Voltage-controlled amp design is hard

audio AGC circuits use a JFET as variable shunt resistor. there are a few tricks to keep the distortion very low.

do you care about signal distortion. variable gain matching between channels speed of gain control change

Mark

nobody interested in the fact you can get a real "ultra-low noise" transconductance VCA with 20MHz bandwidth and 120dB dynamic range for a BUCK?! :O

Yup, the old voltage-divider-to-the-drain trick. With some devices the optimum ratio is near 0.5, but with the 2N7002s I measured awhile back, the ideal ratio was near 1.8-2.0, achieved by running two FETs in series with their gates in parallel, then taking 10k to the upper drain and

100k to the control voltage.

Problem is the Vgs(th) variation is typically 3:1, and the resistance gets less linear at the extremes, so each unit would require individual calibration and there would be no guarantee that the next batch of FETs would even work well enough.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net

They never actually say what the noise performance is. Ashamed of it, I suppose.

Jeroen Belleman

... plus that if you put them in the emitter of a CE amplifier, they form a sort of noisy cascode with the CE transistor, making lots of noise at intermediate gain.

Jeroen Belleman

Claiming "ultra-low noise" when it's not ultra low? That's as outrageous as selling a "tasty, juicy burger" when it's really not!

They claim it's a clone of this with more data on the sheet:

Though who knows if it has equal performance, would have to test. The lowest noise/THD variant sells for $10 on Mouser currently

If bitrex has one we could ask him to measure it. I've got the AD745, but I've never measured it's noise. The 1uV/Hz is pretty easy.

George H.

OK figure 4. Though I have no idea what a dBV is?

0 dBV = 1Vrms? then -100dBV = 10^-5 Vrms over sqrt(20kHz.) ~400 20nV/rtHz? nice... (there's sure to be some 2*pi in there somewhere.)

George H.

I'm gonna order a few parts from that mfgr that seem worthy of testing out, but will surely take a while to get here :(

Right, (100Meg is 1.3uV/rtHz) Gain on the front or back end adds noise in the same 'excess' amount when compared to changing the feedback R.. (or other fancy stuff) I know little of ADP's (but I'm dreaming of calibrating the QE of my spad from linear up through APD to photon counting.) I guess it's better on the back end 'cause you can drive it with more current and keep the impedance lower.

George H.

Thanks. They're doing much the same sort of transconductance things, with one diode's worth of degeneration in the diff pairs. They don't give a noise spec, but that architecture will put them in the 400 nV/sqrt(Hz) class, so they're still well off the pace.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC / Hobbs ElectroOptics 
Optics, Electro-optics, Photonics, Analog Electronics 
Briarcliff Manor NY 10510 

http://electrooptical.net 
http://hobbs-eo.com

The datasheet for the original chip here:

is giving an output noise figure of -98 dBV (typical) at 0 dB gain in a

20-20kHz bandwidth into 20kHz so if I'm not wrong that works out to more like ~90 nV/sqrt(Hz), yeah?

I'd guess for that price tho they're not doing any binning for performance grades, or they'll be from the not-ah-so-good bin.

Am 29.07.2018 um 22:05 schrieb Phil Hobbs:

A friend of mine had excellent results with an AD604, maybe some other part of that family in the IF section of a large dynamic range short wave receiver. It has two sections that could be cascaded.

< >

In a book of U.Rohde, I have seen an interesting AGC stage that looked like a diff amplifier at the first sight.

The input was into the base of the current source, one "diff amp input" was fixed; the other was the AGC voltage input. Only one of the collectors delivered a signal to the output, the other one was shorted to GND.

Diverting the tail current to the shorted collector reduced the gain. That could be run at pretty high current, so noise and large signal behavior were quite good.

cheers, Gerhard

In the ancient days audio was controlled by a LDR with a lightbulb shining on it. Variable resistance...

I know you once remarked LDRs have hysteresis, how about noise?

LDRs are slow... not linear, maybe have aging, production spread, in those days -60dB noise was good enough.

Measure one?

Interesting, thanks. Their gain cell saves a couple of current mirrors vs. a pair of normal OTAs, so they get less shot noise multiplication. OTOH it's only that good at zero input, and degrades pretty fast otherwise--it says 4 dB at 50 uA (1V input in their test circuit). That's a lot less of a problem with APDs than it would be otherwise, because the shot noise current goes up linearly with gain (at best).

Using the very useful rule of thumb that a 3-uA photocurrent has a shot noise of 1 pA/sqrt(Hz), the 20-kHz photocurrent noise with a 1-V signal at a (noiseless) gain of 20 and a feedback resistance of 1M is

E_N = 20* 1 pA *sqrt(50nA/3uA) = 52 uV in 1 Hz

vs. 130 nV at zero signal, and that's not counting the multiplication noise.

So the amplified shot noise goes up quite a lot faster than the amp's, and the 4 dB won't be noticeable. Of course we don't know how the noise changes with frequency.

The symmetry adjustment and tempco of gain could probably be got round, so the only real show-stopper is that it's very thinly stocked--Mouser is the only disty that Octopart shows, and they have only a few hundred of each grade and don't stock reels. Nobody appears to stock the Coolaudio one.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC / Hobbs ElectroOptics 
Optics, Electro-optics, Photonics, Analog Electronics 
Briarcliff Manor NY 10510 

http://electrooptical.net 
http://hobbs-eo.com

Right, that's the Vactrol approach that I mentioned.

They're considerably noisier than metal films, that's for sure. They have a lot of 1/f noise when you apply bias, and the high-gain ones such as CdS or CdSSe multiply the shot noise of their own photocurrent by very high factors.

The gain of a photoconductor is

M = tau/t_transit,

the ratio of the carrier lifetime to the transit time through the device. In CdS this can be 2000x.

Sounds great, but there's a GBW-type tradeoff between gain and speed, which is one reason CdS is so very slow.

Vactrols are really niche devices these days--I doubt that even kids' night lights use them anymore.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC / Hobbs ElectroOptics 
Optics, Electro-optics, Photonics, Analog Electronics 
Briarcliff Manor NY 10510 

http://electrooptical.net 
http://hobbs-eo.com

Thanks. That's a +-5V part, so I'd have to use a 10-dB attenuator in front of it, and then to get a 40 dB gain range, it would be running with an additional 40 dB attenuation at unity gain. So that nice 0.8 nV noise effectively becomes at best 200 nV, which is the basic problem these sorts of chips all have.

The AD approach, which seems basically to be using multi-input faders to smoothly switch between taps of an R-2R network, is imaginative--it should keep the gain tempco under control even at the extremes.

That's the basic transconductance-based VCA circuit. To reduce the noise, you have to degenerate the emitters. More than that, you have to use N diode-connected transistors for the degeneration, or else the signal splits 50:50 regardless of what you do with the bases. Each transistor contributes its own shot noise, so you gain SNR like sqrt(N).

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC / Hobbs ElectroOptics 
Optics, Electro-optics, Photonics, Analog Electronics 
Briarcliff Manor NY 10510 

http://electrooptical.net 
http://hobbs-eo.com