I am trying to measure excess resistance fluctuations in resistors. The circuit itself is very simple. Its a wheatstone bridge excited by a AC drive. the idea is that this drive will modulate the fluctuations to a higher frequency which is then demodulated and measured using a lock-in. The demodulated output should reflect the excess resistance fluctuations which have a '1/f' power spectrum. The circuit consists of a voltage source driving a bridge of identical resistors, the bridge difference is amplified by an instrumentation amplifier(As I do not have to worry about source impedence matching with the amplifier) and then locked at the drive frequency using a lock in. The problem is that the final output power spectrum, shows a sharp roll off of about 20-30 dB/decade instead of the expected 10dB/decade. A sort of filtering is happening in the circuit somewhere, I have kind of isolated the problem to lie in the AD620 in-amp I am using. Could this happen in an in-amp? or in other words are there special precautions to be taken when amplifying ac signals with an instrumentation amplifier? Would using a different instrumentation amp(AD8221) help? P.S: I have taken care of the basic errors that could happen like ensuring a proper return path of DC Bias currents, etc.
Whatever you're doing here is necessarily going to have a very high gain. The AD620 is a reasonably low-noise amplifier with a ~1MHz GBW product. Here's what the gain vs. frequency looks like for different DC gains:
The AD8221 has a similar GBW product that behaves better.
If you are using high gain such as 1000 or 10000 you would expect to see the indicated roll-off in the Johnson-Nyquist noise at 20dB/decade. It should not affect your measurement at lower frequencies of the flicker noise, but you may have to adjust your excitation frequency to optimize the performance. The voltage and current noise corner frequencies of both those amplifiers are less than a few hundred Hz, so something 1KHz or lower might be a good choice.
You can get a flatter response by not trying to get all the gain in a single stage- split the gain into multiple stages.
In your modulated scheme you have to have bandwidth at the amplifier for the desired signal bandwidth plus the modulation frequency.