If you follow the weekly rags, you know this year's buzzword is "multimedia." Everyone with any bandwidth at all has jumped on the full-motion, interactive video bandwagon in hopes of combining text, graphics, video, animation, and sound for applications ranging from education to presentations. I don't know about you, but the year is still young and I'm already suffering from motion sickness.
For the hyperbole to become reality, however, a few key pieces have to be in place, foremost among them data compression. At first, you'd think the hardware that drives multimedia systems -- CD-ROMs, graphics coprocessors, and digital audio subsystems -- would have enough storage and computing horsepower without compression. Not so. Multimedia applications absolutely, positively require fast, efficient compression techniques that provide high compression ratios.
Nor does it matter if the host machine is 80486-, 68040-, RISC-, ASIC-, or transputer-based -- multimedia needs data compression. (I'll withhold for the moment comments about the recent spate of 286-based multimedia announcements.) Consequently, there are a number of players getting into the data compression game with a variety of high-end hardware/software combinations. Intel picked up RCA's DVI (digital-video interactive) a while back, christening the chipset the "i750 video processor." Kodak is pushing its Colorsqueeze image compression software for the Macintosh, EFI its ECOMP program, Next has built compression into its machine using a C Cube CL550-based card, and long-time graphics researcher Michael Barnsley has a fractal image compression system built around custom ASICs and Intel's i960 RISC chip. Standards have evolved, the two most important being the JPEG (Joint Photographics Expert Group) algorithm and the MPEG algorithm from the Motion Picture Experts Group.
So what does compression bring to the party? Consider that while it takes less than 500 bytes of data/second to produce telephone-quality sound, it takes 150,000 bytes/second to produce audio from CD players and more than 22 Mbytes/second for broadcast-quality video! Without data compression, how much music or video frames can you store on your 30-Mbyte hard disk?
Data compression delivers a solution. With DVI, Intel is claiming compression ratios of up to 160:1, enabling up to 72 minutes of full-motion video on a standard CD-ROM. To accomplish this, DVI researchers have come up with a compression algorithm called PLV (short for "Production Level Video") that has i750 hardware assists built into the algorithm. (A document describing the algorithm's data structure is available from the Intel Literature Center; ask for #B4P-05.)
Barnsley's approach is equally impressive. He claims to compress a 190-Kbyte image (with 320 x 200 pixels and 24 bits, or more than 16 million colors/pixel) into a 6-Kbyte file. It then takes less than a second to read, decompress, and display that file as a full-color image on a 25-MHz 80386 PC. Barnsley discussed his fractal image compression at a recent talk sponsored by the Parallel Processing Connection (a Silicon Valley SIG that says it is the "Homebrew Club of the 90s") and demonstrated decompression at 30 frames/second on a standard VGA monitor.
Putting 286 systems aside for the time being, getting into high-end image manipulation and compression isn't cheap. Barnsley's Iterated Systems, for example, offers an $18,000 developer's kit (which includes the i960-based AT-bus compression board, software, training, etc.) that lets developers create compressed applications for distribution to lower-performance, lower-priced multimedia systems. The assumption is that decompression doesn't require the horsepower of compression and that users don't need to store images, just view them.
I guess this is where 80286-based multimedia systems -- which will have 2 Mbytes of RAM, VGA, digital audio sound, 30-Mbyte hard disk, CD-ROM drive, and run Microsoft Windows -- enter the picture as electronic Viewmasters. Still, it seems to me, the 286 is the wrong platform. My guess is that vendors like Tandy, AT&T, CompuAdd, NEC, Zenith, and others chose the 286 not because it offers the best performance or technological approach, but because it is cheap and available from multiple sources. The 386SX (which would make a better multimedia platform even though there's no technological reason for the chip to exist) is faster than the 286 and only $40 to $50 more expensive (around $20 for the 80286 vs. $60 for the 386SX). Granted, education is an extremely price-sensitive market where 286 PCs are already in place and where few schools can or will upgrade just for multimedia. Let's face it, for multimedia the Amiga makes more sense -- dollar-wise and performance-wise -- but then, it has never made it into the broad-based market.
This leads to questions concerning multimedia's chance for success. Can it succeed, or will it, as some have suggested, become the AI of the 90s? The decision by multiple vendors to stake their multimedia future on 80286 PCs probably won't kill the concept or emerging market, but it won't speed its growth either. That's not to suggest multimedia system vendors aren't serious about the market -- after all, they cut down a whole forest of trees for the press releases alone -- but if multimedia has a chance at success, it will be because of the efficiency of the underlying software, particularly compression techniques, and not the hardware that vendors are trying to force on users.
Copyright © 1991, Dr. Dobb's Journal