You can change the frequency or clip a little heatsink on one leg to change the ratio of thermal/voltage effect.
To avoid expectation bias, I would do everything I could think of to eliminate the thermal effects (or at least account for them). With TC's of 50-200ppm I would not expect two randomly picked resistors to match well enough for the necessary resolution. If this is the case flipping the two DUT's should change the phase of the thirds. Running at a higher frequency (beyond the thermal time constant) will also dramatically reduce the effect.
Still bothered by the shorted input giving a reading. I would still look for a ground or "pin1" problem.
BTW in your other test, if you subtract two sets of readings after changing one of the IM frequencies a little most of the non-test related noise will subtract away. You would be left with two closely spaced sets of the real IM components.
Still bothered by the shorted input giving a reading. I would still look for a ground or "pin1" problem.
BTW in your other test, if you subtract two sets of readings after changing one of the IM frequencies a little most of the non-test related noise will subtract away. You would be left with two closely spaced sets of the real IM components.
Last edited:
Yes that would work, but then I don't have any test gear that would do that easily. So the next generation will use a better shield technique. Also I just don't bother showing much of the spectrum below 500hz these days, cleans it up enough for music merchants.
I am pretty comfortable with the current setup having leakage or unbalance around that level. The shield leads are tied to case via a small cap. (But will try lifting and/or shorting one, just to be sure.) This is just a proof of concept test. the final version should be at least 12db better.
I do still every so often screw up my ground schemes. They are a very important part of the work I do. O.T. In one arena I did the main power had to be supplied by two substations as the load was larger than a single one had unused capacity. The original design was to run a ring of ground rods around the perimeter and tie each substation's neutral and safety ground in to the closest point. I wrote the RFI explaining that at those currents the voltage drop would be several volts. This would force all of the buildings actual grounds to all be at different voltages. The solution was to run a cable between the two points of entry and ground that in the middle.
My experience is that devices when produce at the same time match within about 10%. But I will try a lower frequency as that is easier to do.
After I get a feel for what the best resistors are for this test, I should also do the classic wheatstone to see what pops out.
Keep in mind it is 1/2 hour per run.
Scott,
Just tried different ground schemes, no change.
Tried a different test lead dress inside the shielding case and now the shorted inputs are with a few db of the noise floor.
I hope that happicates you. (An English word I learned from a Chinese grad student. (Thus indicating logical is not always right))
Thanks for enforcing the rigor!
Just tried different ground schemes, no change.
Tried a different test lead dress inside the shielding case and now the shorted inputs are with a few db of the noise floor.
I hope that happicates you. (An English word I learned from a Chinese grad student. (Thus indicating logical is not always right))
Thanks for enforcing the rigor!
Simon, I am confused as to what the problem is, and it does not appear to be thermal. Whatever distortion mechanism is creating the 3'rd harmonic distortion, and it could be voltage change of resistance, rather than thermal, is tracking, and the distortion characteristic is stable with different voltage input.
For now.
Actually this has been great, I had never thought of that phasing idea before. Separation of true voltage co-efficient from simple TC issues is a tough problem. But we are already down to -145dB at multiple volts, I don't see any reason to pick one resistor over the other in a phono stage.
If the distortion reduces with signal level then it is not a problem. But there may be components that are at a fixed level not signal dependent, that is in my OPINION what is of interest.
I am currently running the 1% 499 ohm 1/4W metal film pair at 7.5 and 15 volts at 1K and 100hz. This may provide useful data on how to set power limits if the distortion is thermal.
It may also raise other issues.
Simon, you cannot contact me through diyaudio. My email address is: j_curl@earthlink.net. Please use it in future. My messages to and from you are blocked by the moderators.
You can't take that too literally, i.e. they're still there with no signal. That violates causality.
I'm not trying to give you a hard time, John, Walt, and I did a lot of this bridge null stuff in the 80's. It is a very powerful technique that obviates a lot of the sophisticated instruments that you need for direct measurement. With a good external sound card and software to do 1,000,000 + point FFT's you could probably do -180dB with no lab equipment at all.
Which AP are you using
Hello Simon7000
What AP system are you using and what test setup are using, could you do a screen dump of your AP setup.
Regards
Arthur
Hello Simon7000
What AP system are you using and what test setup are using, could you do a screen dump of your AP setup.
Regards
Arthur
Okay "As the signal approaches zero." But then there is always noise present so the signal is never zero for very long. An interesting question as to what effect noise shape has on perception!
AP System 2 FFT analyzer mode, 1K is usually the test signal, balanced 40 ohm out, balanced in. 511 -1023 points for 1K 4095 for 100hz. 4096 averages. If you need more just ask.
Pavel with the resolution I am asking for it takes all day to get a data set!
Will post a bit, but most will get published
I'm quite surprised at that. If the AP does FFT's that slowly you would be far better off taking the data to MATLAB or MATHCAD. You could also do a much better analysis. Larger FFT's give a much better noise floor than averaging small ones. At 96k 4 million points is less than a minute. In an hour you could get 30dB lower noise floor than 4096 4k FFT's. This could easily be set up.
Last edited:
- Status
- Not open for further replies.