36
\$\begingroup\$

My MAX44251 dual op-amp has a very small unwanted 131KHz periodic artifact at the output, seemingly regardless of how it's configured.

My assumption was EMI, but I can't see this 131KHz signal on any other part of the circuit. I've also tested this in multiple buildings, with multiple probes, with all other electronics turned off, and surrounded by foil shielding.

What should I try to remove it? I would like to at least achieve a voltage follower with noise under 1mV.

enter image description here




The chip was originally used in a more complex circuit when I first noticed the problem. BUT, to isolate this issue I made a whole new test PCB with fresh components. I left extra pads to reconfigure the chip in different ways while testing.

enter image description here

Right now it is configured very simply:

schematic

simulate this circuit – Schematic created using CircuitLab

The bypass caps are on the bottom ground plane layer. Vias are hand soldered.

I have observed the effect through both the Agilent 10X passive probe (It's hard to see), and through a probe like the following, with which I can zoom all the way to 2mv/div. Originally, it was observed because the output is fed to a comparator, and the comparator output indicated the input signal amplitude was > the desired 2mV.

enter image description here

The waveform is periodic but kind of strange. Here's a few pics from different angles:

200 ns Stopped

200 ns Stopped

50 ns Free Running

50 ns Free Running

20 ns Free running

20 ns Free running

10 ns Stopped

10 ns Stopped

\$\endgroup\$
  • 6
    \$\begingroup\$ quality schematic, description of test setup, screenshot of observed signal, isolating the problem as far as possible, asking a well-defined question… Your question is pure pleasure to my soul! \$\endgroup\$ – Marcus Müller Dec 1 '16 at 1:13
  • 3
    \$\begingroup\$ What is your power supply? That has the look of a buck or flyback converter ringing to me.... Try it with batteries as your supply instead of whatever you are using. Can you zoom in one one of the spikes? \$\endgroup\$ – Dan Mills Dec 1 '16 at 1:14
  • 1
    \$\begingroup\$ Uh- oh. Look at the datasheet, page 7, second row of figures, figure on the far right. "Input Voltage Noise vs Frequency". There's an ugly spike at... 65 kHz, which happens to be half of your observation, but that graph doesn't even go to 131 kHz. \$\endgroup\$ – Marcus Müller Dec 1 '16 at 1:20
  • 2
    \$\begingroup\$ @DanMills I tried it with +/- 9V from two 9V batteries, the artifact is identical. \$\endgroup\$ – Keegan Jay Dec 1 '16 at 1:31
  • 1
    \$\begingroup\$ @JayKeegan does that Opamp have any active temperature compensation or an automatic offset correction? Looks like the impulse response of something seeing a short impulse to me... if this was a digital system, I'd say there's single-pole IIR being reset every \$\frac1{131\text{ kHz}}\$, but I'm almost certain this opamp is analog. \$\endgroup\$ – Marcus Müller Dec 1 '16 at 1:48
15
\$\begingroup\$

I can't really tell if this is a actually a symptom of what is described in the datasheet:

noise vs freq

Notice how there's a spike that exceeds \$30 \frac{\text{nV}}{\sqrt{\text{Hz}}}\$ at 65kHz – pretty much half of the frequency you're observing your noise at; they didn't characterize up to 131.5kHz, however.

What should I try to remove it? I would like to at least achieve a voltage follower with noise under 1mV.

If you just need a low-bandwidth voltage follower: Use a low-pass filter.

If you need signal up to 65 kHz and above: An RLC notch (band-stop) would probably work best; a quick & lazy design run on my favourite passive filter design tool yielded R=0.16Ω, L=1µH, C=1.5µF as possible configuration.

RLC notch

Note that you could try to use the inverse circuit (RLC bandpass; swap the (L--C) with the R) in the feedback branch of your voltage follower.

| improve this answer | |
\$\endgroup\$
  • 2
    \$\begingroup\$ Thanks for your help looking into this. I'll give the post some time for others to check it out, but I think you are correct. \$\endgroup\$ – Keegan Jay Dec 1 '16 at 1:53
  • 3
    \$\begingroup\$ Wow, that's a nasty thing to have in an opamp, even worse is how poorly the datasheet displays this, I am sure most people would overlook this \$\endgroup\$ – PlasmaHH Dec 1 '16 at 15:23
  • 1
    \$\begingroup\$ @PlasmaHH to be fair, the datasheet does have the figure posted above – but I agree, if you sell something explicitly with a gain*bw of several MHz, you might want to mention there's periodic spurs in spectrum. \$\endgroup\$ – Marcus Müller Dec 1 '16 at 15:42
22
\$\begingroup\$

Note that this is an autozero amplifier (also called chopper stabilized) - many very low offset opamps work by periodically sampling the input offset and injecting a compensating offset to counter drift in the front end. To do this there is an oscillator in the opamp together with a set of analog switches at the input. This can result in clock feedthrough to the output as well as charge injection at the input pins.

Presumably this device is using 131kHz as the switching frequency.

I can't find any detailed information on the Maxim part but here is some info for an Analog Devices part that is probably similar:

Analog Devices Zero drift opamp

If you really need the low offset and drift then they are the best type of devices to use - you may need to Just limit your bandwidth and filter out the clock.

The bandwidth of the auto-zeroing is enough to encompass 1/f noise in CMOS opamps so they can be very low noise for frequencies below 1kHz, a region where CMOS opamps tend to have problems.

If you can't filter out the clock noise see if you can use a conventional part - they will often have a worse drift and offset performance but you can get them better than 100uV offset. You may also have to trade-off input bias current because bipolar input amplifiers are usually better than CMOS for this parameter. Bipolar are usually lower noise as well.

A related problem I have had with a similar Linear Technology part (LTC2051) is that the autozero circuitry can take a very long time to recover from overload when the output saturates - many milliseconds for a part with GBW of many MHz. This makes them unsuitable for any application that saturates as a normal part of its operation such as oscillators or threshold detectors.

| improve this answer | |
\$\endgroup\$
  • \$\begingroup\$ Do all similar zero-drift op amps have such large pulses though? Or is that just the reason this particular op-amp is so cheap? Is the spike amplitude a consequence of this particular configuration? I suppose there is a large overlap between applications requiring very low offset, and applications that only need DC, but 6mV @ 131KHz still seems pretty substantial. \$\endgroup\$ – Keegan Jay Dec 1 '16 at 2:56
  • 3
    \$\begingroup\$ Chopper stabilized or auto-zeroing amplifiers will generally have the possibility of clock feed-through seen as noise on the output. All the vendors claim very low levels of noise. The app note from LT (cds.linear.com/docs/en/lt-journal/LTC2050_1100_Mag.pdf) looks a bit better than yours but not much. \$\endgroup\$ – Kevin White Dec 1 '16 at 3:39
  • 1
    \$\begingroup\$ I have to agree. Use of any analogue part that injects chopper or sampling technology into the analogue path of your signals needs to be applied carefully so that you use it with appropriate bandwidth limiting so that the chopper frequencies are cut out of your usable frequency spectrum. \$\endgroup\$ – Michael Karas Dec 1 '16 at 15:21
6
\$\begingroup\$

I agree with Marcus, the ~130 kHz would be the second harmonic of the Chopper switching frequency ~65 kHz.

A reduced 'Closed Loop Bandwidth' of your Op Amp could result in the second harmonic (~130 kHz) to have a greater magnitude than the first harmonic (~65 kHz), to solve this, as Marcus mentioned, a solution could be adding a passive filter to filter that noise.

There is an article by Art Kay, "1/f Noise and Zero-Drift Amplifiers", that talks about noise in Zero-Drift Op Amps.

If you want to learn more about Op Amp Noise, check out TI Precision Labs for Noise.

| improve this answer | |
\$\endgroup\$
2
\$\begingroup\$

Don't have an answer, but I can tell you, for inspiration, how I would debug this.

First, I would try to solder a bypass cap right to the chip. An 0603 part, 100nF, and use braid to connect to the other pin (for low inductance). You bypass caps are behind relatuvely high inductance vias, and this can make them ineffective for the spikes. The spikes are at 131 kHz, but the frequency contents is much higher, so good bypass matters very much.

This would probably fail :-).

Then I would replace the amp: 1. Analog Devices makes some very low offset trimmed amps. The offset is not as low as in an auto-zero amp, but check it out. Those are a bit more expensive, so check against your budget and offset requirements. Look at AD8615 and similar. The only thing, those get a bit expensive for high-volume consumer stuff.

2 Also, consider a good old instrumental bipolar opamp from burr-brown lineage (Texas Instruments now.). Use the same impedance at both inputs to get rid of the bias current, and make sure the input impedance is low enough that the offset current doesn't matter. Something similar to opa237.

  1. Try a different auto-zero amp, perhaps one with spread-spectrum clock. Again, look at analog devices parts.

Good luck

| improve this answer | |
\$\endgroup\$
  • 1
    \$\begingroup\$ That wire self-inductance is 0.834 nH. Any mathematical way to determine the significance of its effect? \$\endgroup\$ – Keegan Jay Dec 1 '16 at 21:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.