[sdiy] FM Synth Designs

Don Tillman don at till.com
Wed Aug 14 02:37:40 CEST 2002

Yes, the DX-7 was built with 1983 technology, and in 1983 "real time
multiplies" were prohibitively expensive.  

By "real time multiply" I'm referring to a multiply operation that
needs to happen some number of times per audio sample.  That means it
would have to be performed in under 10 uSecs, preferably under 1 uSec.
Faster than that would be even better.  Microprocessors of the day
weren't up to the task since they used microcoded shift-and-add
multplies which took quite a while, plus the other operations they
needed to do.  A 16x16 multiply chip was available from TRW in 1983,
but it cost around $300.00.

The cool thing about the DX-7 is that it's a fully digital synthesizer
with no real time multiplies.  In fact, it does everything with adds
and ROM lookups.  This was the only way they were able to get it out
for a reasonable price.  Very, very clever.

According to the Yamaha patent the VCA's were built by storing the log
of the sine wave in ROM, adding the output level, and performing a ROM
lookup exponentiation.

Note that the original Chowning patent misses this point entirely.  He
goes to the trouble of building a processor that performs shift-and-
add multiplies, and takes about 80 microcycles to do it for one
operator.  The DX-7 implementation is so efficient it can do 96
operators total (16 voices, 6 operators each) in real time.

My wild speculation is that Yamaha didn't need to pay Stanford
anything for the Chowning patent as their implementation was so
different, but they did it anyway to get at all the later CCRMA work.

  -- Don

Don Tillman
Palo Alto, California, USA
don at till.com

More information about the Synth-diy mailing list