Feedback cap voltage?

This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I've quite often seen that the electrolytic capacitor in the negative feedback circuit (to ground) revolves around 220-470uF, most commonly rated at 16V, and recommended to be low-ESR if possible. I've gathered a nice "stockpile" of Rubycon and Sanyo low-ESR electrolytics off some mainboards, but they're all rated at 6.3V.

What would be the risk of blowing the caps in this situation? If my thinking's correct, the voltage across them sould be vaguely equal to the amplifier's input signal voltage...

Thanks in advance,
I've never measured it, but your thinking seems reasonable. Doug Self uses doubled diodes in both directions across the cap, just in case the voltage exceeds the rating. That would certainly happen if something in the amp failed and the output went to the supply rail. At least the cap wouldn't explode. IMO, very low voltage electrolytics are to be avoided. I've serviced a lot of test equipment and audio stuff, and it always seems the high value 6.3V caps have failed first. Either they have high DF (lots of ESR), or leakage, or the value is way off. Even new ones might not be that great. Whenever I see something like 500uF@6.3V I check it very carefully and am often rewarded with a quick repair.
Joined 2007
Paid Member
Hi, No AC should appear across the cap unless you are testing at very low frequencys when the reactance of capacitor becomes significant. The average D.C. voltage is zero EXCEPT under fault conditions.It is permissable to connect reverse biased diode across cap to protect from failure if the output were to put a negative voltage across the cap.The feedback resistors would normally limit any current to safe value but it is good practice to design so that all eventuallities are covered.The voltage seen by the cap also depends on the value (ratio) of the feedback network.All the above also assumes a normal topology on split supplies i.e. not a single rail design.
Given the functioning of a differential pair and negative feedback (in a "conventional" split-supply solid-state amp), i'd expect that the voltages at the inverting and non-inverting input to be quite similar, if not identical.

Thanks for the 2-diode tip, Mr. Hoffman :smash:

The caps I have are 1000-1200-1500uF low-ESR Rubycon ZL/YXG series and Sanyo WG series, recovered from a few Asus socket-370 motherboards.
Depending on the impedances in the feedback network, I would sidestep the question and just use a film cap. I did this to my power amps recently, using a cheap and cheerful polypropylene intended for speaker crossovers, and it was one of the most cost effective upgrades I have ever done - the whole spectrum sounded better, but the bass was totally transformed. On one track that I know really well ("She's Lost Control" from "Unknown Pleasures" by Joy Division) I noticed a sub-bass synth part that I had never been aware of before.
In many amps, you don't need that a big a value, if it is a film cap.

Large values are used to reduce the distortion contribution from the electrolytic, but films have several orders of magnitude less distortion.

For example, if the feedback resistor in series with the cap is 1k (a very typical value), 22uF gives a 22ms time constant, which would be -3dB at 8Hz. 22uF costs a few dollars each, and is maybe an inch in diameter and two long.

If you are making a subwoofer amp, or care about the last fraction of a dB of low bass response, maybe you need an octave more extension, but that is still practical and cheap. But for most purposes, -3dB at 8Hz is just fine, and getting rid of the mud and grunge in the bass is so musically rewarding that you don't miss any roll-off - indeed, in my case the bass sounded much more powerful, as I could hear what notes were being played!

If you have an amp where the feedback resistor is much lower in value, then I agree that this tweak may be impractical.
Conrad Hoffman said:
That brings up another question- Why are the divider values as low as they typically are? It seems like they could be a factor of ten higher without causing a big problem, and the cap could thus be smaller. Yet, almost every amp I've seen uses something like a few kohms and a few hundred ohms.

The feedback network must have low impedance, because of noise rejection.
Series feedback resistor is the second noise entrance after LTP degenerative resistors ... if you use any
the input devices have Cin that causes a destabilizing zero in the 1/β curve, low ||Rfdbk at the inverting input keeps this parasitic's corner frequency well above the loop gain intercept, a small C or series RC on the +in to gnd also improves diff pair front end speed viewed from the feedback loop perspective

a secondary reason could be noise but not many line level sources are even as quiet as a 1 K Ohm R

a final design consideration is that balancing -in ||Rfdbk with +in Rsource gives a minimum in the distortion caused by common mode modulation of the input diff pair C, a properly tirmmed lead cap in the feedback network can cancel the 1st mentioned zero if you need to increase ||Rfdbk in pursuit of matching a higher source Rin at audio frequencies

If electrolytic feedback cap is needed, nonpolar construction is known to have less measurable distortion, also "supersizing" - putting the highpass corner F way below 20 Hz reduces distortion as well as phase shift
single inverse parallel diodes would allow peak voltage across the DC blocking cap of just 600mV or so.
This is equivalent to about 400mVac of signal voltage.

You may be better using a series pair of inverse parallel diodes (4 in all) to allow for a maximum of upto 800mVac of signal voltage across the capacitor.

Check the dissipation in the NFB resistors if the output goes to rail voltage and check the voltage on the base of the inverting input for this condition when the diodes may be passing. I guess a 1k0+47uF will probably be bypassed @ about 10Hz for output voltages at near peak with a series diode.
Could someone simulate this condition?

An alternative is to build it and measure the signal voltage across the blocking cap for the maximum output condition. A load does not need to be connected to check this.
another common "fix" for the problem of latched output causing the release of smoke from electrolytics is to use 2 lytics back to back. of course you have to double the values (a pair of 220uf caps if you want 100uf) and the voltage ratings of the caps should be equal to or higher than the rail voltage. make sure to use a 1uf or 0.1uf bypass as well, since off-the-shelf ESR is somewhat proportional to voltage rating (a 1uf/50v lytic usually has an ESR of about 1 ohm when new, but a 1uf/250v lytic has an ESR of about 6 or 7 ohms or more) (it pays to have an ESR meter...)
a single nonpolar electrolytic will have lower distortion, less inductance than 2 series polar electros - the back-to-back electro sugestion is way beyond its "use by" date today - read Batemann's Capacitor Sound articles

a parallel film cap is a complicated issue, at some point there will be a resonance between the big electro's inductance and the added film C which can cause an impedance peak above what you would see with the electro alone
the parallel resonant peak, long leads, poor layout can render the added film usless or even harmful to high frequency stability - you need the electro's and the film's impedance curves to even sart the analysis
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.