Geddes on distortion measurements

Status
Not open for further replies.
I did notice that distortion levels lower as the test signals lowers. This is one reason why we may say that lower distortion does not mean a better design unless we know exactly what the conditions are.

Additionally, if the test signal is so low that only one bit is toggled during the measurement, it how can it truly represent the DUT? Get the bits to match precisely, and there is virtually no distortion.
 
Excellent work Michael, and interesting conclusions.

The noise signals don't really increase in amplitude with lower signal - did you measure all signals at same attenuation and then re-scale to make the original signals match?
 
Hi



AndrewT said:
Hi Mige,
now that you have the sound card working well at it's new job, would you like to guide us through the process of setting up a generic sound card to do similar data gathering and processing?



Hope Earl doesn't mind.



PART 1

You don't need that much to do this kind of measurements.

1) a measurement software that's working well
2) a capable soundcard
3) an DIY attenuator switch
4) cables
5) known source or DMM for calibration to dBV ( displaying in dBFS and noting a good guess about the range will do it as well )


1)
I am referring to AudioTester PC software here as it is shareware, very capable and easy to work with. You can download it under
www.audiotester.de


2)
The soundcard MUST be a 24 bit one. If it is only 44.1 kHz or working up to 196 kHz doesn't matter .
A ruler flat frequency response is always nice to have but of no concern here ( assuming it is fairly flat within a few dB between 1 - 10 kHz ).
24 bit soundcards differ greatly in their real performance and in their handling. Mine has several LineIN's that are fix gain and some additional ones that are combined Line / Mic / HighZ -IN's with adjustable gain potentiometers. Differential inputs are nice to have but not really necessary.

First step you have to do is to connect your OUT to your IN and find out at what settings you get best results. Start the measurement software ( close all other sound related software first to not have any conflicts ) and play around until you get at least a 90-100 dB of distortion free dynamic range.

Before measurement prepare the software settings:
- Choose " 2D-FFT " the left most big button and the " setup " button next to the " sound on " button . Set 1003 Hz sine wave at different levels there ( start with –3 dB going down / avoid to set to 1 kHz exactly as you get more processing artefacts then ).
- Open the measurement " setup " next to the " start " button and choose " Rife-Vinc3 " FFT Window and " 32768 " FFT points. Set " average " to " 10 ".
- Your soundcard must show up at the bottom and also the setting you have chosen under the pull down menu " options " > "Audio Out Device " / " options " > "Audio Out Parameter " / " options " > "Audio IN Device " / " options " > "Audio IN Parameter "



With my soundcard you need to
- set the digital OUT-level in AudioTester to around –12 dB FS ( "Setup" button )
- make sure the IN-level does not exceed –20 dB FS
- keep the gain potentiometer at around –40 dB max

This way my measurement limits are roughly 110 dB of distortion free dynamic range and resolution down to 100nV ( -140 dBV ) with the MIC-pre at FULL gain.
Keep in mind that my best case of 110 dB translate to poor 18-19 bit and that is reduced at full MIC-gain to 78 dB which translate to VERY poor 13 bit.
Distortion free full 24 bit resolution is utopia as can be seen.

But IF you are not able to achieve at least 90-100 dB of distortion free dynamic range your soundcard isn't good enough to do such measurements I guess.



PART 2

If you found out how to obtain a sufficient large distortion free dynamic range in loop back mode you are almost ready to measure your amps.

My power amps have a gain around 20-30 dB . In addition we would like to have a comfortable way to switch the stimulus in –20 dB steps several times.

The easiest way is to DIY solder a 4 times switch with input resistors of 1k / 10k / 100K / 1M Ohm and use as foot resistor a 100 Ohm potentiometer .
This way you can switch -20 / -40 /-60 /-80 dB easily and fine tune in between as well . With this " all analogue solution " you don't loose resolution at low output levels which is not necessarily the case if you try to attenuate the stimulus with your software mixer ( at least my soundcard kind of " swiches " into low-res / high distortion at some point when attenuating deeply through the software ).

Don't be afraid about the 1M Ohm - in terms of noise and humm and HF pick up this isn't relevant, it's the 100 Ohm foot resistor that counts. Though put the attenuator close to the power amp if using longer cables.
Don't use much less than 1k input resistor as you will create a too heavy load to your soundcard that for sure will produce higher distortions then.

The taper of the pot will be your OUT which you connect to the power amp IN.

Connect a resistive load of 4 or 8 Ohms to the OUT of the power amp and feed the amp OUT signal also back to the IN of your soundcard.
Use your normal speaker cable between load and amp as this is also the real world case and not every power amp likes VERY short cables to the load.


This setup should allow for a first measurement of your power amp. You can set the X-axis and Y-axis according to my plots if you like.
After having finished your first measurement at – lets say the -20 dB position of your attenuator – stop and move down with one click to the arrow at the right side of the plot window.
This way the measurement is frozen and you are ready to do the second one at the –40 dB position of your attenuator. Before starting the new measurement increase the frequency by - lets say - 100 Hz to shift the new graph to the right. Add 20 dB to the value of " pre amplifier gain left / right " in the pull down menu " Options " > " Calibration ". This shifts the new curve up.

Proceed with –60 dB and –80 dB and you are done.

To set the Y-scale to dBV in the pull down menu " options " > " Calibration " you first ought to calibrate. This is done in the pull down menu " Analyze " > " Level Units ". Follow the steps described there or connect a known source and vary the " pre amp gain " setting until the plot shows the right value.



Good Luck .
We all would love to see as many measurements as possible !




Greetings
Michael
 
Hi


soongsc said:
I did notice that distortion levels lower as the test signals lowers. This is one reason why we may say that lower distortion does not mean a better design unless we know exactly what the conditions are.

Additionally, if the test signal is so low that only one bit is toggled during the measurement, it how can it truly represent the DUT? Get the bits to match precisely, and there is virtually no distortion.



Soongsc, MBK had begun this thread because of Earls claim that at low levels ( power ) amps start to have relatively high harmonic distortion due to crossover artefacts – AND that this matters .

He presented a specific method he has developed to look "through the noise" that usually masks that distortion and did also show some example measurements .

My measurements in contrary indicate that well built amps don't suffer from that issue. It would be interesting to find out what is usual nowadays based at measurements of a representative number of devices .


Earl, could you comment on my measurements with respect to the method in comparison with yours and about the results I've got ?


Greetings
Michael
 
Hi


MBK said:
Excellent work Michael, and interesting conclusions.

The noise signals don't really increase in amplitude with lower signal - did you measure all signals at same attenuation and then re-scale to make the original signals match?


MBK, the noise is always at the same absolute level and does NOT depend on the stimulus level. The whole plots ( -40 / -60 / -80 dB ) are shifted upwards manually to increase clarity like Earl suggested in his measurements. ONLY the lowest plot ( -20 dB setting of the attenuator ) is on its right scale here.

What sets the limits in terms of absolute noise floor is mainly the resolution you set the FFT Points to. The higher the frequency resolution ( more FFT Points ) the lower the noise floor - down to below –140 dBFS in the extreme with > 100.000 points.
In our case this does not make any sense as the soundcard itself distorts at around –105 dB below stimulus in the best case . You COULD display a very low noise floor but also the distortion of you card. With the recommended setting of 32.768 FFT points the noise masks the card's distortion figure ( and measurement is way faster as well ).

Hope this makes sense for you.

Greetings
Michael
 
mige0 said:
MBK, the noise is always at the same absolute level and does NOT depend on the stimulus level.

...

Hope this makes sense for you.

That's actually what I assumed, so yes it makes sense to me 😉 . The reason why I asked was, there are two ways to get the graph you showed
a) manual shifting of an increasingly low level signal upward - this is what you did and this is the cleanest solution
b) working with a soundcard preamplifier that gets the lower level signals back up to the same level, and which is attenuated at input of the soundcard for the higher signal levels. Then of course you might get different noise levels. For instance I have a non-defeatable pot at my sound inputs, there is no "always 100%" line in on my card. Of course I can leave this pot at a convenient position and only work with attenuation of the sound out going to the DUT.

Interesting that your sound in must be attenuated -20 dB for best results. On my card I hardly get any distortion improvement on loopback by setting the input to lower levels. But my sound output must be at -10 dB for best results. (software setting, beyond this no further improvement).

One question comes to mind though, for those amps with higher noise signature you showed, this noise might be poorer common mode / PSU etc. noise, and depending on conditions there might be a distortion signature masked by this higher noise.

Still it looks like you said, in your DUTs nothing dramatic seems to occur at lower signal levels.
 
Hi


PMA said:
You should do something with your digital noise, otherwise the results do not seem reliable.

Or are they all 50Hz multiples? Then do something with your measurement setup. These seamingly not important issues do count.


PMA, having worked hard to diminish that effects with NO success I have come to the not so pleasing conclusion that it is REAL measurement.
The soundcard IN's are symmetric and as I don't trust what I haven't made myself, I've even tried a symmetric line receiver in between that has roughly 80-100 dB of CMRR – NO effect. When connecting BOTH soundcard inputs to the same amp OUT there is NO signal – meaning this is NO CMRR issue.

Yes these artefacts ARE multiples of the 50 Hz mains.
They start at roughly 100 Hz –80 dBV so its definitely more than 100 dB below full output of the power amp which isn't sooo bad at all.

Interesting though that the car amp with it's switched PSU ( 12V car battery as power source under normal circumstances ) does behave better in this respect. The car amp was fed by a good laboratory PSU with capacitors in between.
Identical setup for all measurements shown.

Any ideas how to improve the conventional PSU's any further ( the MOSFET amps are already heavily modified ) ?

Greetings
Michael
 
Hi


MBK said:


That's actually what I assumed, so yes it makes sense to me 😉 . The reason why I asked was, there are two ways to get the graph you showed
a) manual shifting of an increasingly low level signal upward - this is what you did and this is the cleanest solution
b) working with a soundcard preamplifier that gets the lower level signals back up to the same level, and which is attenuated at input of the soundcard for the higher signal levels. Then of course you might get different noise levels.


MBK, yes your version a) is the only clean one.
I didn't touch ANYTHING except the attenuator for my measurements - and shifting frequency and " pre amplifier gain left / right " of course .




MBK said:


For instance I have a non-defeatable pot at my sound inputs, there is no "always 100%" line in on my card. Of course I can leave this pot at a convenient position and only work with attenuation of the sound out going to the DUT.

Interesting that your sound in must be attenuated -20 dB for best results. On my card I hardly get any distortion improvement on loopback by setting the input to lower levels. But my sound output must be at -10 dB for best results. (software setting, beyond this no further improvement).


My guess is that the opamps in my soundcard are the limiting factor here.
If forced to their limits they distort and with too low levels they contribute to noise. Its always a compromise - your card might have better opamps OR higher noise floor .
My OUT has it's sweet spot at –12 dBFS ( and roughly an other –20 dB down ) and the IN is at its best at / below -20 dBFS not that much changing if you make use of the MIC pre or not.



MBK said:


One question comes to mind though, for those amps with higher noise signature you showed, this noise might be poorer common mode / PSU etc. noise, and depending on conditions there might be a distortion signature masked by this higher noise.

Still it looks like you said, in your DUTs nothing dramatic seems to occur at lower signal levels.
MBK said:



Agree completely.



Greetings
Michael
 
Hi


PMA said:
Yes, it is not easy. Especially for power levels like 10mW and 100mW.


PMY, to put things into perspective:
the lowest level measurement shown is a 1 kHz sine wave at the power amp out at –70 dBV.

-70 dbV equals 3x10^-4 Volts which equals at the load of 8 Ohm to
1.25x10^-8 Watt
= 0,00 012 5 mW
= 12.5 nW

Its really LOW level stuff we are investigating into, no?


Greetings
Michael
 
mige0 said:
Hi







Hope Earl doesn't mind.



PART 1

You don't need that much to do this kind of measurements.

1) a measurement software that's working well
2) a capable soundcard
3) an DIY attenuator switch
4) cables
5) known source or DMM for calibration to dBV ( displaying in dBFS and noting a good guess about the range will do it as well )


1)
I am referring to AudioTester PC software here as it is shareware, very capable and easy to work with. You can download it under
www.audiotester.de


2)
The soundcard MUST be a 24 bit one. If it is only 44.1 kHz or working up to 196 kHz doesn't matter .
A ruler flat frequency response is always nice to have but of no concern here ( assuming it is fairly flat within a few dB between 1 - 10 kHz ).
24 bit soundcards differ greatly in their real performance and in their handling. Mine has several LineIN's that are fix gain and some additional ones that are combined Line / Mic / HighZ -IN's with adjustable gain potentiometers. Differential inputs are nice to have but not really necessary.

First step you have to do is to connect your OUT to your IN and find out at what settings you get best results. Start the measurement software ( close all other sound related software first to not have any conflicts ) and play around until you get at least a 90-100 dB of distortion free dynamic range.

Before measurement prepare the software settings:
- Choose " 2D-FFT " the left most big button and the " setup " button next to the " sound on " button . Set 1003 Hz sine wave at different levels there ( start with –3 dB going down / avoid to set to 1 kHz exactly as you get more processing artefacts then ).
- Open the measurement " setup " next to the " start " button and choose " Rife-Vinc3 " FFT Window and " 32768 " FFT points. Set " average " to " 10 ".
- Your soundcard must show up at the bottom and also the setting you have chosen under the pull down menu " options " > "Audio Out Device " / " options " > "Audio Out Parameter " / " options " > "Audio IN Device " / " options " > "Audio IN Parameter "



With my soundcard you need to
- set the digital OUT-level in AudioTester to around –12 dB FS ( "Setup" button )
- make sure the IN-level does not exceed –20 dB FS
- keep the gain potentiometer at around –40 dB max

This way my measurement limits are roughly 110 dB of distortion free dynamic range and resolution down to 100nV ( -140 dBV ) with the MIC-pre at FULL gain.
Keep in mind that my best case of 110 dB translate to poor 18-19 bit and that is reduced at full MIC-gain to 78 dB which translate to VERY poor 13 bit.
Distortion free full 24 bit resolution is utopia as can be seen.

But IF you are not able to achieve at least 90-100 dB of distortion free dynamic range your soundcard isn't good enough to do such measurements I guess.



PART 2

If you found out how to obtain a sufficient large distortion free dynamic range in loop back mode you are almost ready to measure your amps.

My power amps have a gain around 20-30 dB . In addition we would like to have a comfortable way to switch the stimulus in –20 dB steps several times.

The easiest way is to DIY solder a 4 times switch with input resistors of 1k / 10k / 100K / 1M Ohm and use as foot resistor a 100 Ohm potentiometer .
This way you can switch -20 / -40 /-60 /-80 dB easily and fine tune in between as well . With this " all analogue solution " you don't loose resolution at low output levels which is not necessarily the case if you try to attenuate the stimulus with your software mixer ( at least my soundcard kind of " swiches " into low-res / high distortion at some point when attenuating deeply through the software ).

Don't be afraid about the 1M Ohm - in terms of noise and humm and HF pick up this isn't relevant, it's the 100 Ohm foot resistor that counts. Though put the attenuator close to the power amp if using longer cables.
Don't use much less than 1k input resistor as you will create a too heavy load to your soundcard that for sure will produce higher distortions then.

The taper of the pot will be your OUT which you connect to the power amp IN.

Connect a resistive load of 4 or 8 Ohms to the OUT of the power amp and feed the amp OUT signal also back to the IN of your soundcard.
Use your normal speaker cable between load and amp as this is also the real world case and not every power amp likes VERY short cables to the load.


This setup should allow for a first measurement of your power amp. You can set the X-axis and Y-axis according to my plots if you like.
After having finished your first measurement at – lets say the -20 dB position of your attenuator – stop and move down with one click to the arrow at the right side of the plot window.
This way the measurement is frozen and you are ready to do the second one at the –40 dB position of your attenuator. Before starting the new measurement increase the frequency by - lets say - 100 Hz to shift the new graph to the right. Add 20 dB to the value of " pre amplifier gain left / right " in the pull down menu " Options " > " Calibration ". This shifts the new curve up.

Proceed with –60 dB and –80 dB and you are done.

To set the Y-scale to dBV in the pull down menu " options " > " Calibration " you first ought to calibrate. This is done in the pull down menu " Analyze " > " Level Units ". Follow the steps described there or connect a known source and vary the " pre amp gain " setting until the plot shows the right value.



Good Luck .
We all would love to see as many measurements as possible !




Greetings
Michael


This is a very nice post, Michael. It will be helpful to many here. Thanks for putting in the time.

Cheers,
Bob
 
mige0 said:
Hi






Soongsc, MBK had begun this thread because of Earls claim that at low levels ( power ) amps start to have relatively high harmonic distortion due to crossover artefacts ?AND that this matters .

He presented a specific method he has developed to look "through the noise" that usually masks that distortion and did also show some example measurements .

My measurements in contrary indicate that well built amps don't suffer from that issue. It would be interesting to find out what is usual nowadays based at measurements of a representative number of devices .


Earl, could you comment on my measurements with respect to the method in comparison with yours and about the results I've got ?


Greetings
Michael
First of all I must confess I am not testing amp very much because I work more with speakers.

I agree with Earl that theoretically the harmonics should increase compared to the relative signal amplitude, but the absolute effect might not be that audible. However, if your method does not do the averaging that Earl explains, then it cannot reveal what Earl is talking about because we cannot compare the results from two different methods. If the AudioTester has some kind of averaging capabilities as we talked about earlier, then we would be able to compare. Additionally, since we also want to take advantage of the full 24bit resolution on the inputs that internal hardware gain must be high enough so that the recorded input can reach close to full scale. This is not possible with most of the sound cards on the market without modification.

The artifacts caused by power supply and the way the amps handle reactive loads have much more effect on sound reproduction. In some of my measurements, using distortion measurements capability in SoundEasy, I discovered that lowering the hamonics in the low frequency range (below 100Hz) improved reproduction quality significantly.

I have two amps with regulated linear power supplies and have seen amps from a new brand called DELIGHT AUDIO that uses switching power supply. But I have not measured them.
 
mige0 said:
Soongsc, MBK had begun this thread because of Earls claim that at low levels ( power ) amps start to have relatively high harmonic distortion due to crossover artefacts – AND that this matters .

He presented a specific method he has developed to look "through the noise" that usually masks that distortion and did also show some example measurements .

My measurements in contrary indicate that well built amps don't suffer from that issue. It would be interesting to find out what is usual nowadays based at measurements of a representative number of devices .


Earl, could you comment on my measurements with respect to the method in comparison with yours and about the results I've got ?


Michael - Excellent work. I'm now back in the US after a couple of (tough!) weeks in Asia.

First, I have never claimed that all amps have audible or significant crossover distortion, only that in the set of common measurements this mechanism is not seen even though it can be highly audible. That any "competent" amp has taken care of this is no surprise to me and is in fact completely consistant with my own experience. But there CAN BE an amp which has large crossover distortion that will test out just as good as any other amp on the common tests, but sound far worse. This is the point.

I do find that you test is lacking the ability to look below the noise and power supply problems which is exactly the situation where my test would shine since it would reject the PS hum, etc. and one could see the true nature of the harmonics at low levels which could be completely obscured in much of your test results.

Remember that I am a strong sceptic of amps sounding different from one another and I have concluded, from experince, that most amps sound fine and indistinguishable from each other (a result shown more than once in blind subjective tests). But I also acknowledge that there are things that one can hear in an amp that do not show up in the measurements - as seen in the typical sales brochures and Rag writups. It is these later interesting, if rare, events that I am addressing in my tests.
 
Hi


gedlee said:



Michael - Excellent work. I'm now back in the US after a couple of (tough!) weeks in Asia.

First, I have never claimed that all amps have audible or significant crossover distortion, only that in the set of common measurements this mechanism is not seen even though it can be highly audible. That any "competent" amp has taken care of this is no surprise to me and is in fact completely consistant with my own experience. But there CAN BE an amp which has large crossover distortion that will test out just as good as any other amp on the common tests, but sound far worse. This is the point.

I do find that you test is lacking the ability to look below the noise and power supply problems which is exactly the situation where my test would shine since it would reject the PS hum, etc. and one could see the true nature of the harmonics at low levels which could be completely obscured in much of your test results.



Earl, thanks for your positive comment and good to have you back at the board.

I would like to perform your method of measurement and sent you 2 or three emails / PM's for the .wav request but it seems as they didn't reach you. Please PM me your valid email once you have same time.

In the measurements ( post #16 ) you didn't tell if the Y-axis is in dBV scale. IF it is, the noise floor in my measurements are only slightly higher as in yours – down at –110 dBV with my measurement – down at –115 dBV with yours.
As I outlined I CAN go lower with no problem BUT the soundcards own distortion sets the limits so I did set the noise to where it just masks the soundcards own distortion.
I did show this measurement limits with loop back mode in its own graphs.

From your measurements one can't tell yours measurement limits.
Best distortion free dynamic range shown in your measurements is –90 dB for high stimulus - mine is roughly the same
Best distortion free dynamic range in shown your measurements is –110 dB for low stimulus - mine is roughly the same


gedlee said:




Remember that I am a strong sceptic of amps sounding different from one another and I have concluded, from experince, that most amps sound fine and indistinguishable from each other (a result shown more than once in blind subjective tests). But I also acknowledge that there are things that one can hear in an amp that do not show up in the measurements - as seen in the typical sales brochures and Rag writups. It is these later interesting, if rare, events that I am addressing in my tests.



The NAIM amps were borrowed for this measurements and I haven't listened to NAIM stuff for many years now.
Interesting thing was, I IMEDIATELY recognised their signature. They really have very strong pro's ( and some con's I don't like that much )


Greetings
Michael
 
mige0 said:
I would like to perform your method of measurement and sent you 2 or three emails / PM's for the .wav request but it seems as they didn't reach you. Please PM me your valid email once you have same time.

In the measurements ( post #16 ) you didn't tell if the Y-axis is in dBV scale. IF it is, the noise floor in my measurements are only slightly higher as in yours – down at –110 dBV with my measurement – down at –115 dBV with yours.
As I outlined I CAN go lower with no problem BUT the soundcards own distortion sets the limits so I did set the noise to where it just masks the soundcards own distortion.
I did show this measurement limits with loop back mode in its own graphs.



I didn't actually bother to scale the data in my examples.

But I think that you missed some of my point. I'm talking about the seveer hum in the PSs shown in most of your plots, not the background noise floor. This hum masks the harmonics in your plots, but the ear would still hear them. My technique would elliminate the hum allowing the harmonics to be seen.

I didn't respond to any of the notes that I got because, as I said, I found an error and wanted to correct it first before sending anything out. Maybe this stay in the US I will get to finish that work. My E-mail is egeddes@gedlee.com.
 
mige0 said:
Hi,





Soongsc, would be great if you could do and post the results here. The more the better...

Greetings
Michael
Most of my tests are related with reactive loads and how they affect sound, this is quite different from the subject. I never look at data alone, but rather try to find relationship between data and audible differences. I will start looking for good amps at some point in time, ones that have high power/volume ratio and some other goodies, and will post measurements when I start doing tests.
 
Hi

gedlee said:


I didn't actually bother to scale the data in my examples.




🙁




gedlee said:


But I think that you missed some of my point. I'm talking about the seveer hum in the PSs shown in most of your plots, not the background noise floor. This hum masks the harmonics in your plots, but the ear would still hear them. My technique would elliminate the hum allowing the harmonics to be seen.





Earl, would like to give your method a try as soon you found some time to do the fixes needed.

In the meantime I made measurements with the MOSFET amp on a well stabilised laboratory PSU to investigate the effect of stabilised versus non stabilised power supply.


An externally hosted image should be here but it was not working when we last tested it.


Here we see the MOSFET amp again with unstabilised PSU like shown in post 179


An externally hosted image should be here but it was not working when we last tested it.


Here we see the MOSFET amp fed by a stabilised laboratory PSU .
The hum is gone and a -95 dB 1st harmonic can be seen that was hidden before .






---------------------------------


Verifying my measurement limits:

An externally hosted image should be here but it was not working when we last tested it.


What can be seen here is that the soundcard's response is completely below in loop back mode.

All settings of the soundcard are left untouched here and the plots are NOT shifted vertically ( the Y-axis dBV scale is valid for BOTH plots )


MikeB, couldn't you show some low level measurements of your symasym?

PMA, couldn't you show some low level measurements of your PM-AB1 ?


Greetings
Michael
 
Michael

These seem to be very clean measurements. Certainly better than any of the amps that I tested - at least with a decent power supply. You see where the power supply problem COULD have abeen a major issue. It could have masked all knids of distortion products which would have been audible. You were able, in this case, to use an external supply. But what if that were not possible? You would not have been able to come to any conclusions about the amp from those tests.
 
Status
Not open for further replies.