opamp input bias compensation circuit

[edanderson]

see page 4 of the 990 pdf, figure 1, the circuit in the dotted line:

http://johnhardyco.com/pdf/990.pdf

i have gotten this to work in a circuit where the gain is constant. the effect is that the output dc offset created by the input device bias current is canceled out. i have yet to test longterm thermal stability, but this seems to be an effective way of reducing the offset, if not to zero then close.

why don't more people implement this type of circuit? in my (thus far limited) experience, it seems to work, the parts cost is low, and i haven't found a large difference other performance aspects. there is another variation on the same idea presented later in the 990 paper. even if you are using a servo (as the MPC-1 does) the nulling circuit seems like a good idea.

thoughts?

ed

 

[bcarso]

[quote author="edanderson"]see page 4 of the 990 pdf, figure 1, the circuit in the dotted line:

http://johnhardyco.com/pdf/990.pdf

i have gotten this to work in a circuit where the gain is constant. the effect is that the output dc offset created by the input device bias current is canceled out. i have yet to test longterm thermal stability, but this seems to be an effective way of reducing the offset, if not to zero then close.

why don't more people implement this type of circuit? in my (thus far limited) experience, it seems to work, the parts cost is low, and i haven't found a large difference other performance aspects. there is another variation on the same idea presented later in the 990 paper. even if you are using a servo (as the MPC-1 does) the nulling circuit seems like a good idea.

thoughts?

ed[/quote]

The input bias currents are the base currents, the emitter currents over (1 + beta), and beta varies by about +0.5% or so per degree Kelvin. In Deane's opamp there is also additional variation with the same sign ( that is, base current going down with increased temperature) due to the negative tempco of the emitters' current source (due there to the negative tempco of the two diodes' forward voltage overcompensating for the transistor Vbe tempco). But the bias current compensation circuit shown provides only a roughly fixed amount, so will only be exact at a given temperature. That may be quite satisfactory for limited temperature and signal self-heating variations.

Note as well that the Hardy compensation depends on there being hardly any common-mode swing, which is pretty well satisfied in the app circuit, especially at high gains below overload. If the input pins move much again the bias compensation is not appropriate, since the voltage across the 1M R's vary but to first order the base currents do not.

There are several IC designs that compensate for bias currents, to make the magnitude as measured outside the chip almost precisely zero over a broad temperature range. This has led some to conclude that bias current noise in these amps is also small, but in fact the compensation circuits make them have as much or more than twice the noise power due to the raw base currents.

For the base currents, and for the sorts of impedances shown in the Hardy application circuit the contribution to noise of the compensation is negligible---the thermal noise current in the 1M R's is small compared to the shot + excess noise in the base currents and the thermal noise currents in the external impedances.

To the extent that the base currents are matched, the traditional strategy of having equal impedances in each input is usually a better approach. If it is necessary to pad out one input, shunting the R with a good-sized C will alleviate the noise source thus introduced, although you won't like the parasitics associated with the big cap sometimes.

In this app, the variation due to the gain setting divider will interfere with the equal Z strategy, although one could contrive a more complicated divider with a constant Z.

 

[recnsci]

[quote author="bcarso"]In this app, the variation due to the gain setting divider will interfere with the equal Z strategy, although one could contrive a more complicated divider with a constant Z.[/quote]

If I understand corectly, Ed is interested in DC offset compensation.
In that case, and ignoring for a moment bias comp. ckt, "gain" leg
is Ac coupled trough 1000u, so only DC path for - input Ibias is trough
10K feedback resistor, thus Ac gain independent. Since Voffset (caused by
Ibaias) is 10K*Ibias, 10K Dc resistance at + input would null that
component of offset voltage (iff Ibias+=Ibias-).
Here, DC resistance at + input is paractically
DCR of transformer (which is probably <100 ohm ). Conecting 10K
in pralell with big cap, and then puting that between lower leg of primary
and ground could help, but harm to audio would much overweight
benefits of reduced DC offset.

Regarding noise, arent noises in LTP
independent stohastic processes, thus uncorelated, thus un-nullable?
(here I ignored a bit noise contribution of tail CCS).

cheerz
ypow

 

[bcarso]

[quote author="recnsci"][quote author="bcarso"]In this app, the variation due to the gain setting divider will interfere with the equal Z strategy, although one could contrive a more complicated divider with a constant Z.[/quote]

If I understand corectly, Ed is interested in DC offset compensation.
In that case, and ignoring for a moment bias comp. ckt, "gain" leg
is Ac coupled trough 1000u, so only DC path for - input Ibias is trough
10K feedback resistor, thus Ac gain independent. Since Voffset (caused by
Ibaias) is 10K*Ibias, 10K Dc resistance at + input would null that
component of offset voltage (iff Ibias+=Ibias-).
Here, DC resistance at + input is paractically
DCR of transformer (which is probably <100 ohm ). Conecting 10K
in pralell with big cap, and then puting that between lower leg of primary
and ground could help, but harm to audio would much overweight
benefits of reduced DC offset.

Regarding noise, arent noises in LTP
independent stohastic processes, thus uncorelated, thus un-nullable?
(here I ignored a bit noise contribution of tail CCS).

cheerz
ypow[/quote]

Agree ypow---I didn't notice the 1000uF. It will have some d.c. leakage btw but constant at least. My remarks would apply if that cap weren't there (which might be a worthwhile part to eliminate).

I would also tend to agree that the harmful effects of the bypassed compensating R in series with the transformer leg would be more that the beneficial effects of offset reduction. Mainly, I would look askance at the parasitic L and R, and stray shunt and coupling L and C to the surrounding parts it would represent.

In the hypothetical case of the compensating R-C in the transformer leg to common, using the outer foil as the grounded side of the cap would alleviate some of this a bit.

Most LTP (by which I assume you mean Long-Tailed-Pair) noise is uncorrelated, certainly. So indeed you can't null it, and I would hope I didn't imply that. There may be a small correlation coefficient dependent on frequency between voltage noise and bias current noise but it's usually not considered. Some overall circuit noise (to which you allude by exception) is partially correlated, which includes the shared emitter current generator (as has also been discussed a bit in the mic pre thread of Val_r's recently). I wanted to draw attention, along the way, to the compensation schemes that reduce the magnitude of bias current but generate a lot more noise in the process---some of the "precision" low-frequency opamps have this defect IIRC. Of course those schemes don't apply to the Hardy configuration.

 

[bcarso]

[quote author="mediatechnology"]Off topic but I have a question for you Brad:

To the extent that the base currents are matched, the traditional strategy of having equal impedances in each input is usually a better approach.

In terms of bipolar monolithic op amps of unknown internal input topology is this almost always true? Or specifically with an OPA27?[/quote]

I'd say that if the opamp has some fancy cancellation scheme, or small (for our purposes) bias currents to begin with (e.g. JFET ot MOS inputs), then I wouldn't bother with equal Z's. In the case of the actively compensated bipolars, I should think that the compensation might be as well under- as over-.

It's interesting how many people throw in a compensating R when it's not doing much besides adding noise. A PhD friend did this on a mic preamp using fancy FET amps (he's really not qualified to design circuits, but as a PhD and especially as a Caltech undergrad he has had a certain...errrr...attitude instilled) and I pointed out that it was a "common mistake". He replied in no uncertain terms that if he made a mistake it was not common! :razz:

 

[bcarso]

[quote author="mediatechnology"] Do you suppose that because there are matched input resistances in the inverting and non-inverting inputs of the integrator I'm indeed seeing first-order Ibias compensation? [/quote]

Probably---not looking at the circuit now, but it works for differential integrators too.

Yeah, I avoid compensating R's if possible, especially if a little offset doesn't do much anyway. There have been a few cost-conscious apps where I used some when a bipolar of the LM324 type was needed for ground/neg rail sensing capability and there was the need for large tau, in one case in a auto-on circuit. I would generally shun that family for passing audio, although at very low closed-loop gains and with some class-A output current pulled to avoid crossover distortion, they can pass mediocre audio---o.k. if the signal is already well-lowpassed.

 

[bcarso]

The comparator thing is indeed another good use when you don't need speed and don't want to bother with hysteresis to suppress oscillation, and when you don't need the wired-AND function of the open collectors of 339/393 type parts.

 

[edanderson]

i should have probably said that my application is a bit different than what is shown in figure 1. note that i am not using the compensation from the MPC-1 circuit, but from figure 1. in my circuit, the opamp that has the compensation circuit hooked up to it is AC coupled from the preceding stage. the opamp has a fixed AC gain of 1, and has an output transformer hanging off the output.

without the compensation circuit, the opamp i am using has relatively stable DC offset at the output of about 0.25V. this means that i should put a cap between the opamp output and the transformer. because it is looking out into the "real world" through the transformer, the cap has to be relatively large to avoid LF rolloff or resonances with the transformer. that requires an electrolytic. but in many cases the signal will swing well below -0.25V, quasi reverse-biasing the cap. a bipolar cap is one solution, but this compensation circuit is another.

so far i have been able to reduce the offset down to zero. the measured effect on THD+N is quite small. even if the DC offset drifts +/- 6mv, that is still less than 1ma in most output transformers, which i doubt would have much effect. i'll find out, in any case, and share what i find.

ed

 

[bcarso]

What in your circuit is providing the d.c. reference at the input? 0.25V is a lot of offset for a unity gain stage!

Or... you do say a.c. gain of 1. Whatever the configuration is, there ought to be a way of making the d.c. offset at the output much smaller than 0.25 V. A schematic would be helpful.

 

[recnsci]

[quote author="bcarso"]
Most LTP (by which I assume you mean Long-Tailed-Pair) noise is uncorrelated, certainly. So indeed you can't null it, and I would hope I didn't imply that.
[/quote]

Problem is I try to read as quckly as possible, and on top of that I tend
to read just few last sentences in post. So, I have no idea where I got
that noise thing from, you didnt imply that.
And I also tend to write without thinking too much, thus I put R-C in
primary insead of secondary.

[quote author="bcarso"]
Agree ypow---I didn't notice the 1000uF. It will have some d.c. leakage btw but constant at least. My remarks would apply if that cap weren't there (which might be a worthwhile part to eliminate).[/quote]

That is my only concern with this toplology. One BIG ugly electro in gain
leg and possibly another on output. Question is, what are remedies for that
beside DC servo (which I tend not to like for nontechnical reasons) ?

Cheerz
urosh

 

[JohnRoberts]

[quote author="mediatechnology"]Off topic but I have a question for you Brad:

To the extent that the base currents are matched, the traditional strategy of having equal impedances in each input is usually a better approach.

In terms of bipolar monolithic op amps of unknown internal input topology is this almost always true? Or specifically with an OPA27?[/quote]

It's straightforward to impute from the data sheets if they have performed input bias compensation. In an uncompensated part, there will be a nominal "input bias current" with a specific polarity depending upon P or N type input devices, with a smaller +/- "input offset current" for expected variance due to process variations. If compensation is performed internally the bias current will be nominally 0, but they usually spec a +/- offset value in the bias position too.

So use of equal value resistance in series with both inputs to cancel out the DC bias caused voltage term is "only" of value when a bias of one polarity is predicted, and sometime not even then if noise is paramount over DC accuracy. Note: If input current is not the same magnitude and polarity any added R can make the offset voltage worse.

JR

 

[bcarso]

[quote author="recnsci"]
That is my only concern with this toplology. One BIG ugly electro in gain
leg and possibly another on output. Question is, what are remedies for that
beside DC servo (which I tend not to like for nontechnical reasons) ?

Cheerz
urosh[/quote]

Nontechnical reasons?

Well, anyway:

To begin with, see to it that the d.c offsets in the source are negligible if you can. Sort of hard with phantom power.

Beyond that, brute force would be just to make extravagant amps with very low offset voltage and negligible bias currents (perhaps temperature-controlled JFET inputs etc.). Not for the faint-of-heart nor liable to be economically feasible for production, but maybe fun as a stunt.

An alternative approach would be to actively stabilize or subtract/divide out the nonideal characteristics of the imperfect capacitors. Feels as if it would lead to infinite regress though. In another thread I joked about some audiophoolish paper/foil/oil caps requiring servos because of their high leakage, but it was a real problem.

I have some other ideas as well, but I'm considering filing on them so discretion is warranted at the moment.

 

[recnsci]

[quote author="bcarso"][quote author="recnsci"]
beside DC servo (which I tend not to like for nontechnical reasons) ?
[/quote]

Nontechnical reasons?
[/quote]

Wierd thing, there are some circuits that I dont like when I see them
on paper, for no aparent reason. DC servo, state variable filter based
EQs are another (****, maybe I have some integrator fobia).

[quote author="bcarso"]
An alternative approach would be to actively stabilize or subtract/divide out the nonideal characteristics of the imperfect capacitors. Feels as if it would lead to infinite regress though.
[/quote]

OK, you lost me here, can you shed some more light on bolded part?

[quote author="bcarso"]
I have some other ideas as well, but I'm considering filing on them so discretion is warranted at the moment.[/quote]

Then hurry with filing :grin:

cheerz
urosh

 

[edanderson]

[quote author="bcarso"]What in your circuit is providing the d.c. reference at the input? 0.25V is a lot of offset for a unity gain stage!

Or... you do say a.c. gain of 1. Whatever the configuration is, there ought to be a way of making the d.c. offset at the output much smaller than 0.25 V. A schematic would be helpful.[/quote]

i don't know if this is ideal or not, but this is what i have built:

the bias comp circuit connects at the inverting input via the 100k resistor. with it in place, the 0.25vdc output offset can be trimmed away and the 470uf bipolar cap at the output can be removed.

bear in mind there are pots hanging between the inverting and non-inverting inputs with RLC chains from the wipers to ground that are not shown. i have fiddled with the values on the bench quite a bit, but so far this is the best configuration i could find.

ed

 

[bcarso]

So the opamp is a 990? As drawn, I don't see where 0.25 V is coming from. Must have to do with all the other not-shown stuff hanging as you say from the n.i. and inverting inputs.