| ergo |
The following is a test I did to help me figure out a new supply for my power amp.
The page looks awful and is too big, so you have to wait while it downloads... sorry...
Still I wanted to publish it and as the time is in very short supply lately this is the best I managed.
http://f5.infonet.ee/ergo/ps/powersupply.html
Anyway as a teaser - this is the spectrum of an output of a type of power supply that is most commonly used - transformer -> diode bridge -> caps -> load
Ergo |
|
|
| ACD |
| However I don't see any test for a PSU with 100nF across the rectifying diodes, and with 1uF to 10 uF decoupling caps across the output.............. |
|
|
| millwood |
nothing unexpected, other than the >30K peak. What is that?
amps usually have pretty good psrr at low frequencies so nothing to worry about.
Also, have you tried the same test on a psu using those fancy soft recovery diodes? |
|
|
| ergo |
Good point ACD. I actually plan to continue with this line of testing. For one I'm waiting my 100'000uF RIFA caps to arrive then I can test how low can I get the ripple....
Still at least for me this series of measurements was a bit of an eye opener. I had not done this kind of spectrum analysis on high current power supply before. It seems that there is a huge amount of **** totalling from the diode bridge + the stuff coming from the wall socket.
The goal is to get the spectrum as pure as possible and then connect the power supply to my amp and listen - at the moment it's only a suspicion but I'm not surprised if the change in sound will be huge.
One thing to test would be an active filter. I'm wondering how much of the high frequency **** evident with simple filters gets through active filter.
Ergo |
|
|
| millwood |
| quote: | Originally posted by ergo
The goal is to get the spectrum as pure as possible and then connect the power supply to my amp and listen - at the moment it's only a suspicion but I'm not surprised if the change in sound will be huge.
Ergo |
probably not a whole lot of differernce but it is nice to try.
How about the idea of regen? what if for example you use the traditional PS to power an amp which generate either DC directly or high frequency AC to be rectified, to then power your real amp?
Wouldn't that help a lot from PSU specturm's point of view? |
|
|
| janneman |
Hi Ergo,
Nice test. The **** you see are actually 50 and 100 Hz harmonics caused by the sawtooth-shape of the voltage on the last cap. The noise, which is at a much lower level, *may* be caused by the diodes, but it can also come from the mains.
You'll note that the harmonics of the 100 Hz rectified signal are higher than the 50 Hz harmonics. There are no surprises here, all is just as theory predicts. When you increase your filter cap, you will probably see that the lower harmonics are lower but that there are more and higher high-order harmonics.
You can get rid of those 50/100 Hz harmonics most effectively with a CLC filter. This will clean up the low level hash as well.
Jan Didden |
|
|
| ACD |
| quote: | | Good point ACD. I actually plan to continue with this line of testing. For one I'm waiting my 100'000uF RIFA caps to arrive then I can test how low can I get the ripple.... |
Just adding bigger and bigger caps doesn't mean that you will get rid of the noise. If you take a look on other supplies in this forum, you'll see that they all have decoupling caps on the output (normally about one 1/100 the value of the main caps) ;)
I'm looking very much forward to see your next test results :) |
|
|
| Eva |
This 31Khz component looks like common mode EMI from the power supply of the computer. Most AT and ATX PSUs operate between 30Khz and 35Khz
It may also come from anywhere inside the computer since there are dozens of EMI sources affecting the soundcard |
|
|
| slowhands |
You have 50 Hz harmonics galore. One very cheap way to reduce them is to put a small (0.22 5w) resistor between the caps, creating a CRC filter. Though it seems crude, it is very effective for low frequency filtering.
As Eva astutely noted, you have noise from your PC power supply too. You may be able to dramatically reduce this with caps across the transformer primary and secondary, typically 0.047 works quite well. This will also filter out power supply and power switch noise on the primary side, and diode noise on the secondary side.
At the load, I would suggest additional filters on each voltage. Conventional wisdom suggests a small electrolytic (100 mF) in parallel with a good film cap (0.1 mF). |
|
|
| Eva |
The only thing that works in order to reduce common mode noise is common mode filtering
Take a big ferrite toroid and simply wind a dozen or so of turns of your signal cable on it before entering the sound card. Use the entire signal cable to do the winding
Look at this example : I have to do this kind of filtering in order to be able to get precise measurements when the ground of one probe is connected to the primary side of a SMPS and the ground of the other is connected to the secondary side. These grounds aren't at the same AC potential due to common mode EMI generated by the own SMPS prototype and the SMPS of the oscilloscope |
|
|
| DSP_Geek |
| quote: | Originally posted by ergo
Good point ACD. I actually plan to continue with this line of testing. For one I'm waiting my 100'000uF RIFA caps to arrive then I can test how low can I get the ripple....
Still at least for me this series of measurements was a bit of an eye opener. I had not done this kind of spectrum analysis on high current power supply before. It seems that there is a huge amount of **** totalling from the diode bridge + the stuff coming from the wall socket.
The goal is to get the spectrum as pure as possible and then connect the power supply to my amp and listen - at the moment it's only a suspicion but I'm not surprised if the change in sound will be huge.
One thing to test would be an active filter. I'm wondering how much of the high frequency **** evident with simple filters gets through active filter.
Ergo |
This is really good work, meticulous and clear. Congratulations. As Jan Didden pointed out, the higher frequency crud you see is harmonics from the waveform on the filter cap, which the larger series inductor is mostly successful at removing. The 50 Hz and its harmonics are likely caused by a small imbalance in the diode bridge, but it's sufficiently lower than the 100 Hz stuff that the effect is negligeable.
I suspect the 1 ohm series resistors aren't doing much when you throw in even the smaller inductor, so it might be amusing to remove them and see the effect - amplifiers already generate enough heat!
By active filter, do you mean a regulator or capacitance multiplier?
Francois. |
|
|
| PMA |
| I am sure that the spectrum shown (it is a power supply spectrum) is nothing special. Amplifiers have PSR (power supply rejection) and better the amp better the PSR. I speak about construction, earthing, wiring, too, not only about circuit design. |
|
|
| PMA |
I have attached an example of the amp's spectrum. Class A (great current consumption from the PSU) at 4Vrms output voltage into 4 Ohm.
In the first part (under 1kHz) there are PSU residuals. Similar to ergo's image, suppressed by the PSR factor. From 1kHz the THD spectrum can be seen. The last part (10kHz - 20kHz) is an error of the method - noise shaping. |
|
|
| ergo |
Thanks all guys for sharing your comments.
PMA - I know the amps have quite a bit of PSR, but the gut feeling of mine tells that this might still be a point where an improvement can be had. Especially I'm suspicious of how this kind of power supply **** is affecting the typical push pull output stage... In the first stages there are either resistors or current mirrors between the "signal path" and supply lines + usually additional RC filter etc, but the output stage has exactly this kind of supply voltage full of **** on its drain (collector) or source (emitter). Aint it possible that with music signal and especially the low signals this influences the signal - distorts or modulates it?
**
As for the origin of all this **** - well I'm not convinced that it all comes from diode bridge. I was just reading an article from AudioXPress by Charles Hansen: "The AC Power Line and Audio Equipment" He is showing with measurements in the article series that there is a considerable amount of 50HZ (60Hz inUSA) harmonics present already in the wall socket due to distortion mechanism in generators etc. Plus there is a huge amount of EMI flying around these days.
I will try to find a small transformer which had I think some 6V secondary. Then I can send it right to soundcard and take a "measurement of spectrum right out of the wall socket. :) Should be interesting...
***
The peak around 32KHz is yes quite surely the PC power supply, but I dont quite know how it gets into my measurement. It has to be over the air as if I make a loopback test or if I measure the output of my cd with this soundcard the same peak is not there.
Regards,
Ergo |
|
|
| Eva |
The waveform present in the mains is rich on harmonics but take into account that the diode bridge conducts only about 30% of the time so these harmonics are not likely to be shown at the output
The spectrum on the output comes from the fact of charging a capacitor through sharp pulses repeated at 100Hz having about 30% duty cycle and 333% crest factor [10A pulses for 3A output, it's as noisy as a 50Hz flyback converter]
The 31Khz stuff is mostly not radiated through the air, it's conducted through mains wires and appears as an AC potential between earth and the center tap of your transformer
Take into account that big 50-60Hz transformers show very high primary-secondary parasitistic capacitances and, while providing enough isolation at 50-60Hz, they act like a high frequency AC short so the output of your transformer is shorted to mains at high frequencies
The SMPS of your PC makes both mains wires fluctuate [both in phase] from ground/earth at 31Khz. This is called 'common mode conducted EMI'
This fluctuation propagates through mains wiring and passes freely through your transformer and ends as a fluctuation between the center tap and the ground of the computer [earthed or not] effectively creating an AC ground loop on your measurement
You have two ways to cure the problem. A crude incomplete solution is to use full bridge diodes for each supply so the transformer is nearly isolated from equipment ground for 70% of the time, but common mode signal will be still here during diode conduction. A professional solution is to use a suitable common mode filter on the primary or the secondary side of the transformers in all audio equipment and maybe enhancing the poor common mode filtering usually present on computer SMPSs
I allways use common mode filtering since the effects of common mode EMI on audio circits are easily seen on the oscilloscope |
|
|
|
