Bob Cordell Interview: Power Supplies

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Another thing that may influence the role of power supply filtering and quality is if the amplifier is of a minimalist design. A number of designers feel that the best sound is obtained with the simplest possible signal path. In some cases this leads to designs that have less power supply rejection than a more complex design with many transistors. Thus, it may turn out that great attention to the power supply will make a bigger difference in these types of designs.

I am not in the minimalist camp, and I will spend a transistor to improve power supply rejection without half a thought, and I have a strong goal to achieve very good PSRR out to fairly high frequencies.

Bob
 
AKSA said:

These caps have two purposes; one to smooth the power pulses coming off the rectifier (and these are pretty savage), and the other to accommodate speaker earth return current. If both these currents coexist in the same cap, then intermodulation results, so I like to use two caps per rail, separate them with a small resistor (typically 0R22, or even a tiny inductor around half a millihenry) so that the two currents are separated.

For intermodulation to occur, a nonlinearity is necessary. Otherwise, the superposition principle applies.
And if there was a built-in diode or whatever in a cap, 0r22 would not linearize it.

Gerhard
 
Bob Cordell said:


Another thing. If the VA rating is primarily an indicator of maximum core temperature, then different transformers of the same VA rating could have vastly different useable d.c. rectified current capabilities and regulation in the real world. For example, copper is now very expensive. A transformer manufacturer could probably arrive at the same thermal VA rating by using less copper and more iron, and maybe arranging that iron to dissipate heat better. The resulting transformer would have significantly more winding resistance and would not have as good regulation, even though it had the same VA.

Copper loss dominates at maximum load while iron loss dominates in no-load situations. An optimal transformer is usually designed so that copper and core losses are equal for the intended application. So a transformer for a class A amplifier should have relatively more copper than a transformer for something that is spec'ed for PMPO.

Gerhard
 

me:
> For a typical tube amp with Vplate of several hundred Volts, 1mF should be enough.
>Stored energy goes up with the square of the voltage.

Tube_Dude:
If that is obviously true from a energy point of vue , we must not forget thet the rail capacitor is the path that close the signal to ground.

So this parasitic series impedance must be the minimum that can be practically implemented...

Ideally it must be 0 Ohms at all signal frequencies...;)

For an amplifier at 300V/100mA, the impedance level is 100 times higher than for an amp with 30V/1A, so, everything else equal, we could afford a much higher power supply source impedance.

Gerhard
 
gerhard said:


For an amplifier at 300V/100mA, the impedance level is 100 times higher than for an amp with 30V/1A, so, everything else equal, we could afford a much higher power supply source impedance.

Gerhard


Good point. I use about 1000 uF at 500V in my KT-88 tube amp, so that would correspond to something on the order of 100,000 uF/rail in my solid state amp, not far off.

Bob
 
Bob Cordell said:

I am not in the minimalist camp, and I will spend a transistor to improve power supply rejection without half a thought, and I have a strong goal to achieve very good PSRR out to fairly high frequencies.

Bob

Bob, can you tell any number, what level of PSRR is recommendable ? Is -90dB a good number, or should it be even more ? (getting difficult)

Mike
 
AX tech editor
Joined 2002
Paid Member
AKSA said:
[snip]However, topology must play a part. These caps have two purposes; one to smooth the power pulses coming off the rectifier (and these are pretty savage), and the other to accommodate speaker earth return current. If both these currents coexist in the same cap, then intermodulation results, so I like to use two caps per rail, separate them with a small resistor (typically 0R22, or even a tiny inductor around half a millihenry) so that the two currents are separated.

Cheers,

Hugh


Hugh,

I have a problem with that. The speaker current ultimately comes from the first cap, the one closest to the rectifier. It therefore also needs to return to that. The ripple current may divide in some way between the caps but you can't say that one current goes to one cap and another to the other cap.

Jan Didden
 
janneman said:
Hugh,

I have a problem with that. The speaker current ultimately comes from the first cap, the one closest to the rectifier. It therefore also needs to return to that. The ripple current may divide in some way between the caps but you can't say that one current goes to one cap and another to the other cap.

Jan Didden

The speaker current comes from the transformer, it has nothing to do with the Filter Caps, contrary to popular notions, it's return loop terminates at the transformer. It's all about the loop, this much is true.

Regards, Mike.
 
MikeB said:


Bob, can you tell any number, what level of PSRR is recommendable ? Is -90dB a good number, or should it be even more ? (getting difficult)

Mike


The simple answer is as much as you can get without going to heroics that throw you off course.

There are many topological decisions you can make that affect PSRR. Always think carefully about where the current flows when the supply line changes voltage. Think about where you are dumping that current - are you unintentionally dumping it into a ground that wants to be quiet?

Of course, use current sources with a nice high impedance output and a supply-independent current-setting mechanism. Cascoding can be helpful.

Also be mindful of how you do feedback compensation. Common Miller compensation has rather inferior power supply rejection.

Differential circuits tend to have better PSRR than single-ended circuits.

My use of MOSFETs "forces" me to do something good: I always use a boosted B++ and B-- for the input and driver circuits so that I don't waste precious high-current main-supply voltage on gate drive and VAS headroom needs. This means that I cannot take the usual shortcut of just taking the main supply rails and R-C filtering them to get the rails for the input and driver circuits.

I always derive the boosted supplies with extra small a.c. windings connected at the ends of the main transformer windings and then feeding the resulting larger a.c. to a separate bridge rectifier to supply the raw boosted supply voltages. Those extra windings can be provided by a small extra transformer with two isolated secondaries on the order of 10-20V each, or they can be extra windings placed on the main torroidal power transformer.

Bob
 
Bob Cordell said:

Those extra windings can be provided by a small extra transformer with two isolated secondaries on the order of 10-20V each, or they can be extra windings placed on the main torroidal power transformer.

I'd prefer the small extra transformer for the "small signal" stages, and a 2-chamber non-toroidal type, because it has much less capacitance to the bad outside world.

Gerhard
 
Originally posted by MikeBettinger
The speaker current comes from the transformer, it has nothing to do with the Filter Caps, contrary to popular notions, it's return loop terminates at the transformer. It's all about the loop, this much is true.

If you see it this way, you must continue the loop at least to the generator in the next power plant.
The conduction cycle in the transformer can be surprisingly short, especially if you have heroic capacitor banks. The transformer will deliver current only when its secondary voltage is larger than the instantaneous capacitor voltage + rectifier drop. But when it delivers, it will be like feeding a short circuit. Add ultrafast rectifiers and you have a serious EMC problem.

regards, Gerhard
 
gerhard said:


If you see it this way, you must continue the loop at least to the generator in the next power plant.
The conduction cycle in the transformer can be surprisingly short, especially if you have heroic capacitor banks. The transformer will deliver current only when its secondary voltage is larger than the instantaneous capacitor voltage + rectifier drop. But when it delivers, it will be like feeding a short circuit. Add ultrafast rectifiers and you have a serious EMC problem.

regards, Gerhard

Correct.

Bob
 
gerhard said:


If you see it this way, you must continue the loop at least to the generator in the next power plant.
The conduction cycle in the transformer can be surprisingly short, especially if you have heroic capacitor banks. The transformer will deliver current only when its secondary voltage is larger than the instantaneous capacitor voltage + rectifier drop. But when it delivers, it will be like feeding a short circuit. Add ultrafast rectifiers and you have a serious EMC problem.

regards, Gerhard

Nope, I see the loop as starting and ending with the secondary winding. But then I see the power as being drawn from the secondaries with the filters being there only to prop up under heavy load. Charging and discharging the caps is is a slow process relative to the demands of the output transistors. The conduction cycle is a bit more complex than it seems and very much affected by the layout.

Why should one place the filter charging currents in series with the supply and the returns for the gain stages? Seriously.

The differences in approaches are measurable, audible and easy to verify.

EMC is minimized through controlling the loop (again) and is also easily measured (relatively).


Regards, Mike
 
AX tech editor
Joined 2002
Paid Member
MikeBettinger said:
[snip]But then I see the power as being drawn from the secondaries with the filters being there only to prop up under heavy load. [snip]


But, Mike, the secondaries only deliver power for 20% of the time (less with higher cap values), so it is really the caps that deliver the power on an ongoing basis and it is the secondaries that prop up the caps now and then.

Jan Didden
 
janneman said:



But, Mike, the secondaries only deliver power for 20% of the time (less with higher cap values), so it is really the caps that deliver the power on an ongoing basis and it is the secondaries that prop up the caps now and then.

Jan Didden


I agree with you, Jan. However, I'm having a little bit of trouble visualizing Mike's approach.

I prefer to have some large capacitance (e.g., 1000 uF + 1 uF film + 0.1 uF ceramic) very close to the output transistors to force a tight loop for circulation of the output transistor currents for this very reason. Those currents are high, have fast edges, and are highly non-linear. As mentioned earlier, an X-capacitor combination from rail to rail can also help in this regard, since you really want to try to sum the positive and negative half-cycle Class-AB currents back to a linear current before they travel very far. And a little bit of impedance in the +/- lines back to the main reservoir capacitors can actually be a good thing. Then, the rectifiers replenish the reservoir capacitors, and the reservoir capacitors replenish the local output stage storage capacitors.

Bob
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.