3
\$\begingroup\$

Say I need to make a very low-noise signal inverting schematic. For that I came up with the schematic below - the U2 is a low-noise opamp in an inverting configuration, and the schematic around U1 is a rail-splitter typically seen in guitar pedals - it provides the "ground" level for the U2, thus converting a unipolar power supply into a bipolar.

The problem is that simulation shows that removing the C2 filtering capacitor drastically improves the noise characteristics for some reason. When C2 is connected, there is a very clear spike at around 20kHz, while removing it removes the peak (graphs are also shown below).

So, the question is: what causes this spike? Why can't I add additional filtering using C2 capacitor (even the rail-splitter TLE2426 datasheet suggests leaving only the C1 capacitor)? It does not even improve the noise characteristics at ~1kHz frequency range, so what is the point in it? In guitar pedals the noise characteristics are not so important, but still, why is it there then in guitar pedals if it does not improve characteristics at any frequencies? Any suggestions on should I remove the C2 capacitor in my case or how should I reimplement the rail-splitter?

P.S. If it is not clear: I've used the same schematic for both tests, the only difference is the C2 capacitor. In both cases I tested the full schematic shown below with both U1 and U2, these are not separate schematics! The input and output are the U2 input and output, U1 only provides the mid-rail voltage. I've just tried to remove the C2 capacitor from this schematic.

Schematic and graphs

\$\endgroup\$
3
\$\begingroup\$

As @brhans said, most of opamps become unstable in unity-gain configuration if they drive capacitive load. That is the case, when you put large 10uF C2. According to TL072 datasheet, there is 128 Ohm output resistor, which makes RC-circuit in conjuction with C2. It makes additional phase shift, which at some frequency converts negative feedback into positive, leading to oscillation. Why do you want to use C2?

If the purpose is to additionally filter noise, then, as @brhans suggests, use resistor in serise with C2. Yes, it will increase DC ouput resestance of splitter, but it will decouple large capacitance from feedback looop and removes oscillation. Also, you can select opamp that is 'able to drive unlimited capacitive loads', AD826, for example.

\$\endgroup\$
  • \$\begingroup\$ I think that C2 is necessary to eliminate any induced noise, is it not? The low-ohm resistance in series with the C2 seems to solve the problem for me, however being not a perferct solution. AD826, by the way, still makes the spike appear, but at 300k frequency rather than 20k. Still not a perfect solution. \$\endgroup\$ – sx107 Jun 8 '18 at 12:11
  • \$\begingroup\$ @sx107, what is 'induced noise'? In my opinion, C2 will filter noise, produced by U1 opamp (by the way, AD826 has lower voltage noise than TL072). If it not necessary (noise already acceptable for the application, and load doesn't make current pulses) do not use C2 at all. Also, I guess the spike at 20 kHz (or at 300 kHz) may appear as simulation artefact. Netherless, most opamps in unity-gain configuration do tend to oscillate being loaded by large capacitance. So, real measurements are worth doing them. \$\endgroup\$ – Eugene K Jun 9 '18 at 18:24
0
\$\begingroup\$

That cap is potentially very important, as it reduces the effect of noise on the power supply line. In the worst case, if the output of the circuit drives a high-power load, the entire system may oscillate as the effects of driving the load feed back into the input.

As to the noise spike, I haven't a clue. However, it would be exceedingly nice if you would run the two circuits using the same op amp and the same circuit. That is, inverter vs inverter or follower vs follower. As it stands your results are not directly comparable.

\$\endgroup\$
  • 1
    \$\begingroup\$ Looks to me like we're seeing the TL072 unstable & oscillating due to the large capacitive load. I wouldn't describe a sharp spike like that as 'noise'. \$\endgroup\$ – brhans Jun 8 '18 at 2:50
  • \$\begingroup\$ @brhans I've tested this schematic with different opamps and with the TLE2426 rail-splitter IC. The results are the same, adding a capacitor between ground ("negative rail") makes that spike appear. I've just tested it with a LM317 regulator and the spike does not appear in that case, but that is maybe because Proteus simulates LM317 as an ideal regulator. \$\endgroup\$ – sx107 Jun 8 '18 at 3:00
  • \$\begingroup\$ @WhatRoughBeast I've tested the same schematic for both cases, U1 only provides the mid-rail voltage. The only difference is the C2 capacitor. U1 and U2 form the full schematic, with or without the C2 capacitor. \$\endgroup\$ – sx107 Jun 8 '18 at 3:02
  • 1
    \$\begingroup\$ @sx107 - many (most?) opamps don't like directly driving large capacitive loads. If you must have C2 then I'd suggest a resistor between it and the opamp - maybe 10R or so. \$\endgroup\$ – brhans Jun 8 '18 at 3:04
  • \$\begingroup\$ @brhans Doesn't that reduce the rail splitter's output resistance? In case of this schematic it does not change a lot, but if I will, for example, use an RC filter then it will affect the frequency characteristics a lot. \$\endgroup\$ – sx107 Jun 8 '18 at 3:19

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.