signal at far higher
multiples than the Nyquist rate as much as 256 times the bandwidth
yields a significant reduction in system components and, eventually,
costs. Harris shows a Cubic digital receiver designed in the university's RF lab.
Holman's talk on Tuesday night will be tied to a continuously running demonstration of a multichannel sound system that, according to DSP World promotions, "pushes beyond the limits of today's technology." The system is something that Holman would hesitate to call "10.1 channel," but certainly his new company, TMH Corp. (Los Angeles), will be demonstrating the realistic effects that could be produced with a full circle of speakers. Two of them are arranged at plus/minus 30 degrees to the listener; another two are at plus/minus 60 degrees; two are at plus/minus 110 degrees; and two are at plus/minus 150 degrees. There will be two subwoofers and an additional pair, making up what Holman calls a "high-wide front." Using material collected at Lucasfilms and the School of Cinema-Television at USC, where Holman holds a professorship, the demonstration is promoted as a "seriously fun romp that challenges the way engineers think and feel about DSP."
Audio entertainment is in fact but one of the applications and concepts to be promoted at DSP World . And, like the conference, this week's Signals Focus section keys in on a number of contemporary themes in digital signal processing.
Multirate signal processing is one of the more glamorous and attention-getting themes. Fred Harris, professor of electrical and computer engineering at San Diego State University, has done work in this area that is among the most visible and impressive efforts. The Cubic Signal Processing Chair at the university Harris holds a number of patents on digital receiver technology, and Harris lectures on DSP applications. Clients include the Navy Ocean Systems Center, Lockheed, Hughes, Sylvania-GTE, Rockwell, Hewlett-Packard and Motorola.
In his contribution to the section, Harris' tutorial on multirate signal processing uses CD and DVD audio as prime examples. He notes that with contemporary audio processing-DVD players for example-the system must accommodate signals sampled at 96 kHz as well as at 44.1 kHz. To juggle the multiple sample rates, the DSP-based system will actually sample or resample the analog signal, not just at Nyquist criteria, two times the effective bandwidth to prevent aliasing, but at many times-64 or 256 times-the effective bandwidth. The advantage is a dramatic reduction in the number of system components, and an eventual reduction in cost. A digitally oversampled signal is, in fact, easier to filter than a lower-frequency signal in the analog realm.
In sigma-delta audio D/A converters, for example, the sample rate of the digital audio stream is synthetically raised 64, 128 or 256 times from its original 44.1 kHz. The digital audio stream then resembles an audio signal sampled at approximately 12.5 MHz. Changing the sample rate this way effectively elevates the base frequencies of the quantization noise out of the audio stream, where it is easier to filter, and drastically compresses the changes in amplitude between one sample and the next. Consequently, a 1-bit switched-capacitor D/A-essentially, a fast charge pump-is all that is required to restore the original analog signal.
After multirate processing, one of the most intriguing developments at the DSP World Spring Conference is the new DSP architectures. Engineers are curious about whether very-long-word-instruction (VLIW) and other wide-word architectures such as Siemens' Carmel, or superscalar machines like Analog Devices' TigerSharc, will do a better job of servicing their new applications. Interestingly, benchmarking experts Jennifer Eyre and Jeff Bier of Berkeley Design Technology Inc. (Berkeley, Calif.) point out that the TigerSharc is more like a VLIW machine in the compile-time choices it makes.
General-purpose DSPs are continually challenged by other devices-microprocessors and FPGAs, for example-for their ability to perform fast math functions. DSPs hold their own against microprocessors on many operations, such as digital modulation of cell-phone RF signals, or motor control (acceleration/deceleration) functions. But programmable logic seems to have a performance edge in encoding/decoding functions-especially with mathematically dictated expansion codes such as CRCs and error-correction codes.
Rob Weinstein of Memec Design Services (Mesa, Ariz.) and Anil Telikepalli of Xilinx Inc. (San Jose, Calif.) explain how programmable logic can be used to implement the convoluted coding used in digital video broadcasts. Implementing communications and DSP functions in FPGAs is another major theme of DSP World.
Not coincidently, wideband CDMA makes use of convolutional encoding. The contribution from Marc Barberis of Synopsys (Mountain View, Calif.) describes the design and operation of a wideband code-division-multiple-access handset. From the time it is powered on, writes Barberis, the handset must search for a pilot signal and coded data in carefully specified channels using rake receivers with diversity antennas. The encoding on 60-MHz channels that will elevate the data-reception rate for voice-over-Internet Protocol (VoIP) is another new and glamorous application of DSP. Ross Mitchell and Ken Unger of British Columbia-based HotHaus describe some of the algorithms a part like the Texas Instruments TMS320C5410 must process in order to packetize digitized voice for the Internet-and still maintain quality of service. They note that the "carriage of voice over packet-data networks actually embodies numerous complex technologies, including packet voice, fax relay, signaling, network-protocol stacks, call agents and network-management applications."
Mitchell and Unger note that the algorithms that must be supported by the DSP on a per-channel basis include those for voice-such as echo cancellation, automatic gin control, comfort-noise detection-as well as fax, modems, DTMF signaling and packet management. The last includes packet assembly, concatenation, delay equalization and jitter management. They point out that a 100-Mips processor actually will do VoIP processing for five channels.
Meanwhile, no DSP system can be implemented without decent software-development tools to analyze and optimize trade-offs in DSP product design. Phil Radtke and Arda Erol of Spectrum Signal Processing (Burnaby, B.C.) explain how a trace tool-one that visualizes the actual execution cycles of DSPs in a multiprocessor system-can tweak and improve system performance.