🚀 go-pugleaf

[Hope no one gets this twice -- original posting vanished into the aether]

This was drafted 12/87 in response to a posting of Bill Mayhew's, and it 
became a mini-treatise, including a dogmatic-audiophile anecdote. Having 
stumbled on it again I belatedly post. Topics: merits of off-chip op amps;
stereo separation issue (and anecdote); frequency-dependent nonlinearity.

1. Merits of off-chip op amps for audio circuits.

From: wtm@neoucom.UUCP (Bill Mayhew)
Subject: Re: NAK decks (degraded sound)
Date: 5 Dec 87 17:06:35 GMT

> ...One of the better things about the [Teac] C3RX is that its Dbx 
> encoder / decoder is built from "discrete" ICs using TL072s. 
> (Not exactly a terrible op-amp.)  The tracking using a 
> non-monolithic Dbx circuit is much better.

Background first. The (Texas Instruments) TL07X series is
representative of the "mixed-process" monolithic op amps that emerged
in the mid-1970s as low-price, relatively high-performance successors
to all-bipolar chips like the (Fairchild) 741 and (National) LM 101.

Texas Instruments (TL06X/TL07X/TL08X) and National Semiconductor
(LFX55/LFX56/LFX57) used modified bipolar fabrication processes that
allowed good-quality junction FETs also, while RCA (CA31X0 series) used
an MOS/bipolar process. All aimed to deploy field-effect transistors 
where advantageous in the op amps, especially in input stages.

Now for those interested, dbx gain-control subsystems contain as an
essential component a bipolar-transistor variable-gain circuit. This 
particular circuit is the subject of the original Blackmer patent that 
underlies this whole product line. (Dave Blackmer is the db in dbx, or 
so I was told by one of his former associates. I wrote a thesis on 
variable-gain circuits and Blackmer's was one of many that I looked at 
in some detail.)

This particular gain-control circuit uses one subcircuit for the
positive excursion of the audio signal and one for the negative. 
A FAST op amp is needed in one particular spot (for each of the 
stereo channels) to guarantee that no glitch occurs on the 
transition as the signal crosses zero.  Another requirement on op 
amps in dbx circuitry is low DC error (offset) in the gain-control 
path, for gain tracking in companding modes. 

Although I'm not sure exactly what Bill had in mind, from these
considerations it does not follow that an outboard JFET/bipolar 
op amp like the TL072 is necessarily better than on-chip (presumably 
all-bipolar) op amps in the critical dbx core circuits. In practice 
it is much easier to get good performance if you design the op amp to 
fit the application --as a designer would do with an on-chip op amp --
rather than use an off-the-shelf one like TL072, good though it 
assuredly is in many applications. 

2. The stereo separation issue

I know that some audiophiles rail about how one cannot possibly get
good stereo separation when both channels of circuitry share a common
monolithic substrate. Before addressing this let me give an anecdote
from my experience.  A guy lectured me about stereo separation
in a record store in Berkeley a couple of years ago and I'll never forget
it.  It was the kind of commentary that gives some audiophiles such a 
bad name: pseudotechnical, hip, and full of jargon. I bet it impresses
the hell out of all of their friends. If these folks were content to 
be honest, and simply say that they *hear* less separation than they 
would like from a certain product (if indeed that's even true), they
would have my respect. But when instead they feel bound to explain it 
in terms they are really not facile with, or (worse, the disciple 
syndrome -- this is how Fascism takes root) seizing on some fashionable
and speculative explanation out of one of the (proudly) 
limited-circulation magazines that serve as mouthpieces not only for 
the insightful but also for the fatuous (simply being unpublishable
in the major media is not always a sign of genius, I feel bound to 
remind), they are going too far.

But back to my audiophile. Rather than tell me his impressions or 
experience, this guy was lecturing me that two-channel chips *never* 
have significant stereo separation since the single chip substrate is 
incapable of good electrical isolation between audio paths. Note the 
purely technical character of this assertion. Also, never say never. 
Actual electrical isolation depends on a gamut of things including
basic physics, circuit, and chip layout, and designers have a lot of
control over it. In contemporary work on oversampling data converters 
we are being surprised at just how much on-chip isolation we can obtain
with straightforward measures (it's even more critical there since we
put large analog and *digital* systems on the same die). I have measured
audio analog isolations of 110-120 dB, the limit of my HP spectrum 
analyzer when augmented and coddled, between amplifiers on a silicon 
substrate, with careful layout.  But my deeper point is that my 
loquacious audiophile might even have been right and he STILL wouldn't
have known it (although he assuredly would have "known" it), since he 
was talking about something he really had little clue about.

[This recalls an even more than usually inane "analog-versus-digital"
claim cited on the net months ago. It seems that no less an authority
than _The New Republic_ had quoted "several audiophiles with $100,000
systems" as declaring solemnly that digital audio used discrete-time
samples and was therefore "fundamentally incapable" of capturing
continuous-waveform audio. One wants, as Paul Fussell wrote, to weep.
Accordingly, I fully expect grave doubts to be expressed in similarly
informed quarters against equipment containing two-channel chips
(even digital ones!) because "it is fundamentally impossible for a
single integrated-circuit substrate to keep separate etc. etc."]

To finish the anecdote, I indicated polite skepticism about this stereo 
separation impossibililty, and he continued about a product he had
bought where all the chips were ripped out and retrofitted with 
discrete components (taking the premise to its logical conclusion,
I suppose). It was only when he started boasting of his new CD player 
with its "256:1 oversampling!" that I excused myself.

I was with musical friends at the time. One of them said to me, "why do 
audiophiles get so serious about mere *sound*? I listen for the 
performance," a good point that I hear a lot from musicians. (I could 
answer, but it would be cynical, uncharitable, and maybe wrong besides.)

3. Second-order electrical behavior

Engineers learn about things like gain, slew rate (the rate at which
the output voltage can change; if you try to exceed it you can suffer,
in audio, "transient intermodulation distortion"), frequency response,
stability, and (if they go to a progressive school) noise.

Op amps, being designed for negative-feedback applications, depend on 
lots of internal gain to make their closed-loop circuits behave the way
the designers intended.  Yet op amps are at heart DC devices, and above 
DC their gain quickly begins rolling off with frequency. It rolls off 
(in off-the-shelf unity-gain-stable op amps) normally in a single-pole 
response to unity at some frequency (typically 1 MHz for bipolar-input 
and 5 MHz for FET-input op amps).  This means that at 20 kHz, the 
open-loop gain is respectively 50 or 250, not all that big. This 
open-loop gain is the negative-feedback correction factor that 
suppresses several nasties -- in particular, distortion in the op amp 
input stage. As soon as the AC voltage excursion across the differential
inputs (which would always be negligible were the open-loop gain large)
exceeds some number of millivolts, appreciable nonlinearity develops in 
the op amp's voltage-transfer curve and harmonic and IM distortion 
result (steadily, not just in transients).

A former co-worker of mine at Signetics Analog, the venerable
wideband-amplifier expert Ralph Lovelace (who also grows flavorful 
tomatoes), designed a "video" op amp with extremely high bandwidth,
close to physical limits (can't find the data right now or I'd
quote the part number). Not as fast but still enhanced in bandwidth,
and tailored in other respects for audio (low input noise -- less
than 0.6 uV in 20 kHz from low-impedance sources -- and high output
current), is the Signetics 5534 family.  Walter Jung (_Audio IC Op-Amp 
Applications_, Howard Sams paperback) sings praises of the 5534 family
(only one reason, I'm sure, why the Signetics marketing folks sent free
copies of the book to anyone they could). Other op amps have emerged
more recently with the stringent audio applications specifically in mind.

The upshot of this is that common off-the-shelf bipolar-input op amps 
like the 741 are virtually useless for handling large-signal audio
(a few years ago I calculated just the *harmonic* distortion at audio 
frequencies with 741-class op amps (including LM 101 series; LM 124 
series; RC 4136; etc.) and it was horrific. It can be expressed solely
in terms of closed-loop output voltage for a given op amp, as I recall,
even when frequency compensation is relaxed to enhance bandwidth.)

Common FET-input op amps are better, since both their bandwidth tends to
be greater and their linear range for differential input voltages is
much larger than for normal bipolar-input op amps. It is not even 
unreasonable to use extra-wideband (or "video") op amps, especially if 
you are trying to reject out-of-band signals at even higher frequencies 
that are mixed in with the audio, a situation that occurs routinely in 
the active-filter stages of all CD and DAT players. As far as this
frequency-dependent nonlinearity is concerned, what is directly critical 
in op amps is not the "unity-gain" bandwidth (at which open-loop gain 
drops to 1); nor the "power bandwidth" at which slew limiting is reached
(although that is also an upper bound, it is rarely the main problem); 
nor the open-loop gain at DC; but rather the gain at the highest input 
frequency (whether audio or interference); the resulting "error voltage"
across the op-amp's differential inputs; and the nonlinear distortion 
referred-to-input that this yields.

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