Poor CMRR Results with AD620

I've been working for a while with the AD620 for a differential audio application. One test I'm using to determine how effective the circuit is at rejecting common mode noise is to directly couple my signal generator to both inputs of the AD620. I'm feeding a 500 mVRMS 1 KHz signal into both inputs and listening/measuring the output of the AD620 to see if it's attenuating it as much as expected. Unfortunately I can easily hear the 1 KHz tone in my audio path.

I realize the datasheet spec's for CMRR are based on DC to 60 Hz, but based on Fig 14 of the datasheet, I should still be able to achieve 80 dB of attenuation at 1 KHz. So far I'm not getting anywhere near that. The circuit I'm using looks like this. I've tried stripping out C64 and C65 and it helps, but the result is still nowhere near -80dB. I'm the one that layed out the PCB, and I'm pretty confident (although not positive) in the grounding that's being provided since I know that can be a factor.

Any insight would be appreciated.

  • One additional question. Based on my numbers, -80 dB from a 500 mV input signal works out to 50uV. For the gain resistor I'm using (750 ohm), this works out to a gain value of 66. Does this gain value still need to be applied against the 50uV for an expected result of 3.3mV? I'm not clear on whether the gain value is applicable to the common mode signal or whether I can safely assume that this gain number applies ONLY to differential signals.

    Just want to clarify if I'm understanding the expected result properly.

  • I can't help you with your problem (have you tried to measure the signal with a multimeter instead of listening?)

    But I can answer you the second question. The CMRR is defined as Ad / Ac. The ratio of differential mode gain to common mode gain, so, yes, the 66dB gain applies and you should get 3.3mV.

  • Thanks gentlemen. I'm connecting a Keithley 2015 as my signal source, so the output impedance of that is set to 'high' (which I believe is > 10K). I've un-populated all the diodes, C64, C65, C68, C69 thinking the same thing......that these components must be creating an imbalance, but I detect zero improvement. My application is purely audio frequencies, so Harryh: I'm not sure what kind of frequency range you are referring to when you are talking about the diode capacitance. I'm assuming you mean frequencies larger than the audio band?

    I think my original expectation was un-realistic. I was thinking that I could put a really large common mode signal into it (like 1V RMS) and expect a full 130 dB of attenuation on it so that I couldn't even hear this in my audio path. What I'm finding is that the attenuation available is very frequency dependent (as I realize the datasheet shows), and that the gain I have set (66) still gets applied to the common mode signal after the attenuation. As a result, the largest my common mode signal can be is around 100 mV for it to be un-audible in my audio path. No where near what I was hoping for.

  • 0
    •  Analog Employees 
    on Mar 13, 2014 8:57 PM

    Hi ByronS,

    There are a couple things that I think are being missed here. First of all, your RFI filter has a differential cutoff of less than 2kHz, which will attenuate audio frequencies.

    As shown in the specification table, the low frequency CMRR of the AD620 and similar instrumentation amplifiers tends to increase with gain, so the CMRR at a gain of 66 will be closer to 110dB min (A grade) rather than 73dB. But as you pointed out, it decreases at higher frequencies. Here's a SPICE simulation of the typical behavior:

    To interpret this, a 1Vrms signal at 1kHz applied directly to the inputs would typically cause a 5uVrms error signal referred to input (RTI), or about 330uVrms seen at the output (RTO). Which is a reduction of 106dB vs the signal amplitude (RTI or RTO).

    Unfortunately, system constraints tend to reduce the performance further. Capacitor tolerances typically aren't very tight, and as Harry mentioned, imbalances in the diode capacitance will add to the problem too. Even if they seem small (and diode capacitance isn't always small), you're looking for upwards of 100dB matching, so the slightest mismatch can end up making a big difference in the CMRR. Here's a simulation what happens if you have ±10% capacitors. It will be 20dB better at high frequencies if you have 1% caps.

    You may not be able to find 200pF of imbalance in your circuit, but this result is fundamental. There is a different amount of attenuation for the positive and negative paths, which results in a differential signal. AD620 doesn't reject differential signals, it amplifies them. I am willing to bet you can find 2pF of imbalance, even when the RFI filter caps are removed, and that already brings you down to ~86dB at 1kHz.

    Luckily there are a couple of things that I think can help here. Since the AD620 released a couple decades ago, there have been several improvements to instrumentation amplifiers. I would advise you to look at the AD8421. It uses a different pinout designed to optimize CMRR vs frequency, it has higher dc CMRR, lower noise, and lower distortion than the AD620, and most of all it has integrated over-voltage protection. This will allow you to reduce the value of your input protection resistors and improve system CMRR. (Reducing the resistor values is the easiest way to desensitize the circuit to capacitance imbalances.) There is a nifty circuit for RFI filtering in the AD8421 data sheet using chip ferrite beads to achieve wider bandwidth and lower noise, while still getting similar RF attenuation. You can adjust that circuit as needed. If you still want to go for higher performance, you might try replacing the three individual capacitors of the RFI filter with an X2Y capacitor, which are inherently better-matched.

    Best regards,


  • 0
    •  Super User 
    on Mar 16, 2014 5:34 PM


         Where I work we're very happy with the SSM2019 -- although I haven't actually measured its CM response, this mic preamp has proven itself in installations with 200' cable runs.  I do recommend enough input RF filtering to keep out nearby AM stations.  We use 390 uH surface-mount inductors with 1 nF to gnd.

         Best regards,