THE MAGAZINE FOR COMPUTER APPLICATIONS
Circuit Cellar Online offers articles illustrating creative solutions
and unique applications through complete projects, practical
tutorials, and useful design techniques.

TWENTY YEARS AGO, TODAY

by Tom Cantrell

Start ı Big Bang DSP ı Compiler is King? ı Once More, With Feeling ı Sources and PDF

Though many are familiar with the story of the microprocessor, few may recall the early days of DSP.[1] Itıs illuminating (and timely, being roughly the twentieth anniversary) to take a look back. Frankly, itıs also amusing.

For instance, Intel is known for being first in micros, and we know they were responsible for a number of other important inventions, such as DRAM, EPROM, and flash memory. However, they also came up with a number of other firsts that turned out to be lasts, including LCD watches, bubble memory, and, letıs not forget, the DSP.

What? You never designed in the venerable Intel 2920 (see Figure 1) introduced in 1979? This puppy was by definition a signal processor (actually, a digital analog computer), seeing as the inputs and outputs were analog!

Figure 1ıIntroduced in 1979, the Intel 2920 was one of the first DSP chips. With only a 192-instruction EPROM and 40-word data RAM, it was perhaps a case of "too little, too soon," and acceptance was lukewarm at best.

Actually, you can dimly discern the underpinnings of modern DSPs in the 2920 if you look close enough. For instance, it had a "very-short, long-instruction-word" architecture with each 24-bit instruction comprised of separate parallel operations for ALU, barrel shifter (left shift 2 to right shift 13), and analog conversion (successive approximation done in software). The 2920 also featured conditional execution (loads, adds, and subs), which is now all the rage.

The good news is the 2920 was a single-chip DSP, with all program and data memory on-board. The bad news is there wasnıt much of itı192 ı 24-bit EPROM program memory and 40 ı 25-bit RAM.

With a maximum of 192 instructions, the 2920 didnıt even have a branch instruction per se. Instead, you would just string out your code (inlining they call it now) and end it with the EOP instruction, which would reset the program counter to zero. There werenıt any inner loops in 2920 apps, because there werenıt any outer loopsıjust "the loop."

Lack of a multiplier was a rather key omission, so if you sprung for the most expensive 2.5-MIP version of the part, you might be able to handle up to 10 kHz. Need more bandwidth? Make your program shorter. Needless to say, there was no C compiler. Even those used to dealing with the baroque programming called for by other Intel chips did plenty of head scratching over lines of code like ADD DAR, KM2, R0, CND6. Guess it was a good thing you only had to keep 192 of them straight.

Alhough the 2920 never sold worth beans, it was arguably the first commercial DSP. However, there were other contenders, including the then semi-major player AMI (until it was laid low by the mid-80s video-game blowout). They announced their S2811 signal-processing peripheral in 1978 but had process problems that ultimately stopped the show. Meanwhile, NEC announced their own UPD7720 in 1980 but, no doubt preoccupied with DRAMs and such, failed to get the tools together or mount an effective promotion effort.

In terms of actual delivery, the first place ribbon should probably go to Bell Labs who disclosed their DSP1 at the same time as NEC. Itıs not completely surprising that they came up with a good chip, considering their legacy of names like Nyquist, Bode, and Shannon on the payroll. Talk about interesting chat around the water cooler! Of course, it also helps to get designed in when headquarters (AT&T) happens to own the phone business. The DSP1 quickly found its way into a variety of telecommunication equipment apps and arguably sparked the first wave of digitization of the network.

Circuit Cellar provides up-to-date information for engineers. Visit www.circuitcellar.com for more information and additional articles.
For subscription information, call (860) 875-2199, subscribe@circuitcellar.com or subscribe online. ıCircuit Cellar, the Magazine for Computer Applications. Posted with permission.