Why not feed the PSU output via a suitable capacitor straight into a sound card Ed? Might save the front end signal conditioning you are doing (I'm taking a strong hint here from Scott's MM approach and other articles I've read on measuring PSU noise using a sound card). There was an article on the TNT webzine a few years ago on an LM317 noise investigation - measurements by sound card also.
That distortion profile looks very similar to the one Johnc124 posted yesterday while investigating the LM833P and its quasi output.
http://www.diyaudio.com/forums/anal...opamp-compensation-technique.html#post4407724
http://www.diyaudio.com/forums/anal...opamp-compensation-technique.html#post4407724
Even the venerable Radford ZD22/HD250 manual includes a simple two transistor battry powered high gain amp to investigate noise and hum.
I sometimes think we make to much of the minutia and overlook the bigger picture in audio. Loss noise regs are great, its keeping the noise at the same level at the destination end of the PSU thats difficult.
Hi Jan, yes indeed, I expected somebody to mention the 'masking' word - but the idea I was submitting was in relation to feedback. Maybe what I am saying that feedback reduces masking. Is this why some of us don't like feedback. Look at what JC said not long ago:
A friend of mine alerted me to that post and since then I have been more present here. He knows this is a topic that engages me.
But back to the topic, let's stay with the 'devil's 7th' as the ideal subject. I have the LM3878T Trans-Amp here and I now need to get my hands on an LM3875T Gainclone amp and do comparative distortions, same power etc, and compare with a focus what the 7th does relative to 6th and 8th. I have built a lot of LM3875T Gainclones (see JLTi Integrated Hybrid Amplifier). They sound quite good, but the Trans-Amp version is so much sweeter, but needs the right speaker of course, and I do have the right speaker in the Elsinore Mk6.
If conventional feedback compresses/flattens out the harmonics like the 7th and the surrounding 6th and 8th, then the 7th is part of the reason we don't like hearing what feedback does, because, as you say, we don't have neighbouring 6th and 8th to balance/mask it out. But using the same circuit in Trans-Amp mode and we now still get low distortion, but the 5th, 7th, 9th etc is masked better and now the same chip sounds sweeter.
But since as pointed out here, that these are basically power opamps, then why not find a way to use regular opamps in the same way, as an alternative way of achieving gain. But you can not reference to ground is a sticking point.
Imagine current driving an RIAA network using Scott's AD797, wouldn't that be interesting? Could work out rather well, maybe even special?
The output is floating, the Sum & Buffer part is the challenge.
Cheers, Joe
.
Last edited:
If the distortions are low, then really, none of that matters. Audible issues arise elsewhere, if one doesn't design competently.
Interestingly, in this thread 'engineers' are critisized for doing THD measurements to show that feedback makes amps better.
Here, THD measurements are used to prove that NFB is flawed.
As one of my former military bosses once said: 'If you don't know where you're going, any road will take you there'.
Jan
Here, THD measurements are used to prove that NFB is flawed.
As one of my former military bosses once said: 'If you don't know where you're going, any road will take you there'.
Jan
Could work out rather well, maybe even special?
The output is floating, the Sum & Buffer part is the challenge.
Cheers, Joe
.
I really wonder what problem this is trying to solve, and the only special thing I can see about this circuit is its sensitivity to impedance/ impedance variations in the subsequent circuitry. This is bound to offset the RIAA correction in plural ways.
Well l doubt whether typically audible, but Baxandall showed years ago that NFB can exaggerate higher order harmonics, that's for sure.
I don't know why op-amps tend to have higher even-series than odd-series distortion spectra. But that's common to see IME. Would also be interested to know.........?
Last edited:
This is fully rectified by Bob Cordell. Baxandall's example was a single non-linear FET stage. Bob showed that with any reasonable linear amplifier, you only need a bit of NFB to push ALL harmonics to below audibility.
Jan
Around these parts we have some folks producing non-GMO, organically grown, cold pressed canola oil. I confess the $15/l price has kept me from trying it. It is greenish like a good olive oil.
I suspect that since most opamps are not symmetrical (ie that are of the 'blameless' variety), second order effects are not suppressed as they are in fully balanced designs.
Let's think about the low order harmonics for a minute:-
2nds will be happening once on either half cycle but not on both half cycles
3rds will be happening every half cycle
If your circuit is fully balanced, the thirds are likely to be more prominent
If it's not balanced, it could be that on one half the distortion is different.
Very interesting topic by the way.
Let's think about the low order harmonics for a minute:-
2nds will be happening once on either half cycle but not on both half cycles
3rds will be happening every half cycle
If your circuit is fully balanced, the thirds are likely to be more prominent
If it's not balanced, it could be that on one half the distortion is different.
Very interesting topic by the way.
Bon,
Since shunt regulators became more popular a few years back one can now design units with equivalent impedances of a few milliohms. As 600 milliohms has noise about 100 picovolts per square root of Hz. there is a need to go very low noise in the measurement system.
Now 4 decent opamps in parallel will get down to about 500 pV/rtHz. Using a multiplier to correlate two sets should provide another 30 to 60 db of sensitivity.
As my design goal is to get 160 dB dynamic range in a complete system power supply noise is an issue.
My earlier bit about resistors shows the way to keep their contribution below that. A bit on capacitors shows their issue is microphonics, so that becomes a mechanical design issue, but again their contribution can be below the design goal.
After I finish my power supply bit, I may do an oscillator design. I do not know if it will be DSP or analog based. I do have the design concept. It will have a quadrature master oscillator and quadrature slaves at the 2nd, 3rd, etc harmonics. These will be mixed in to null distortion components detected at the output. To adjust the harmonics will require a method to monitor them. The method may be to use phase locked quadrature vector amplifiers after a notch filter. I suspect using this technique will allow distortion products low enough to measure the distortion of any audio band A/D converter.
Now one of the real issues will be the output noise level. If the output is band limited it may be possible to keep the noise level low enough. Fortunately this is not as big an issue as for testing A/D converters multiple samples may be averaged to reduce the noise contribution to the measurement.
Last edited:
I don't think getting to those really low noise levels with paralleled opamps is too practical. I posed some time ago a really low noise power supply desing. I also posted the circuit for the preamp I used to validate it. I used a transformer input As I remeber the equivaleny input noise resistance was about 5 Ohms. I'm sure there are an assortment of MC input transformers available that could achieve this. If they are yesterday's audio news they would be cheap as well.
A suitable input coupling cap is also an issue. Jim Williams used a big wet slug tantalum ($500 new but cheaper in the surplus market) and it will be an issue if its internal resistive impedance is in the same magnitude as the noise impedance you are measuring. Using cross correlation to remove the noise of the preamps is a good idea. It will take a lot of averages and time to get a significant noise reduction. There is a lot on using cross correlation to measure ultra low phase noise oscillators.
I'm not sure why you want to go to such complexity to make a low distortion oscillator. For around $60 you can get one of Victor's oscillators from eBay and its at least 10 dB lower distortion that any ADC made. There are some other efforts documented on DIYaudio that are as low or lower, however they are a lot harder to make.
A suitable input coupling cap is also an issue. Jim Williams used a big wet slug tantalum ($500 new but cheaper in the surplus market) and it will be an issue if its internal resistive impedance is in the same magnitude as the noise impedance you are measuring. Using cross correlation to remove the noise of the preamps is a good idea. It will take a lot of averages and time to get a significant noise reduction. There is a lot on using cross correlation to measure ultra low phase noise oscillators.
I'm not sure why you want to go to such complexity to make a low distortion oscillator. For around $60 you can get one of Victor's oscillators from eBay and its at least 10 dB lower distortion that any ADC made. There are some other efforts documented on DIYaudio that are as low or lower, however they are a lot harder to make.
That is about the ratio of the energy contained in a stretched rubber band to that of 2 kg of TNT. That'll be quite a dance hall.
Last edited:
The Baxandall's study concentrated on a very special case, one should be careful in generalizing the results, the oft cited Boyk/Sussman article is a little more general.
Most IC processes are not fully symmetric, the beta/VA product is usually different (sometimes quite different) for NPN vs PNP. The dual DSL drivers do pretty good used fully differential at audio frequencies for seconds.
Last edited:
- Status
- Not open for further replies.