Hello all.
Today I have started taking THD measurements of my amplifier with a simple setup: soundcard output -> amp -> dummy load -> voltage divider followed by AES-17 filter -> soundcard input
The results are not too bad, although some tweaking is required in order to get the optimum voltage level to the soundcard for every power you measure.
So far I have only measured 1KHz tones at various power levels and with a couple of load resistors.
Supply is 500VA transformer that gives (at idle) +/-50V DC with 9900uF per rail (a little bit too low), so the maximum power at 5 ohm before clipping is around 207W.
10 ohm load:
-----------------------------------
1W -> 0.023% THD
10W -> 0.033% THD
63W -> 0.17% THD
103W -> 0.5% THD
5 ohm load:
-----------------------------------
1W -> 0.037% THD
10W -> 0.07% THD
40W -> 0.19% THD
115W -> 0.2% THD
160W -> 0.51% THD
207W -> 0.65% THD
It is clear that THD increases with power level, and that at the same power output, distortion is lower with lighter load, so I suspect from the power supply, that is very weak and causes too much ripple.
Do you have any suggestions looking at this measurements?
BTW: the lower THD value obtained with a loopback cable from output to input of soundcard is 0.01% at that frequency. Should I substract that value directly from the measurements or what?
Thanks!
Today I have started taking THD measurements of my amplifier with a simple setup: soundcard output -> amp -> dummy load -> voltage divider followed by AES-17 filter -> soundcard input
The results are not too bad, although some tweaking is required in order to get the optimum voltage level to the soundcard for every power you measure.
So far I have only measured 1KHz tones at various power levels and with a couple of load resistors.
Supply is 500VA transformer that gives (at idle) +/-50V DC with 9900uF per rail (a little bit too low), so the maximum power at 5 ohm before clipping is around 207W.
10 ohm load:
-----------------------------------
1W -> 0.023% THD
10W -> 0.033% THD
63W -> 0.17% THD
103W -> 0.5% THD
5 ohm load:
-----------------------------------
1W -> 0.037% THD
10W -> 0.07% THD
40W -> 0.19% THD
115W -> 0.2% THD
160W -> 0.51% THD
207W -> 0.65% THD
It is clear that THD increases with power level, and that at the same power output, distortion is lower with lighter load, so I suspect from the power supply, that is very weak and causes too much ripple.
Do you have any suggestions looking at this measurements?
BTW: the lower THD value obtained with a loopback cable from output to input of soundcard is 0.01% at that frequency. Should I substract that value directly from the measurements or what?
Thanks!
Yes, but now I have checked, it is not an AES-17, but an approximation made by 3 stage RC filter, described in a IRF application note.
The values are: 470 ohm - 470nF - 680 ohm - 2.2nF - 1kohm - 1nF
Previously, a 1k / 100 ohm pot adjustable voltage divider has been connected in order to have the optimum output level for the soundcard to measure correctly for variable output power levels.
Best regards
The values are: 470 ohm - 470nF - 680 ohm - 2.2nF - 1kohm - 1nF
Previously, a 1k / 100 ohm pot adjustable voltage divider has been connected in order to have the optimum output level for the soundcard to measure correctly for variable output power levels.
Best regards
The output (L C) filter could be responsible for the icressed distortion at higher load levels. The inductor is not vary linear at higher load levels, so this could be an explanation. Supply pumping chould also be it. You could make at test with different capasitors on the supply rails, to se if it makes any difference.
NFB is taken (only) after output filter ;-)
I have not observed power supply pumping during all my tests even at higher power levels. But what I do have is quite a bit (perhaps 6 to 8Vpp at the max. power measured) of supply ripple due to not very large capacitors.
I will also play with dead-time, perhaps it is excessive. Now it is implemented by 6.8 ohm gate resistor in antiparallel with shottky, and with a 3k3 resistor in antiparallel with a 1n4148 at each IR2113's input, so they form a time constant with its input capacitance only during rising edge.
Perhaps I will lower that 3k3 resistors and see if distortion gets lower, but after that I will have to test the amp at full power for quite a while to be sure that it is still reliable! ;-)
Charles what do you think about these numbers? Are they too bad for a triangle-based amp?
Best regards,
Pierre
I have not observed power supply pumping during all my tests even at higher power levels. But what I do have is quite a bit (perhaps 6 to 8Vpp at the max. power measured) of supply ripple due to not very large capacitors.
I will also play with dead-time, perhaps it is excessive. Now it is implemented by 6.8 ohm gate resistor in antiparallel with shottky, and with a 3k3 resistor in antiparallel with a 1n4148 at each IR2113's input, so they form a time constant with its input capacitance only during rising edge.
Perhaps I will lower that 3k3 resistors and see if distortion gets lower, but after that I will have to test the amp at full power for quite a while to be sure that it is still reliable! ;-)
Charles what do you think about these numbers? Are they too bad for a triangle-based amp?
Best regards,
Pierre
Here's a small update.
I have tried with other module and reduced the dead-time resistor, noticing a very small reduction in THD. But the output ripple is no longer a nice sine wave but has peaks, I suppose some shoot-through is occuring.
I have been able to reduce the distortion measurements with a little of care while doing the tests, but what I have noticed is that THD increases with increased input frequency. I suppose that this is because my feedback network is designed to have about 31dB loop gain at up to 2KHz, then it starts decreasing. In fact, my 100Hz figures are noticeable better than what I post before (that was at 1 KHz).
Is it better to have lower feedback gain but in a wider band? Anyway, is there any other way to reduce distortion once that I have checked that dead-time is not the culprit?
Perhaps the gate resistors are too low (6.8ohm) and they are switching too fast causing more EMI problems.
If you have any ideas, they are welcome!
I have tried with other module and reduced the dead-time resistor, noticing a very small reduction in THD. But the output ripple is no longer a nice sine wave but has peaks, I suppose some shoot-through is occuring.
I have been able to reduce the distortion measurements with a little of care while doing the tests, but what I have noticed is that THD increases with increased input frequency. I suppose that this is because my feedback network is designed to have about 31dB loop gain at up to 2KHz, then it starts decreasing. In fact, my 100Hz figures are noticeable better than what I post before (that was at 1 KHz).
Is it better to have lower feedback gain but in a wider band? Anyway, is there any other way to reduce distortion once that I have checked that dead-time is not the culprit?
Perhaps the gate resistors are too low (6.8ohm) and they are switching too fast causing more EMI problems.
If you have any ideas, they are welcome!
hi Pierre,
some time ago i've done THD measurement of the UcD like amp (300-1000khz switching) without any low pass filter before my old audigy..no problem up to 0.002%=audigy loopback THD. However my old subamp design (80khz) has more need for such filter, hence the internal low pass filter of the sound card quite enough for DIY THD measurement if the switching higher 300khz?
some time ago i've done THD measurement of the UcD like amp (300-1000khz switching) without any low pass filter before my old audigy..no problem up to 0.002%=audigy loopback THD. However my old subamp design (80khz) has more need for such filter, hence the internal low pass filter of the sound card quite enough for DIY THD measurement if the switching higher 300khz?
Salut Pierre
The figures are not bad at all given that it is
1.) your first one and
2.) a single first-order loop topology
What I particularily like are the figures at low input signal levels. The famous TacT MillenniuM for instance has 0.1% THD, BTW !!!!
A means to improve THD performance would be the use of a higher-order NFB loop.
Also your output filter core might indeed be a responsible factor for the increase in THD.
If you are not content with your THD measurements have a look at this bargain:
http://stereophile.com/amplificationreviews/205krell/index4.html
Some more ? Your's still better than this one:
http://stereophile.com/amplificationreviews/729/index8.html
This is one that is in the same ballpark as yours:
http://stereophile.com/amplificationreviews/442/index6.html
Regards
Charles
The figures are not bad at all given that it is
1.) your first one and
2.) a single first-order loop topology
What I particularily like are the figures at low input signal levels. The famous TacT MillenniuM for instance has 0.1% THD, BTW !!!!
A means to improve THD performance would be the use of a higher-order NFB loop.
Also your output filter core might indeed be a responsible factor for the increase in THD.
If you are not content with your THD measurements have a look at this bargain:
http://stereophile.com/amplificationreviews/205krell/index4.html
Some more ? Your's still better than this one:
http://stereophile.com/amplificationreviews/729/index8.html
This is one that is in the same ballpark as yours:
http://stereophile.com/amplificationreviews/442/index6.html
Regards
Charles
Thanks, Charles. for that encouraging words!
At 100Hz and 1W output my THD is around 0.015% into 5 ohms.
As I have observed a dependency of THD with frequency, I will try to re-design the NFB so it is more constant across the audio band, while trying to have it as high as possible. So the gain has to be reduced to, say, 20dB (now I have almost 30dB).
That should decrease THD, right?
I have only 9900 uF per rail by now, perhaps that's increasing THD a lot. Seems possible?
The coil shouldn't be causing much THD IMHO, it is dimensioned up to about 30A, while I have around 0.35% THD at 3.5A RMS output.
I measured also the noise floor, and when I get rid of ground loops, etc, I can have less than 100dB of noise at every frequency in the audio band with the input shorted.
Best regards,
Pierre
At 100Hz and 1W output my THD is around 0.015% into 5 ohms.
As I have observed a dependency of THD with frequency, I will try to re-design the NFB so it is more constant across the audio band, while trying to have it as high as possible. So the gain has to be reduced to, say, 20dB (now I have almost 30dB).
That should decrease THD, right?
I have only 9900 uF per rail by now, perhaps that's increasing THD a lot. Seems possible?
The coil shouldn't be causing much THD IMHO, it is dimensioned up to about 30A, while I have around 0.35% THD at 3.5A RMS output.
I measured also the noise floor, and when I get rid of ground loops, etc, I can have less than 100dB of noise at every frequency in the audio band with the input shorted.
Best regards,
Pierre
THD measurements
Bonjour Pierre
It would be helpful to get some THD+N vs frequency measurements at say 100W or so. To see whether your distortion rises with frequency, and also to see whether your distortion is harmonic or anharmonic.
If you have the facilities to do a 20-20kHz sweep and post the resulting graph, please do. Alternatively you can graph it yourself from a number of single frequency measurements.
I don't know how you are measuring the distortion, but I assume it's a THD+N measurement as these are most common.
I found with my open loop 'digital' amps (PowerDAC) that the non-linearities in the output inductor were a major source of THD and that this THD rose with frequency - so I changed to air-cored coils. Bridge deadtime is another major cause of distortion.
Another thought that occurs to me is that you may have switching jitter which (and I stand corrected on this) would create anharmonic distortion with a broad frequency spectrum, rather like noise. You could get a rough clue of the nature of the distortion by simply doing an unweighted S/N measurement. If it seems higher than what you would expect you might have a jitter problem.
I'm surprised that the feedback doesn't correct the distortion more, especially since you say you've taken it from after the LPF. Possibly your open-loop distortion is worse than you think? It could be useful to try doing some open loop distortion measurements (you'll probably have to make some temporary mods to the amp get a sensible fixed throughput gain to get this right). If your measurement setup facilitates real time continuous distortion measurements you could then do some tweaking and see what makes it better and what makes it worse.
Hope this helps.
Regards
John
Bonjour Pierre
It would be helpful to get some THD+N vs frequency measurements at say 100W or so. To see whether your distortion rises with frequency, and also to see whether your distortion is harmonic or anharmonic.
If you have the facilities to do a 20-20kHz sweep and post the resulting graph, please do. Alternatively you can graph it yourself from a number of single frequency measurements.
I don't know how you are measuring the distortion, but I assume it's a THD+N measurement as these are most common.
I found with my open loop 'digital' amps (PowerDAC) that the non-linearities in the output inductor were a major source of THD and that this THD rose with frequency - so I changed to air-cored coils. Bridge deadtime is another major cause of distortion.
Another thought that occurs to me is that you may have switching jitter which (and I stand corrected on this) would create anharmonic distortion with a broad frequency spectrum, rather like noise. You could get a rough clue of the nature of the distortion by simply doing an unweighted S/N measurement. If it seems higher than what you would expect you might have a jitter problem.
I'm surprised that the feedback doesn't correct the distortion more, especially since you say you've taken it from after the LPF. Possibly your open-loop distortion is worse than you think? It could be useful to try doing some open loop distortion measurements (you'll probably have to make some temporary mods to the amp get a sensible fixed throughput gain to get this right). If your measurement setup facilitates real time continuous distortion measurements you could then do some tweaking and see what makes it better and what makes it worse.
Hope this helps.
Regards
John
Thanks.
In fact, my setup is rather crappy, a soundblaster and some nice programs.
One of them, by the way, allows real-time THD measurements, so it can be quite useful to find the cause.
About S/N: I measured it this way: I adjusted everything so I could get full power measurement reading 0dB with the PC, and then I shorted the input and saw the spectrum. It was always below -100dB. In fact, no matter if it was on or off, so it is almost below the soundcard noise level.
From my simulations, the loop gain is about 30dB up to about 1.5KHz, dropping to about 13dB at 20KHz. Not too much, so I will try to increase it a little bit by reducing the gain without compromising estability. Simulations achieve about 40dB up to a -3dB point at around 1.8k, going down to 21dB at 20KHz, so the difference is about 7-9dB, should be noticeable.
And about dead-time, I tried to reduce it and got almost no improvement, only increased peaks at the output. I will try to increase switching time a little bit (now 20ns), perhaps EMI is causing increased distortion.
I'll keep you posted, thanks for your ideas!
In fact, my setup is rather crappy, a soundblaster and some nice programs.
One of them, by the way, allows real-time THD measurements, so it can be quite useful to find the cause.
About S/N: I measured it this way: I adjusted everything so I could get full power measurement reading 0dB with the PC, and then I shorted the input and saw the spectrum. It was always below -100dB. In fact, no matter if it was on or off, so it is almost below the soundcard noise level.
From my simulations, the loop gain is about 30dB up to about 1.5KHz, dropping to about 13dB at 20KHz. Not too much, so I will try to increase it a little bit by reducing the gain without compromising estability. Simulations achieve about 40dB up to a -3dB point at around 1.8k, going down to 21dB at 20KHz, so the difference is about 7-9dB, should be noticeable.
And about dead-time, I tried to reduce it and got almost no improvement, only increased peaks at the output. I will try to increase switching time a little bit (now 20ns), perhaps EMI is causing increased distortion.
I'll keep you posted, thanks for your ideas!
Pierre: Also thinking aloud: If you have or can borrow a fast hi speed scope (at least 150MHz, preferably 500MHz) you could check out the PWM waveform at your output stage phase node. It's possible you are getting ringing or some other hf artifact which are causing false switching above a certain power level.
Having said that, on my 'open loopers' I've always found the THD very insensitive to hf junk on the waveform, but if the contamination is severe enough it can cause mis-switching and in the extreme, shootthru and MOSFET funerals. In which case a snubber is called for and I can advise you further on the design procedure.
Good luck
John
Having said that, on my 'open loopers' I've always found the THD very insensitive to hf junk on the waveform, but if the contamination is severe enough it can cause mis-switching and in the extreme, shootthru and MOSFET funerals. In which case a snubber is called for and I can advise you further on the design procedure.
Good luck
John
Thanks, John.
I had already measured the switching waveform with a 100MHz oscilloscope and carefully at the mosfet pins with a short ground tip and I must say that the waveform observed seems extracted from a book. When you start to modulate, at the highest power levels a little of overshoot starts to appear, but nothing above 15%.
Now my design seems quite robust, and I think shoot-through is not a problem, although in the earlier phase of the amp a lot of mosfets died until I debugged the layout and selected proper devices.
Thanks, anyway.
Pierre
I had already measured the switching waveform with a 100MHz oscilloscope and carefully at the mosfet pins with a short ground tip and I must say that the waveform observed seems extracted from a book. When you start to modulate, at the highest power levels a little of overshoot starts to appear, but nothing above 15%.
Now my design seems quite robust, and I think shoot-through is not a problem, although in the earlier phase of the amp a lot of mosfets died until I debugged the layout and selected proper devices.
Thanks, anyway.
Pierre
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.