opamp circuit offsets

that is a key point to this method

if the frequencies are harmonically related, then you have to worry about the phase.

E2_Offsets.jpg

Every 250ms you'll get an answer, after 2 seconds all 16 possible combinations will have cycled by.

Tried that. It peaks at +-6 mV. The correct answer is 8.

The four sines never peak simultaneously.

Then use square waves. Or offset the sine waves by 90 degrees so they will peak simultaneously (cuz cos(0) = 1, so cos(0.5 * 0) + cos(1 * 0) + cos(2 * 0) + cos(4 * 0) = 4).

It's more fun - and easier - to use four unrelates sines and let it run.

But then you have to wait.

Or you could use the step command, once for each source:

Version 4 SHEET 1 880 680 WIRE 320 64 96 64 WIRE 480 64 400 64 WIRE 96 112 96 64 WIRE 96 224 96 192 FLAG 96 224 0 FLAG 480 64 Vbob SYMBOL voltage 96 96 R0 SYMATTR InstName V1 SYMATTR Value {bob} SYMBOL voltage 416 64 R90 WINDOW 0 -32 56 VBottom 2 WINDOW 3 32 56 VTop 2 SYMATTR InstName V2 SYMATTR Value {sue} TEXT 72 296 Left 2 !.step param bob list 0 1 2 TEXT 62 316 Left 2 !.tran 10u TEXT 72 272 Left 2 !.step param sue list -0.25 0.25

Doesn't even need to be a .TRAN, .DC also has stepping capability. ...Jim Thompson

Try cos(ft) then - that starts with the main in phase peak.

I think for what you want to do the requirement to search all extrema is that that periods must be mutually prime so that they have quite a long repeat cycle with every fractional phase represented.

Waveforms peaking at pi/2 for (2n+1)pi/2 respectively

huh? those four are harmonically related.

Amusing that we've been jawing for three days about how to save time on a 5-minute job. ;)

Cheers

Phil Hobbs

To find a worst-case max value of offset of which the chances of ever seeing in the real world are likely asymptotically close to zero.

But hey Gray code is cool I hadn't thought about that in a while

What aspect of op amp input offsets would make you think they are not jointly normally distributed? The input offsets are combined in a very simple linear equation so they have no impact on one another.

You have a lot of answers showing you how to get 8mV systematically, but the question I would ask is whether that's at all a meaningful figure to design around.

I don't think "2mV worst-case offset" is at all a hard limit; some opamps list "guaranteed maximums" where I assume they're testing every part to spec before it ships - I don't see such a guarantee in the ADA4817 datasheet. If they're just sampling it could be anything really, with a decreasing amount of probability. It also doesn't cost 15 bucks

Basically my point is that intuitively it seems way more likely that you'd get a straight up defective single unit that threw the whole thing out of whack by 50mV than four sloppy-but-in-spec parts that all conspired in just the right way to give you the "derived" 8mV figure, and I bet the grunt math would bear this out. So like, why is this a meaningful figure for design.

Do we know that op amp input offsets are normally distributed, though? Usually in the context of worst-case design for a single amp it doesn't really matter what the exact distribution of the offset is.

If every offset source is assumed to be jointly normal iid here then yeah, every partial sum is also jointly normal iid and the final output-referred offset is jointly normal iid as well, and the linear summation all works great. It would also work fine if you had some other weird distribution for the offset sources V1 and V2 and just summed those, as they're not conditionally dependent on anything and you could just theoretically convolve the distributions to get the distribution of the sum.

However, the output-referred offsets of U4 and U6 are conditionally dependent on what the offsets of U1 and U5 are, and if those offsets aren't actually jointly normal I think that implies a problem for the mathematical justification of the sum you wrote down.

In any case in thinking about it's probably moot, at least for simulation/pen and paper there doesn't seem to be much other option other than to make assumptions about the distribution off the offsets you're pulling from for any particular part as that stuff's not included in any model, you'd need real-world data, and if the manufacturer knows that it's something other than normal they likely aren't telling.

It probably is "other-than-normal" if they pull some of the samples that are the good shit for "guaranteed" parts.

It is indeed a dumb idea to do in simulation without real parts though, as I seemed to have been assuming there was some kind of random variable actually in the model to discover. Which there isn't. It's all right there.

Of course >:-} ...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

Thinking outside the box... producing elegant solutions.

I don't think ADI would ever ship a part with 2 mV offset; they probably guardband that. Some parts have histograms of the production distribution, and never show outliers past the max spec. Well, they wouldn't.

Defective singles are bad and need to be replaced. That is rare.

If I assume, say, +-5 mV I'll probably never see one past that. +-8, never.

I need to trim the unit to +-1 mV with some guardband, so the question is how much trim range I should plan for.