Haha! Yes!! It does!! Thank you!! I just now saw this, but hadn't been following this thread. What a pleasant surprise!
I wish that tankcircuitnoise (and everyone else) would also wander over to here:
http://www.diyaudio.com/forums/power-supplies/106648-paralleling-film-caps-electrolytic-caps.html
I haven't had time to get back to it, yet, but I'm still excited about paralleling caps without any conductors in common, so that the total ESL of the paralleled cap/conductor combinations might be reduced in the same (algebraic) way that the total resistance of parallel resistances gets reduced, which I have read (in Ott's latest EMC book) can't properly occur if there is any MUTUAL inductance; ergo my belief that paralleling the conductors as well as the caps might provide a possibly-significant benefit.
I started thinking about that for cases where multiple parallel decoupling caps might need to be used across one set of pins (when planes are NOT being used, as is the case for a lot of DIY PCBs) in order to lower both the ESL and the ESR so that the (local) "power supply" impedance seen by the power pins of an amp chip, for example, could be made low-enough, to a high-enough frequency, to provide for good-enough transient response [ f = 1 / ( (π) (risetime) ), I think ].
THEN I realized that MAYBE it would also be a good idea to use multiple parallel copies of the power and ground rail traces, ALL the way from the rectifiers to the load's power pins! i.e. EACH smoothing cap (or each subset of smoothing caps) would have its own rails, all the way from the rectifier bridge (or maybe after the first big cap) to the load. And actually, for low inductance connections, in general, couldn't we just use multiple parallel traces, or, in the case of a board interconnect, a ribbon cable with multiple conducors for each rail and ground?
Anyway, I guess those things should be discussed in the thread at the link I gave, rather than here, unless they're applicable here too of course.
Cheers,
Tom
Last edited:
I agree,
forcing currents to go via a higher impendance path can create more problems.
Thanks
-Antonio
Or read this appNote from AnalogDevices. The idea of currents taking ALL possible paths at DC, yet becoming more concentrated under the outgoing wire as frequency
increased, was a real eye-opener for me. And while reading a Schaum's Outline book
on E&M, to be told coax-cables began to be effective at about 50KHz: surprising.
Apparently near 50KHz, the inductance in paths far away from the shield becomes
enough to "encourage" return-currents to "choose" the coax-shield. Its all part of
minimizing the stored energy, I think.
Anyway, after reading this ADI Brokaw AppNote, I began to understand planes,
and even to use slits as further "encouraging" of return currents. Just do not route
ANY Any any fast signals over slits.
http://www.analog.com/static/import...14948960492698455131755584673020828AN_345.pdf
increased, was a real eye-opener for me. And while reading a Schaum's Outline book
on E&M, to be told coax-cables began to be effective at about 50KHz: surprising.
Apparently near 50KHz, the inductance in paths far away from the shield becomes
enough to "encourage" return-currents to "choose" the coax-shield. Its all part of
minimizing the stored energy, I think.
Anyway, after reading this ADI Brokaw AppNote, I began to understand planes,
and even to use slits as further "encouraging" of return currents. Just do not route
ANY Any any fast signals over slits.
http://www.analog.com/static/import...14948960492698455131755584673020828AN_345.pdf
Hi tankcircuitnoise,
I've used -ECdesigns'- stepped rectifers concept in both low and high power application with success. It can use 3 or more stepped diodes. I attach one of his diagrams:
https://picasaweb.google.com/lh/photo/rTYyMt2-7BSOVSKNLQ9H3tMTjNZETYmyPJy0liipFm0?feat=directlink
Of course parts and values can be changed to your specific needs...
What do you think?
I found it brilliant.
Cheers,
M.
Interesting. The form that I gave is equivalent to
Bw * Tr = 0.318 (i.e. = 1 / Pi )
I wonder what causes the difference.
The equation that I gave came from Henry W. Ott's famous book, "Electromagnetic Compatibility Engineering". I didn't see the derivation and am too old to remember how to re-derive it.
I'll take a ride in this long thread - a really nice one - to add my 2 cent:
If it's audible, it should be measureable.
It seems obvious that anything here should be related to the changes in the output devices voltages (Vce, Vds or Vpk) due to the finite supply admittance - specially regarding the tracks and wire inductances. Transients could have even more impact. Sure, that includes the current return path and eventual decouplers too.
These voltage changes would impact the output via the finite impedance of the output devices (early effect, capacitances and the like).
Some of this effect is cancelled by negative feedback, but this is also finite. And there's also the possibility of any of these impedances / admittances to be non-linear, adding insult to injury.
I was just thinking about a way to measure this. Sure, distortion, spectrum analyzers and stuff. But also there should be a way to provoke the effect.
Maybe a good way to test this would be adding series resistors and/or inductors to the PS rails - between reservoir capacitors and rail decouplers and/or between decouplers and output devices - and watching the effect with a test signal. The test signal could be a sinusoidal, where the distortion effects can be more easily measured, or a damped sinusoidal, to get the effect of a transient.
But it would tell a lot to watch the rail voltage right before the output device (collector, drain or plate). I'd bet some inductance here would cause a short and deep drop. If driven far enough, this drop would disturb the output device quite a bit. I wouldn't be surprised if some conditions actually saturated the output device, even when the output is far away from the rails.
Sadly, I don't have the resources yet to do this test. But, just in case someone luckier has good stuff and time in his/her hands, and wants to figure this out, here goes my suggestion.
Best regards all,
Emerson
If it's audible, it should be measureable.
It seems obvious that anything here should be related to the changes in the output devices voltages (Vce, Vds or Vpk) due to the finite supply admittance - specially regarding the tracks and wire inductances. Transients could have even more impact. Sure, that includes the current return path and eventual decouplers too.
These voltage changes would impact the output via the finite impedance of the output devices (early effect, capacitances and the like).
Some of this effect is cancelled by negative feedback, but this is also finite. And there's also the possibility of any of these impedances / admittances to be non-linear, adding insult to injury.
I was just thinking about a way to measure this. Sure, distortion, spectrum analyzers and stuff. But also there should be a way to provoke the effect.
Maybe a good way to test this would be adding series resistors and/or inductors to the PS rails - between reservoir capacitors and rail decouplers and/or between decouplers and output devices - and watching the effect with a test signal. The test signal could be a sinusoidal, where the distortion effects can be more easily measured, or a damped sinusoidal, to get the effect of a transient.
But it would tell a lot to watch the rail voltage right before the output device (collector, drain or plate). I'd bet some inductance here would cause a short and deep drop. If driven far enough, this drop would disturb the output device quite a bit. I wouldn't be surprised if some conditions actually saturated the output device, even when the output is far away from the rails.
Sadly, I don't have the resources yet to do this test. But, just in case someone luckier has good stuff and time in his/her hands, and wants to figure this out, here goes my suggestion.
Best regards all,
Emerson
The main issue for me in this case is the fact, that I don't find model circuits/equivalent circuits (respectively P-Spice models) for the main transformers, electrolytic caps and the mains like this
Transformer - Wikipedia, the free encyclopedia
and this:
http://www.scholar.de/studenten/lernmaterial/download/trafoersatzschaltungen.pdf
If I would have this for each transformer and other power supply parts of an certainly amp model so as the individual curvature include clipping/distortion effects of the individual mains socket in the wall, it would be an easy task to intestigate the different effects and reasons for the sonic differences while listening tests by a p-spice simulation.
Last edited:
I agree, but I don't think we need to go that far to find something.
What I believe that should be found is the relationship between rails sagging and rising distortion. Surely, the transformer, the mains, etc., all affect sagging in a "long" fashion (>~10ms), by their ability to recharge the reservoir caps. But, in the end, it's just sagging - if the rails drop 1V, the output devices will have 1V less, no matter which part of the supply chain lost this V.
So my point of view is that, if some transformer or capacitor is said to be better, my guess is that it reduces sagging, so output devices suffer less.
Or, from another angle: whatever good the cap/transformer/rectifier/mains combination is, we can make it 1 ohm worse, or 1mH worse, or both, and already have a clue of the mechanism Bob is searching. If you're simulating, you can also make it 1 ohm or 1mH better...
With the complete models, then we can tell how much.
Best regards,
Emerson
Here is one that I used:
Spice Component and Circuit Modeling and Simulation
That link has my downloadable LT-spice (text file) power transformer model, and a link to the paper that I used to derive it.
I made it as simple to use as possible. It calculates the derived transformer parameters from some simple transformer measurements that can be taken, which are then entered in the model.
Last edited:
Let's compute the PowerSupplyRejection of an Emitter Follower, to be modeled
as Rout of the transistor in parallel with gm*vbe Ie_source, with that R+I driving
an 10_ohm load.
Assume an Early voltage of 150 volts (that is, all the transistor curves have a tilt
that intercepts at 150 volts. Let's run the transistor at 50 volts and 5 amps (that
is equivalent to 10_ohm load). The tilt of the transistor output VI is computed as
(150volt+50volt) / 5amps = 200volt/5amp = 40_ohms.
Our device model is thus
--------*--------*--------------------- V_60DC
..........| .........|
.....40_ohm ..big_gm
..........|......... |
Ampout*-------*-------------- 10_ohm load.
And we have a 5:1 voltage divider (we just ignore the big_gm)
Thus 10 volts of V_60DC becomes 2 volts at the speaker.
With P = V^2/R, that's 2*2/10 = 0.4 watts of whatever trash is on rail.
With 1 volt of ringing caused by 120Hertz diode_turnoff (a few microseconds),
impulsing the series resonant network of transformer inductance and rectifier
junction capacitance, we get that trash attenuated only by 5:1 and then
appearing across the speaker. That powerlever is 0.2 * 0.2 / 10 = 4 milliWatts.
Cure? lots of VDD quieting,
and/or cascoding the output emitter-follower with another device a fixed
voltage above the collector. That keeps the output device at a constant Vce,
even as V_60DC experiences lots of trashing from power-line, rectifiers,
or from the other audio power channel's current demands.
tank
as Rout of the transistor in parallel with gm*vbe Ie_source, with that R+I driving
an 10_ohm load.
Assume an Early voltage of 150 volts (that is, all the transistor curves have a tilt
that intercepts at 150 volts. Let's run the transistor at 50 volts and 5 amps (that
is equivalent to 10_ohm load). The tilt of the transistor output VI is computed as
(150volt+50volt) / 5amps = 200volt/5amp = 40_ohms.
Our device model is thus
--------*--------*--------------------- V_60DC
..........| .........|
.....40_ohm ..big_gm
..........|......... |
Ampout*-------*-------------- 10_ohm load.
And we have a 5:1 voltage divider (we just ignore the big_gm)
Thus 10 volts of V_60DC becomes 2 volts at the speaker.
With P = V^2/R, that's 2*2/10 = 0.4 watts of whatever trash is on rail.
With 1 volt of ringing caused by 120Hertz diode_turnoff (a few microseconds),
impulsing the series resonant network of transformer inductance and rectifier
junction capacitance, we get that trash attenuated only by 5:1 and then
appearing across the speaker. That powerlever is 0.2 * 0.2 / 10 = 4 milliWatts.
Cure? lots of VDD quieting,
and/or cascoding the output emitter-follower with another device a fixed
voltage above the collector. That keeps the output device at a constant Vce,
even as V_60DC experiences lots of trashing from power-line, rectifiers,
or from the other audio power channel's current demands.
tank
Last edited:
How much more likely does it become that such ideas gain credence when otherwise respectable and respected designers treat them as worthy of discussion?
No reason at all.
Yes. People will say anything to get attention. Quod erat demonstrandum.
Tank,
Hard to ignore the gm*Vbe dominant term with a followers 100% feedback.
Once the output moves there will be a corresponding current correction as a result of the changing Vbe.
Interesting to try and determine the audible effects of power supply sag, if it were so straight forward then just throwing power at it should improve it (more capacitance, regulators or cascodes). There are for example re-charging effects taking place after a burst especially with multistage filters. But I suspect there are other areas which are possibly even more adversely effected, for example Vbe multipliers or any circuit where there can be a large capacitance across nodes which get significantly perturbed during large slow transients.
Fun stuff to think about, we just need easier day jobs.
Hope this helps
-Antonio
Hard to ignore the gm*Vbe dominant term with a followers 100% feedback.
Once the output moves there will be a corresponding current correction as a result of the changing Vbe.
Interesting to try and determine the audible effects of power supply sag, if it were so straight forward then just throwing power at it should improve it (more capacitance, regulators or cascodes). There are for example re-charging effects taking place after a burst especially with multistage filters. But I suspect there are other areas which are possibly even more adversely effected, for example Vbe multipliers or any circuit where there can be a large capacitance across nodes which get significantly perturbed during large slow transients.
Fun stuff to think about, we just need easier day jobs.
Hope this helps
-Antonio
Hi Emerson,
Sorry I took so long to get back to this thread with some thoughts. Darned taxes - yes, I'm a procrastinator. I see you've already gotten some good answers.
What you have said is largely on target. Since this thread was started, I've given power supplies a lot more thought and written some of that down in my book "Designing Audio Power Amplifiers."
Early effect in the output stage definitely cannot be ignored, but is substantially mitigated in many Triple output stage topologies. This is yet another reason to use Triple output stages rather than Double output stages.
Sadly, many amp designs suffer from poor power supply rejection, often unnecessarily. This opens the door to ripple and other garbage on the supply lines. Clamping the signal swing with Baker clamps before the output hits the rails also helps mitigate the sound of a poor power supply when clipping happens.
In John Atkinson's reviews, I like the fact that he usually shows 50 Hz spectral analysis and am often amazed at how many significant harmonic and mains lines there are in the spectrum.
The thing with high-end reservoir capacitors and such is that what they presumably do in helping the sound should be measurable on the rails in many cases if properly measured. Equally important, If instead of one 50,000uF cap you use two 25,000uF caps in a Pi network separated by as little as 0.1 ohm you will do far more good than a single exotic reservoir capacitor. Of course, reservoir capacitors should usually be bypassed with a smaller capacitor that is more adept at filtering high frequencies. A case can even be made for shunting the rails with a heavy Zobel network to deter resonances (electrolytics themselves can exhibit resonances).
I believe that if one does a good job in these respects and a good job managing RFI on the line side, there is little need for any exotic power line cord (whose use is often of questionable value in the first place).
Cheers,
Bob
There seems to be plenty of room to discuss still.
I've just finished reading all articles someone (sorry, can't find the original post anymore) brought here a few years ago:
More about decoupling and bypass capacitors, power-supply rails, models, and their interactions (Part 5.6)
I think this brings us a lot closer to the answer. As expected, it's all about the supply changing and the amp letting the change reach the speakers. The article makes me think the real issue is not if the rails sag too low, but too fast. Well, of course too low is an issue, but too obvious, and not in the scope of Bob's question. What I can sum up:
1.Output stages, specially AB/B class ones, draw quite complex current waveforms from the PS, with lots of high frequency content.
2.Caps, tracks, wires, etc., inductances and capacitances form resonant circuits that, if not correctly damped, will ring after these high frequency components.
3.Amps are increasingly susceptible to rail's higher frequencies stuff. We have more parasitic capacitances and less feedback (if there's any).
Apart from the resonances mentioned in the articles, I would add the series inductance between the output devices and decoupler cap (or anything else closer to the output devices).
Did we get something?
Best regards,
Emerson
Hey, good to see you're watching, and the looooooong thread with its several links seem to point people to a good direction!
This is in my list
. A friend recommended this one and, after reading the table of contents, I decided it's a must-have.Yes. And I guess this can be reasonably simulated, since most cap differences lie in ESR, ESL and leakage. Though some present nonlinearities too, which would be harder to model - but, IMHO, seem lest important than ESR and ESL.
Maybe further than deterring resonances, flattening out PS impedance to a couple straight lines (1st order down at LF, 1st order up at HF, plane in between), to satisfy the most avid?
Best regards,
Emerson
What is the best power supply design for packaging 3-amplifiers in one chassis?
I build 3-channel amplifiers in one chassis driven by a 3-way digital crossover in order to direct-drive one 3-way speaker: woofer(20-80hz); midbass(80-1.4KHz); tweeter. The driver circuit topology is identical for all three amps. The number of output transistors varies by function: 8-pairs for woofers; 4-pairs for midbass and tweeter.
I currently have a separate electronically regulated +/- 52V power supply on each amplifier driver PCB, and one large 1,500VA +/- 45V shared CRC supply common-ground shared for all output stages. The output stage is on a separate PCB screwed to the heatsink. The driver PCBs are on 1.5" stand-offs attached to the output PCB in order to isolate high output current noise and heat. This packaging also makes it easy to exchange experimental driver circuits. Heavy wire creates a common star ground on the CRC ground bus bar.
Lately I have been wondering if a separate L-C for each amplifier coming off one common rectified C would produce cleaner, more isolated sound. (common_C) + (local_LC) filtering.
What is the best power supply design for packaging 3-amplifiers in one chassis?
I build 3-channel amplifiers in one chassis driven by a 3-way digital crossover in order to direct-drive one 3-way speaker: woofer(20-80hz); midbass(80-1.4KHz); tweeter. The driver circuit topology is identical for all three amps. The number of output transistors varies by function: 8-pairs for woofers; 4-pairs for midbass and tweeter.
I currently have a separate electronically regulated +/- 52V power supply on each amplifier driver PCB, and one large 1,500VA +/- 45V shared CRC supply common-ground shared for all output stages. The output stage is on a separate PCB screwed to the heatsink. The driver PCBs are on 1.5" stand-offs attached to the output PCB in order to isolate high output current noise and heat. This packaging also makes it easy to exchange experimental driver circuits. Heavy wire creates a common star ground on the CRC ground bus bar.
Lately I have been wondering if a separate L-C for each amplifier coming off one common rectified C would produce cleaner, more isolated sound. (common_C) + (local_LC) filtering.
What is the best power supply design for packaging 3-amplifiers in one chassis?
Very nice Tom! I've down loaded it and will play with it over the next few days!
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.