HELP, I have non-linear gain in two channels

What's the Q of the stages? What type of bpf? What frequency ranges, DC to 1Hz, or DC to 0.5C +- 3dB

Martin

Poke around with an oscilloscope. One of the stages is clipping... typical for band-pass filters (may have internal node gains approaching "Q", depending on filter architecture).

...Jim Thompson

How big is the variation?

If it is bandpass, then it is AC. If it is AC, why bias does matter?

Before searching into the complicated problems, I would check for the simple stuff. It could be the power frequency or some other noise picked up by the channels in the different proportion. It could also be the effect of cross coupling between the channels. It could be the bad voltage references for the ADCs.

Vladimir Vassilevsky DSP and Mixed Signal Design Consultant

I wouldn't have thought it possible that 2 channels of 9 opamps and shed loads of passive components are ever going to match within 1%. If you're plotting the deferens on a cycle by cycle basis then tiny phase shifts will give big errors. Even with an RMS calculation for each channel there will be differences, due to the numerous capacitors changing their values slightly with differing signal levels. God forbid you using anything remotely like ceramic caps. These can offer a (useful?) 2X value change for a 2X signal change.

I agree with Jim, you are likely seeing the effects of clipping.... another possibility, see what the GBW product is on the op-amps you're using and if it is sufficient for the application.

darn Q's always get in the way, Crom

More info

Hello again

I ran a test today. I connected the signal from one of the band pass channels to both AD converters so they receive the same signal. The gain difference was almost totally gone. I hade only a small gain difference which I think is the ADCs linearity error. The dynamic gain is acceptable.

So the dynamic gain "fault" is located to the band pass filters.

And, yes, I use ceramic capacitors.

I can't se any clipping.

It may not be readily apparent in the output. Check internal nodes.

...Jim Thompson

(just my luck for you to have used hand selected, non commercial, NPO types in a computer optimised R.F. BP arrangement, running at 1V5, so ignore the following. Others may find it of passing interest :)

'Low K','COG', 'NP0', ceramics are nice but only run a few hundred pF and can be fine for many R.F uses. Leaves the 'medium K' and 'high K' types which get worse all the way up to about 100nF. I've seen a lot of talk here but no hard numbers, so stuck a 100nF (worst case, high K) 100nF across a variable DC supply and measured it. Problems immediately turned up as I couldn't finger a 100nF cap'. They were all showing 90nF. Turns out they are "100nF" but sunshine (rare in UK) was warming the room up to 24degC and the appalling tempco of these things had caused the value shift. (body heat from 2 fingers dropped C value to 65nF!). DC bias voltage C value (at 24degC)

Cap' was so p*** poor, that for confirmation I knocked up this L.F. voltage controlled oscillator. ___ ,---|___|---, | 22k | | | | | | 1|\\ 2 | o----| >O---o-----o Frequency Out | |/ | CD40106 (hex schmidtt) | (12V supply) --- --- 470nF 100k | ___ | 0-12V o-----|___|----o DC Control | | --- o --- | | 100nF ceramic === === 0V 0V (created by AACircuit v1.28 beta 10/06/04

Control V Frequency

(As a final insult, the cap' ESR measured 47ohms at 1kHz, suggesting any filter resistors need to be >4k7). Basically for filter use, I'd be incined to use -anything- other than ceramics

I still love ceramic caps though. They're small, cheap and are excellent for decoupling use.

Thank youy for your interesting reply :-) I will try to change to other kind of caps.