paralleling film caps with electrolytic caps

[gorgon53]

disregarding connectionprobs something like a low impedanz "flattened" coaxial cable, "sandwiched" stripline a.s.o would be best. Paralelling will help to bring inductans further down. A cable with twisted wires alternately connected +-+- a.s.o. and the thinnest possible isolation should be good enough in most audioapplication. In low voltage applications the best "cables" can be made up of magnetwire. Pay attention to providing the smallest possible looparea at the caps connetion because this inductans can easely swamp low a otherwise low cableinductanc. In a PS with high current high-C input the looparea transformer-rectifier-first cap can induce significant strayfield. Keep it as short as possible. How critical the loop thereafter is depends mostly on load requirements. Pay attention to possible ringing and make sure that sufficient damping and "load-sided" C is provided. A "C" of 0.1uF wont necessarely cut it.

[gorgon53]

correction to #320
...reduce inductance... should read: minimize inductance...

[gorgon53]

gootee, I am all with you on your last post, and for me in DIY there is no such thing as overkill, lol.
Geez, I am working on a practically zero-impedanc tube power amp to current induced OPT distortion minimized without the otherwise needed nasty feedback from the output.
I am about to sim a totally overkill amp giving pre-transformer thd in the ppm range. If I want in reality get anything even close to what I simmed I will need first of all a totally overkill powersupply. In the process of designing such a beast I "invented" a new type of powersupply wich I will hopefully be able to load up on Duncans PSU-calculator
side for public discussion. In the limited currentrange of a tube class-B amp this PS
is capable of giving better regulation than what is possible with L-input (without ringing)
using the same caps. It has also the advantage of not producing overvoltage at low load. Furthermore it has good pf and remarkable good efficience as compared to C-input and maybe even L or 100Hz resonated L (because of the smaller resistans po.
But there are also downsides: it needs really big caps,has fairly high resonant currents and need a electrically small but mechanically only a bit smaller than L-input choke with a critical and almost dc-undepended inductanc. This means big airgap and lozza electromagnetic strays wich will need shielding. The choke is also the bottleneck and makes a industrial maufacuring process very expensiv. Thats why I decided on making it public instead of patenting. It should be well suited for certain DIY projets. Simmed
with Duncan PSU designer I was able to get a dc-regulation of around 2-max2.5 % for 100-400mA at 250V (this includes extensiv filtering to get full load ripple output to only a few hundred uV).
Overkill?
Offcourse, but why bother with DIY if you could buy something like it

[DSP_Geek]

Originally Posted by mjf

quote:For board-to-board power supply cabling (for simple boards without planes), I would now have to recommend using something like a ribbon cable, with many conductors, with many alternating V+ and Gnd pairs (if single supply), or many V+/Gnd/V- triples (if dual supply)

here is a foto. it shows a small pa - amp, the psu board (+v 0 -v) is connected with a "ribbon cable" to the poweramp boards (the +-15v are made with res and z-diodes on the preamp board).

McCormick Audio amplifiers featured (don't know if they still do) smallish power supply filter capacitors at each output transistor. The trace is a few centimetres long, if that, so inductance is minimized. The only problem would be IR drop from rectifier pulses going through the wires from the rectifier to the filter caps.

[gootee]

Originally Posted by gorgon53

gootee, I am all with you on your last post, and for me in DIY there is no such thing as overkill, lol.
Geez, I am working on a practically zero-impedanc tube power amp to current induced OPT distortion minimized without the otherwise needed nasty feedback from the output.
I am about to sim a totally overkill amp giving pre-transformer thd in the ppm range. If I want in reality get anything even close to what I simmed I will need first of all a totally overkill powersupply. In the process of designing such a beast I "invented" a new type of powersupply wich I will hopefully be able to load up on Duncans PSU-calculator
side for public discussion. In the limited currentrange of a tube class-B amp this PS
is capable of giving better regulation than what is possible with L-input (without ringing)
using the same caps. It has also the advantage of not producing overvoltage at low load. Furthermore it has good pf and remarkable good efficience as compared to C-input and maybe even L or 100Hz resonated L (because of the smaller resistans po.
But there are also downsides: it needs really big caps,has fairly high resonant currents and need a electrically small but mechanically only a bit smaller than L-input choke with a critical and almost dc-undepended inductanc. This means big airgap and lozza electromagnetic strays wich will need shielding. The choke is also the bottleneck and makes a industrial maufacuring process very expensiv. Thats why I decided on making it public instead of patenting. It should be well suited for certain DIY projets. Simmed
with Duncan PSU designer I was able to get a dc-regulation of around 2-max2.5 % for 100-400mA at 250V (this includes extensiv filtering to get full load ripple output to only a few hundred uV).
Overkill?
Offcourse, but why bother with DIY if you could buy something like it

Wow. Coool. I can hardly wait to see all of that!

[gootee]

Originally Posted by gorgon53

disregarding connectionprobs something like a low impedanz "flattened" coaxial cable, "sandwiched" stripline a.s.o would be best. Paralelling will help to bring inductans further down. A cable with twisted wires alternately connected +-+- a.s.o. and the thinnest possible isolation should be good enough in most audioapplication. In low voltage applications the best "cables" can be made up of magnetwire. Pay attention to providing the smallest possible looparea at the caps connetion because this inductans can easely swamp low a otherwise low cableinductanc. In a PS with high current high-C input the looparea transformer-rectifier-first cap can induce significant strayfield. Keep it as short as possible. How critical the loop thereafter is depends mostly on load requirements. Pay attention to possible ringing and make sure that sufficient damping and "load-sided" C is provided. A "C" of 0.1uF wont necessarely cut it.

Yup.

[gootee]

Originally Posted by gorgon53

Any means that force the loopcurrent-induced magnetic flux to travel ALL THE WAY through the smallest possible looparea will reduce inductance, stray and pickup of external fields. Reducing the loop area is the most efficient thing to do. Unluckely most caps are seldom built the way to accomplish that the best possible way.

Concur. Faraday's Law (and Maxwell's Equations) can be a bitch.

It also applies to emitting stray fields (as you alluded-to somewhere else, regarding power supply front ends' wiring, I think).

The two things that I probably repeat the most, on diyaudio.com, are to minimize ALL enclosed loop areas and to not share ground conductors.

[gootee]

Originally Posted by AndrewT

I would like to think about it from a slightly different direction.
The capacitor filter not only filters the 50Hz and it's harmonics, it also filters the HF trash that comes in along with the distorted sinewave power signal.

Little bits of carefully placed parasitic inductance and capacitance can enhance the HF filtering action.

I'll try to describe an example.
Take the first, high Zo, layout.
Reduce the inductance of the input and output cables slightly by adopting twisted wiring as close to the end capacitors as physically possible.
Place the 3 parallel caps alongside each other. Couple them together with a straight linking wire across the 3 +ve terminals and another straight wire across the 3 -ve terminals. The wiring from cap1 to 2 has a little inductance. The wiring from cap2 to cap3 has a little inductance. Add this inductance to the model.

I would expect the overall LF filtering to be similar to Gootee's first model. I would expect the HF filtering to be markedly improved compared to either of Gootee's models.

I have a 10cap bank for a KSA100. that makes up a +-75mF /channel. The 4 cap to cap links must have some effect.

Further, the effect of the first cap will have less to nothing of an influence on the sound quality coming from the speakers. The last cap in the string will have a greater influence on final sound quality. Effective decoupling at the amplifier and particularly at the output devices and VAS stage may completely dominate the sound quality now that real DC is emanating from the 10cap smoothing bank.
I think that a more accurate model of the smoothing bank will reveal advantages that Gottee's first model misses.

Sorry. No (unless I didn't model what you said). I added L and R for an inch of trace between the caps' connections to the rails (two in power rail and two in ground rail), for the un-paralleled model, and at high frequencies (1 MHz to 1 GHz), the filtering stays 0.4 dB worse than the original un-paralleled model, and 9.6 dB worse that the paralleled model. At low frequencies (up to somewhere between 20 and 100 kHz), it is 1.8 dB worse than the original un-paralleled model and 4.8 dB worse than the paralleled model.

I really don't even like to think of them as "filter caps", any more. They are "current reservoirs".

And remember that the inductances you wanted to add would get voltages induced across them, by any time-varying current that flows through them, and other voltages (which would sum with the inductive ones) induced across them because of their resistances.

It now seems that "voltage filtering" in a linear power supply is just not the best way to think about what the caps do. It's as unenlightening as thinking about the voltage, at all, in a linear PSU. The voltage is almost constant. The signal we hear is the CURRENT, from the PSU. The current is where the action is. The PSU rail voltage dips are directly controlled by the transient load current flows and the decoupling and reservoir capacitor sizes, ideally, but in reality are also significantly affected by the inductances of the conductors and the inductances of the capacitor sizes/geometries (there may be basically no inductance in most capacitors, except that which is due to conductor lengths, i.e. basically pin spacing).

We can probably also usually think that the rectifier output current just charges the capacitors, and the load current all comes from the capacitors. Yes, some of the 100 Hz or 120 Hz voltage ripple gets past the reservoir caps. But that is probably a small effect, usually, or can be made to be small, compared to the voltage fluctuations that are caused by the load's current draws across the inductance (and resistance) of the rails.

-----

Something that occurred to me, earlier today, about the 100/120Hz ripple voltage in a linear power supply: It seems like maybe we could just disconnect (or shunt-away) the rectifier output, whenever the capacitors are "full", i.e. when the voltage is already at the desired level. And then when the voltage dips, slightly, due to the load drawing some current, we could allow the rectifier to provide JUST enough current to recharge the reservoir caps. That way, NO 100/120Hz ripple voltage should EVER get past the reservoir caps. And knowing the sizes of the caps in advance and sensing the voltage differential to be corrected, as it occurs, we should be able to precisely meter-in the correct number of electrons, with the correct speed profile, exactly when needed. Of course, a good implementation would start to crack open the current valve whenever the voltage just started to dip, and voila, perfect constant DC voltage should result (AT the caps, at least). Maybe that means using a smart preregulator of some sort. We could probably also sense the voltage farther downstream (at the load), with a pair of conductors that didn't carry much current (so their sensing wouldn't be affected much by their parasitic inductances and resistances), and anticipate the effects of the inductances and resistances of the main current-carrying rails, and adjust the cap-recharge valve's dynamic response accordingly. It seems like a fairly-straightforward control-system problem, if a decent basic hardware control mechanism could be implemented. [Are there any 3rd or 4th year BSEE students lurking, who are taking (or have just taken) classical feedback control theory? A closed-form mathematical expression of the problem and solution would be nice.] Then again, who cares about anything but what happens at the point of load (and also, nothing interesting is happening when the caps are staying full, anyway)? So maybe the preceding idea should be applied at the decoupling caps or the point of load, instead of worrying about the ripple at the main caps. Or, we could do both. <smile>

[gorgon53]

gootee, tnx for pointing out the emmitting field.
By "strays" I thincking of the emmitting field. Sorry, I was a bit sloppy. When I use a foreign language I often pick the first word that comes into my mind.

It also came to my mind that running a STRAIGTH RUN unshielded conductors to and from a shunt cap emmit and pick up AFTER your "perfect" cap. So, a 90deg bend nearest after the cap connection may be a good thing. Is it of importance in audio? Depends....
Anyway, I find it good to be aware of that too. I was not, until I nosedived into industrial heater equippement. There about everything needs more attention the higher
the power times frequency product gets. Alltough losses are mostly easier to "see" or measure than at low power/low frequency and other things arent, but results wont stay unnoticed if things are not done the rigth way. Paralleling stuff can be a very "enligthning" expirience as is a "bent" coppertube when the manetic field gets squeezed
to much

[gootee]

Originally Posted by gootee

Sorry. No (unless I didn't model what you said). I added L and R for an inch of trace between the caps' connections to the rails (two in power rail and two in ground rail), for the un-paralleled model, and at high frequencies (1 MHz to 1 GHz), the filtering stays 0.4 dB worse than the original un-paralleled model, and 9.6 dB worse that the paralleled model. At low frequencies (up to somewhere between 20 and 100 kHz), it is 1.8 dB worse than the original un-paralleled model and 4.8 dB worse than the paralleled model.

CORRECTION!

AndrewT was correct.

I made a rather-egregious error, here. Instead of looking at the Load Voltage vs Frequency plot with an AC voltage source at the input, I mistakenly looked at the plot of Impedance vs Frequency as seen by the load (with a AC current source at the load).

As AndrewT asserted, adding impedance models between the capacitors does, of course, increase the effective filtering of input voltages as seen by the load. If one inch of conductor is modeled between the capacitors, with 15 nH and 1 mOhm, then at 100 kHz there is an additional 6.7 dB of voltage attenuation between input and load, and the difference increases as frequency increases.

Attached are ltspice schematics and some saved plot-settings files, so that anyone can play around with the simulations.