| UnixMan |
Hi everybody,
I was wandering about an idea. Among tube lovers, it is widely accepted that in audio gears tube rectifiers generally gives better results (better sound) than sand state rectifiers, even when coupled with massive CLC filtering.
Indeed, there are some valid technical reasons behind this. Tube rectifiers are not affected by the (direct and reverse) switching noise, plus their transfer characteristic is such as to drastically reduce the rectification noise magnitude and spectral bandwidth thanks to the "softer" and "smoother" currents that flows into the filter capacitors.
Unfortunately, tube rectifiers have all sorts of well known drawbacks, too.
So I was wondering: why not trying to "emulate" the behaviour of tube rectifiers replacing the simple diodes with some active solid state circuitry?
Thinking about it, it became obvious that using some active circuitry it should be possible to hunt for (and achieve) even better results than the mere "emulation" of a tube rectifier. Better results which of course can include some degree of embedded voltage (pre-)regulation, too.
Possibly (but not necessarily, given the different goals) this can be sort of like the synchronous rectifiers used in some SMPS.
Then something came to my mind... I remember a long time ago I have read somewhere about someone who have written an article on some magazine (maybe Glass Audio, or MJ, or some other one... or was it the JAES? can't remember) proposing to replace diodes with MOSFETs in audio PSUs. That is, possibly something similar to what I was wondering about (don't really know, I never got to read the article itself, only heard about it).
Now the questions for you guys are:
what do you think about the idea?
did anybody tried to do something like that?
does anybody know about the aforementioned article?
Thanks a lot! |
|
|
| poobah |
Seems like there would be some painless ways of doing this with special taps, "extensions" more precisely, of the power trans secondary... something on the order of 10 Volts higher (per leg) to drive the gates. The trick being that you would have to avoid waking up the body diode of the FET.
You would need a clear picture of the conducting angle to make sure you had the right amount of gate drive over the range of input voltages and load currents.
I 'spose you could test this easily enough with a second trans with a 1.1 step up ratio...or so. Actually, you could phase 2 heater windings on the secondary to pull this off... just for testing. |
|
|
| Eva |
Put a 10 to 100 ohm resistor in series with a 1N4007 diode and you'll have a nice tube rectifier.
Active rectification with MOSFETs is not an "improvement" because the body diode of the MOSFET is used, and it's much like a 1N4007 or a bit faster.
Also, diode behaviour is strongly mistified (usually by people that talks a lot about vague things like "noise" and "switching" but has never looked at the real thing with an oscilloscope). There is a thread showing actual current and voltage waveforms that I think it's worth reading :
http://www.diyaudio.com/forums/show...10&pagenumber=5
(The most interesting stuff starts at page 5 or 6). |
|
|
| poobah |
| These people need a P-channel tube... super-cooled anode surrounded by holes! - :D |
|
|
| SY |
| quote: | | Put a 10 to 100 ohm resistor in series with a 1N4007 diode and you'll have a nice tube rectifier. |
Yes. I've done precisely that and when spectra were taken, the 1N4007 with series resistor showed less noise on the rail than either a GZ34 or a diode-connected 572B when the resistor was adjusted to give an equal voltage output. |
|
|
| peranders |
Eva, you can do some tricks with Infinoen's BSP, BST IC's. One trick is to connect a powerswitch backwards and let the internal charge pumps make the mosfet conducting with the wrong polarity.
I'm not sure this will work as rectifier but it sure works as a super diode with no losses! We use it commercial products with good results.
This is a real smart idea I would never thought of myself :idea:
http://www.diyaudio.com/forums/show...3690&highlight= |
|
|
| UnixMan |
Hi poobah,
| quote: |
Seems like there would be some painless ways of doing this with special taps, "extensions" more precisely, of the power trans secondary... something on the order of 10 Volts higher (per leg) to drive the gates. The trick being that you would |
mmmh... not sure if I got it right (could you be so kind to post some rough sketch?).
BTW, if I understood it correctly it looks a lot like (it is?) a self-driven synchronous rectifier... and indeed it sounds like an interesting idea to start with! :cool:
(I guess you'd need a center-tapped secondary... or do u think this can be arranged as a bridge, too?)
But then... there would be any real advantage here? If I understood it correctly, the MOSFETs are operated as switches, thus I guess that currents (and noise) will not be much different from what you would get from a conventional (diode) rectifier.
As I said, my goals are pretty different from what have pushed synchronous rectifiers into the SMPS world. I couldn't care less about efficiency (which is what SR are all about).
I do care mostly about reduction of rectification noise, and would welcome some degree of voltage regulation at the rectifier level (so that the following massive filtering will decouple it as much as possible from the signal path).
To fully implement my idea, the active devices (be that MOSFETs, BJTs or whatever happen to be the most appropriate ones for the task) should be operated in their linear region (dissipating fairly amounts of power) at least in some part of the cycle.
In the device control network there must be some R-C to limit the current slew-rate (that is, to limit the bandwith of the current pulses), as well as some voltage reference (and comparators, etc) to gradually reduce and eventually cut-off the charging current when the voltage across the filter cap approaches the set value (which of course must be fairly less than what you would get from a "normal" rectifier).
This way I _hope_ to get something /better/ than a silicon diode with a series resistor! ;)
...which BTW will work quite well if you can afford to have a high enough R value in the circuit; I did that using an integrated bridge -> _470R_ -> 100uF (-> 30H -> 330uF) on some 380V/20mA PSU for the pre/driver section of my last "baby" (a 6C33 SET), and it is as noiseless as it could be... but try doing the same on the 200V/500mA+ PSU for the output tubes!
With this respect, I have to admit I do care a bit about efficiency... it should not end up producing more heat than the room heating system! :clown: |
|
|
| poobah |
Paulo,
I don't know if I would bother with all this work for rectification. See what SY & EVA say above. Regular diodes can work well. It is a curious idea. Also, I haven't totally thought this thru... just a crude idea and a VERY crude drawing!
I think this would work in bridge or center tapped. Play with the idea... I would use sand... |
|
|
| UnixMan |
| quote: | Originally posted by SY
Yes. I've done precisely that and when spectra were taken, the 1N4007 with series resistor showed less noise on the rail than either a GZ34 or a diode-connected 572B when the resistor was adjusted to give an equal voltage output. |
sure... see my previous post. :)
Problem is, with a fixed resistance in series with a diode you will not only limit the slew-rate of the pulse, but the max peak current as well. To be effective, this trick require to waste a sensible fraction of the power on the resistor, and will soon become impractical :hot: as the required PSU output current increase.
The active rectifier/regulator I'm proposing would dynamically adjust its series impedance during the cycle to limit the charging current slew-rate without necessarily limiting the max peak current.
On average (I hope!) it should be a bit more power efficient (though possibly a bit more noisy) than an "equivalent" D+R.
And for sure it would be much more efficient than a D+R followed by a conventional linear voltage regulator... in fact, I guess that (being partly "switching" by its very nature) it should also be more efficient than any conventional linear voltage regulator alone.
Of course, if it's followed by any serious (CLC and/or CRC) filter, it will provide no regulation against fast load fluctuations (well, unless some more tricks are used to also sense the overall PSU output voltage - that is, a feedback loop taken from the last capacitor of the chain...).
But this is a very desired feature indeed! I personally believe that any voltage regulator directly interacting with the audio circuit destroys the sound quality. :smash:
| quote: | Originally posted by poobah
I don't know if I would bother with all this work for rectification. See what SY & |
see above... it would not be just for rectification!
Moreover I have to say that when it comes to hi-quality audio, all of the PSU -including rectifiers- deserve the maximum attention.
Having done a few experiments with different (sand) rectifiers applied to a couple of amplifiers (a SET and a sand), I still can't believe how _MUCH_ these can affect the overall sound, even when they are followed by a ( by all means! :D ) massive CLC filter (330uF+30H+560uF+470uF+47uF) in one case or when they are used in a hi-NFB SS design which is supposed to (does!) have a huge PSRR. :eek:
BTW: now that the idea should be clearer, are there any ideas/suggestion for possible implementation, anyone?
P.S.: Thanks for the drawing. |
|
|
| Eva |
peranders:
This switch IC trick is nice. However, what I was trying to point out is that a MOSFET connected to a transformer as a synchronous rectifier has to be turned off just before the input voltage tries to fall below the output voltage. This forces the body diode to conduct for a brief period of time and to show its reverse recovery characteristic, thus there is no EMI improvement over a normal diode (although there is an obvious dissipation improvement).
I'm employing synchronous rectification in my 15V 120A PSU so I know this behaviour quite well. At full load I'm turning my MOSFETs off approx. 500ns before they are asked to stop conducting and I'm turning them back on approx. 800ns after they are asked to start conducting back. Body diode behaviour is crucial during these glitches (particularly due to reverse recovery) altough additional conduction losses are negligible considering that the switching period lasts approx. 16us. |
|
|
| peranders |
I'll think this type of rectification is good _only_ when you need much current at low voltages and high efficiencies. Otherwise, is it worth the trouble :scratch:
The polarity protection I pointed out is brilliant! With BTS660 you have 50-60 A super diode with no losses! Amazing! At 20 A you'll get 2.8 Watts and a voltage drop of 0.4 V. Compare this to a regular diode! Bu the nicest thing is when you have even small currents say 10 A. The other good thing is that the package is TO220, pretty small to be a 60 A "diode". |
|
|
| mirlo |
| you can avoid body diode conduction by series connecting two mosfets back to back so that the body diodes oppose each other, tieing the gates together. |
|
|
| Eva |
Employing these integrated switches for low current applications may be disadvantageous because the internal charge pump and control logic draw some current by itself.
Consider a 50mA load. It would cause approx. 20mW dissipation in a small schottky diode, but one of these switches operated at 12V and drawing 3mA on his own would be already wasting 36mW (at the expense of added functionality, though). Much higher average load currents are required to make it worth the effort, particularly in battery-operated applications. |
|
|
| poobah |
Paolo,
I have derived a topology that will achieve ALL of your requests. It borrows heavily from resonant SMPS design. I have not derived a control strategy or transfer function... and this would be most difficult. Nor do I think it would be practical to implement; I do not agree with your assesment that post supply regulation is so bad for the sound... a capacitor can fix that.
However, if you are genuinely interested in the mental exercise, I will post a drawing and an explanantion. |
|
|
| UnixMan |
| quote: | Originally posted by poobah
I have derived a topology that will achieve ALL of your requests. It borrows heavily from resonant SMPS design.
|
Wow, that's a great news! :happy1:
| quote: |
I do not agree with your assesment that post supply regulation is so bad for the sound... a capacitor can fix that.
|
so here is were we somewhat disagree... I think that some "serious" CLC or CRC low-pass filtering (with Ft < 20Hz) is needed if you want to completely "decouple" the audio circuit from the regulator, and that would be no less than a good ripple filter... ;)
Of course, how much this can really matters to the sound is:
a) very "system" dependent... e.g. it's likely that results would be different for a feedbackless SET rather than a fully balanced hi-NFB SS circuit (I have to admit that for the sake of briefness I have been quite a bit too drastic on my previous statement... :rolleyes: );
b) as for anything related to "sound quality", much of a subjective issue...
...thus arguing on that would be completely pointless. :)
| quote: |
However, if you are genuinely interested in the mental exercise, I will post a drawing and an explanantion. |
...sure I am!!! :yes: |
|
|
| poobah |
| Here is the schematic: |
|
|
| poobah |
OK,
I have shown only one side of the circuit... it must be doubled for full wave.
Conditions: Steady state, Inductor is "dry" (no current), both switches are open, we start at bottom of sine wave (theta = 0).
1. Switch 1 is turned ON sometime AFTER the primary voltage is greater than the voltage of C1. Because the inductor has zero current, the current will build in a SOFT manner. Volatge on C1 will build because of charging.
2. When the voltage at the primary passes the peak of the sine wave, it will fall until Vprimary = VC1, at this point; S1 is opened and S2 closes.
3. Current in the inductor will decay to zero, The tail of the decay will be SOFT.
4. When the current in L1 equals zero switch 2 is Opened.
Wait and then; back to step 1:
** C1 and L1 must be chosen to have a resonant frequency... maybe about 10 or 20 times more than 120 Hz.
** L1 must must run "dry" between cycles.
You are on your own to do the State-Space math -
:D |
|
|
| poobah |
Oh,
By the way, I would use a buck derived smps before I would do all this work...
:D |
|
|
| UnixMan |
| quote: | Originally posted by poobah
Here is the schematic: |
Oho!
This may be somewhat like what I had in mind... or it may be not (well, I guess it all depends on the L/C values and on what you put in the control unit! :D ).
Anyhow, I was not thinking about a resonant circuit, and in my mind the "switches" were connected to a C-input filter rather than to some LC.
Moreover, to keep noise as low as possible I was not thinking about operating the devices like "pure switches" but rather operating them in a "hybrid" fashion (with a bit of fantasy we could name it "soft-switching" mode :clown: ), i.e. letting them be in their linear region for part of the cycle (somewhat like in a "class C" amplifier).
BTW: I don't know much about all the many SMPS techniques... what are the principle of operation and main (dis)advantages of a resonant SMPS? |
|
|
| poobah |
Well... only something to consider... mental exercise; but you DID ask.
As far as resonant switchmode is concerned: I have not built any. I read the documentation for curiousity's sake. There are 4 factosr that are key: inductor currents are brought to zero, capacitor voltage are brought to zero (OR zero dv/dt), Switches are opened or closed at points of zero current or zero voltage. These factors decrease switching losses. ON SEMI & others have many write-ups and there are some devices available for some topologies. Control and math is complicated... not impossible; but NOT easy!
Cheers, |
|
|
| UnixMan |
Oops... I replyed too early! :D
| quote: | Originally posted by poobah
OK,
I have shown only one side of the circuit... it must be doubled for full wave.
Conditions: Steady state, Inductor is "dry" (no current), both switches are open,
[...]
|
OK, except that I guess you meant "secondary" where you've written "primary", I think I got it.
Really very interesting, indeed!
| quote: |
You are on your own to do the State-Space math - :D
|
yeah, that sounds like a though job... :rolleyes:
BTW: doesn't it exists any dedicated driver chip for this type of SMPS as they do exists e.g. for driving synchronous rectifiers? :radar: That could made things much simpler and avoid much of the work... :-?
| quote: | Originally posted by poobah
Oh,
By the way, I would use a buck derived smps before I would do all this work...
:D |
what I don't like about SMPSs is that (for several good reasons, when it comes to their goals...) they are operating at relatively high frequency and typically are terrible wide-spectrum noise generators... exactly the opposite of what I'm looking for... :clown:
mmmh... would it be feasible to modify one such (resonant type) SMPS to operate at much lower frequencies from a normal mains PT? :scratch: probably not... :( |
|
|
| poobah |
I don't think there any chips for THIS design. Power factor correction and efficiency is what drives the market.
Some intersting areas of study would be "ferroresonant transformers" and also "magnetic amplifers"... these utilise the effect of magnetic saturation in interesting ways. Low efficiency, but if you're building SE amps... who cares?
:D |
|
|
| Eva |
The design that poobah has proposed operates at line frequency (100Hz or 120Hz), not at high frequencies.
L1 may as well be the leakage inductance of the mains transformer, maybe with some additional inductance in series.
The control scheme won't work because S1 can't be opened until L1 current has decreased to zero due to transformer leakage inductance (that may be even bigger than L1 and there is no way to clamp it), so S2 is not required.
So S1 and S2 may be replaced by a single thyristor in place of S1, that would open automatically when inductor current reaches zero. Then the output voltage would be controlled through the firing angle of that thyristor and we would have something very similar to the PSU found in Carver's amplifiers. An alternative control scheme may be based in skipping mains cycles, this is quite common in microcontroller-driven electric heaters. |
|
|
| poobah |
Eva,
Could a thyrister be put in the center tap leg. and then just use diodes (or a tube if you must) on the "ends" of the primary. Then use simple control of the conduction angle? |
|
|
| Eva |
| Then you would need a thyristor in the center leg and one diode in each other leg, thus making the current flow through two diode drops. On the other hand, one thyristor in each leg produces just a single diode drop. Remember that a thyristor is almost like a conventional diode, but in order to turn on it doesn't only require to be forward biased but also a trigger gate pulse. |
|
|
| poobah |
Yes, of course, I am only thinking of a way to simplify the gate/trigger drive for the thyrister. 2 diodes drops is small a for a tube amp. In this way, the gate drive circuitry could be ground referenced perhaps.
hmmmmmm..... |
|
|
| Eva |
| In these high voltage circumstances, the most comfortable way to trigger a thyristor is through a small pulse transformer. A single one may be used to trigger both, since the reverse-biased one will ignore the pulses. |
|
|
| poobah |
Paolo,
Are you following along here? |
|
|
| UnixMan |
| quote: | Originally posted by poobah
I don't think there any chips for THIS design. Power factor correction and efficiency is what drives the market. |
ok, I should really go and read some docs about resonant SMPS... :rolleyes:
but if I got it right from what you have said about them, then a "standard" resonant SMPS would perfectly meet our needs ( with the extra bonus of good efficiency! :) ) as long as the resonant frequency is kept low enough.
If the switches are only operated when currents are null, I guess there should be no switching noise. Then, the only possible source of noise should be from the charging current itself. If the resonant frequency is low enough, the spectrum of such currents should be limited to low frequencies which are easily filtered and will not easily "spread out" of the PSU... am I right? :confused:
| quote: | Some intersting areas of study would be "ferroresonant transformers" and also "magnetic amplifers"... these utilise the effect of magnetic saturation in interesting ways. Low efficiency, but if you're building SE amps... who cares? :D
|
that may be interesting, too... but I guess it would require specially made "irons" that I'm afraid would be rather difficult to get. Moreover, quite likely it would be heavy, bulky and... expensive. :-$ |
|
|
| UnixMan |
| quote: | Originally posted by poobah
Paolo,
Are you following along here? |
Yeah, I'm trying to keep up... this thread is becoming really very interesting!!! :cheerful:
...it's just that you're posting faster than I can get to read it! :D |
|
|
| poobah |
Well,
I am guessing, maybe incorrectly, that your biggest goal was to achieve regulation very early is the supply. Pre-regulate rather than post-regulate.
Eva's idea is great... my inductor, L1, is already hiding in the secondary of the transformer; and is not required. This elimanates the need for Switch 2 also.
:D |
|
|
| UnixMan |
Hi Eva,
| quote: | Originally posted by Eva
The design that poobah has proposed operates at line frequency (100Hz or 120Hz), not at high frequencies. |
well, sure... but if I understood it correctly, the switches can be operated at a higher frequency, depending on the resonant frequency of the L/C. :confused:
| quote: | | [...]So S1 and S2 may be replaced by a single thyristor in place of S1, that would open automatically when inductor current reaches zero. Then the output voltage would be controlled through the firing angle of that thyristor |
Wow, this sounds like a really GREAT idea!!! :cheerful: |
|
|
| poobah |
This would give you your "pre"-regulation.
I would put the SCR in the center tap, you coould then used sand/snubbers or a tube for rectification on the "ends/legs" of the secondary.
I would use some heater supply to power a small circuit (sand) to: 1. control (regulate) your output voltage through conduction angle, 2. provide a B+ delay to prevent cathode stripping and output pops.
You may still need some small inductance where I have drawn L1... it depends on the leakage inductance of the secondary... with Eva's realization you do not need switch 2.
This could be the new rage? |
|
|
| UnixMan |
| quote: | Originally posted by poobah
I am guessing, maybe incorrectly, that your biggest goal was to achieve regulation very early is the supply. Pre-regulate rather than post-regulate.
|
well.. yes, this definitely was (has become? I've moved a bit along the line... :D) one of the main goals.
As it should be clear by now, :D the other main goal was to keep the bandwidth of the whole "rectification noise" as narrow (low frequency) as possible... and Eva's idea seems promising also for that department! :cool:
| quote: | Eva's idea is great... my inductor, L1, is already hiding in the secondary of the transformer; and is not required. This elimanates the need for Switch 2 also.
:D |
Absolutely!!! :yes:
I was beginning to draw some rough sketch of Eva's idea as I have understood it. But now it's late and have to go... here is night and I still have to have dinner for tonight! :cannotbe: |
|
|
| UnixMan |
| quote: | Originally posted by poobah
I would use some heater supply to power a small circuit (sand) to: 1. control (regulate) your output voltage through conduction angle, 2. provide a B+ delay to prevent cathode stripping and output pops. |
Well, in some cases the idea is good, but in general I would like more a "self-booting" circuit (not requiring another external supply for the control circuit), if possible...
| quote: | | This could be the new rage? |
why not? indeed it offers some nice features...
BTW, let' sum up. Here is a rough sketch of the idea, as I have understood it; is that right?
The next step will be to design the "control unit" circuit. |
|
|
| poobah |
| I would use the SCR in the center tap and 2 diodes in the normal way for each leg of the secondary. Then use a little bit of rectifed heater power for control circuit. |
|
|
| Eva |
In order to trigger a thyristor, a small current should be made to flow from gate to cathode, so you should connect the secondaries of the pulse transformer between these legs. There is no point in inserting the secondary windings of the main transformer in the trigger loop.
As a control circuit you may consider the good old TL494 synchronized to the 50/60Hz waveform and with its internal clock programmed for 40Hz or so operation. The TL494 will be fine since it employs trailing-edge blanking (the pulses grow backwards from the end of the cycle as duty cycle is increased, instead of growing from the beggining as usual). Voltage mode control with enough frequency compensation will do the trick, and there is still a free op-amp in the IC that allows to implement some degree of output current limiting. |
|
|
| Eva |
| I forgot to mention that you are going to need some kind of high frequency pulse generator with on/off in order to feed the trigger transformer. A LM393 comparator may do the job (Nat.Semi. datasheet contains some hints about that). |
|
|
| UnixMan |
| quote: | Originally posted by poobah
I would use the SCR in the center tap and 2 diodes in the normal way for each leg of the secondary. Then use a little bit of rectifed heater power for control circuit. |
Yeah, that would made things a bit simpler, and would also avoid the need for a transformer to drive the thyristor, an CR coupling being probably enough.
But then it would require a center-tapped PT, while in the long run I would like to change this to a bridged setup in order to be able to use "normal" PT. |
|
|
| poobah |
Well...
Either method would work... the method with 2 SCRS would be more efficient. The center tap method avoids the pulse tranformer and requires some careful thought about grounding.
I believe nearly all transformers with voltages for tube B+ will have center taps.
I would focus on the control strategy and topology... the fine points of design. Then, decide which SCR layout makes the most sense to you.
Just a thought, there may be silicon modules for light dimming that may be useful for AC synch and control of "firing angle"... might save you a lot of parts.
:D |
|
|
| UnixMan |
| quote: | Originally posted by Eva
In order to trigger a thyristor, a small current should be made to flow from gate to cathode, so you should connect the secondaries of the pulse transformer between these legs. |
OK... BTW, I would say that the return path was there already (via the ground connection). :confused:
| quote: | | There is no point in inserting the secondary windings of the main transformer in the trigger loop. |
how would you sync the PWM driver to the mains frequency, then? :confused:
| quote: | As a control circuit you may consider the good old TL494 synchronized to the 50/60Hz waveform and with its internal clock programmed for 40Hz or so operation. The TL494 will be fine since it
|
I'll have a look at it...
| quote: | | I forgot to mention that you are going to need some kind of high frequency pulse generator with on/off in order to feed the trigger transformer. A LM393 comparator may do the job |
that could be done... but wouldn't a simple CR do the trick? |
|
|
| UnixMan |
| quote: | Originally posted by poobah
I believe nearly all transformers with voltages for tube B+ will have center taps.
|
except the one I have. :D
Indeed, in my case (having only ~220Vac on the secondary) I've been a bit dumb not choosing a center-tapped PT, even if I was planning on using sand anyway. :rolleyes:
But if you consider a PSU for tubes which requires several hundred or even over 1 KV dc, using a center-tap PT would be rather troublesome, or at least quite expensive.
| quote: | | I would focus on the control strategy and topology... the fine points of design. Then, decide which SCR layout makes the most sense to you. |
agreed.
| quote: | Just a thought, there may be silicon modules for light dimming that may be useful for AC synch and control of "firing angle"... might save you a lot of parts.
:D |
mmmh... intersting idea... it's definitely worth some search... |
|
|
| poobah |
In any topology you will need a second low voltage winding to derive "house keeping power" for the start-up control. This winding can provide synch.
If you have no center tap, you are forced to use 2 diodes and 2 scr's. |
|
|
| Lars Clausen |
| quote: | | We use it commercial products with good results. |
Per Anders: I have always been wondering where you work in the daytime :)
All the best from
Lars |
|
|
| Eva |
peranders:
Ouch!!
I have seen these BTSxxx ICs priced anywhere from 3 euro/piece the lowest current models to 12 euro/piece the 8 miliohm version, so I think that I will continue with my discrete solutions... |
|
|
|
