RFC: fully floating buffer with OV protection and gain control

[xiphmont]

Hi folks,

Everyone seems to have designed a buffer at one point or another, all slightly different in order to fit the specific application.

The point of this specific design is to act as a floating balanced buffer for testing, both to sample the differential voltages within a test circuit and then drive an A/D box. It includes overvoltage protection that will clamp the input to the opamp to just past VCC/VEE when powered and to approximately ground when not powered. Input attenuation is pre-clamp, allowing the circut to be used at up to 400Vp-p [theoretical]. Of course, being a floating circuit, there must be a shared ground potential somewhere-- just not in the signal path.

I have a few question on which I'd appreciate comments.

This is a probably an appropriate circuit to feed balanced-input A/D boxen. But most A/D boxes are both unbalanced and have a 'common' ground across all the inputs. Do I gain anything by having the fully differential output into an unbalanced input? I could just as easily avoid ground loops by having each buffer circuit 'float' on the signal ground provided by that A/D input. Or should I embrace the suck and just put up with the ground loop potential on such examples of consumer equipment?

Assume the opamps are OPA[4]132. Given the large-ish value resistors on the input, is the BJT overvoltage protection still necessary? My guess is yes due to the 2^13ohm FET input impedence. I don't want to accidentally pop an opamp by having the wrong range selected when I hook it up to an amp. Not that I love the opamp, but it could feed through to the much more expensive A/D box...

Any other thoughts?

[xiphmont]

In case it wasn't clear what I meant by "I could just as easily avoid ground loops by having each buffer circuit 'float' on the signal ground provided by that A/D input.", I've attached another pic.

Again, simulation and limited bench time only tell me so much... I know both circuits work, I'm just looking for obvious recommended improvements as well as where to eliminate unneccessary overkill.

[KSTR]

Hi Xiphmont,

To make some statements one must make some assumptions.

I assume you have the buffer's PSU GND connected to the A/D-GND or a potential nearby, and that the buffer GND is not directly connected to safety earth (only via a loop breaker circuit). Then you would not gain anything from the servo output, in fact it only adds distortion. In that case you'd be better off with the conventional subtractor as the output stage, connecting it's REF input to the A/D GND (which kills GND-loop hazard). This is your "floating buffer" in your second drawing.

When the A/D is fully differential, again there is not much profit, then two simple subtractors would work better, this time REF being GND of the buffer supply. Or a simple subtractor and an inverter after it. Of course no output should be shorted to GND, then. The primary goal of the servo ouput is to maintain constant gain into both balanced or unbalanced inputs (one line shorted to GND), which need not bother us here as we have full control over the situation.

The input stage doesn't eliminate common mode signals (common mode gain = 1, and that's why you need the subtractor), so if you want to probe signals at, say, 50V common mode you must divide down the common mode signal (resistors from the subtractor opamp's +IN and -IN nodes to GND). This comes at the cost of increased noise gain. This also assumes that the PSU for the input stage is floating and it's "GND" is connected to a potential nearby the voltages you want to probe. Still you have the problem that you cannot exceed the input amp safe ranges, differentially also (and you would need to divide down the outputs or change gain of the subtractors anyway to not overrun the A/D). If that is required (eg measuring the output of bridged power amp), then input dividers are needed (EDIT: I now see you have those already), but then in turn you could divide down to GND as well, reducing common mode "on the fly" and then would not need a separate floating supply and the dividers in the second stage.

This all very much depends of what signals you want to measure, if you really want to do millivolts at 400V common mode it's gonna be tricky.

Regrading protection, those 1M, though adding quite a bit of noise, would do the job (and please add some filter caps, both common mode and differential), but clamp diodes to the supplies are never overkill (simple 1N4148 will do, no need for more complex circuitry).

You also must make sure that there actually is a DC path for the input bias currents back to the supplies of the input amps (via GND, normally), as little as it may be, but you need it. For higher gains AC coupling is often a better choice (then you need distinct bias path resistors) because of offset/drift issues.

All in all, welcome to instrumentation, can really be a big can of worms, sometimes.

You might want to check some AppNotes on instrumentation at Analog Devices, Linear, TI, National, etc, and the corresponding parts for more info.

And have a look at the THAT1600 and THAT1200, those are advanced line drivers and receivers.

Last not least, don't forget plain simple signal transformers (output maybe buffered, depends), those are safe, easy to use and work very well in many AC situations at high common mode voltages. For example I use them to measure PSU rail ripple, even actually listening to it. Need some precautions, though (no DC, not even for a brief moment, or many good xformers will be degraded severly, getting magnetized).

- Klaus

[xiphmont]

Originally posted by KSTR
Hi Xiphmont,

To make some statements one must make some assumptions.

Of course. I didn't go too overboard because I half expected no one would care to answer :-) Buffer design being old hat to the experts and all...

Originally posted by KSTR
I assume you have the buffer's PSU GND connected to the A/D-GND or a potential nearby, and that the buffer GND is not directly connected to safety earth (only via a loop breaker circuit). Then you would not gain anything from the servo output, in fact it only adds distortion. In that case you'd be better off with the conventional subtractor as the output stage, connecting it's REF input to the A/D GND (which kills GND-loop hazard). This is your "floating buffer" in your second drawing.

Yes, that correctly matches my unstated assumptions, and answers my question directly. Thank you!

Originally posted by KSTR
When the A/D is fully differential, again there is not much profit, then two simple subtractors would work better, this time REF being GND of the buffer supply. Or a simple subtractor and an inverter after it. Of course no output should be shorted to GND, then. The primary goal of the servo ouput is to maintain constant gain into both balanced or unbalanced inputs (one line shorted to GND), which need not bother us here as we have full control over the situation.

That would have been my next question :-)

Originally posted by KSTR

This all very much depends of what signals you want to measure, if you really want to do millivolts at 400V common mode it's gonna be tricky.

The goal is to handle signals where common mode is typically in the same range (or smaller) than what I need to measure, eg, be able to use the same buffer implementation to look at an amp's line input and power output. The PSU will be floating-- this is intended to be battery powered. Easy to float batteries :-) However, I intended more than one buffer to use the same batteries, so that will limit the common-mode use. I have to accept some limitations somewhere :-)

Originally posted by KSTR
Regrading protection, those 1M, though adding quite a bit of noise, would do the job (and please add some filter caps, both common mode and differential), but clamp diodes to the supplies are never overkill (simple 1N4148 will do, no need for more complex circuitry).

As envisioned, the circuit would run on +/- 6V-- zeners that small leak current like a sieve. I had wanted to maintain as high an input impedence as possible. thus the attempt to do clamping via BJTs.

[edit: I'm not sure what you mean by the clamp diodes-- I had assumed you were talking about zeners, but the 4148 isn't a zener, so... Do you just mean stacking a few diodes and relying on the forward junction voltage? It seems like that would add some strong nonlinear distortion.]

The 1M inputs were, again, an attempt to keep input impedence high despite the attenuation circuit sitting in front of the input buffer. Would the self noise of 1M resistors really be that much of a problem if chosen properly (eg, Vishay/Dale CMF55 series).

Originally posted by KSTR
You also must make sure that there actually is a DC path for the input bias currents back to the supplies of the input amps (via GND, normally), as little as it may be, but you need it. For higher gains AC coupling is often a better choice (then you need distinct bias path resistors) because of offset/drift issues.

I was still considering the possibility of dropping the BJT clamping/1M resistors and just using coupling caps and bias path resistors, yes. I had not forgotten about the input bias path, but I think i also have not thought about it hard enough.

[KSTR]

Those 1N4148 are switching diodes (low capacitance/leakage) and are meant to go to the PSU rails, so they are reverse biased under normal conditions. You could also use the reverse biased diode of a JFET, which is even better in terms of leakage.

Battery power is a wise decision, keeps away lots of trouble

With the 1Meg inputs, the unavoidable thermal noise will be dominant (and construction dependent noise add to that) once you go below a certain signal level. As long you aren't to measure millivolts or below, this will probably not be a factor (for distortion measurements, frequency response etc).

You could do a building block approach (already there with those colored shades you made) and make different input modules for different purposes.

On safety thing to consider: check that the A/D input can withstand a constant DC of your supply voltage in case of a fault/overload condition. Most sound card inputs (do you use a sound card) are AC coupled and have protection to survive 6V DC or so (nevertheless I burnt 2 channels of mine lately, with "only" 10V AC, but I still have 22 cannels left ;-)

- Klaus

[xiphmont]

Originally posted by KSTR
Those 1N4148 are switching diodes (low capacitance/leakage) and are meant to go to the PSU rails, so they are reverse biased under normal conditions. You could also use the reverse biased diode of a JFET, which is even better in terms of leakage.

Oh, oh, I understand now. That makes perfect sense. And I'd felt so clever w.r.t. the transistors.

Originally posted by KSTR

With the 1Meg inputs, the unavoidable thermal noise will be dominant (and construction dependent noise add to that) once you go below a certain signal level. As long you aren't to measure millivolts or below, this will probably not be a factor (for distortion measurements, frequency response etc).

...there is plenty of use to having as much signal depth as possible. The resistors I like to use (the aforementioned CMF55 series) list a noise rating of .1uV/V, but I'm guessing that's not what you're talking about. You're talking about a constant level of self-sourced noise current/voltage, correct? And the quoted rating is proportional introduced noise.

Originally posted by KSTR

You could do a building block approach (already there with those colored shades you made) and make different input modules for different purposes.

Not at all a bad idea. Now that I know eBay is a good source of cheap Elma type 6 knockoffs and I can have as many supermultideck switches as I want for a pittance....

Originally posted by KSTR

On safety thing to consider: check that the A/D input can withstand a constant DC of your supply voltage in case of a fault/overload condition. Most sound card inputs (do you use a sound card) are AC coupled and have protection to survive 6V DC or so (nevertheless I burnt 2 channels of mine lately, with "only" 10V AC, but I still have 22 cannels left ;-)

- Klaus

I'm using some old eMagic USB boxen I bought years ago. They're eight channels each, decent for the era and reliable. They may well be AC coupled, but if so, their F1 is insanely low; they can produce and sample 2Hz at full amplitude.

That said, I had chosen the VCC/VEE of +/-6V with exactly this in mind; a worst case output of 6V. Given that these are cosumer units (and so have a line level of -10dBu) I was seriously considering running the buffer at +/-4.5v for an extra level of safety. That would still let an OPA2132 swing +/-3V.

[xiphmont]

OK, went and looked it up...

Assuming my displayed bins are 25Hz wide on a linear scale, a 1M resistor is generating .64 uV RMS white noise per bin, or -124 dB. Before adding any gain, that's more noise than measured from my A/D box. That was worth knowing!

I get to step back and think about that attenuator now.

Thanks for all the help, Klaus.

[xiphmont]

Round of questions #2:

Using 10k input resistors buys me another 20dB of noise floor. Given the application (reasonably good test and measurement using commercial computer sound hardware), I think this is probably pretty good. 10k resistors have a noise floor of -158dB/sqrt(Hz). A good 16 bit sound card is about -149dB/sqrt(Hz). My A/D box only manages about -155dB/sqrt(Hz) in 24 bit mode (which isn't that good but it was one of the very first USB A/D boxes.) If I use 2W units, that still gives margin for gracefully dealing with 120V line inputs. Because.. well the curious might want to watch the mains at some point.

Second: Using clamping diodes would dump current into the power rails on overvoltage. This application is battery powered; reverse-charging batteries is uusually bad. And when unpowered, the rails will be floating. The biased BJT clamp I came up with dumps current to ground (which will still be there when the batteries are disconnected). Is this reasoning valid, ie, that the transistor approach might be the correct choice for batteries?

Thanks again.

[KSTR]

I think your clamp circuit is fine, from that POV (which I seem to have missed).

The 10k would do fine also. My soundcards have 5k series R and Zin(AC) of 10k and there is no signficant noise penalty. One can still use synchronous averaging to get the noise down further, if required (and it will also kill the hum/spuriae which swamp those thermal/device noise levels often anyway).

Um, use 2kV-rated resistors and maybe even gas-discharge overvoltage clamps if you really want to connect to the mains

DISCLAIMER for the innocent readers just lurking here: DON'T EVEN THINK OF TRYING THIS IF YOU'R NOT FAMILIAR WITH THE SAFETY RULES.

Have fun.
- Klaus.

[xiphmont]

Originally posted by KSTR

Um, use 2kV-rated resistors and maybe even gas-discharge overvoltage clamps if you really want to connect to the mains

DISCLAIMER for the innocent readers just lurking here: DON'T EVEN THINK OF TRYING THIS IF YOU'R NOT FAMILIAR WITH THE SAFETY RULES.

Have fun.
- Klaus.

Yes, I wasn't seriously suggesting doing that without considerably more overbuilding/safety design incorporated into the implementation. The 120v design criteria was a theoretical capability of just the attenuator and op-amp protection should I decide to go for the additional capability of inspecting mains. The circuit as shown is not safe enough to do that! It is only good enough for when 'everything is going right' and then only with careful component selection.

Hadn't meant to potentially endager the innocent!