Replacing 220uF cap

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

Gustav

Well-known member
Joined
Jun 4, 2004
Messages
2,271
Location
DK
I am working on a PCB for this schematic drawn up for a summing mixer...

https://www.dropbox.com/s/a2ec6qx3m9djvdi/Majestic%20Buzz%20-%20schematic%20REV%201.0.pdf

The transformer is 600 Ohm, so I don't understand the 220uF NP cap,which would give a corner frequency of 1,2Hz, unless I am misunderstanding something (and not counting the 1uF cap in parallel, which I am also a bit puzzled by). I would prefer to find a film-cap, but those are pretty expensive at 220uF, so couldn't I replace the 220uF NP with a lower value film-cap (600Ohms and a corner frequency of 20HZ gives me a value of about 13uF for example).

Or am I completely misunderstanding the function?

Alternatively, some googling has revealed that replacing the NP with two polarized back to back in series and adding a small value NP in parrallel should work. I will try it out, but does anyone have experience with how this affect the sound of a circuit?

Gustav
 
The corner frequency is where the -3db point is, typically. So you don't really want to be down 3dB at 20hz. I imagine the 220uF is bypassed with a 1uF film to improve higher frequency response, since so many people are afraid of using 'lytics in audio settings.
 
getting the corner f down low is perhaps about phase consistency within the audio band. and bypassing electros with a film cap can in some cases also cause an audible "phasey" sound.

so, all in all, you'll have to "test drive" the build before comitting to it.

can you hear a call of a breadbord?
 
tv said:
getting the corner f down low is perhaps about phase consistency within the audio band. and bypassing electros with a film cap can in some cases also cause an audible "phasey" sound.

so, all in all, you'll have to "test drive" the build before comitting to it.

can you hear a call of a breadbord?

Of course - I just wanted to make sure I understood a fraction of what was going on before doing so :)

Actually, its just a pain in the ass to source the 220uF bipolars here, and if it werent, I wouldn't have thought twice about this.

Gustav
 
in absence of a true bipolar, try a "frankencap" made of two regular cheapo electros and a filmcap, like in the pic

one will be roughly 236uF, the other 201uF, but as elco tolerances are all over the place, you really shouldn't care much, just plug-and-play and see which one works out best for you.

that's the best unscientific answer I can muster

 

Attachments

  • frankencap220uF.png
    frankencap220uF.png
    4 KB · Views: 64
I would recommend the serial configuration, because in this circuit there will likely be a small but indeterminate DC offset coming out of each op-amp. (I suspect this to be the cause of the LF bump mentioned. :) )

The main reason NP caps are used is when the average DC level at one end is both + and - at different times with respect to the other, or the absolute DC polarity cannot be determined at design time. They are also used sometimes when there is no DC at all to bias the cap. The serial configuration should work in both situations, but the parallel would not work well in the former case.

--mark
 
In this case you can use any polarized one, and put them with both + to TX, see that primary of the TX is floating so nothing happens here, is just like having the frankencap with the TX inside.

To R3 and R4 I'd put a choke in parallel, something in the order of 22µH I guess, they are there to deal with bus capacitance to avoid oscilations at high freq but it could not be in ideal condition, the config resistor/inductor in parallel deals with the capacitance at high freq that is where it affects.

JS
 
Thanks so much guys - Care to dissect the schematic from start to finish? I would surely learn a lot....

Even things like - why do we use resistors across the inputs, and why 22K, I'm  coming to learn it as an operational application rather than see the why. Its a bad habit on my part.


Gustav
 
why do we use resistors across the inputs, and why 22K

They are not "across" the inputs.

> "frankencap" ..., like in the pic

The parallel version will conduct DC quite well.

> 600 Ohm, so I don't understand the 220uF

It's not 600 ohms. It is reflected load (600 ohms or probably higher) *in parallel with* the transformer's inductance. Taking a simplistic approximation: 5 Henries at 20Hz is about 600 ohms, so pencil the transformer as 5 Henries.

A simple "600 ohms 20Hz" approximation leads to about 15uFd. 15uFd with 5H series-resonates at 18Hz with reactance about 580 ohms. If the series resistance is just the opamp, that's "zero" ohms. A good 600r winding may have another 30 ohms self-resistance. With 30 ohms series, Q is 19(!) and peaking is 25dB(!!) at 20Hz.

This design suggests a 220uFd cap, series-resonant at (assumed 5 Henries) 5Hz. Reactance here is about 150 ohms. If the series resistance is just the opamp, that's "zero" ohms. A good 600r winding may have another 30 ohms self-resistance. Resonant Q is about 150/30 or 5. A Q of 5 is likely to be a dB or two up at two octaves away from resonance.... your 20Hz response is *rising* not flat or falling.

It actually proposes an added 75 ohms. Series R is around 100 ohms, Q is about 1.5, which isn't a big deal if it is well below the band. 5Hz should be out of the way for most speech/music audio. However the 100 ohms series, *into true 600r load*, is over 1dB of loss. Possibly moot, possibly confusing if you line-up tone unloaded then throw 600 ohms on. (Old hands would expect this, actually 6dB sag; but new-hands grow up with negligible output impedances.)

You can get a neat Butterworth (or Chevy) response _if_ you precisely control the inductance and all resistances. But iron-core inductance is a loose thing, and output loading can be all over the place. Also an iron-core with just-enough inductance is liable to be high distortion. We "usually" aim for excess inductance, EXCESS capacitance, get the resonance far below the audio band. The impedance scaling also allows less added resistance to damp the Q.
 
PRR said:
why do we use resistors across the inputs, and why 22K

They are not "across" the inputs.

> "frankencap" ..., like in the pic

The parallel version will conduct DC quite well.

> 600 Ohm, so I don't understand the 220uF

It's not 600 ohms. It is reflected load (600 ohms or probably higher) *in parallel with* the transformer's inductance. Taking a simplistic approximation: 5 Henries at 20Hz is about 600 ohms, so pencil the transformer as 5 Henries.

A simple "600 ohms 20Hz" approximation leads to about 15uFd. 15uFd with 5H series-resonates at 18Hz with reactance about 580 ohms. If the series resistance is just the opamp, that's "zero" ohms. A good 600r winding may have another 30 ohms self-resistance. With 30 ohms series, Q is 19(!) and peaking is 25dB(!!) at 20Hz.

This design suggests a 220uFd cap, series-resonant at (assumed 5 Henries) 5Hz. Reactance here is about 150 ohms. If the series resistance is just the opamp, that's "zero" ohms. A good 600r winding may have another 30 ohms self-resistance. Resonant Q is about 150/30 or 5. A Q of 5 is likely to be a dB or two up at two octaves away from resonance.... your 20Hz response is *rising* not flat or falling.

It actually proposes an added 75 ohms. Series R is around 100 ohms, Q is about 1.5, which isn't a big deal if it is well below the band. 5Hz should be out of the way for most speech/music audio. However the 100 ohms series, *into true 600r load*, is over 1dB of loss. Possibly moot, possibly confusing if you line-up tone unloaded then throw 600 ohms on. (Old hands would expect this, actually 6dB sag; but new-hands grow up with negligible output impedances.)

You can get a neat Butterworth (or Chevy) response _if_ you precisely control the inductance and all resistances. But iron-core inductance is a loose thing, and output loading can be all over the place. Also an iron-core with just-enough inductance is liable to be high distortion. We "usually" aim for excess inductance, EXCESS capacitance, get the resonance far below the audio band. The impedance scaling also allows less added resistance to damp the Q.

Will have to read this more than once armed with google - thanks so much, PRR!

Edit:And I used the term "across",  as they are in series on each input "across the board" (Thats my rationalization, and I'm sticking to it).

Gustav
 
for the 220uF question, here's a couple of thoughts:

You have a 220uF on each side of the transformer primary - but a capacitor in one side is enough to prevent DC current from flowing through the transformer, which is what we want and need.

And as you have two capacitors effectively in series in the schematic, they are behaving like a single 110uF.

which means that if you put two 220uF standard electrolytics in series, plus to plus, you have a non-polar 110uF in effect like the original circuit. Put this in one of the sides, and omit the other - you're free of DC anyway

A fancy thing to do here would be to polarize the in-between-caps +-to-+-point by dragging it through a 1M resistor to +15V. This makes even poor electrolytics sound pretty good, and prolongs operational life. I think SSL did this a lot in their 9000-series? Or was it FocusRite-for-neve?

And btw - thanks PRR. Always enjoying complex stuff made simple.

Jakob E.
 
PRR said:
It's not 600 ohms. It is reflected load (600 ohms or probably higher) *in parallel with* the transformer's inductance. Taking a simplistic approximation: 5 Henries at 20Hz is about 600 ohms, so pencil the transformer as 5 Henries.

A simple "600 ohms 20Hz" approximation leads to about 15uFd. 15uFd with 5H series-resonates at 18Hz with reactance about 580 ohms. If the series resistance is just the opamp, that's "zero" ohms. A good 600r winding may have another 30 ohms self-resistance. With 30 ohms series, Q is 19(!) and peaking is 25dB(!!) at 20Hz.

This design suggests a 220uFd cap, series-resonant at (assumed 5 Henries) 5Hz. Reactance here is about 150 ohms. If the series resistance is just the opamp, that's "zero" ohms. A good 600r winding may have another 30 ohms self-resistance. Resonant Q is about 150/30 or 5. A Q of 5 is likely to be a dB or two up at two octaves away from resonance.... your 20Hz response is *rising* not flat or falling.

It actually proposes an added 75 ohms. Series R is around 100 ohms, Q is about 1.5, which isn't a big deal if it is well below the band. 5Hz should be out of the way for most speech/music audio. However the 100 ohms series, *into true 600r load*, is over 1dB of loss. Possibly moot, possibly confusing if you line-up tone unloaded then throw 600 ohms on. (Old hands would expect this, actually 6dB sag; but new-hands grow up with negligible output impedances.)
And btw - thanks PRR. Always enjoying complex stuff made simple
Du.uuh!  This is simple?  :eek:

My $0.02 is that the Inductance varies in level too.  #42 in Audio Transformer Inductance http://www.groupdiy.com/index.php?topic=47515.msg598916#msg598916  Look for more PRR stuff there and also a link I post to a rotten SRS transformer datasheet which illustrates the problems with mismatching.

You can get a neat Butterworth (or Chevy) response _if_ you precisely control the inductance and all resistances.
I used to do a lot of this in my Calrec mike days with the Zobels.  A juggling act cos the level, length of mike cable, phase of the moon .. all varied.  My target was Ford rather than Chevy.  :)
 
> This is simple? 
> ....the Inductance varies in level too.


Yes; but Gustav's questions suggest first-order understanding. Like he knows 600r only needs 10uFd or 50uFd, so why 220uFd?

Meaning he hasn't (yet) thought of inductance.

Let us let him pretend the transformer inductance is "constant" for a while.

The problem of transformer inductance varying with level is, IMHO, Nth-order undestanding. It hurts my brain. I really do not want to think about it.

What do we do when an important parameter is poorly controlled? We see if we can rig the rest of the system so it "does not matter".

Here in Maine we get snow. One week, 2 inches, which won't stop a Miata. Another week, 27 inches, too much even for the Blizzak-equiped Accord. How do I make it "not matter"? I have a 4WD 3-ton truck with lift-kit. And plow.

The hundreds of uFd is the electronic analog of a 3-ton truck.

The exact value isn't important. I could use a 5-ton truck or 10-ton truck, except cost and size may be problems. A 2-ton would work, but I know I could not plow the worst snow with a 1-ton Jeep (CJ3, not the monsters of today). Then the depth of the snow "would matter".

The 3-ton cost me $1,600 (it's very beat!). Many hundreds of uFd can be had for 1% of that price.

With several tons of truck or caps, the varying value of snow or inductance "does not matter".
 
we can try and reverse the xfmr on paper just for kicks,

from the Sowter site it looks like an EI 21 core as per the dims,

probably a sq stack,

Mag Met has 21 EI with a 1/2 inch tongue, so sq stack means A-cm = 1.61

34 dbu is 39 volts so lets get the turns,

B = 100,000,000 x Volts-RMS/4 x 1.61 cm x 20 hz x Turns,

steel cores run at 16,000 gauss all day, and all night long, so Faraday says that

16,000 = 25,000,000 x 39 volts/1.61 cm x 20 Hz x Turns,

B max = 16,000 = 30,279,503/Turns

Turns = 30,279,503/16,000 = 1892 Turns,

seems like a lot of turns for a 600 wind but remember we have a steel core so we need those turns to make up for a lack of perm,

now we can get Henries since we know  the core,

Mag Met says that for 21 EI,  H = 0.2431 x 10^-8 x K1 x Turns^2 x permeability,

K1 is stacking = 0.9 for 14 mil lams laced 1 x 1,

H = 0.007831947 x Perm,

now we have to guess at the perm, because it is all over the place depending on the steel and where we run it at, do we use u-initial, u-max, u-average?

inductance varies so much with freq and level, don't let it get you down, just pick a typical number, something between 10,000 and 20,000, lets try 10,000,

H = 0.00783 x 10,000 = 78 Henries,

seems high for a 600 wind, where did we go wrong?

ahh, read the fine print, Sowter is giving us level and distortion specs at 50 Hz,

do we need to re-calculate the whole enchillada? heck no, we have linear equations,

to get specs at 20 Hz we just divide, 20 Hz / 50 Hz = 0.4,

Turns = 30,279,503/16,000 = 1892 Turns @ 20 Hz,

so Turns at 50 Hz = 1892 x 0.4 = 757 Turns

remember when we said the turns count looked high? now it looks real good for a 600.

now we can get inductance, i bet it looks more reasonable also,

uh oh, inductance formula is quadratic in turns, what do we do? re calculate?

no, too lazy,  just square our 0.4 correction factor, 0.4^2 = 0.16,

78 Henries x 0.16 = 12.48, now were talkin,

so if somebody drops the xfmr during shipping, we have about 10 Henries

that will stay at 10 Henries even down to 5 hz, not so with Supermalloy,

can Sowter get 757 turns on that core? sure.

can they get twice that plus shields? we need a pri and sec, ya know,

mean length turns for 21 EI is 3 inches, so we have 757 x 3" = 189 feet of wire for the pri or sec,

what is DCR? 65 ohms? what wire will give 65 ohms at 189 feet?

just check the wire table, looks like # 35, can we get 757 turns of #35 wire on a 21 EI?

sure, can we get 1514 turns for pri and sec?

sure, #35 = 25,000 turns per sq inch, 5/16 x 13/16 widow area = 0.254 sq inch,

0.254 x 25,000 = 6,347 turns, but wait, we have copper foil  shields and yellow tape and KP and Nomex in there, and some sloppy winding machines with imperfect wire, and a nylon bobbin with flanges (paper wound? we have margins),  so we only get about 0.4 times as much copper as predicted, so 0.4 x 6,347 = 2,539 Turns

room to spare, that means we can use thick enamel and hi-pot the thing to 5,000 volts,

ok now as far as inductance and level, ever see a B-H curve?

take the derivative of any point on there and see how many different values you get,

this is why Fred Forsell hates transformers, that weird curve, non linear as heck.

add in a little air to the core and the curve changes,

go past 2000 Hz and you have no curve.


so we just over design a bit and open up a pack of ciggys, or pray. or both.

things could be worse, you could do so much heroin that you wake up in a morgue with a toe tag, this happened to Skip Spence of Moby grape fame,

"In fact he actually ODed once and they had him in the morgue in San Jose with a tag on his toe. All of a sudden he got up and asked for a glass of water."  :eek:

Happy 4th of July to whom it may apply,
 

Attachments

  • hot-dogs-grill_635x250_1372863341.jpg
    hot-dogs-grill_635x250_1372863341.jpg
    35.1 KB · Views: 1
Here's a sim with the naive guess around 22uFd, the proposed 220uFd, and several guesses for inductance (which could be the one winding at different levels or after hard shipping).

A: 20uFd 5H == +7.9dB @20Hz, +1.3dB@42Hz
B: 220uFd 5H == +0.4dB @20Hz, +0.1dB@42Hz
C: 220uFd 12.48H == +0.2dB @20Hz, +0.05dB@42Hz
D: 220uFd 78H == +0.03dB @20Hz, +0.008dB@42Hz

We often pretend audio goes to 20Hz. Debatable, but we like to see it plotted. Realistically few monitors run much below 50Hz, and few instruments below 42Hz (piano has a note 28Hz but anything less than a Bosendorfer Grand you can't hear it for the overtones; large organ is always a special-case).

Case A has "real bad" response at 20Hz, and the 42Hz error is almost significant.

Case B is a hair up at 20Hz, but in any large chain of amplifiers the -0.1dB droops in each stage add-up to negate the -0.4dB error. And that's my wild-guess of an inductance barely-good enough to call Hi-Fi and 600 ohms.

C.J.'s scratchings give other values probably more in line with "typical good" 600 ohm windings. 12H or 78H (against 220uFd) give "don't care" errors; in particular we don't-care if truth is 12H or 78H, or even 42H on Sunday and 200H on Thursday. The "over-size" "3-ton" cap has pushed all errors far below the audio band.
 

Attachments

  • L-C-out.gif
    L-C-out.gif
    15.4 KB · Views: 22
PRR said:
Here's a sim with the naive guess around 22uFd, the proposed 220uFd, and several guesses for inductance (which could be the one winding at different levels or after hard shipping).

... loadsa technical stuff
Foul PRR!  No sign of a load.  ;D

Gotta assume some load.  And if its a 600R application, need some source R too.

Gustav, you can use the load (& source) to tame the evil peaking that PRR shows.
 
Like Ricardo, I will have to say I don't find this simple, but I will revisit the first chapters in my electronics beginners books and get back to this thread with a glass of red wine the next time my wife works the hospital night shift :) On the first pass, it was hard enough understanding Ohms law, and the more that sticks and find practical applications, the more use I have of the excellent information provided here.

I finally (sort of) figured out the 22K resistors too  :-[

Gustav
 
> Foul PRR!  No sign of a load.

10K load is no difference (0.2dB @ 5Hz).

600r load takes a couple dB off the subsonic peak, but also a good 1dB off the over-all level. I call it a wash.

75r load gives lovely non-resonant curve -0.2dB @ 20Hz, but *6dB down over-all*.

Nothing foul here.

(too hot here to post all plots: here's 10K and 1K.)
 

Attachments

  • L-C-out2.gif
    L-C-out2.gif
    9.5 KB · Views: 16
Back
Top