Film shunt for electrolytic cathode bypass caps?

[soapfoot]

Working on a batch of 8 REDD.47 preamps, schematic here:

http://www.ladislavbrezovnik.com/diy/dripredd47/REDD47_Schematic.pdf

I've managed to find film capacitors for every location except the two cathode bypass caps, C2 and C6, which are 100µf @ 6V or better.  Even my power supply filters are film caps in these builds-- perhaps overkill, but for future maintenance concerns alone I deemed it worthwhile.

Since I couldn't find any cost-effective 100µf film caps of any voltage (space is not a limiting factor in this case), I spec'd some Sprague Atom 100µF @ 100V (readily available from one of my suppliers) and moved on.

Re-reading Jung and Bateman's articles about DF and DA performance (and their audible manifestations) in capacitors for audio applications has gotten me curious. Since I know barely enough to be dangerous, I was hoping some of the more experienced folks here could chime in.

Jung suggests film bypasses of electrolytics used in cathode bypass applications [here, page 3], but I seem to recall hearing some say that film bypasses can actually degrade (rather than improve) performance in certain cases. I'm not prepared to trust my own intuition on the issue, so I ask the esteemed panel here:

1) Should polystyrene film bypasses to these electrolytics be used in your opinion/experience?  If so, what value?  1% of total? .1%?

2) What influence should I expect capacitor DF and DA to have in a cathode bypass application in a tube preamp?  Non-existent, mild, moderate, or large?

3) Would a pair of non-polar 220µF in series, bypassed with a polystyrene film shunt, be even better (as suggested in both Jung and Bateman)?

Thanks for any info or experience!

 

[JohnRoberts]

Arghhh I recall all the arm waving about DA the first time around .

If you look at the mechanism for dielectric absorption (distributed capacitance) it can be modeled as multiple capacitors with small Rs in series with each cap. DA only expresses when there is  a changing terminal voltage across that cap. I am not a tube guy but a bypass cap presumably does not have a changing terminal voltage.

DF represents HF behavior and could make a subtle difference in a bypass application.

Back in the '70s I did my own bench work to make sense of this stuff and found measurable differences in some applications with electrolytic caps that could be improved by paralleling that cap with a better dielectric cap. In my bench tests I found that I had to use a roughly 1/10 value in parallel to make the composite measure good. 

FWIW modern electrolytic caps have evolved to deliver much lower ESR (for use in switching PS), So I wouldn't trust my old bench tests using old capacitors (and perhaps some old articles) .

Circuit design should not be about opinions. If the electrolytic caps create a problem identify and address that.

It's bad enough to design based on our own opinions, don't design based on other people's opinions, unless they are the target customers.  8)

JR

 

[emrr]

You shouldn't 'expect' anything, though I suspect film bypasses may help with increasing ESR as electrolytics age.  You should see if you can hear it, using clip leads to take a bypass around an electrolytic in and out of circuit.  Non-polars should be 'better'.  Consider when building REDD47's that you are straying far from the original sound already with modern tweaks like this (and modern transformers/resistors); you probably already have.  Today's electrolytic will already be so much better than the original.  10% is the bypass size I've heard thrown about the most, practically speaking try whatever fits size/cost. 

 

[soapfoot]

Consider when building REDD47's that you are straying far from the original sound already with modern tweaks like this (and modern transformers/resistors); you probably already have.

Understood, and agreed. Duplication of the exact original sound of Abbey Road circa 60s is much less important to me than making an excellent-sounding tube preamplifier! These were designed as high-fidelity devices with the best parts they had available at the time, and I'd like to continue that philosophy in my builds.

 

[emrr]

The impossible question, when you can't compare, would be:

What happens when you build an old circuit with new parts?

When discussing electrolytics, it reminds me of all the unhappy customers who had their early SS Neve/RCA/etc rebuilt and thought the improvements from new caps destroyed the character they liked.  I don't at all expect this to happen with a few cathode bypass tweaks, but it's worth having in the back of your mind when spec'ing clone builds.  Abbey Road engineers might lash one up with todays parts and decide they need to change something in either the circuit or the parts selection. 

Have fun with it!

 

[soapfoot]

Agreed!  Which is also why I shy away from the term "clone." Clone, to me, implies "identical genetic material," which in my world extends all the way down to the component level!  In these builds, I'm copying the circuit, but the implementation is of course quite different (regulated DC heaters, different component types, Sowter transformers, etc).

We've built two of these already as prototypes, with regular Sprague electrolytic bypass caps at C2 and C6 being the only electrolyitcs in the units.

The performance is excellent--very much a go-to on lots of stuff over our Neves, etc-- but on a few delicate acoustic sources (really nice pianos, etc) there is just a hint of thickness in the lower midrange that isn't always appropriate. And by 'thickness' I mean just a hair of obscured lower-mid detail, maybe, in comparison to something like a 1084... "congestion" might be a better descriptor than "thickness," although it might be overstating the case slightly. I'm talking about the ability to hear deeply into details of the bass strings of a piano, for instance.

Again, just a hint... certainly wouldn't kick them out of bed on almost ANY source. On brass and woodwinds, they usually simply can't be beat by anything in our studio (Neve, SSL, API, Hardy, etc).

If I could lick that one tiny quibble, these would be fierce.  We've already decided we want to have a total of ten, hence this second batch of 8. So obviously we like the prototype units well enough. I just want to push them as far as possible...

...and I got to looking at those two big fat electrolytic caps right there, and wondering...

 

[emrr]

I'd strap films into the existing prototypes and see what you think.  I expect any differences to exist in the highs.  I don't know what the stock REDD47 size cathode bypass cap is, or if you've upsized them, but I'd also listen to them upsized 2-10x.  That would do more to improve bottom end linearity. 

 

[soapfoot]

Thanks Doug, you've really got me thinking. Doing the math for the stock value of 100µF at C2, here's what I get:

f = 1/(2 * 3.14 * Rk * Ck)

Where f = the corner frequency, Rk = the cathode resistor, and Ck = the cathode bypass capacitor.

Using the stock values of 1k for Rk and 100µF for Ck, the equation solves as f = circa 1.59 Hz

That gives us a decade and change before we're dealing with any relevant frequencies. Should that be enough to get our phase and everything right?

If I substitute 220µF instead, the math works out to about 0.72 Hz, which would give us a solid decade plus an octave (and change) before audio.

In light of these numbers, do you see a benefit to increasing C2? Sorry as I realize this thread has turned a bit from a broad conceptual question into something more project-specific.

 

[abbey road d enfer]

Thanks Doug, you've really got me thinking. Doing the math for the stock value of 100µF at C2, here's what I get:

f = 1/(2 * 3.14 * Rk * Ck)

Where f = the corner frequency, Rk = the cathode resistor, and Ck = the cathode bypass capacitor.

Using the stock values of 1k for Rk and 100µF for Ck, the equation solves as f = circa 1.59 Hz

That gives us a decade and change before we're dealing with any relevant frequencies. Should that be enough to get our phase and everything right?

If I substitute 220µF instead, the math works out to about 0.72 Hz, which would give us a solid decade plus an octave (and change) before audio.

In light of these numbers, do you see a benefit to increasing C2? Sorry as I realize this thread has turned a bit from a broad conceptual question into something more project-specific.

Your basic calculation is wrong, because the capacitor has to bypass both the cathode resistor AND the cathode output resistance, which it sees in parallels.
The cathode output resistance is more or less equal to 1/S. For a 12AX7, that's about 700 ohms. So you must enter 500 ohms in your formula. That's the reason why the Ancients said "target at 1/10th the desired -3dB frequency". Don't worry about phase. You're dealing with minimum-phase circuitry here, so phase is just a by-product of frequency response. get the frequency response right and you will get correct phase response for free.

 

[soapfoot]

I see, thanks for your help.

So I'd need to figure out transconductance for the EF86, right?

Then 1/S = the cathode output resistance, which is seen in parallel to the 1K bias resistor.

Transconductance in an EF86 looks to be 2.2 micromhos, 1.75 minimum, in typical circumstances, according to the data. I'll get to work.

 

[soapfoot]

My math for an EF86 assuming 2 micromhos transconductance:

1/0.000002 = 500k

Placing this in parallel with the 1k gives us 998 ohms, which seems functionally identical to the 1K.

Am I missing something, or does the low transconductance of the EF86 make its cathode output resistance fairly irrelevant?

However, transconductance for the ECC88 is listed as 6750 minimum. That's somewhere around 150 ohms, which parallel with the 200 ohm biasing resistor gives us about 85 ohms total impedance.

With the stock 100µF bypass cap, that works out to be around 18.5 Hz as the cutoff. That seems borderline to me.

Increasing to 220µF gets us down to 8.4 Hz on the output stage, which seems safe.

 

[abbey road d enfer]

My math for an EF86 assuming 2 micromhos transconductance:

2.2mmhos not umhos

1/0.002 = 500 ohms

Placing this in parallel with the 1k gives us 998 500 ohms, which seems functionally identical to the 1K.

 

[soapfoot]

got it!!  I'll do my math again. I really do appreciate the help.

 

[soapfoot]

1k in parallel with 500R = 333 ohms.

1/ (333 * 3.14 * 2 * .0001) = 4.78 Hz.  Seems safe enough, if my new calculations are correct. Now I'll work on the parallel ECC88 triodes

 

[soapfoot]

OK for the 6DJ8/ECC88:

Transconductance listed as 6750 µmhos minimum.  Strapping them in parallel doubles it to 13500 minimum (probably more like 14500 typical). Let's assume a "high" value of 15000.

1/.015 = 66.7

In parallel with the 200R cathode bias resistor, that puts us down around 50R.

With a 100µF cap, that puts our cutoff at around 31.85 Hz.  Could that be right? If so, I think a bigger bypass is in order!

 

[abbey road d enfer]

OK for the 6DJ8/ECC88:

Transconductance listed as 6750 µmhos minimum.  Strapping them in parallel doubles it to 13500 minimum (probably more like 14500 typical). Let's assume a "high" value of 15000.

1/.015 = 66.7

In parallel with the 200R cathode bias resistor, that puts us down around 50R.

With a 100µF cap, that puts our cutoff at around 31.85 Hz.  Could that be right? If so, I think a bigger bypass is in order!

In fact there is a correction factor due to the presence of a plate resistor, so the actual value may be closer to 100R. Anyway I would use a larger value. When the EMI designers created the REDD47, 'lytic caps were bulky so they made a compromise, counting on the help of overall NFB to improve the LF response. But today 470uF or even 1000uF caps of similar size to the original caps are available. I would choose at least 470uF with the highest life expectancy at 105°C.

 

[soapfoot]

In fact there is a correction factor due to the presence of a plate resistor, so the actual value may be closer to 100R. Anyway I would use a larger value. When the EMI designers created the REDD47, 'lytic caps were bulky so they made a compromise, counting on the help of overall NFB to improve the LF response. But today 470uF or even 1000uF caps of similar size to the original caps are available. I would choose at least 470uF with the highest life expectancy at 105°C.

I absolutely will do that. There is plenty of room.  Thanks a million for your help.  I might increase the EF86 cap overall for good measure, too.

I may try two high-temp nonpolar 470µF in series, shunted with a 2µF film for the EF86 bypass (as per Bateman's and Jung's findings), and two 2200µF nonpolar in series for the 6DJ8 (maybe I'll find some kind of 10µF film shunt to use there, too, or try a 2µF, or maybe skip the film shunts altogether).

I've also flirted with the idea of LED bias, but I'm not sure I want to bring these all the way into the 21st century, particularly since I have not tried that before and we're already sort of past the "prototyping" stage and are now in the phase of building the rest of the ten we want to outfit our studio.

 

[abbey road d enfer]

I may try two high-temp nonpolar 470µF in series, shunted with a 2µF film for the EF86 bypass (as per Bateman's and Jung's findings), and two 2200µF nonpolar in series for the 6DJ8 (maybe I'll find some kind of 10µF film shunt to use there, too, or try a 2µF, or maybe skip the film shunts altogether).

Don't do that! As much as it's justified for coupling capacitors that are subjected to zero DC, in the case of cathode res, the polarity is well defined. Non-polarised lytics are already to caps of opposite polarity in series (difference is that by construction they share a common electrode). One half of the caps would be permanently reverse-biased, which is not good. A standard high-quality polarized lytic is a better choice.

I've also flirted with the idea of LED bias, but I'm not sure I want to bring these all the way into the 21st century,

Not a bad idea per se, but it doesn't leave much flexibility in terms of voltage and variations of the valve's characteristics.

 

[soapfoot]

Good advice, thanks! I will heed it.

Do you have any thoughts on voltage rating?  Would it be good to go super overkill? It looks on most data sheets like tan∂, ESR and other measureables improve somewhat as voltage rating increases.