[MindsMirror]

 
There are still a few thing that may be worth trying, however. What happens if you use 16.5 ohms serial resistors (2x33 ohms in parallel), or if you add the 33 ohms resistors to the sound card input instead (as close to the amplifier as possible) ?

With the splitter plugged directly into the amp, headphones directly into the splitter, and the line input connected to the splitter through 33Ohms, the distortion starts at 1Vrms. This time, unplugging the line input+resistor does not remove the audible distortion, so it seems like adding the resistor the the line input as you suggested prevents the interaction that was happening before. The first harmonic in the electrical measurement is at -60dB, but I would say that the audible distortion I can hear coming from the headphone is much higher than that. That seems to suggest that there is high distortion coming from the driver, and some of it is fed back to the amp/measurement.
 
I don't know if I mentioned this, or if it's relevant and would affect anything, but when I was testing with the series resistance between the amp and headphones, I only connected and measured the left channel, the right channel was open. I'm using alligator wires to connect the resistors, and the ends are almost too fat to fit on the right channel ring, so I didn't try it in case it would short the amp.

 

[Greenears]

  With the splitter plugged directly into the amp, headphones directly into the splitter, and the line input connected to the splitter through 33Ohms, the distortion starts at 1Vrms. This time, unplugging the line input+resistor does not remove the audible distortion, so it seems like adding the resistor the the line input as you suggested prevents the interaction that was happening before. The first harmonic in the electrical measurement is at -60dB, but I would say that the audible distortion I can hear coming from the headphone is much higher than that. That seems to suggest that there is high distortion coming from the driver, and some of it is fed back to the amp/measurement.
 
I don't know if I mentioned this, or if it's relevant and would affect anything, but when I was testing with the series resistance between the amp and headphones, I only connected and measured the left channel, the right channel was open. I'm using alligator wires to connect the resistors, and the ends are almost too fat to fit on the right channel ring, so I didn't try it in case it would short the amp.

I'm clueless why you're doing this but there must be a reason
 
Anyhoo - basic electrical engineering.  If I understand you correctly, the 33 ohm is on the split branch in series (IE inline between the inner "hot" connector of RCA to line-in with the outer "ground" shield of the RCA not touched at all) going directly to the line input of a measurement sound card.
 
If so .... line in is usually high resistance, many k ohm.   So the 33 ohm won't make a jot of difference no matter what is going on with the source and sink, which you can check by adding/removing it and repeating the measurement.   That is because you add the 33 to the many k ohm of line-in resistance which will make no noticeable difference to the total resistance on that branch, or any other branch.  If this is not the case and you see a big jump in measurement, you have a short, cold solder joint or something else wrong with your setup.

 

[MindsMirror]

 
I'm clueless why you're doing this but there must be a reason
 
Anyhoo - basic electrical engineering.  If I understand you correctly, the 33 ohm is on the split branch in series (IE inline between the inner "hot" connector of RCA to line-in with the outer "ground" shield of the RCA not touched at all) going directly to the line input of a measurement sound card.
 
If so .... line in is usually high resistance, many k ohm.   So the 33 ohm won't make a jot of difference no matter what is going on with the source and sink, which you can check by adding/removing it and repeating the measurement.   That is because you add the 33 to the many k ohm of line-in resistance which will make no noticeable difference to the total resistance on that branch, or any other branch.  If this is not the case and you see a big jump in measurement, you have a short, cold solder joint or something else wrong with your setup.

Have you even read any of the other posts in the thread? We have already established that there is probably something wrong with the setup. I am well aware that according to basic electrical engineering, 33 ohms is insignificant against the resistance of the line input, but I can tell you that when I add a 33 ohm resistor in front of it does decrease the distortion. Basic electrical engineering does not explain where the distortion came from, in fact it usually ignores distortion all together and assumes ideal sources and components, so obviously it's not that simple.

 

[stv014]

If so .... line in is usually high resistance, many k ohm.   So the 33 ohm won't make a jot of difference no matter what is going on with the source and sink, which you can check by adding/removing it and repeating the measurement.

 
Actually, the input can possibly become low impedance under certain conditions:
- if there is a small parallel capacitance on it (for example added intentionally to filter out RF interference, or it could be just parasitic capacitance from cables etc.), then it has low impedance at very high frequencies, which may make the amplifier unstable
- if the input voltage is too high, depending on the design of the input; this seems unlikely, though, since the problem also appears with the line input which is known to be usable up to 2 Vrms, not to mention it would also have lead to very high output current, possibly damaging the sound card and/or amplifier
 

  This time, unplugging the line input+resistor does not remove the audible distortion, so it seems like adding the resistor the the line input as you suggested prevents the interaction that was happening before. The first harmonic in the electrical measurement is at -60dB, but I would say that the audible distortion I can hear coming from the headphone is much higher than that.

 
So, adding the resistors to the sound card input reduced the distortion from worse than -30 dB to -60 dB ? And there is no high distortion either when the resistors are used as the load, but not anywhere else ? That would agree with the theory that it could be an amplifier stability issue.

 

[MindsMirror]

 
So, adding the resistors to the sound card input reduced the distortion from worse than -30 dB to -60 dB ? And there is no high distortion either when the resistors are used as the load, but not anywhere else ? That would agree with the theory that it could be an amplifier stability issue.

Okay, if that is the problem, is there anything I can actually do about it?

 

[stv014]

Well, if there is no problem when only the headphones are connected to the Magni, and the loopback setup is fixed by adding serial resistors to the input, then there is not really anything that needs to be done, except maybe to find out and confirm exactly what was causing the distortion.

 

[Greenears]

   
Actually, the input can possibly become low impedance under certain conditions:
- if there is a small parallel capacitance on it (for example added intentionally to filter out RF interference, or it could be just parasitic capacitance from cables etc.), then it has low impedance at very high frequencies, which may make the amplifier unstable
 

 
Hmmm.   Interesting idea but seems unlikely to me. 
 
First of all surely this would only happen at high frequency well outside the audio band, and THD+N measurement must be filtered to the audio band, or usually A weighted.  Narrow the bandwidth window to rule this out?  Try just a pure sine sweep 20 Hz - 20 kHZ and also try the sweep at various voltage levels.  Presumably that would expose any frequency dependence.
 
Another thing to try is to use crocodile clips or whatever to shunt around the 33 ohm resistor, so you can try with and without 33 ohm without jiggling any cables.  Or replace the 33 ohm with a wire or switch or whatever.  A poor connection is much more likely to me.
 
I don't think we fully understand the schematic here and the nature of the setup.  The most likely explanation is the line in is not actually a line in its a mic in or has some optional extra input gain stage and the cause is a simple overload.
 
Anyhow, good luck!

 

[MindsMirror]

But there is still a problem, the distortion still happens at about 0.8V with the headphones plugged directly into the Magni. The loop back measurement without the resistor in line with the line input apparently just made it start at a lower voltage. With the resistor in line with the Magni's output, the headphones can go significantly louder without the same audible distortion.

 

[Greenears]

Sorry I don't know what a magni is and I don't have a clear vision of your circuit. I believe a simple sine sweep will tell all. I feel like it is a simple problem to solve but need some data. Best of luck.

 

[stv014]

  Hmmm.   Interesting idea but seems unlikely to me. 
 
First of all surely this would only happen at high frequency well outside the audio band, and THD+N measurement must be filtered to the audio band, or usually A weighted.  Narrow the bandwidth window to rule this out?  Try just a pure sine sweep 20 Hz - 20 kHZ and also try the sweep at various voltage levels.  Presumably that would expose any frequency dependence.

 
While there is not enough data to confirm that amplifier instability is what is causing the problem, it is actually possible for RF oscillation to degrade performance even in the audio band, as the amplifier has to output a very high frequency signal at near clipping level. The graph below is only from a simulated feedback loop, but it shows similar effects (green: stable, red: unstable):

 

Originally Posted by Greenears /img/forum/go_quote.gif
 
I don't think we fully understand the schematic here and the nature of the setup.  The most likely explanation is the line in is not actually a line in its a mic in or has some optional extra input gain stage and the cause is a simple overload.

 
In that case, the increased distortion would not also be audible on the headphones, though. And, according to the OP, the distortion does not appear when the load (headphones) is replaced with a pair of 33 ohms resistors.

 

[stv014]

Sorry I don't know what a magni is

 
It is a Schiit amplifier, that has apparently been discontinued, and replaced by the Magni 2.

 

[stv014]

  But there is still a problem, the distortion still happens at about 0.8V with the headphones plugged directly into the Magni. The loop back measurement without the resistor in line with the line input apparently just made it start at a lower voltage. With the resistor in line with the Magni's output, the headphones can go significantly louder without the same audible distortion.

 
It would be best if the tests could be repeated with another low impedance amplifier, like the O2. Other than that, more loopback measurements could be performed with samples I will post later, although I am not sure if these would reveal anything important that is not already known.
 
Edit: for the additional measurements, copy this to a simple text file:

 sine 0.25 2.3 1500 0 0.5 0 sine 0.25 2.3 2000 0 0 0.5 mls 3 2.1 1001000100100000 0.25 0.25 sine 5.5 10.5 16 1 30000 1 0 0.5 0.5 sine 16.4 0.1 60 0 0.01 0.01 sine 16.5 4 60 0.01 60 1 0 1 1 sine 20.5 0.1 60 0 1 1 sine 20.9 0.1 1000 0 0.01 0.01 sine 21 4 1000 0.01 1000 1 0 1 1 sine 25 0.1 1000 0 1 1 sine 25.4 0.1 15000 0 0.01 0.01 sine 25.5 4 15000 0.01 15000 1 0 1 1 sine 29.5 0.1 15000 0 1 1 jtest 30 0 1

Then generate a WAV file with the 'testgen' utility from this package (replace amptest.txt with whatever the text file was saved as), using the following command:

 testgen.exe amptest.txt amptest.wav 96000 24

 
If you post loopback recordings of this in various configurations (with and without the serial resistors, with resistors as the load, etc.), I can analyze them. I recommend using the line input, and setting the playback volume so that 0 dBFS would be about 1.5 Vrms on the load.

 

[Greenears]

I took at look at the Magni2 specs online (they don't post Magni1) and it's a conventional <0.2 ohm output impedance.  So it's a standard voltage mode amplifier with high gain loop.  I really doubt there can be high frequency instability that is so well understood and easy to stabilize by design with phase margin etc.  Frankly if there is oscillation it's probably just defective.
 
I don't see any need for extra resistors.  Try completely different cables with no resistors also to rule out a bad cable.  Cheap $5 cable will do just fine.  Determine max V and max I (at the point where you just notice distortion beginning to increase) and stay under 90% of those values.  If it's not loud enough at those levels return it.  Amplifiers are simple I see people overthink this on forums way too much.
 
Unfortunately they don't spec the max V and I on the spec sheet.  Annoying but that's par for the course with amplifiers.

 

[MindsMirror]

   
It would be best if the tests could be repeated with another low impedance amplifier, like the O2. Other than that, more loopback measurements could be performed with samples I will post later, although I am not sure if these would reveal anything important that is not already known.
 
Edit: for the additional measurements, copy this to a simple text file:

 sine 0.25 2.3 1500 0 0.5 0sine 0.25 2.3 2000 0 0 0.5mls 3 2.1 1001000100100000 0.25 0.25sine 5.5 10.5 16 1 30000 1 0 0.5 0.5sine 16.4 0.1 60 0 0.01 0.01sine 16.5 4 60 0.01 60 1 0 1 1sine 20.5 0.1 60 0 1 1sine 20.9 0.1 1000 0 0.01 0.01sine 21 4 1000 0.01 1000 1 0 1 1sine 25 0.1 1000 0 1 1sine 25.4 0.1 15000 0 0.01 0.01sine 25.5 4 15000 0.01 15000 1 0 1 1sine 29.5 0.1 15000 0 1 1jtest 30 0 1

Then generate a WAV file with the 'testgen' utility from this package (replace amptest.txt with whatever the text file was saved as), using the following command:

 testgen.exe amptest.txt amptest.wav 96000 24

 
If you post loopback recordings of this in various configurations (with and without the serial resistors, with resistors as the load, etc.), I can analyze them. I recommend using the line input, and setting the playback volume so that 0 dBFS would be about 1.5 Vrms on the load.

Here are some test results.
https://drive.google.com/folderview?id=0B-uizCXMY0QBfkJWMUw5YzdaSUc5S01zeWdoeUczMWhYTWJXeExpbzlhaUxjekUxN2c3Umc&usp=sharing
 
I made a couple of 33 Ohm cables (resistors in series with the left and right channel wires) so I didn't have to use alligator clips to attach everything.
 
All of the tests are measured with the soundblaster Z line input, which is connected to the splitter via one of my 33 ohm cable. The headphone are also connected to the splitter. In the test 1 files, the splitter is connected directly to the amp, in the test 2 files it is connected to the amp via the second 33 ohm cable. I did the tests at two different volumes. All of the recorded files are with the line input at unity gain, so 0dBFS in the files is always 2.2V, you can see in the recorded signal how loud the amp was set. One of the tests has the amp set with 0dBFS at about 1.5V, the other is just below 2.2V, higher than that and the line input will clip.
 
Let me know if any other test configurations would be useful. Should I do one with the line input connected by a regular cable? Unfortunately I don't have access to another low impedance amp, The impedance on the sound card's amp is about 28 Ohm.

 

[stv014]

  I took at look at the Magni2 specs online (they don't post Magni1) and it's a conventional <0.2 ohm output impedance.  So it's a standard voltage mode amplifier with high gain loop.  I really doubt there can be high frequency instability that is so well understood and easy to stabilize by design with phase margin etc.

 
Actually, the Magni/Magni2 do not seem to use the "standard" method of stabilizing the feedback loop with dominant pole compensation, as noted by the FAQ:
 

 
What’s the big deal about “constant feedback?”
Magni 2 and Magni 2 Uber use a gain stage that has an open-loop bandwidth greater than 20kHz. That means that the feedback is constant across the entire audio band. By comparison, most op-amps start rolling off at 10-100Hz, which means there could be 20-30dB less feedback at high gain.