opamp circuit offsets

Probably, worst case won't happen in a million production units. The probabilities matter too, and the sine wave thing hints at that.

If I did all gaussians, I might never see the worst case.

Sines are fun, too.

Just make beep boop machine music boxes for a living, even many mV of offset probably won't objectively affect the sound that much. However it may decrease the subjective dynamic range and analog "warmness" of the sound so it's probably best to use op amps with at least a 1GHz GBW and dip them in liquid oxygen to recalibrate any offset irregularities

Behringer just released a new budget synthesizer with an all-analog synthesis path and one of the first comments from a musician who'd got their hands on one was "Yeah it's analog, but it doesn't sound like...analog enough"

output.

Input offset of an op amp is an intrinsic value, it doesn't depend on an environmental factor other than the ones which are controlled (temperature, supply voltage...). If the apparent offset of the op amp varies with input drive, then you aren't measuring input offset, you are measuring some other feature of the device. In the OP's circuit the inputs are combined linearly according to circuit theory. I'm not clear what you are trying to describe.

If you are talking about maximum absolute value of input offset, then the distribution doesn't matter. Any part outside that range is defective by definition. If you want to know the distribution of the output offset, this is a simple mathematical relation to the offsets specified for each device. Some data sheets show actual measured distributions of input offset which is always normally distributed in my experience.

I really don't get where you are trying to go with this other than to say that the basic assumptions in use are not valid, but you haven't shown any evidence of this. You are just speculating based on your lack of knowledge, perhaps?

That's hilarious. As in, wrong.

We don't really care what the particulars of the input offsets are, though. If I understand the problem correctly all we are interested in knowing is how much the voltage at "DIFF" will, on average, deviate from

Is that correct?

Ok, well, if they're always iid random variables which are normally distributed in your experience then ok, a simple linear sum of the distributions of output-referred offsets to compute the distribution of the sum is mathematically valid, because the Gaussian is its own Fourier transform.

But that's the only case. If they are not then the sum is not mathematically justified, and I don't think you can even just immediately convolve the distributions because the output offsets of two of the amps are conditionally dependent on the outputs of the other two.

Yes, the input-referred offset voltages are all independent of each other and nothing affects that. But the figure we're interested in is a sum of the outputs.

I guess what I'm trying to say is that it it seems you used superposition to arrive at that linear equation, which is fine, if the assumptions that allow you to use superposition with random variables hold, which is that all the input offsets are Gaussian. If they aren't then I believe in this circuit all bets are off. You say they should always be so, then OK, all is well.

You can use superposition to evaluate the distribution of the sum from an arbitrary linear (arithmetic) sum of distributions of random variables in one very specific, very Gaussian case.

Indeed, and if they would only get a real simulator that is designed to solves this type of problem.

-- Kevin Aylward

I posted a very simple, easy to implement, even in LTspice, method... but it was ignored... Larkin just can't bear that I post trivial solutions to his whiny bloviations >:-} ...Jim Thompson

The only problem is a lack of understanding of mathematics. It's hard to use a simulator properly if you don't understand the math involved in your problem.

Are you aware that your signature is not recognized as a signature? I believe it needs to have nothing else on the signature identification line other than the two dashes and a space.

'The format of the sigdashes line is "^-- $", ie it consists of two dashes and a trailing space only.'

Sure. But you have to get the equation right first.

That's the problem with equations. Once you write them, and put the schematic aside, they become detatched from reality and take on a life of their own. Errors are no longer visible, and they propagate.

I've seen so many engineers proudly present a page of equations that concluded in obvious nonsense. The equations are text, symbols, simple-minded mechanical devices, no longer reality.

K so what is it?

You have lost me. The input offsets are what will give you a non-zero output when the inputs are grounded. DIFF is the output.

I'm not really worried about the "distribution". I would be concerned with the max spread of the outputs given the max spread of the four

I don't know if JL needs to specify something more narrow that will cover a large number of cases. That would require rather simple statistics really given the simplicity of the linear equation defining the output.

I don't need to convolve anything to get the maximum output. Linear combinations are simple.

A simple linear combination.

As I've said, I'm not worried about the distributions, only maximum. That is easy to calculate. I think if you want to calculate the distribution, a Gaussian distribution of the inputs is perfectly reasonable. Even if for some reason the input distributions are not perfectly Gaussian, it isn't likely to be far off.

I don't care; that's why I used Spice.

But symmetry requires that V1 and V2 have the same gain to DIFF.

He is right. I mixed up the inputs, not that it matters, because the offsets are all equivalent, V1=V2=V3=V4. I find it hilarious that he is using a simulation to get an approximate answer to such a simple design rather than do some 10th grade math.

Vout = (6 V1 - 2 V3 + 6 V2 - 2 V4) / 4

Really? Who here can't figure this out in a minute? If the gain of the circuit isn't understood, how could anyone possibly be able to design and use it? Maybe Sloman is right, JL doesn't so much design circuits as stumble around with them until something eventually works?

Yeah. It does seem that vast numbers are using spice, but, essentially, have no knowledge of electrons or sums.

Nope. I might look into it...

-- Kevin Aylward

How ironic that you would be doing, then, an analysis based on worst-case numbers for the op amp, the simple equation, rather than insisting on a published distribution of offset voltages in the datasheet.

Ignoring the real data, and analyzing, instead, the abstraction of 'upper limit' and 'lower limit' as published, means you're using the equation and not real data. No biggie, though; even though the approach is flawed, it fails safe. So, it's a reasonable check, even though it lacks predictive accuracy.

ve 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.

The kluge is linear constant coeff and looks symmetrical top to bottom, to me, so you only need to compute the constant DC output to V1 and V4, separa tely, V1-1mv V4=0, V1=0 V4=1mV. Then 2x sum of abs value of those coe ffs, 2x(|k1|+|k4|), with k1= Vout/V1 etc.. gets you whatever you want in the way of worst case, typical, 3 sigma, or whatever you're satisfied with.

Using a digital computer to simulate an analog computer to obtain an approximate answer to a problem which (given certain assumptions) can be directly evaluated in closed-form is definitely worth some geek-cred, though