I'm sure you're right about using a non-polarized cap but they cost more than the whole off-the-peg Chinese/Indian crossover that I'm trying to hack!
In an active circuit you can bias the mid-point of back-to-back electrolytics to the appropriate rail with 1M+ resistor, putting both caps into a more linear part of their curve, though adding a small amount of leakage current of course.
As noted in various posts, an aluminium electro is a diode with a knee around 1.5V, so if operating about 1V or so there is no reason for them to ever be 'formed'.
The problem with back-to-back (seperate or integrated (bipolar) is that for 2 x 1V = 2V of signal neither is a diode so you get the series capacitance: i.e. two 22uF gives you 11uF.
But above 1.5V in any direction one becomes a diode, so you suddenly get 22uF, with a diode. I.e the capacitance varies non-linearly with the voltages we are most likely to listen to.
So for any audio work I would suggest you bypass them as much as possible with film, and think about stacking them so they never hit that 'knee' of the diode.
Recently as a feedback capacitor driven at 1k/33k = 1/33 of the output voltage of max 50V, I decided that the max voltage of 50/33 = 1.5V was better handled by replacing the 47uF (non-polar) cap with two 100uF in series (not back-to-back) and a 820n bypass (which I had spare).
Perhaps do a spice simulation or measure the max voltage across the cap in usage, and see how many volts it actually sees, and then decide how to avoid that diode 'knee', as I don't think it's an aid to audio quality 😀
The problem with back-to-back (seperate or integrated (bipolar) is that for 2 x 1V = 2V of signal neither is a diode so you get the series capacitance: i.e. two 22uF gives you 11uF.
But above 1.5V in any direction one becomes a diode, so you suddenly get 22uF, with a diode. I.e the capacitance varies non-linearly with the voltages we are most likely to listen to.
So for any audio work I would suggest you bypass them as much as possible with film, and think about stacking them so they never hit that 'knee' of the diode.
Recently as a feedback capacitor driven at 1k/33k = 1/33 of the output voltage of max 50V, I decided that the max voltage of 50/33 = 1.5V was better handled by replacing the 47uF (non-polar) cap with two 100uF in series (not back-to-back) and a 820n bypass (which I had spare).
Perhaps do a spice simulation or measure the max voltage across the cap in usage, and see how many volts it actually sees, and then decide how to avoid that diode 'knee', as I don't think it's an aid to audio quality 😀
Only temporarily, until the node where the cathodes are tied together has settled to the lowest potential. It is inelegant and it won't do musical transients any good.
Like this in fact:
(resistor is 1M)
I measured this with bias, no bias, floating and with the electrolytics in series rather than back-to-back:
So the biasing has a small effect (this is steady state, transients presumably are greatly helped). The test was at 250Hz, leading to 5dB attenuation so upto about 5V across the caps, 10% of their voltage rating.
0.0004% isn't bad around the 6dBV level (2Vrms) all things considered....
Here's some more detail on the harmonics, showing predominantly 3rd:
Last edited: