PS tests

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I have been contemplating tests for power supply quality as used in Audio. Sure, measuring ripple, but what is it that really makes a sonic difference? What SHOULD we measure?

OK, Noise is pretty straight forward. I use TrueRTA software spectrum analyzer to look at the noise. This lets me see the effects of different diodes, suppression caps and regulator noise. Very insightful. Static resistive load, drive the probes through a simple voltage divider. If you have not done this, you may be very surprised at the level of harmonics produced.

What I was thinking about is how to measure the effect the supply has with transient loads. Basically, transient regulation. How much effect does one design have over another in resupplying the caps that supply the instantaneous current into the load? I wonder what the relationship between supply impedance and the size of the at-load capacitance needs to be, or if other parastatic effects nullify the extremes many have gone too.

I want to use some objective measures of expected quality.
If the latest fancy shunt regulator sounds better than a cheap three terminal, what measurement quantifies that?

What is the sonic difference between using a high efficiency switching supply as a pre-regulator rather than a linear supply and big caps? I want to quantify this.
 
Test to show effect of PS

On a stereo amp with shared power supply, drive one channel hard into a dummy load while listening to the undriven channel through a speaker. On many designs you may be horrified by the low level distorted audio audible. Much of this is leaking through the power supply.
OK, monoblocks solve the crosstalk, but distortion caused by "loss" into the PS is still happening. All designs I've seen use the power supply as a path to or from the load.
 
Administrator
Joined 2007
Paid Member
On a stereo amp with shared power supply, drive one channel hard into a dummy load while listening to the undriven channel through a speaker. On many designs you may be horrified by the low level distorted audio audible. Much of this is leaking through the power supply.
OK, monoblocks solve the crosstalk, but distortion caused by "loss" into the PS is still happening. All designs I've seen use the power supply as a path to or from the load.

:)

Read it all to get the context,
http://www.diyaudio.com/forums/soli...-lin-topology-nfb-tappings-2.html#post1624677
 
I have been contemplating tests for power supply quality as used in Audio. Sure, measuring ripple, but what is it that really makes a sonic difference? What SHOULD we measure?

OK, Noise is pretty straight forward. I use TrueRTA software spectrum analyzer to look at the noise. This lets me see the effects of different diodes, suppression caps and regulator noise. Very insightful. Static resistive load, drive the probes through a simple voltage divider. If you have not done this, you may be very surprised at the level of harmonics produced.

What I was thinking about is how to measure the effect the supply has with transient loads. Basically, transient regulation. How much effect does one design have over another in resupplying the caps that supply the instantaneous current into the load? I wonder what the relationship between supply impedance and the size of the at-load capacitance needs to be, or if other parastatic effects nullify the extremes many have gone too.

I want to use some objective measures of expected quality.
If the latest fancy shunt regulator sounds better than a cheap three terminal, what measurement quantifies that?

What is the sonic difference between using a high efficiency switching supply as a pre-regulator rather than a linear supply and big caps? I want to quantify this.

Great idea!

Transient response is very important and would be interesting to test, especially if you changed things and compared the results. But what to change? And what and how to measure?

I haven't thought any of this through but some things come to mind:

You might want to use a special test fixture, maybe using a MOSFET switch, maybe somewhat similarly to the way it was done with the one shown in Figure 11 at Capacitor Characteristics .

It would be interesting to see the rise time (and a plot vs time) of a transient current that was suddenly allowed to flow through a 4 Ohm or 8 Ohm load (and the voltage across the load), after switching (for example) the maximum available voltage across the load. It would be good to know the maximum available current slew rate, and how long it could be sustained, etc.

On a related note, it would also be interesting to measure the output impedance of the supply as a function of frequency, up to at least a few hundred kHz, possibly for different output-power levels. (The rise-time in the previous paragraph would be related to frequency by f = 1 / (Pi x trise). I guess you would want to measure impedance to a frequency somewhat beyond that.)

For all of those tests, it would be interesting to see how the results changed for changes in the power supply itself. But it would also be extremely intreresting, to me at least, and since you would already have most of the setup in place, to test different lengths and configurations of power supply rails and different types and sizes of decoupling capacitors at the load, with various current-pulse demands by the load. Maybe you could determine, as you mentioned, whether or not certain extremely good qualities in the power supply would be "wasted" to varying degrees if the power distribution were done in particular ways.

I guess some might think that you wouldn't need to bother with some of that, since we can calculate it all, off line. But it would still be very interesting to actually see what happened. And of course calculations or simulations are both only as good as the model being used. So there would be some real value in actual testing.
 
Last edited:
Good thoughts. I'll chew on these for a while as I finish up a few other projects. I had not thought about cross channel effects. I was thinking more of low level power supplies, but I bet it applies there too. What got me back to this was building a buffer, aka CMoy head amp to drive the 600 Ohm input to my e-mu I use for speaker measurement. I have been using batteries, but was considering a line supply. It took me a long time to get the residual noise and distortion below 130dB of my test rig, so introducing a line supply became a topic for deeper consideration. So, simple 3 terminal? , External switcher and a rail splitter? Sulster ( my apologies, I bet I have that wrong) or Jung and so on.

Z out for supply over freq. Very good idea. We do it for power amps as a unit, we should do it for the supply. Construction can sure make a difference. Changing the power and ground system of my B&K st140 made obvious improvements. REALLY obvious. I think they did something themselves in later units.

The ability to have a PC generate odd but repeatable pulses has opened a few doors. I found how an amp deals with initial wave fronts to vary a lot. I suspect the power supply including the physical construction and treatment of source and return to be part of it. Why does my old Rotel sound so darn good? Or my Creek?, yet my Hafler measures far better.

I have followed cap development since Walt Jung's original article in Audio about ESR and DF helped me solve a mystery problem at my work. A multi-million dollar problem solved by changing from a ten cent electrolytic to a 50 cent foil-film in the right place. The Phd engineers did not believe me until I made the change and proved it. (It was a half nano-second bump in Dcc causing a Z80 to NMI).
 
The moment you write "sounds better", you are chasing a ghost, building a supply only suitable for a particular combination of listener preference and amp (and maybe source and speakers too).

Otherwise, objectively it would seem logical to reduce ripple at the amplification stage and respective output transistors so these are the appropriate points to measure ripple. Once you have these measurements you can either try to assign cause or blame, and/or simply build up or add additional filtration to provide more smoothing.

Point being, in a power amp the ripple caused by the amp itself greatly exceeds the ripple present from most other factors, assuming a certain level of diligence to follow typical, time tested designs.
 
Well to get a power supply to have a "good" sound, it shall not influence the output at all.

Noise and ripple is something you can not do that much about beyond a reasonable level. The amplifier itself must have some immunity against these factors.

The interesting thing is the interaction between the load (amplifier) and the PSU.

I would be most worried at frequencies around and below the cut-off frequency of the PSU regulation loop
 
I agree, a perfect amp would have no sound of it's own. Being able to characterize what actually contributes to most amps HAVING sound is the goal. I was surprised at the level of mains harmonics I was seeing on various supplies when build a test rig and looking at my various low power amps outputs, both regulated and unregulated supplies. I was looking at the output at 1 watt with a 1K tone for the simple static test into a pure resistive load. 120Hz may have been -90dB, but 240 or 480 Hz at -60! How much of this is easily dealt with by diode selection, suppressor cap selection, and so on. It is oft said to lose the 1N4000 series or similar high power and put in high speed Schottky. OK, I want to see a picture on my scope or spectrum analyzer.

Load interaction with a complex load I do suspect to be relevant. It should be no surprise the cleaner the power supply looked, the less character the amp had. Mind you, I am starting with pretty good amps. Of the unregulated power amps, I have a H-120, Rotel 840, Creek, Amior, and a home built FET. I wish I had a Bryston for reference. I have a handful of low voltage regulated supplies from my Heath bench, switchers, old 3 terminals and battery. The battery was of course harmonically perfectly clean, but I would not say the best sounding on my RA-1 head amp and Grado's. I prefer objective measurements to "sounds better" When they correlate with subjective ones, I figure I am in business as I will then know what to build to. Of course, the physical implementation effects the parastatics and we must look at "all things being equal". If yo u are wondering, I like the Creek and Rotel. I was surprised how much character I could ascribe to the Hafler. The Amior is, well lets say it is a good thing they went back to building boom boxes. It may become a box and transformer for a gain clone. Great place to play with power amp supply differences.

Lots of ideas here and I will need to think about how to test some of them. I think it worthwhile.
 
Well to get a power supply to have a "good" sound, it shall not influence the output at all.

Noise and ripple is something you can not do that much about beyond a reasonable level. The amplifier itself must have some immunity against these factors.

The interesting thing is the interaction between the load (amplifier) and the PSU.

I would be most worried at frequencies around and below the cut-off frequency of the PSU regulation loop

Yes, the interesting thing IS the interaction between the load (amplifier) and the PSU. But so much depends on the power distribution network in the amplifier, where the rails' and the ground returns' inductances and the decoupling capacitors' abilities to accurately supply the transient currents have to be there, no matter how good the power supply might be. So what do we actually need to test, for the power supply alone?

What are a power supply's requirements and the desired specifications, for this application? Answering those questions would tell us what requirements and specs to test, and might also suggest how to test them.

The power supply must be able to accurately provide all of the music signal that drives the speakers, as a possibly-large and dynamic CURRENT signal, with high fidelity. I like to try to keep in mind that the high-power "signal path", which is the only one that we actually hear, is, after all, current that comes straight through the power supply and the power amplification devices, which are really just "precision current valves" (i.e. controllable resistances) that are controlled by the small-signal portions of the amp (which is also influenced by the power supply). They don't pull or push anything; they simply "allow" current to flow from higher to lower voltage locations, when they are opened and closed to varying degrees. The decoupling caps must provide all of the current for at least the start of every transient, since the inductance of the power rails makes it literally impossible for the power supply to get the current moving in time to meet the demand accurately.

With that type of picture in mind, what does the power supply really need to do, and, what are the most-important performance qualities that it should have, and with what actual specifications? i.e. What are the requirements and the desired specifications, for this application???

Answering those questions would tell us what requirements and specs to test, and would probably also suggest how to test them.

Personally, when designing a power amplifier, I think that it makes more sense to think of "the power supply" as including everything up to the active devices' power and ground pins, since THAT is where I hope to see the "perfect" impedance vs frequency plots, and that way I can design the decoupling and rails' inductances and the bulk/smoothing caps (and regulation, if used), etc, all together as a single system (which needs to be able to allow current to dump into some of the pins and swallow whatever current comes out of some of the other pins, all with perfect accuracy, on demand). I just think it's much easier to know and define the requirements for (and later test/measure) what is needed at the power and ground pins of the active devices, than at some "artificial" point called "the psu output" that is basically in the middle of the power and ground rails of the system as a whole.

That might seem unhelpful, at first, but I'm hoping that it might help to separate-out what is required of the PSU, alone.

Sorry to have blathered-on for so long about all of that.

Cheers,

Tom
 
Last edited:
Tom, you are about as brief as I. :) We agree completely, the power supply is part of the signal path and everything must be looked at as a whole. I was trying to be a little simpler to start thinking what really basic practices we assume have what really basic effects, like diode selection, or would returning to pi filters have better performance with respect to harmonic suppression but cause transient regulation degradation.

I bet you have seen the same thing, the cap right on the heat sink at the transistor being high a high ESR type. Least we forget for most cap technologies, capacitance is not always the largest variable parameter of the part. Caps are hard. Thermal, voltage and frequency variable RL circuits effected by humidity that have some amount of capacitance. Yea, that's a cap. Z5U or X7R's used as integrators, tant's used as coupling.

I am intrigued with the idea of raising DCC above chassis ground by 10 to 100 Ohms. I have not seen that before. I have seen them only AC coupled. Where can I get more on that subject?
 
Tom, you are about as brief as I. :) We agree completely, the power supply is part of the signal path and everything must be looked at as a whole. I was trying to be a little simpler to start thinking what really basic practices we assume have what really basic effects, like diode selection, or would returning to pi filters have better performance with respect to harmonic suppression but cause transient regulation degradation.

Sounds like a good idea. One thing I've noticed is that so many people seem to focus on keeping the power supply voltages clean and it seems like they're forgetting that the current is the signal. The decoupling caps can't even respond to transient demands for current unless the voltage across them dips. We can increase their values to make the dips as small as desired, of course. But why? What are the trade-offs?

I bet you have seen the same thing, the cap right on the heat sink at the transistor being high a high ESR type. Least we forget for most cap technologies, capacitance is not always the largest variable parameter of the part. Caps are hard. Thermal, voltage and frequency variable RL circuits effected by humidity that have some amount of capacitance. Yea, that's a cap. Z5U or X7R's used as integrators, tant's used as coupling.

Yes! Caps are where it's at! Caps and the inductances of the traces and wiring...

This is very interesting:

Cornell Dubilier Electronics

I wish they had some small electrolytics, there.

I am intrigued with the idea of raising DCC above chassis ground by 10 to 100 Ohms. I have not seen that before. I have seen them only AC coupled. Where can I get more on that subject?

I have seen some discussions here, about that. I would try searching for "star ground". But I'm guessing that if the grounding topology and layout are done correctly then you probably wouldn't need to use that technique. I've been wrong, before, though.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.