Native Signal Processing

How Intel & Microsoft
Failed to Communicate

Francis Vale


Personal computing is about to hit a fork in the technology road. We are at the onset of 'ubiquitous computing". This new technical direction is being propelled by the next generation of wireless communications, like Personal Communications Services (PCS).

At the heart of these new wireless computing devices will not be Intel PC CPU's, but rather, digital signal processing chips. And as a consequence, this new DSP-dominated path might be hiding a corporate sinkhole for both Microsoft and Intel.

DSPs process radio wave transmissions far more effectively and easily than traditional computer CPUs. Motorola, for example recently unveiled its 66 MIPS, 24 bit, 56300 DSP core engine, aimed at the high end digital cellular arena. Analog Devices is also in the telecommunications game, with its Oak DSP core that is ideal for V.34 modems. DSPs are also superb at communications echo-cancellation and noise reduction, as well as doing data encryption.

Today, conventional system technology requires a dedicated receiver tuned exclusively to a specific band. But DSP-based radio systems are appearing which let a single unit simultaneously handle over 60 plus channels. The ultimate industry goal is for DSP-based systems to replace hardware-oriented radio frequency technology with an all digital format, which many call 'digital radio', or 'software radio'. As its name implies, a software radio seeks to replace all the hardware radio functionality with intelligent algorithms, processed by speedy DSPs.

Multimedia processing performance on DSPs also surpasses conventional CPUs. A DSP also excels at multimedia data compression tasks. And leaps forward in DSP performance are happening much faster than in traditional CPUs, including the RISC units. Finally, DSPs are much cheaper. In 1995, Motorola's MVP DSP chip cost about $232 per thousand MIPS, vs. an Intel Pentium's one hundred or so MIPS which cost around $400. With this kind of price/performance advantage, the DSP chip is well on its way to becoming the mainstay of mobile computing.

The big question is, will DSPs just stay in the signal processing background? Or will new DSP chip generations evolve into something else; something at a higher user interaction level? Will users continue to place a premium on having traditional software applications, like word processing? Or will users instead want slick video/speech personal interaction systems? Will these mobile, multimedia interaction devices become the new definition of a PC?

Obviously, the success of DSP-based, mobile 'PCs' would not please either Intel and Microsoft. Both companies have a lot to lose if DSP-based PCs were to redefine computing and gain significant market share.

Indeed, if such DSP-enabled systems were to become popular mass market goods, the two PC giants might, as a defensive move, throw their lot in with the consumer/entertainment industry and its highly restricted notion of computing New 'WinTel' machines might appear at a cost of about $500, but would likely suffer from limited functionality. Neither company would want to cannibalize its fat margin sales coming from $2,000 - $4,000 PC systems. (The new 'Internet appliances' we are hearing so much about might be similarly kludged, but we will have to wait and see.)

From an 'open' systems viewpoint, such marketing collusion between the two PC giants and consumer electronics companies would create the worst of all possible 'closed' computing worlds. Herein lies a potential example of 'digital convergence' run amok. (This innocuous term makes these tectonic cybercrunches sound like some sort of industry love-in.)

Meanwhile, neither Intel nor Microsoft is standing still, waiting for the other DSP shoe to drop. For example, witness Intel's attempts to move software developers' attentions away from DSPs and onto the next generation of Pentium/P6/P7 chips. And last year, Microsoft also scuttled its DSP software support efforts in favor of native Win95/NT audio drivers.

Intel thinks its ever faster Pentiums, and especially its new P6/P7 chips, (NOTE: This article was written in Feb. 1996) are fast enough to take over many DSP functions.
Moreover, Intel says that some parts of the multimedia application have to run on the host CPU anyway, so why not put it all on there? Intel calls this DSP-off loading onto its CPUs, 'NSP', or Native Signal Processing. In essence, NSP is Intel's idea of a 'software DSP.'

Being no fool about future multimedia directions, Intel wants to keep the DSP foxes out of the PC henhouse any which way it can. But a $10 or $20 DSP chip, with its tremendous multiplication capabilities, can run past a Pentium any old day. So Intel's performance riposte to the DSP's native performance was its new P6 with integrated, special multimedia features; e.g., instructions that assist in MPEG-video decompression.

These P6 MPEG features notwithstanding, digital audio processing is where most of the PC market is at, especially for DSP-enabled speech. Intel therefore initially targeted complex, real time, digital audio processing PC applications (games, in particular) as the first candidate for NSP. For its Native Audio features, Intel used IA-SPOX, a real-time kernel that drives the audio portion of NSP. To reach IA-SPOX, developers had to go through an Intel-specified API (application program interface).

But where there is a software API on a computer, there is Microsoft. The resulting Clash of The Titans was predictable: Last year, Microsoft essentially ordered third party developers not to write applications that used NSP's IA-SPOX services. Instead, Microsoft developed its own APIs for handling this complex data type. Rather than using a real time kernel like IA-SPOX to provide the services, Microsoft set out to provide efficient drivers for all the popular sound cards.

Microsoft also provided a high level audio 'mixing' panel for developers to access, instead of the much lower level NSP APIs of Intel. All of these Microsoft enhancements eventually found their way into Windows95/NT. TAPI (telephony) applications can also use the audio API instead of a speech processing DSP. This new audio API is now part of Microsoft's larger set of (Direct-X) multimedia APIs.

Microsoft's audible karate kick to NSP resulted in Intel suddenly going quiet last summer. But Intel did not accede the NSP fight so easily. Although NSP initially ran only under Windows 3.x , Intel was busily porting away to Win95/NT. And to start the new year off with a bang, Intel announced a NSP software library with more than 200 functions and algorithms. These functions and features include fast Fourier Transforms, finite-impulse response filters, 2-D image processing, vector manipulation routines, and speech recognition primitives. Meanwhile, Intel dropped its Native Audio work, where most of the NSP/Microsoft API fracas all started.

For its part, Microsoft has also been busily coding away, fixing up its Direct X multimedia APIs so they can run at ring 0 of the NT/Win95 operating system, instead of at the current ring 3. At ring 3, latency problems still occur as the application software has to pass through more layers of the OS to get at the underlying hardware. These MS efforts are being driven in large part by games developers, for whom processing speed is king (which is why MS DOS-based games refuse to die, much to the displeasure of Microsoft.)

But regardless of the API winner in this DSP chameleon war, the fact remains: A seismic computing shift is obviously imminent. Multimedia chips with hundreds of millions of transistors, screaming along at one trillion operations per second, will become the heart of tomorrow's 'PC'. True DSP processing, and not some pseudo-DSP, will be an integrated part of these new computing engines.

For all of Intel's and Microsoft's API efforts, neither company has correctly positioned itself for the next computing era. The Intel CPU, so long the PC industry standard, will be relegated to an ancillary processing role in this new mobile, multimedia world order. And if DSP-dominated devices are to become the new personal computing future, where does that leave Microsoft? Despite MS NT's portability to non-Intel CPUs, Microsoft has bet the OS/application farm on Intel PC chips. If the Intel CPU land should go fallow, then Microsoft is in store for a profitability famine.

Copyright 1996, Francis Vale All Rights Reserved, 21st

21st, The VXM Network,