@hifitown You mention reforming ancient paper caps above. What is your empirical data and general experience on this process?
I have tried leaving leaky paper caps with DC voltage applied for a week and the DC leakage doesn't go down and quite often rises!
I save such rejects for AC applications where they aren't rejects anymore but possibly decent crossover caps, per the thesis of this post.
In such cases, my evaluation moves into the subjective realm of musical/sonic aesthetics and away from the cold numbers of lab test gear.
Are you using AC to reform? Perhaps relying on a different metric than DC leakage to track progress?
Hey Joe,
This thread is getting long and I think I may have gotten some of your messages mixed up chronologically. I read that amazing response to my post with the empirical measurements and I need to write a proper response to that, because I mostly agree with what you were saying.
Anyway, to answer this particular question, this is an important one because it dives right down into the ethical value and practical usability of a lot of vintage Western Electric gear, especially crossover networks, which is what this thread was originally aimed at.
You’re totally right about reforming capacitors. This is sketchy territory, and I’ve got to say Walt was generally not a fan of it either. However, over the years, both other experimenters and even the vintage computer crowd have demonstrated that reforming electrolytics, especially wet and dry electrolytics in power supplies, can absolutely work under the right conditions. But that is a high-voltage DC phenomenon. This distinction is extremely important.
What I did not post in my empirical findings were some of the interesting measurements I encountered while testing a completely leaky Western Electric 437A paraffin paper capacitor. On a Sencore at voltage, the thing was basically shot. Just like you said, you could probably stand there and “reform” it all day long and it still wouldn’t survive much above a few volts DC before leaking like a sieve.
But here’s the interesting part.
These capacitors are sitting inside some of the most expensive and revered crossover networks ever made, like the WE 757A networks, all soldered together in arrays specifically to keep ESR low, and people are still using them every day without touching them. So clearly something different is going on in this application.
When I actually put one into circuit and listened to it in a normal crossover role, for example roughly 4 µF feeding an Altec 802 horn driver as a high-pass, it worked fine. More than fine, actually. It sounded pretty damn good.
So why is this possible?
Well, the capacitor in a crossover is often only blocking a few volts of AC, not sitting under continuous high-voltage DC stress like it would in a power supply or coupling application. Completely different use-case scenario.
When I threw that same capacitor onto DATS and compared it against my control capacitor, which was a brand-new Solen polypropylene from Parts Express that leaks
nothing, the Western Electric measured horribly by conventional standards. It had drifted upward by roughly a full microfarad and would have been completely unacceptable in a high-voltage DC application.
Originally, Western Electric rated those paraffin caps fairly conservatively, around 200 VDC continuous / 250V AC if I remember correctly, but they were never intended to sit continuously at high DC potential. These were essentially coupling and signal capacitors.
Anyway, the 437As are frequently some-leaky. You see them sold that way on eBay all the time. You absolutely cannot assume they are “good” simply because they are NOS.
But for crossover use, apparently that leakage often does not matter much.
What fascinated me was the actual measured behavior. The reason I eventually formed those conceptual charts that I posted was because I could not directly visualize the actual action of the leaky capacitor itself. But the measurements strongly s
uggested that the leaky WE cap was producing the smoothest and most relaxed transition behavior of all three extremes.
At the opposite end was the Solen polypropylene, which measured almost like a straight wire electrically.
There is clearly some lower threshold where the capacitor becomes too compromised. But what surprised me is that these Western Electric paper caps continue functioning remarkably well in crossover or filter service even when they would completely fail a conventional leakage test at anything much
above maybe 20-30 volts DC, if that.
Fortunately, in crossover use, you almost never even see voltages anywhere near that, even at peaks.
Yes, sonically, the effect is fascinating. The tonal balance of the leaky capacitor is actually excellent. Warm, but not overly warm. Extremely easy to listen to. No harsh edge whatsoever.
Compared to something in the middle category, like an absolutely perfect oil capacitor or, beyond that, a good 1960s Mylar film cap, the severely leaky, beyond-saving, completely failed wax capacitor sounds a little bit less exact. Possibly slightly lower in apparent resolution, depending on the application, and very much more relaxed around the edges. You could almost describe it as scuffed or feathered, just like the chart tries to imply.
The only real way to hear these differences is exactly what you’re doing: swapping capacitors directly in and out and listening.
Honestly, I encourage you to try exactly this experiment yourself. Take a truly leaky capacitor, the kind that absolutely will not reform properly under DC, but one that still retains roughly correct capacitance value. For example, a nominal 4 µF cap that has maybe drifted to 5 or 5.5 µF but is not completely gone or shorted of course.
As long as the capacitor is still actually functioning, I think you’ll hear exactly what I’m talking about.
And yes, to answer your question directly, in low-voltage crossover AC use, leakage simply may not matter nearly as much as people assume. Reforming itself, however, is fundamentally a DC electrochemical process.
I don’t believe AC can reform a capacitor in the same way because of the reversing polarity involved, although I could certainly be wrong and that’s another rabbit hole worth investigating. (Beefheart?? Where are you??)
Isnt this stuff already known, just not ibn the audio domain?? Yeah I think so.
The engineering term I’ve been loosely applying to some of this is “slew rate.” I don’t think that is necessarily the proper term in this exact context, but it points in the right direction. Engineers certainly deal with rise time, fall time, recharge and discharge behavior, and switching behavior in capacitors in power supply design. That has nothing directly to do with this type of audio crossover use, but if engineers have to take the charge and discharge action of a capacitor into account in one context, then I don’t think it is crazy to suspect that some low-frequency audio-related version of that behavior can also matter here.
So we’re not really talking about superstition, voodoo, or mojo. We’re talking about electrical action. DATS gives us a small taste of this with its low-voltage, in vitro testing, but putting it into a really useful visualization for the audio spectrum is still difficult.
That is where I think 24-bit FFT analysis may eventually knock it out of the park.
Why 24-bit? Well, 8-bit FFT has been around forever. 12-bit FFT has been around too, but historically that meant
very expensive test equipment, and stil an order of magnitude lower in resolution.. Now almost anybody can put together a modest PC-based rig and do high-resolution 24-bit analysis. I think that is what may eventually blow a lot of the skeptics, critics, and naysayers out of the water on things like this, whether we’re talking about wire, tubes, or capacitors.
At that point, it may be all over but the shouting. The only argument left will be the same one we are having now: what actually sounds best?
And that’s where I think there is some educating to do, if you will. The overly bright sheen of modern gear using electrically perfect components is not necessarily the more accurate sound.
I really feel like Bell Labs would have understood this. If you look at speakers like the 755 and 728B, they were not trying to create etched, hyper-detailed, artificially bright sound. They were trying to deliver presence, warmth, and intelligibility. That is not inaccurate or untrue to the signal being fed to them. In many ways, it may be exactly the opposite.
I hate having to explain all of this purely in words because I can actually show the measured difference between a leaky Western Electric capacitor and a perfectly functioning polypropylene capacitor with ordinary modern test equipment.
And honestly, DATS itself is still fairly primitive. I suspect moving toward higher-resolution 24-bit FFT-based analysis on a PC rig is eventually going to make it much easier to directly visualize what these capacitors are actually doing to the audio signal itself, rather than only viewing them through ESR, phase angle, and other indirect electrical metrics.
![IMG_20260620_184128301_HDR[1].webp](/data/attachments/121/121898-94eeda8ab7cebaa783795a6704c7e662.jpg?hash=t1w5PRzB7n)
![IMG_20260620_184218938_HDR[1].webp](/data/attachments/121/121899-19e4b006b743b910a9f3c05e17923f3f.jpg?hash=7ZhK1HtSxy)
