low-distortion gain control

Looks good. Been a long time since I used an MDAC but they're handy. Only fly in the ointment is the 12nV/rtHz noise. Quieter would be nicer.

For audio - indeed even 35kHz I'd probably choose the 20 cent 5532 dual op-amp. For DC < whoops ! > precision you'd want something finer. TI's Burr Brown division also makes some fine op-amps but I just get to use commodity parts these days, so I'm not the man to advise.

Graham

They're just pushing their LT1468 opamp. Hmm, not a bad choice!

--
 Thanks,
    - Win

I could probably live with 40ppm rather than 25pm (this isn't an audio app, but rather for instrumentation), but what would be a good choice, better than the 5532 and not as painful to use as an AD797 or LT1115? I don't want high Ib, In, Is, Pd, nor do I need the 0.9nV input noise. But 20V/us and 110MHz GBW should help insure low distortion at 30kHz.

- an hour passes -

I just looked at the LT1468 suggested in the appnote in Fred's post, it's an improvement over the AD797. Consider, same 2.5ppm distortion at 30kHz, yet 3nA Ib instead of 250nA, and half the supply dissipation. I mean, 1/4 watt in an 8-soic is pretty severe, if it's not necessary, and I'll have 30 of them in this project.

Both opamps use the bootstrapped-output-stage-driver trick, including the distortion-canceling capacitor Analog Devices talks about (page 8), but with opposite polarity transistors (a more detailed schematic of the LT1468 appears in George Feliz' 3-page article in Linear Technology Magazine, Nov 1998). I also like the LT1468's more reasonable price.

Next I'll examine the pga2310 and pga2311 suggested by ban (although zipper-noise doesn't worry me, in my instrument the gain settings will be fixed), one of those may trump the dac8811 + LT1468 combination. At first glance the 30kHz distortion looks 10x higher, but we'll see.

--
 Thanks,
    - Win

I'm a little bit concerned with the zipper-noise, that would be heard when using it for audio. For unipolar instrumentation it seems to be OK. OP627, fast enough that the glitches(fig.9 -30mV) don't upset the input. If you want a good audio gain control have a look at PGA2311 or PGA2310

These are classes better, stereo, much more dynamic and control range and you can change the gain at 0-xssings, which avoids the zipper-noise.

--
ciao Ban
Bordighera, Italy

For pure audio the PGA range is indeed the preferred choice.

Actually, as I understand it, it doesn't matter where gain changes are made. If you make an instantaneous gain change at Vpk then you get the largest deltaV. Zero-crossing yields zero deltaV but yields instead the highest possible delta (dV/dt). I'll bet Mr Fourier says they're all but equivalent. I.e. - no free lunch.

Ramping gain up / down exponentially / log - whatever should be quietest for zipper noise. Looks like I'm going to have to work on this myself soon.

Graham

Going by the TI data sheet, the 5534/2 aint that good at 35k, 0.004% approx, at AV=1 It's almost good enough for my hearing though,.

The AD797 claims to be 0.0003% and the LT1115 dont show graphs going up above 20K.

martin

Funny, we're doing something very similar right now: a 16-bit bipolar dac, feeding a 10-bit 2q mdac to scale the range of the whole thing, steps from +-10v to +-2mv. We're probably going to use a DAC8831 feeding an AD5432, both serial. We'll use slow cmos chopamps for high analog precision.

We looked at using digital pots as the gain scalers, but they often seem to have scairy linearity behavior near the bottom tap. The r-2r dacs don't guarantee the absolute division accuracy when the code is, say, 0000000001 (lsb only, divide by 1024) either, but they still look like a safer bet.

John

Funny you should ask:

Selecting Op Amps for Precision 16-Bit DACs ? Design Note 214

Aha! You failed to see they take the measurements with a 20kHz filter, which of course reduces the distortion harmonics for fundamentals over 10kHz. See the plot going down? Very fishy, until you see the note.

Indeed. Perhaps I shouldn't use the word "gain control" because I'm just setting the operating level for a precision capacitance sensor.

--
 Thanks,
    - Win

I think I saw that someone had designed an audio compressor around the 2310, possibly on the prodigi forum. But long time ago, I was trying to use parallel DACS in Eq's, the biggest problem was charge injection zipper noise, especially when the MSB flips. I wonder how TI do it?

martin

In article , martin griffith wrote: [....]

Try the LT1028's data sheet. The LT1115 and the LT1028 are really the same part.

--
--
kensmith@rahul.net   forging knowledge

I wonder because the graphs show at 20kHz less distortion than the 8811. If you want to use a fine external amp the Wolfson WM8816 is similar to the TI device, but requires external opamps. The big question if you want to make a dB-linear attenuator with 0.5dB or bigger steps, the approach will be much easier software wise. Also I do not like Figure8 of the dac 8811, if you want more than 60dB attenuation. Depends all on your application.

--
ciao Ban
Bordighera, Italy

Yeah, I had another look, that filter must cut in quite strongly. The built-in opamp is not optimized for BW, but noise. And has only 3MHz GBW.

Ahh, quite different from audio, where also very low frequencies matter. With 35kHz you are well above the noise corner of cmos-opamps, which are avoided for audio. The Wolfson is IMHO the best digital attenuator for Audio use, especially in balanced operation. You can achieve >120dB dynamic range.

--
ciao Ban
Bordighera, Italy

I looked at the WM8816 a couple of years ago with Barry Porter, for an audio project, but ISTR that there was a latch up problem,mentioned in athe data sheet. any way Baz had laid out the board wrong, and we had to hand wire with kynar wire the chips to the board. This destroyed them. They seemed very static sensitive

Have they improved the ic at all in this respect, I'd love to try them again. Not having a internal opamp is most excellent.

martin

Yes, but beware of both when comparing to the AD797. Note that the LT data sheets don't show wideband noise (as in above the audio band). They actually have a rather large noise peak around 5-6MHz, you need to be careful about this. The AD797 does not have this "feature", it's well-behaved noisewise everywhere. BTW, I learned about this from the AD797 designer, and as a disclaimer I also work at ADI by day.

Steve

This is where the internal design of the part really counts.

I'll be doing level control in DSP so I don't have that particular problem.

Graham

LT1028 has low voltage noise and huge current noise, so works well in very low-impedance circuits, probably not a good choice to back-end a current-mode dec.

John

In article , Stephan Goldstein wrote: [....]

Yes, the noise peak can be well below that point depending on the exact unit you pick. At about 100KHz, they show about 3nV/sqrt(Hz). The noise increase is due to the fall of gain in the first stage letting the second stage's noise show up.

I does strike me as odd that no-one has brought out op-amps much below

1nV/sqrt(Hz). It seems that there would be enough applications that wouldn't mind the higher current etc that would be needed to make it happen. It is after all, mostly a question of input transistor size and current.

--
--
kensmith@rahul.net   forging knowledge

I wrote already that I like the OPA627, but I got the prefix wrong. This amp will probably do better in this application, the distortion is maybe 0.001% at 35kHz in unity gain, but it is quite costly. At least the datasheet shows a filter only at 80kHz, so it is possible to predict the performance.

--
ciao Ban
Bordighera, Italy