Laboratory Amplifier

[MaxFrister]

Laboratory Amplifier


I’m working on a general purpose lab amplifier.  It started with the designs from The Art of Electronics (TAOE) vol 2 and 3 but I’ve made a few changes in requirements.

I have a few questions and would also be interested in any criticisms upon the current design.

Design goals include:

• Input impedance of 10M-ohms
• Fully differential (isolated) input
• DC offset to +/-10V from a single control
• Output in-phase with input

the original book designs provided:

• Calibrated and variable amplification
• Bandwidth from DC to 100kHz
• Bandwidth limiting
• Fault tolerant to input of +/-150V
• Output impedance of 50-ohms

Questions:

• Is 10M input impedance achievable with the AD620 or will the offset currents result in an uncorrectable offset?
• The AD620 instrument amp requires a DC ground reference for “truely differential” inputs to keep them from floating to the rails.  R2 and R3 attempt to provide it but adversely effect the 10M input impedance.  Is there a better way?
• C15 and C16 are included in TAOE vol 2 but not 3 (which also feature different front end opamps, LF411 and OPA627).   Why are they there in 2 and gone in 3?
• TAOE has the offset adjustment for the LF411 referenced to V+ but the National datasheet says to reference it to V- (source or sink current into the long-tail pair).  Does it make any difference?
• R13 and C24 are not in TAOE Vol 2 but are in Vol 3 (again the op amps have changed from LF411 to OPA627).   The text say they are there for stability but I’m not clear on how those values were arrived at?  Are they necessary for the LF411 as well?

By the way, I’m planning on using LF411 ($0.00 each)  instead of OPA627 ($25.00 each).

[Conrad Hoffman]

Haven't seen the original circuit. Don't know why they made the changes between revisions, but I'd expect the performance of a "jelly-bean" 411 and the fairly exotic 627 to be quite different. The input currents are a factor of 10 different, 5 pA for the 627 and 50 pA for the 411. Then you have the 620 instrumentation amp at 1 nA. What did they use originally? All opamps need to be referenced to ground on the inputs for differential operation. You might have to select opamps to work with 10M, or use the better opamp. Look at page 17 of the 620 data sheet where they use 100 kohm to reference the inputs. I think this amp is best suited for lower impedance sources. BTW, most lab amps would use two BNCs, with the shells at ground and matched paths to the opamp inputs. If you want the high frequency common mode rejection to be any good, the signal paths, including probes, have to match. Also, paint or keep the diode protection network in the dark.

[awallin]

an open hardware modernized version of the SRS 560/570 would be a nice project!
https://www.thinksrs.com/products/sr560.html

at least some years ago the schematic was available from SRS.

[jh15]

My lab space is quite crowded, would like to have a lab amplifier

[MaxFrister]

The original text-book example is similar to what I've given but omits the differential input.

I've made a number of changes to the first revision.

• Replaced the in-amp with an ad8220.  This has impressive specs, reasonable cost, but is smaller than I usually like.
• Added 2 input bnc jacks referenced to ground.  I left the original single isolated BNC since I often just need to convert from high impedance to 50-ohm without amplification.
• Changed the protection diodes with low leakage JFETs.  I didn't know about the light-sensitivity of 1n4148.
• Fixed the +/-10V part of the circuit.

I've been looking for the schematic for the SR560.  A fragment is available in TAOE.  It appears to set the differential input impedance to 200Meg!

[RandallMcRee]

No expert, me, but it does appear that you have four different ways to set the output offset voltage....!?

Maybe one, only is needed. Perhaps, considering gain, two?

Randall

[MaxFrister]

My intention is just to bring one of the DC offset pots out to the front panel; the remainder are internal trim pots.  They may not all be necessary but it is much easier to omit filling a board footprint then to try to wedge a circuit on later.

There is actually a question whether it is more useful to have the DC offset before or after the gain stage.  If you are using it to null out an offset in the input, then before would be more helpful but if you are trying to set an output bias point then after would.

[Kleinstein]

The AD8220 is still relatively slow. For a gain of 10 or 100 it may still be acceptable.

Unless for a very high impedance input, I see no need for the very expensive OPA627. The still expensive OPA827 might be better in many aspects and also an OPA141, OPA172 should be still reasonably good and affordable, though not as cheap as an LF411.

For really low noise one might consider a different input protection, e.g. with depletion more MOSFETs instead of the simple resistors.
For the protection it gets a little tricky if one want's to allow relatively high voltage (e.g. a gain of 1) and still low leakage.  Leakage get lower at a reduced voltage. So one might have to compromise here. BAV199 diodes could be an alternative to JFETs.

A really universal amplifier is kind of tricky. It gets considerably easier of one allows a reduced range, like a little less BW, limited input range (e.g. +-1 V or even less). Also DC performance could still be an issue.

Another point to consider might be AC coupling at the input.

[Conrad Hoffman]

Depending on what you want to do with it, maybe 1 meg inputs would be better, compensated for capacitance just like a Tektronix input. I use their differential amp plug-ins constantly. At work we have the ADA400A, which is a bit pricey and harder to use, do to the same. You might look at the user manual for that to see how they arranged the block diagram. No schematic. I've also used the SRS preamp and it was OK, but I prefer knobs over digital button control. You might also look at the manuals for the Tek 1A7A and 7A22, which do have schematics, to see how they did it. The performance is probably nowhere near as good as you can do today with opamps. FWIW, I've never had good luck with pricey purpose-designed instrumentation opamps. Rolling my own using the classic design of three good opamps, or even adding FETs to the inputs, has always served me better.

[David Hess]

Input impedance of 10M-ohms[/li][/list]

1 megohm inputs would allow using standard oscilloscope probes.  See below about true differential inputs.

Output in-phase with input

If the input is differential, then the output can be in-phase or reverse-phase.

Is 10M input impedance achievable with the AD620 or will the offset currents result in an uncorrectable offset?

Not really, the offset might be correctable however the input current noise will overwhelm the input voltage noise so a lower input bias part should be selected.  The AD620 would even be marginal with 1 megohm inputs.

The AD620 instrument amp requires a DC ground reference for “truely differential” inputs to keep them from floating to the rails.  R2 and R3 attempt to provide it but adversely effect the 10M input impedance.  Is there a better way?

These are only required for AC coupled applications.  See below about true differential inputs.

The "ground reference" is the common mode input voltage.

C15 and C16 are included in TAOE vol 2 but not 3 (which also feature different front end opamps, LF411 and OPA627).   Why are they there in 2 and gone in 3?

The bypass capacitors are a good idea for various reasons.  They remove the low pass filter produced from the series resistance and capacitance of the protection network and input and also remove high frequency noise from the series resistors.

TAOE has the offset adjustment for the LF411 referenced to V+ but the National datasheet says to reference it to V- (source or sink current into the long-tail pair).  Does it make any difference?

The LF411 offset voltage adjustment circuit needs to go to the negative supply.  TAOE had a mistake.

R13 and C24 are not in TAOE Vol 2 but are in Vol 3 (again the op amps have changed from LF411 to OPA627).   The text say they are there for stability but I’m not clear on how those values were arrived at?  Are they necessary for the LF411 as well?

The OPA627 is 16 times faster than the LF411 so R13 and C24 keep it stable despite the added phase lag caused by the LT1010.  R13 and C24 are not be required for the slower LF411.


Consider using one BNC for each input with only a 10 megohm resistor to ground for each one.  Or even better, use 1 megohm resistors so that standard oscilloscope probes can be used.  Add jumpers in series with the shunt resistors so that the inputs can be configured as "infinite" resistance inputs.

Take a look at the design of the Tektronix AM502 for some ideas.

You might also look at the manuals for the Tek 1A7A and 7A22, which do have schematics, to see how they did it. The performance is probably nowhere near as good as you can do today with opamps.

Matching the performance of the 5A22/7A22/AM502 would be a challenge even today.  One problem is that operational amplifiers configured for shunt feedback have poor overload recovery which would need to be taken into account.

[MaxFrister]

Thank you all for the great suggestions.

Regarding the design, I have no illusion that I can match the specs of the extraordinarily well designed SRS and Tek gear.  My goals are much more modest:

1. Learn something, and
2. Build a useful bit of lab gear.

Regarding adjusting the offset of an LF411 using V+, I tried it last night.  Nope, it doesn't work.  I don't think I would have tried that experiment with a $25 op amp.

I'll spend a bit more time comparing op amps but then settle on my choices.  I hate choosing op amps; there are a billion of them and the specs are always incomparable.  For the lowest noise and highest bandwidth, I suspect the front end should be discrete anyway.

I didn't know about the am502.  I thought about putting my design in a 500 module, but didn't like idea of roar of the TM500 mainframe and the difficulty of working on that physical format.

I like the idea of matching scope probe impedance and optional AC coupling. 

In what situations is the infinite impedance input useful?  Does it occur often enough that the switch should be brought to the front panel?  I am trying to avoid feature creep....

I have also considered including a calibrated attenuator that would make it easier to attach high impedance sources to  the spectrum analyzer.

[Conrad Hoffman]

As far as using offset null pins, do what the data sheet says and nothing else. Using those pins can often degrade other properties, so if you don't need to, don't. I find infinite impedance more useful on DVMs for metrology work, and in 44 years of using scopes, have never wished I had it. Remember that at higher frequencies the input capacitance will load things more than the DCR anyway. IMO, the most useful thing you can learn from the Tek schematics is how they did the input attenuators and compensated the probes. Don't underestimate the value of those bandwidth filters, as when you apply high gain you'll see nothing useful without them. That also tells you that trying to build a very high bandwidth circuit for high gain applications is probably a waste of time.

[macboy]

If you are looking for only 100 kHz, you might want to have a good hard look at some of the audio power amplifiers offered by TI (originally NS), such as LM3886, LM3875, LM1875. These are essentially power opamps. The LM3886 can be powered from over 80 V (+/- 40) and can be loaded with as little as 8 Ohms at that voltage. The LM1875 is lower power, but is a simple small 5 pin TO-220.  All of these devices have SOA and thermal protections built in. Another interesting device is TI (Burr Brown) OPA549, which has a current limit input pin to program the output current limit from 0 to 10 A using a resistor or a voltage. That would be a great feature for a bench amplifier used as a four-quadrant power supply, don't you think? Its GBWP is quite a bit lower than the National devices but still good enough for 100 kHz. 

[David Hess]

Regarding adjusting the offset of an LF411 using V+, I tried it last night.  Nope, it doesn't work.  I don't think I would have tried that experiment with a $25 op amp.

The LF411 datasheet is old enough that it includes a full schematic which show the offset null connections which is what I used to verify how they should be used.

I'll spend a bit more time comparing op amps but then settle on my choices.  I hate choosing op amps; there are a billion of them and the specs are always incomparable.  For the lowest noise and highest bandwidth, I suspect the front end should be discrete anyway.

It takes experience to recognize the different types of operational amplifiers.  A JFET input operational amplifier would generally be the best choice for the input because of noise and input bias current but a discrete design has some advantages for input protection, controlling input capacitance, and coupling between the inputs.

The LT1102 JFET instrumentation amplifier is one of the very few suitable parts for the input amplifier but check out its common mode rejection ratio at 50kHz of less than 40dB versus 100dB for the Tektronix AM502.  An AMP02 does a little better and would be another good choice for a simple design.

Getting better performance will require using JFET operational amplifiers or discrete JFETs.

I didn't know about the am502.  I thought about putting my design in a 500 module, but didn't like idea of roar of the TM500 mainframe and the difficulty of working on that physical format.

I do not know about a "roar" but I would not suggest building a TM500 module either.  The 1, 2, and 3 bay TM500 mainframes have no fan.

I like the idea of matching scope probe impedance and optional AC coupling.

The big advantage of being able to use x10 oscilloscope probes is that they provide a lower input capacitance and shielding.  The problem is that their mismatch compromises common mode rejection.

With that in mind, I would make the 1 megohm input shunts trimmable so that a pair of standard x10 probes can be used without compromising common mode rejection.  The way Tektronix achieved this was to make special adjustable x10 probes.

In what situations is the infinite impedance input useful?  Does it occur often enough that the switch should be brought to the front panel?  I am trying to avoid feature creep....

The AM502 just has a jumper on the inside.  Trying to switch such a high impedance point is difficult to do without adding leakage or coupling.

Don't underestimate the value of those bandwidth filters, as when you apply high gain you'll see nothing useful without them. That also tells you that trying to build a very high bandwidth circuit for high gain applications is probably a waste of time.

The need for bandwidth limiting becomes very apparent when using the 5A22/7A22/AM502 at high gain or high sensitivity with an oscilloscope.  Note that these three instruments are all the same but the AM502 is intended for stand alone operation with a separate instrument.  An AM502 would have been paired with an AF501 variable bandpass filter to make spot noise measurements before FFT spectrum analyzers became available.

Noise reduction can also be implemented through the averaging and high resolution functions on a DSO.

[duak]

I could envisage cases where 10M (or even higher) Zin is desireable, but the attenuation of the probes is not.  I have an idea for a Hi-z input circuit where the 1M resistor is selectively bootstrapped.  ie., the bottom end is either tied to common to give 1M Zin or is driven by a percentage of the input signal by a buffer.  If the bootstrap signal is 90%, you get a 10M input impedance.  An advantage is that the actual signal line is not switched so the little PCB leaf switches Tek used aren't needed.  Separate buffer amps would be needed, but then a less expensive INA could be used.

Cheers,

[David Hess]

Using a x10 probe is definitely a problem for noise limited measurements; it immediately raises the input noise by 10 times and that is not even including the increase in Johnson noise from the increase in resistance.

On the other hand, a x10 probe isolates the cable capacitance which may be more important than noise and it increases the common mode and differential input range by 10 times.  The old Tektronix 7A13 differential amplifier even has a "x10 Vc" switch which adds x10 attenuation to the input and x10 gain to the amplifier so the sensitivity stays the same but the input common mode range increases by 10 times; this allows 10mV/div sensitivity over a +/-100 volt common mode range.

The above points to a serious limitation in most differential input amplifier designs; increasing the common mode range by attenuating the inputs commensurately increases the input noise.  Low input noise with high common mode input range is possible by simply using a design with high input common mode range which is what the Tektronix 7A13 does with an input stage which operates from +/-50 volt supplies for a base +/-10 volt input common mode range.  Typical oscilloscope FET input stages have a common mode input range of +/-250 millivolts which may indicate why the Rigol DS1000Z series has so many signal fidelity issues.

Operational amplifiers used as followers or a differential amplifier can do pretty well in this respect at low frequencies; +/-10 volts with +/-15 volt supplies is feasible and there are some parts which can do much better than this without an exotic circuit design like the LTC6090/LTC6091 although being a CMOS part, its 1/f noise is pretty poor and its AC common mode rejection is not any better than the instrumentation amplifiers we have discussed.  Also, differential input protection is still required despite the high common mode input voltage range.

[MaxFrister]

Thanks for all the insightful comments.

I've completed rev. 3.  It has a number of minor changes including fixing the roll-off circuit and adding an ac input mode.   

Included is a preliminary pcb layout; it will likely change as I try to source the parts.

[duak]

Nice design - should work well.

A few little things:

1.) if you're thinking to use X10 probes, I would add a balancing pot between the input resistors R101 & R102 of say, 1K with the wiper tied to ground.  This would compensate for any differences in the resistances of the probes and improve low frequency common mode rejection.

2.) I would isolate the power supply voltages for the first (INA) stage from the output stage with a series R and a bypass caps to common.  The output stage could draw enough current while driving a 50 ohm load to couple enough signal through the common power supplies to degrade the INA's performance. 
There are a number of paths:
 - the INA itself - look at the PSRR graphs,
 - the input protection diodes' reverse capacitance,
 - the offset control and opamp buffer,
 - the layout

My rule of thumb was a holdover from the bad old days of vacuum tube or discrete transistor amplifiers where one never powered more than two stages from one decoupling network.  Three stages almost always oscillated in weird ways eg., motorboating.  Opamps are much better but aren't perfect.  A colleague had something like three 40+ dB gain stages powered from the same point and it gave him grief until he realized what was up.

3.) Feature creep warning:  I'd implement a 3rd order filter as I found a sharper filter to be more useful.

Cheers,

[MaxFrister]

...
1.) if you're thinking to use X10 probes, I would add a balancing pot between the input resistors R101 & R102 of say, 1K with the wiper tied to ground.  This would compensate for any differences in the resistances of the probes and improve low frequency common mode rejection.

Good idea, but see answer to #3.

2.) I would isolate the power supply voltages for the first (INA) stage from the output stage with a series R and a bypass caps to common.  The output stage could draw enough current while driving a 50 ohm load to couple enough signal through the common power supplies to degrade the INA's performance. 

Good idea.  I think I can accomplish it without major damage to the existing layout.  Where would you recommend I set the -3dB point on the filter?  Would you include the offset control for the INA on the filtered supply?

3.) Feature creep warning:  I'd implement a 3rd order filter as I found a sharper filter to be more useful.

For my purposes, I think first-order filter is okay.  Notice that that third gain stage is already complicated: Summing junction for dc offset, gain of 6dB to 50-ohm impedance output,  low-pass filter, and output buffer.


Feature creep?  I'm good at that.  I've learned from experience that at some point I have to say "good enough" and move on to the next stage -- ordering the pcb in this case.  Otherwise, the project keeps expanding in scope until (a) I can not longer accomplish it, or (b) I lose interest.  But here is my wish list for version 2

• Relay switches to minimize noise pickup from wires running to the front panel
• MCU for external computer control
• Digital pots to eliminate the expensive 10-turn pots and MCU control
• Meter on output for setting dc offset (the original had this).
• Clipping indicator light
• Uncalibrated gain indicator light
• Reverse phase switch

and of course lower noise design.

[duak]

Max, regarding the power supply filtering for the INA.  You're using 10u caps already.  If the series R is 22R, the corner frequency will be 800 Hz and will give a minimal DC drop.

I would definitely power U102 and perhaps U103 and U104 from the filtered INA supplies.

I just noticed that U204, the -5V buffer, has a 10uF bypass cap on its output.  Most opamps with large capacitive loads are unstable without a series resistor.  U205, the +5V reference is probably stable with a capacitive load but should be checked.

Also, RV101 will be exceedingly touchy on the higher gain settings because of the high gain of the following stages.  Just thinking out loud here, what purpose does it serve?  I can sort of see a need for canceling out an offset before being further amplified.

Cheers,

[MaxFrister]

The board is in the mail -- or at least on a slow boat.  It is hard to complain about 10 boards for $11 delivered.



I made a few more tweaks to the layout and added more resistors to allow for calibration.   I'll post an update in a few weeks or months after testing.

I may very well leave RV101 off the board if it does not seem necessary or helpful.

[MaxFrister]

I finally got around to putting this in a chassis.  For power I was able to recycle a defective torroidial transfromer from my Datron 1081. 

The amp performs mostly as expected.  There were some issues with pcb footprints.  The LF411 is too slow so I did not reach my design bandwidth of 100Khz at +46dB.  The front end is wrong; my attempt to make it symmetric around ground is incorrect.  Right now, neither AC coupling nor oscilloscope probe compensation work as intended.

I probably should add some labels to the controls.