opamp circuit offsets

Just put 1 mV in series with each noninverting input in turn. That'll give you the partial derivatives, so it's simple to figure out what the worst case is (plus or minus the datasheet maximum, signs chosen so they all add up.

Cheers

Phil Hobbs

Binary. Assign each offset to a bit position, then count. ...Jim Thompson

--
| James E.Thompson                                 |    mens     | 
| Analog Innovations                               |     et      | 
| Analog/Mixed-Signal ASIC's and Discrete Systems  |    manus    | 
| STV, Queen Creek, AZ 85142    Skype: skypeanalog |             | 
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  | 
| E-mail Icon at http://www.analog-innovations.com |    1962     | 

     Thinking outside the box... producing elegant solutions.

Too much work.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

Too much work.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

On Apr 15, 2017, John Larkin wrote (in article):

This approach is an application of multisine test signals, where one synthesizes a fairly random wideband signal using a finite set of sine-wave test tones.

There are many ways to choose tone frequencies and their amplitudes. There is a large literature. Lots in IEEE Trans on Instrumentation and Measurement.

One common dodge is to make the amplitudes of the test tones vary with frequency such that the SNR of all tones are the same despite the differentiating effect of say a transformer or the like.

NIST uses multisines generated from a Josephson-junction voltage source to measure known-property near-gaussian noise to calibrate Johnson-noise based absolute thermometers. For your application, I?d chose at least five test tones whose periods are mutually prime.

Joe Gwinn

I only need four offset sources. I want them to occasionally peak simultaneously, positive and negative.

F and 5F seem like a nice pair to start with.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

Make the offset sources Gaussian noise uncorrelated for each trial, run a bunch of parallel simulations and assume the Central Limit Theorem to find the mean and variance of the Gaussian distribution of the sum of random variables to your required confidence, then work it "backwards" to find the best fit mean and variance for the unknown distribution you're pulling "samples" from

I will assume that's too much work, though. :(

Or multitone sine wave signals if you want to strictly limit bandwidth or whatevs

Add 'em all up, worse case, then take the sqrt of that number. (I guess that doesn't work.)

I use a lot of dual opamps, I've always wondered if the offset's would be better if I made one inverting and the other non. (usually after the first stage one care's less about the offset, unless it's a filter.)

George H.

Yeah, I think that cookbook method only works given certain assumptions for a single opamp you have per-section real world stats on from the datasheet, not for a blob of stuff that you don't...

We should find out!

Particularly since it won't solve the problem as stated. ;)

Cheers

Phil Hobbs

FWIW, a situation somewhat similar to Larkin's recently manifested itself in my own little world. For my case, 100, 200, and 400 Hz tones were arbitrarily selected to "spice things up," as they say. Your advice leads me to believe that 157, 257, and 457 Hz are a better bet. :0)

Thank you,

--
Don Kuenz KB7RPU

My statistics classes were a long while ago. And perhaps I'm not understanding the problem from the description - these are _simulated_ offset sources, right? To evaluate the statistical properties of the structure in simulation?

Or are these "DC offset sources" intended to be part of the thing IRL? I don't know what this thing is supposed to do, so I have no idea.

It seems to me at least like some sort of statistical problem where you're attempting to infer the worst-case probability of some value for a particular sample from the population when all you have to work with is averages.

If your worst case input offset is 2 mV on each amp, the worst case offset on the output will be 8 mV. The math is very simple if you understand how op amps work which I know you do. Why is the max offset so hard to calculate?

The input gain of the first stage is 3. The inverter would have a gain of 2 from the offset input shown. So the total offset gain contribution of these two stages is 3 + 3 + 2 since the first stage is fed through the second. The two sets of inputs in parallel make the total gain from the amplifiers 16. The summing network feeding the VCVS has a gain of

0.25 making the total a gain of 4. With 2 mV offset at each input with the correct polarity will give 8 mV offset max.

Since two of the inputs contribute much more to the total than the other two, I would be inclined to use just the two main contributors to determine an estimate of a "typical" offset value. I think 5 mV would not be an absurd value, but I would be more comfortable with 6 mV. The question is *how* typical will this be? Do you want 95%, 99% or 99.99%? I can't see much use for the first two cases and I have little use for unqualified typical values. Your simulation data seems to show around

95% confidence with a 6 mV offset, but it's hard to eyeball. Certainly it's not 99% at 6 mV.

--

Rick C

That'll work, but beware, the sine waveform spends little time near zero, and lots of time near peaks. The statistics are all wrong, so it'll only give worst-case, not expected RMS, results. Four sines (2-3-5-7 frequencies are incommensurate) as error signals interacting will still tell you worst-case, and still won't tell you expected RMS. So, you'll get excellent yield of items at ten times the accuracy predicted.

If your offsets had a gaussian distribution (likely), the full treatment would include a random sample from the '100 thousand normal deviants' chapter of a good statistics reference.

It's a linear circuit so the maximum should be at the limit of offset voltages in one direction or the other. Fundamental Theorem of linear programming, IIRC.

So square wave +/-Vos at each input and halve the frequency on each source to effectively get a binary counter that exercises each and every boundary condition (Vos limit) combination. 4 op amps so there are only 16 combinations, so not much output waveform to examine.

--sp

--
Best regards,  
Spehro Pefhany

That will give the super-super pessimistic figure, yeah? I guess when the glass is half-empty it's half empty.

I guess I thought he was looking for the more "typical" value, though?

Well, I had my engineer hat rather than sales hat on (actually I was wearing my fashionable "Ohioans for Concealed Carry" baseball cap, but that's another story).

--sp

--
Best regards,  
Spehro Pefhany

One way-out-of-nominal IC in a circuit seems like happenstance that can be prepared for in the design phase, but like four stinker op-amps all in the same subcircuit? That sounds like enemy action!

Or at least should happen infrequently enough that it's OK to write it off as a stillborn board in the test phase...