Frantically Seeking VCA

[Norm Dresner]

Several (but not too many) months ago I saw a schematic for a voltage
controlled amplifier (VCA) circuit that claimed to have very low distortion
in the audio range (~.01% or less). It's my vague recollection that it was
in a magazine but it may have been on a web-site too. It's also my very
vague recollection that it might have been by Bob Pease of NS.

All pointers, hints, suggestions, recollections, URLS, etc graciously
accepted.

TIA
Norm

 

[Joerg]

Hi Norm,

Sorry, don't know about this one but Bob Pease responds to email.

There may be one interesting device to look at: Analog Devices AD603.
It's for RF but might be useful at audio. This chips does 40dB range
which can be controlled in a dB-linear fashion via an input voltage. The
downside for audio is that you probably need a follower up front as it
has a 100 ohms input impedance.

I designed lots of amps with this chip and it always did what I wanted
it to.

Regards, Joerg

 

[Max Hauser]

Several (but not too many) months ago I saw a schematic
for a voltage controlled amplifier (VCA) circuit that claimed
to have very low distortion in the audio range (~.01% or less).

There are not really very many ways, fundamentally, to do it at that level
of linearity.

If the actual gains you want are switched rather than fully continuous, that
helps a lot.

Linear Technology has a continuous "video fader" product and some digitally
set CMOS PGAs of good linearity.

 

[Mike Diack]

Several (but not too many) months ago I saw a schematic for a voltage
controlled amplifier (VCA) circuit that claimed to have very low distortion
in the audio range (~.01% or less). It's my vague recollection that it was
in a magazine but it may have been on a web-site too. It's also my very
vague recollection that it might have been by Bob Pease of NS.

All pointers, hints, suggestions, recollections, URLS, etc graciously
accepted.

TIA
Norm

www.thatcorp.com

 

[Ben Bradley]

Several (but not too many) months ago I saw a schematic for a voltage
controlled amplifier (VCA) circuit that claimed to have very low distortion
in the audio range (~.01% or less).

Well, it certainly wasn't a CA3080 or CA3280 based circuit, was it?


Was it basically some chip, or did it involve discrete components
(other than the usual resistors and caps, and maybe an opamp on the
input and output?)

It's my vague recollection that it was
in a magazine but it may have been on a web-site too. It's also my very
vague recollection that it might have been by Bob Pease of NS.

That's all of course "pretty vague" and I haven't seen anything
like that lately...

All pointers, hints, suggestions, recollections, URLS, etc graciously
accepted.

There's the chips by THAT corp, http://thatcorp.com/ which seems t
make some high-end products in this area, and also Analog Devices
(http://analog.com) still makes some VCA's they got with the SSM line.
I think at least some of these should have specs like that.

 

[Norm Dresner]

Well, it certainly wasn't a CA3080 or CA3280 based circuit, was it?


Was it basically some chip, or did it involve discrete components
(other than the usual resistors and caps, and maybe an opamp on the
input and output?)


That's all of course "pretty vague" and I haven't seen anything
like that lately...


There's the chips by THAT corp, http://thatcorp.com/ which seems t
make some high-end products in this area, and also Analog Devices
(http://analog.com) still makes some VCA's they got with the SSM line.
I think at least some of these should have specs like that.

It's my recollection that it involved an Op Amp or two and a bunch of
discrete transistors and the usual passive stuff.

 

[John S. Dyson]

Well, it certainly wasn't a CA3080 or CA3280 based circuit, was it?


Was it basically some chip, or did it involve discrete components
(other than the usual resistors and caps, and maybe an opamp on the
input and output?)

For the 0.01% or less distortion level, the variable transconductance
scheme with BJTs isn't going to be very successful. The 'thatcorp'
technology (or the approx equvalent SSM stuff from AD) can come close
to 0.01%, perhaps exceeding it. (It is probably possible
to cancel the distortion for the simple variable gm schemes, but it would
be difficult to cancel the distortion as well as a thatcorp device
might achieve.) The Analog Devices SSM2018 might/might-not be
readily available... The Thatcorp THAT2180 is good for audio
applications, but might not meet your 0.01% in real world
conditions, even though it's specs are pretty good. (Most all of the
variable gain BJT devices utilize the log characteristics of the
device, but there is a difference in peformance based upon circuit
topology.)

Likely, the most successful scheme for incredibly low distortion
over a very wide bandwidth would be a switched resistor thing (similar
to a D/A), and Analog Devices has a device targeted to IF AGC applications,
but I don't know if it will work well for audio (with low 1/F noise.) There
might be some problems in audio if the AGC device doesn't have fine
enough gain changes. (AD8334 is an example of the 'switched resistor'
scheme -- I don't know the technical name for it.)

My guess is that the most likely successful device for audio would
be a selected (high quality) thatcorp component. In medium production
quantities, it is about $5.00, and a discrete implementation with
such high performance would be 'challenging' . Even though Analog
Devices also has similar component families, their availability has
been questionable.

John

 

[John S. Dyson]

It's my recollection that it involved an Op Amp or two and a bunch of
discrete transistors and the usual passive stuff.

Given a few OpAmps are used in conjunction with 'discrete' BJTs,
you'd likely need to use well matched pairs (e.g. Analog Devices
MAT01, MAT03, MAT04 or ThatCorp matched transistors), and it would
take some serious design, debugging and verification to replicate
the performance of the Thatcorp VGA devices.

If possible, it is higher quality to do the signal processing with
a DSP or other kind of CPU (assuming that you have the signal in
a digitized form, and you aren't designing a gain control element
for guaranteeing signal levels before A/D conversion.) Of course,
if you are designing a last chance peak limiter before an A/D conversion
(trading some peak limiting AGC aciton for the more obnoxious clipping),
then there is little that can be done in software.

In my experiments, I have found that the flexibility of doing audio
signal processing in software can help to better maintain quality,
to provide more degrees of freedom, and provide more precise performance...

John

 

[Paul Burke]

Given a few OpAmps are used in conjunction with 'discrete' BJTs,
you'd likely need to use well matched pairs (e.g. Analog Devices
MAT01, MAT03, MAT04 or ThatCorp matched transistors), and it would
take some serious design, debugging and verification to replicate
the performance of the Thatcorp VGA devices.

The old PMI apps book had a design based on the MAT-04 and OP27. I used
it years ago, and it worked very well. Maybe AD has the design lying
around somewhere.

Paul Burke

 

[Norm Dresner]

For the 0.01% or less distortion level, the variable transconductance
scheme with BJTs isn't going to be very successful. The 'thatcorp'
technology (or the approx equvalent SSM stuff from AD) can come close
to 0.01%, perhaps exceeding it. (It is probably possible
to cancel the distortion for the simple variable gm schemes, but it would
be difficult to cancel the distortion as well as a thatcorp device
might achieve.) The Analog Devices SSM2018 might/might-not be
readily available... The Thatcorp THAT2180 is good for audio
applications, but might not meet your 0.01% in real world
conditions, even though it's specs are pretty good. (Most all of the
variable gain BJT devices utilize the log characteristics of the
device, but there is a difference in peformance based upon circuit
topology.)

I've already e-mailed THAT asking about a small-quantity buy since their
2180 looks quite promising.

Norm

 

[Norm Dresner]

The old PMI apps book had a design based on the MAT-04 and OP27. I used
it years ago, and it worked very well. Maybe AD has the design lying
around somewhere.

Now that's going to take some serious digging to find the book but it should
be interesting.

 

[Norm Dresner]

The old PMI apps book had a design based on the MAT-04 and OP27. I used
it years ago, and it worked very well. Maybe AD has the design lying
around somewhere.

Hey, I found it. In the 1990 PMI Analog Databook, p 9-41. It _only_ uses a
MAT-04, an OP-27, and a pair of OP-41's. Oh, yeah, and 1% metal film
resistors and a 100 uF non-polarized tantalum cap. The writeup says it
shoudl have "a distortion level of 0.03% over a wide range of control
voltages."

The THAT2180 is looking quite good.

Thanks

Norm

 

[Norm Dresner]

Hey, I found it. In the 1990 PMI Analog Databook, p 9-41. It _only_ uses a
MAT-04, an OP-27, and a pair of OP-41's. Oh, yeah, and 1% metal film
resistors and a 100 uF non-polarized tantalum cap. The writeup says it
shoudl have "a distortion level of 0.03% over a wide range of control
voltages."

Just another thought: The MAT-04 is four NPN transistors in a single
package. Does anyone know about the matching for an MPQ2222? It's got to
be cheaper and more readily available. Not that I'd seriously do this
much work to replace a ~$5-10 IC, but there's the curiosity in my
engineering side that's pushing thru.

Norm

 

[Walter Harley]

I've already e-mailed THAT asking about a small-quantity buy since their
2180 looks quite promising.

I talked with them recently; their minimum buy is one tube, which is either
25 DIPs or around 100 (if I remember correctly, which I probably don't)
SMD's.

They do say that "there may be some unauthorized folks selling smaller
quantities." I have been considering ordering a tube of these myself, for a
project I'm cooking up, so perhaps I might become one of those unauthorized
folks if you need

However, I was thinking of ordering the B grade, which is .02%, not the more
expensive .01% grade. And I'd be getting DIPs, to prototype with. If you
need A-grade SMD's, plan on spending a chunk of money...

 

[Norm Dresner]

I talked with them recently; their minimum buy is one tube, which is either
25 DIPs or around 100 (if I remember correctly, which I probably don't)
SMD's.

They do say that "there may be some unauthorized folks selling smaller
quantities." I have been considering ordering a tube of these myself, for a
project I'm cooking up, so perhaps I might become one of those unauthorized
folks if you need

However, I was thinking of ordering the B grade, which is .02%, not the more
expensive .01% grade. And I'd be getting DIPs, to prototype with. If you
need A-grade SMD's, plan on spending a chunk of money...

I got pretty much the same e-mail from them explaining that 1 tube of 25 was
the minimum. But it appears that they're available in 2's and 5's from
http://www.profusionplc.com/cgi-bin/gex/pcatdtl?ipartno=THAT2180LB

They're in the UK. Since my son lives there [but is here visiting right
now] I'll probably give him a few bucks to buy some parts and ship them back
to me. The A-version is seeing for 10.6GBP in 2's and 8.5GBP in 5's while
the B-version is 6.0GBP for 2's and 4.9GBP for 5's.

Norm

 

[John Dyson]

uses

Just another thought: The MAT-04 is four NPN transistors in a single
package. Does anyone know about the matching for an MPQ2222? It's got to
be cheaper and more readily available. Not that I'd seriously do this
much work to replace a ~$5-10 IC, but there's the curiosity in my
engineering side that's pushing thru.

I don't think that you'll see the desired matching from devices like the
MPQ2222. There is no specficiation for matching, and the component
might even be built from multiple chips. Even if it is built from a single
chip, the matching could be +-25mv or worse, which is of little use for
you.

This 'Gain Control' amplifier thing is ugly, no matter how it is done. It
is fortunate that matching and component characteristics nowadays can
be impressive.

When looking at old tube designs (or even old FET designs), they tend to
be quite ugly, and amazing that they had performed as well as they did.
Early on, I had even seen the use of the dynamic resistance of diodes (and
using ad-hoc techinques to remove the thump.)

If absolute precision isn't needed, and super fast attack/release times
aren't needed, and some drift is allowed, then the Lamp(LED) along
with resistive photocell scheme can be quite nice.

The technology today can be very repeatable and very precise. Such
approaches (like the That stuff, or some of the SSM stuff from AD) do
as well as someone can build.

Except for input AGC (before the 18-20bit A/D converter), I prefer
using DSP techniques for gain control. For before the A/D converter, then
all I'd tend to use would be a peak limiter which would avoid harsh
distortion
on peaks. With the large bit depth, then the recording can be done with
plenty of headroom.

Nowadays, my approach would be (in order of preference, but also with
prejudice): (For audio)

1) DSP!!!
2) Gilbert or That gain control element.
2A) Switched resistor scheme??? ( I don't know how well it might work.)
3) Photo resistor
4) FET
5) Dynamic resistance or simple variable gm method.

The photo resistor can be done with very low distortion, but might
have more limited lifetime and more drift. The FET is mostly just
useful for peak limiting, but matching across channels across a wide
gain control range isn't very good.

A WORST approach is to use the dynamic reisstance of a diode or simple
gm change for a transistor. The linearity will tend to suck badly (without
some compensation.) By the time that one would apply excellent distortion
compensation, then the Gilbert or That gain control cell would be best.

John

 

[Ian Buckner]

I don't think that you'll see the desired matching from devices like the
MPQ2222. There is no specficiation for matching, and the component
might even be built from multiple chips. Even if it is built from a single
chip, the matching could be +-25mv or worse, which is of little use for
you.

This 'Gain Control' amplifier thing is ugly, no matter how it is done. It
is fortunate that matching and component characteristics nowadays can
be impressive.

When looking at old tube designs (or even old FET designs), they tend to
be quite ugly, and amazing that they had performed as well as they did.
Early on, I had even seen the use of the dynamic resistance of diodes (and
using ad-hoc techinques to remove the thump.)

If absolute precision isn't needed, and super fast attack/release times
aren't needed, and some drift is allowed, then the Lamp(LED) along
with resistive photocell scheme can be quite nice.

The technology today can be very repeatable and very precise. Such
approaches (like the That stuff, or some of the SSM stuff from AD) do
as well as someone can build.

snip<

The photo resistor can be done with very low distortion, but might
have more limited lifetime and more drift. The FET is mostly just
useful for peak limiting, but matching across channels across a wide
gain control range isn't very good.

A WORST approach is to use the dynamic reisstance of a diode or simple
gm change for a transistor. The linearity will tend to suck badly (without
some compensation.) By the time that one would apply excellent distortion
compensation, then the Gilbert or That gain control cell would be best.

John

I've seen an AGC loop/variable gain done with an LED illuminating a
matched pair of photoresistors. One of them was connected in a DC
(no signal) loop, used to set the LED current to force that resistance
to the required value. The match between the resistors provides a
reasonably well controlled and very linear gain change in the signal
path. Obviously, this only works well over the range of resistance
where the pair are well matched. You also have to be wary of the
response time variations of LDRs at different illumination levels.

Regards
Ian

 

[Walter Harley]

I got pretty much the same e-mail from them explaining that 1 tube of 25 was
the minimum. But it appears that they're available in 2's and 5's from
http://www.profusionplc.com/cgi-bin/gex/pcatdtl?ipartno=THAT2180LB

They're in the UK. Since my son lives there [but is here visiting right
now] I'll probably give him a few bucks to buy some parts and ship them back
to me. The A-version is seeing for 10.6GBP in 2's and 8.5GBP in 5's while
the B-version is 6.0GBP for 2's and 4.9GBP for 5's.

That's right, I'd forgotten about ProfusionPLC.

No need to enlist your son; Profusion will ship directly to the US, and they
are responsive to web and email orders. ISTR that the last time I ordered
from them (some power audio MOSFETs) there was actually NO shipping cost (!)
and the goods arrived faster than they generally do within the US. I was a
very satisfied customer.

 

[ddwyer]

I've seen an AGC loop/variable gain done with an LED illuminating a
matched pair of photoresistors. One of them was connected in a DC
(no signal) loop, used to set the LED current to force that resistance
to the required value. The match between the resistors provides a
reasonably well controlled and very linear gain change in the signal
path. Obviously, this only works well over the range of resistance
where the pair are well matched. You also have to be wary of the
response time variations of LDRs at different illumination levels.

Regards
Ian

There is a led driven photoresistor inthe Farnell catalog.
The variable gain amp can be considered as a multiplier with an
amplifier, the LTC part is expensive, an op amp + led/photoresistor is
low cost.
Precision is not required low distortion is usually the critical
parameter.
Separate the level detection fronm the variable gain element solves the
proble for many applications.

 

[John S. Dyson]

There is a led driven photoresistor inthe Farnell catalog.
The variable gain amp can be considered as a multiplier with an
amplifier, the LTC part is expensive, an op amp + led/photoresistor is
low cost.
Precision is not required low distortion is usually the critical
parameter.
Separate the level detection fronm the variable gain element solves the
proble for many applications.

It is true that in many applications, 'precision' isn't incredibly
important, and distortion does tend to be much more important, however
there are some gain control applications where sub-dB tracking between
channels is important. Tracking tends to be easier over narrow gain
control ranges (e.g. 10-20dB) vs. very wide ranges that might have been
useful for DBX noise reduction type applications. The Gilbert/That
type approaches allow for excellent tracking.

When using gain control elements that have 'excellent' tracking, then
there might be more topologies for practical and useful designs. For
example, (using a FET gain control), the feedback topologies are easier
to deal with than the feedforward designs. For the very repeatable
Gilbert or 'That' designs, the feedfoward topologies become less
troublesome. (Feedforward topologies don't provide the self-correcting
abilities.)

I still strongly suggest using DSP techniques if possible. There
are numerous opportunities for mitigating the effects of gain modulation
and much easier implementation of dynamic attack/decay times.

One of my experimental methods uses a running FFT with an appropriate
windowing function. (This is very easy to do, but there are some design
considerations that aren't initially obvious, or there will be alot
of glitches.) The FFT method can do some amazing gain control, and a
small amount of compression (or expansion) can result in near optimal
results. Using the FFT approach is like having a 32 to 1024 band
compressor (or expander.)

John