Path: utzoo!mnetor!uunet!lll-winken!lll-lcc!ames!pasteur!trinity!max
From: max@trinity.uucp (Max Hauser)
Newsgroups: sci.electronics
Subject: The notorious 8038 VCO (was Re: VCO waveform generators)
Message-ID: <1116@pasteur.Berkeley.Edu>
Date: 1 Mar 88 01:30:03 GMT
References: <18403@topaz.rutgers.edu> <2547@genrad.UUCP>
Sender: news@pasteur.Berkeley.Edu
Reply-To: max@eros.UUCP (Max Hauser)
Organization: UC Berkeley, IC design
Lines: 50
Summary: A bas les 8038s!

Now that someone, with the best of intentions, has brought up the 8038
VCO chip, I'd like to offer some negative comments about it based on my
experience. I deliberately omitted the 8038 from my earlier response
recommending other chips.

The 8038 is somewhat notorious among people who've dealt with a
lot of monolithic oscillator circuits. For one thing it runs hot,
and in the normal mode of operation, on-chip elements factor in
the oscillation-frequency equation, and those elements have a 
substantial TC. Therefore the oscillation frequency tends to drift.
Try powering it on with an audible oscillation frequency and you can
hear it drift noticeably.

The 8038 was about the first "complete" function-generator chip when
it came to market circa 1971. Competition at Exar subsequently
produced the 205, 2206 and 2208 devices, with the 2206, like the 
Intersil 8038, yielding sine waves as well as triangles. I have used 
all of these devices and in my experience the 8038 required more 
components, a second objection besides the instability.

A third objection concerns how it shapes the sinusoids, a subject I
know a little about. Modern monolithic triangle-sine shapers work, 
generally, by taking an almost-right natural nonlinearity and playing
with it to obtain a DC distortion curve shaped like part of a
sinusoid (abundant references available on request). This has been
done both in bipolar and in MOS technology, as well as earlier in
discrete components, and if I recall it is done with differential
transistor pairs in the Exar 2206. The Intersil 8038 however uses the
competing approach of piecewise-nonlinear approximation for the 
sinusoid shaping circuit, a venerable method from analog-computer
technology and very handy for arbitrary waveshapes (although in the case
of sinusoids, more complicated and critical than alternatives). Piecewise-
linear shaping approximates a sine function with a series of linear
slopes that change at breakpoints. It yields cuspate (nondifferentiable)
errors in the transfer curve even if done very well; however the 8038
employs lateral-PNP transistors, if I recall, for the breakpoint
network and therefore has a low speed limit. Distortion of the sinusoids
was visible on a scope -- not to mention audible -- in the kilohertz.

So if you actually need sinusoids from the VCO, let me entreat you to
consider alternatives to the 8038. There are also some newer chips put
out with electronic music service in mind, but I have not used them.

Epilog: after developing a bad taste for the 8038 in the early 1970s,
I saw later that Jim Roberge's graduate analog-hacking course at MIT
gave students 8038 VCOs and asked them, as the first lab exercise, to,
as someone put it, "find six lies in the spec sheet of this chip."
Marvelous!  (Undergrad engineering courses often study how to design 
things, while graduate ones often study how not to.)

Max Hauser / max@eros.berkeley.edu / ...{!decvax}!ucbvax!eros!max