And I got -166 on the third!
No one has asked about DC with the AC superimposed on it anyone care to guess what that does?
So at line levels we are at -190dB or less, what on earth can this have to do with circuit performance? DC bias makes seconds.
Ikoflexer - You have to take into account the noise BW of the FFT bin before a final number.
Last edited:
Yes,
But Demian only sent me good resistors. The bad ones can be bad enough to actually matter.
What I find interesting is that 25 cent resistors often have lower noise and less distortion than the very expensive ones.
The other interesting item to note is that my measurements pretty much mirror "Golden" ears experience. So how did they get their ratings by listening to things that should not really be showing?
DC bias does make evens.
Attached is a typical carbon comp, not the worst of the lot. Notice it is used at 62 mw. It gets worse by 12-15 db at rated power.
Very dangerous line of reasoning. I also thought you were finding cheap resistors that were just as good, does that mean JC can punt the ones he uses?
BTW do you know were the skirt on the 1K in that plot comes from?
Thank you Scott. Do you really think it will make such a huge difference?
Here's a little experiment I just ran. Measured a 0.780mV RMS 100kHz sine wave from a signal generator with the Fluke 8920A true rms voltmeter. Fed this signal to a HP3585A spectrum analyzer. The fundamental peaked at about 10.7dBm and the noise floor was around -80.7dBm. When Ref Level Volt it said the peak was at about 760mV, and the marker on the noise floor showed about 24uV. These numbers fit my previous calculation. In this case the noise floor was at 0.76/10^(90/20) which is about 24uV. Why would the noise BW of the FFT bin not have to be taken in account with the HP 3585A?
Thanks for your patience.
Very dangerous line of reasoning. I also thought you were finding cheap resistors that were just as good, does that mean JC can punt the ones he uses?
BTW do you know were the skirt on the 1K in that plot comes from?
Not really a dangerous line as long as you realize an infinite number of monkeys hacking at typewriters will... Some audio experts will get it right some wrong, it is the long term track record of some golden ears that is interesting. Ignoring interesting things may be just as silly as blindly following.
Most likely there are ten resistors in a circuit possibly 100 so some care is required.
As to what resistors JC uses, it turned out the 10 cent Resistas were better than most even the $10 ones!
Yes I did find some cheapies that work very well, but when I got samples I got some as bulk and some as taped. They were listed with different countries of origin and behaved differently. I also found items listed as by different manufacturers that looked and behaved the same. So at my level where I do not buy from the manufacturers, I would have a distributor hold inventory for me while I test a sample.
As to the skirt, Scott give yourself a gold star. I was wondering if anyone would notice. I assume you also noticed the higher noise level.
I thought the spike at 1kHz WAS the fundamental. Looking back in the thread I found this
"The fundamental is at 0db which is 15.81 volts. The spike at 1k is the residual after trimming for the lowest value with a 10 turn 20 ohm equivalent trim pot at the base of the 4 1k resistors in the bridge. Trim balance has some effect on the even order distortions but almost none on the odd orders. The residual changes during the hour or so it takes to get the readings."
which apparently explains why the 1kHz spike does not go up to 0dB, which I thought it should. So my calculations were for 15.81V = -92dB as I took the height of the 1kHz spike from the images posted as the fundamental. But you say that the spike is a residual which I don't understand. I can guarantee you that most people reading this thread don't understand that.
Can you point me to some documentation that explains how you get 176dB dynamic range? I realize that your target audience is those 10 people in the world that are well versed in all this, but thanks in advance for not ignoring a beginner.
"The fundamental is at 0db which is 15.81 volts. The spike at 1k is the residual after trimming for the lowest value with a 10 turn 20 ohm equivalent trim pot at the base of the 4 1k resistors in the bridge. Trim balance has some effect on the even order distortions but almost none on the odd orders. The residual changes during the hour or so it takes to get the readings."
which apparently explains why the 1kHz spike does not go up to 0dB, which I thought it should. So my calculations were for 15.81V = -92dB as I took the height of the 1kHz spike from the images posted as the fundamental. But you say that the spike is a residual which I don't understand. I can guarantee you that most people reading this thread don't understand that.
Can you point me to some documentation that explains how you get 176dB dynamic range? I realize that your target audience is those 10 people in the world that are well versed in all this, but thanks in advance for not ignoring a beginner.
Iko, I hope this gets through. First, do you understand how a THD tester works? A THD tester, like my ST1700B (actually this one that I am using is Demian's) NULLS the Fundamental test tone, 1KHz in this case, and then measures the percentage of the residual left. IF the NULLED fundamental is equal or greater than the rest of the residual, you just can't measure lower than the NULLED fundamental, without help. This 'help' is usually a spectrum analyzer, either FFT or other based. The spectrum analyzer sees the NULLED fundamental, and usually sets it as the reference for it. Then the spectrum analyzer goes to work, measuring each harmonic, etc. and gives it a value. Now, what about noise? IF you signal average the continuous signal, you can essentially remove the noise to a very low level. The tradeoff is the time taken to average. This is WHY I use a HP 3563 with my modified 1700B. This appears to be essentially what Simon uses. It is possible, just difficult.
Thanks John. I also hope that I'm not too annoying, but you did say that you're partly here to teach, and I'm willing to learn.
Well, I did think that perhaps he's using a notch filter but couldn't find any mention of it. That's also why I asked for the experiment setup. Looked back in the thread but couldn't find it. It was also hard for me to accept that a -92 notch filter was used.
If anyone has a link to the details of the experiment setup please point to it.
Well, I did think that perhaps he's using a notch filter but couldn't find any mention of it. That's also why I asked for the experiment setup. Looked back in the thread but couldn't find it. It was also hard for me to accept that a -92 notch filter was used.
If anyone has a link to the details of the experiment setup please point to it.
Actually the distortion measurement here is a little different. The fundamental is applied across the bridge network. The bridge is balanced to remove the fundamental present at the bridge terminals. the full potential is present across the network and the individual resistors. the Non linear components will appear across the network amplified by having the nonlinear elements (resistors) on opposite sides of the bridge.
There are still the following challenges- The harmonics of the source are still present. They will have a level and phase relationship with the fundamental that could boost or reduce/cancel them. Ideally the source should be free or at a level that is at least 10 dB better than the dut. Second, the differential measurement here is very challenging, you are looking at parts per billion signals and even a small mismatch in stray capacitance can overwhelm the measurement. I would drive from a isolated balanced source with the center tap tied to the shield/common on the differential to reduce the common mode signal.
The approach my Radiometer CLT-1 uses starts with a very low distortion excitation and a very narrow band filter (10 Hz at 30 KHz) to measure the same distortion. Its much harder to build but more tolerant of environment in use and more flexible (up to 500V excitation). And its really large. We seem to be getting similar results. I'm going to look into anything that can improve my instrument at this stage.
There are still the following challenges- The harmonics of the source are still present. They will have a level and phase relationship with the fundamental that could boost or reduce/cancel them. Ideally the source should be free or at a level that is at least 10 dB better than the dut. Second, the differential measurement here is very challenging, you are looking at parts per billion signals and even a small mismatch in stray capacitance can overwhelm the measurement. I would drive from a isolated balanced source with the center tap tied to the shield/common on the differential to reduce the common mode signal.
The approach my Radiometer CLT-1 uses starts with a very low distortion excitation and a very narrow band filter (10 Hz at 30 KHz) to measure the same distortion. Its much harder to build but more tolerant of environment in use and more flexible (up to 500V excitation). And its really large. We seem to be getting similar results. I'm going to look into anything that can improve my instrument at this stage.
No,
But everyone serious should subscribe anyways, as it is the last audio only technology magazine. Of course if you are a manufacturer you should every so often place an ad, can't hurt.
Also I sent you 4 almost inexpensive resistors that test very well. I could not get the exact values and they are not ROHS. Let me know if you play with them.
Scott,
I may have been a bit rude about the skirt. I assume you already know that the increase in noise floor is excess noise due to the current through the resistor under test. It has the property of amplitude is proportional to 1/f. As this produces more energy at low frequencies these then mix with the test tone. As you know in a non linear system there are residuals at f1 + f2 and f1 - f2 these are what form the skirt. Although I do not show the frequency response quite low enough to see that at the very lowest frequencies the noise level rises again, but it does show low enough that it is clear it is not continuing to decrease.
BUT a carbon comp resistor is also microphonic to a small extent, so there may be some of that showing also. (Self generated thermal noise source.)
Some of the other resistors also had this problem. I just got in a kit for review that uses some of the high harmonic distortion resistors in the feedback loop. For more money they give you the bulk metal ones. The listening panel will of course not know which amp they are listening to or how it is built. But I expect it will be rated well, based on previous even harmonic rich examples.
Oh If anyone wants to repost the new image or the carbon comp image as a PDF on this thread only they have my permission.
Iko
I try to answer most questions, but I was curious to see if you would look at the rest of the thread or if someone else would guide you.
My design goal is 160db of range, a single resistor at -165 would not be a problem, however ten of them could be! Some of the resistors do not even come close to meeting the -160 goal and if you figure 24 bits of audio is 147db about 1/2 of the resistors measured do not achieve that.
As to the test circuit
Don't forget the source distortion is also decreased by the nulling. A bad null is still 70db. Bad nulls occur most with high tempco parts since they do not measure well, it is less of a problem. Good parts can null to -120 so if the test source is only clean to -90 or .003% really good results can be had!
The capacitance is to a certain extent handled by wiring practices and low test values on the resistors.
The wirewounds do not null as low as other types which in my OPINION is due to inductance.
I have done measurements with the best null versus the worst and although things change they do not change as much as device variability.
Same type resistors from a large batch may vary by 6 db on the third harmonic, but the same resistors retested from the start to the end of run (3 weeks) tested the same.
As with any process "feel"s involved, there sometimes is an artifact that shows up at 10K, I know from many measurements that it is the instrument not the DUT. BTY 32K samples 4K averages
Last edited:
I only get tin stars these days.
I published this in 1982, it was an EDN cover article on the AD524. I could make a perfect pink noise generator with a bridge of 3 MF and one carbon, but now I can't find a carbon comp bad enough to repeat the circuit.A couple of other things, you don't listen to resistors i.e. there are no amplifier circuits that would support seperability of -147dB resistor distortion from the -110 to -120 dB of the best amplifiers.
Actually very few resistors in a typical circuit see full (or any) signal swing, usually only 2 or 3.
I don't see any concensus on resistors to support a "mesurement = sound quality" stance, there are plenty of people who won't give up their naked Vishays.
There is little evidence that anything more than the thermal properties of the resistors is being measured. In that case normal thermal management would handle the problem and the square law of power vs I or V would reduce the distortions dramatically. Already it is normal to make the feedback resistor in a PA 2-5W.
Ah, thank you! I did look but couldn't find it. The thread is large and the search function can be hit and miss. I'll go back and look for some more details at the beginning of the resistor testing discussion.
ikoflexer - On your FFT question. This is a common point of confusion, tones at a single frequency are measured directly but noise is measured across the whole FFT bin. For instance a 64K FFT at 96k sampling frequency has a bin width of ~1.43 Hertz so for noise you need to divide by the sqrt or 1.43 to get nV/rt-Hz. Continuous time spectrum analysers have the same but different problem. Here you need to consider the resolution BW and video BW settings. Many analysers have an automatic feature to compute nV/rt-Hz and then the single tone numbers are "wrong".
- Status
- Not open for further replies.