Hi all,

the attached is the oip3 simulation result of a Gm-C filter, it's confusing that the IM3 is away from 3× slope line, and not smooth at all, could anyone explain it to me the reason?

Thanks in advance

Thanks for the reply,
As the IM3 could be buried in the noise when the simulation started from a lower level, it still couldn't give the correct slope at this power level.

And I tried to involve less intermodulation products during the simulation, but still results the similar result, thus to say, the higher order intermodulation products don't give much influence as we expected.

And do you think it's because of the gm-C structure, because other circuits will not give such kind of results, for example VGA.

Hi,


I guess we can't know unless you post the schematic. There is some discussion on this forum about IM3 results showing the wrong slope even at low power when using BSIM models, but this is for transistors biased in the deep triode region so it might not apply to your circuit. If you have PSP models, you could try using them.

What do you mean by buried in the noise? There isn't any noise in the simulation. If you are talking about the simulator accuracy, you can try increasing the accuracy setting to conservative for example.

If you say that the slope is wrong even at low power levels then it shouldn't be due to high-order harmonics.


regards,
Aaron

Hi BR,


that would be true in measurement, but in simulation, noise has to be simulated seperately. You only see circuit noise when you run noise simulations like "noise", "pnoise", or "transient noise". I'm not too sure about the whole matter, but I believe the "noise floor" you see in simulation is due to numerical inaccuracy related to the limited number of bits used in calculations. I don't remember the numbers, but I believe that "noise floor" is a lot lower than the -50 dBm I see in your plot.


cheers,
Aaron

Here is what I found through a quick search online...

http://www.designers-guide.org/books/dg-spice/chA.pdf

Why don't you post the plot with the input starting from a much lower value...btw, seems that the numerical noise floor may be due to the point where your accuracy setting tells the simulator that the tolerance is acceptable enough and stops the simulation.


Aaron