Hi,
low order even harmonics at -110db to -120db would be innocuous, but high order harmonics sound about as bad as multiplying them by their order number to comparee with say 2nd harmonic i.e. 9th order harmonic sounds as bad as 4.5 times greater 2nd harmonic to our ears.
Crossover distortion is VERY peaky with again high order distortion but over a very short part of the overall waveform. A straight % measurement under values the the sound effect by about the ratio of whole wavelength of the fundamental to the % time that the crossover peak shows on the residual, about 10 to 30 times as bad.
Combining these two effects together, short time scale and high order harmonics, means that the low level measurements of crossover distortion should be increased by between 20db and 40 db and this correction factor means apparently inaudible distortion can have a profound effect on sound quality far worse than other low order distortions at -70 to -80 db.
In my view we should strive to identify some types of distortion at very low levels and then decide if they need to be eliminated or reduced. We cannot make that type of decision unless we armed with the evidence.
low order even harmonics at -110db to -120db would be innocuous, but high order harmonics sound about as bad as multiplying them by their order number to comparee with say 2nd harmonic i.e. 9th order harmonic sounds as bad as 4.5 times greater 2nd harmonic to our ears.
Crossover distortion is VERY peaky with again high order distortion but over a very short part of the overall waveform. A straight % measurement under values the the sound effect by about the ratio of whole wavelength of the fundamental to the % time that the crossover peak shows on the residual, about 10 to 30 times as bad.
Combining these two effects together, short time scale and high order harmonics, means that the low level measurements of crossover distortion should be increased by between 20db and 40 db and this correction factor means apparently inaudible distortion can have a profound effect on sound quality far worse than other low order distortions at -70 to -80 db.
In my view we should strive to identify some types of distortion at very low levels and then decide if they need to be eliminated or reduced. We cannot make that type of decision unless we armed with the evidence.
Re: As an experiment .....
Yes, OPA627 is an exceptional op-amp, but it's too expensive for me to justify using it for anything but extremely special projects. However, while it does have very low THD, it's nothing special when it comes to noise.
I'll just turn them off when recording a sample - eliminate them completely.mwmkravchenko said:
I hadn't considered magnetic shielding. I suppose more tests are in order.sam9 said:
I did do a few quick experiments with oscillators, but came to the conclusion that the the effort needed to make one with low enough distortion was higher than cleaning up the output from from a soundcard.mzzj said:
I know datasheets are always... economical with the truth, which is why I actually tested a bunch of op-amps. The best one was the OPA227, which is what I used.Lars Clausen said:
Yes, OPA627 is an exceptional op-amp, but it's too expensive for me to justify using it for anything but extremely special projects. However, while it does have very low THD, it's nothing special when it comes to noise.
A number of reasons:darkfenriz said:
- While -120dB may be inaudible, it's not quite proven definitively that it is never audible, so it seems only sensible to minimize distortion just to be sure.
- So I can show off
- To challenge myself and learn more.
- Who said I would only ever be using this to develop audio equipment?
It makes sense to measure distortions with -120dB level, especially for high-order harmonics.
Several measurements can be seen here. The text is in Czech, but every engineer would understand ...
http://web.telecom.cz/macura/spectra/xa2es.html
Several measurements can be seen here. The text is in Czech, but every engineer would understand ...
http://web.telecom.cz/macura/spectra/xa2es.html
For electromagnetic fields steel is worse than aluminium. What matters here is conductivity. Thickness is not so important, quite thin layer of aluminium and you already have -200dB attenuation.AudioWizard said:
Magnetic fields are completely different beast, they are usually low frequency and need ferromagnetic material for shielding. Faraday gage has little or no help at all against these.
And then there is conducted emissions left to deal, sometimes rather nasty. Battery powered equipment helps a lot.
mzzj said:
One of the other problems with "mu-metal" was that bending it distorted the electro-magnetic characteristics.
FWIW, I have found copper tape (from 3M) to be effective in some of my instrumentation projects. I had a 12 bit ADC which was giving me "bigger than LSB" errors until I placed a ribbon of copper tape across the top of the device and tacked it to the analog ground plane.
Electric fields (E-fields) are shielded at the surface of the conducting material so aluminum foil works as well as copper or tin plated steel...copper and tin plate are easier to solder to.
For magnetic fields (H-fields) ferrous material work the best vs price.
But copper, because of it's high conductivity actually works better than steel only below about 60 Hz
I have no data on aluminum but it may show the same results as copper (but is difficult to solder.)
Brass (copper alloy) may be easier to find and is stiffer for forming shapes.
For magnetic fields (H-fields) ferrous material work the best vs price.
But copper, because of it's high conductivity actually works better than steel only below about 60 Hz
I have no data on aluminum but it may show the same results as copper (but is difficult to solder.)
Brass (copper alloy) may be easier to find and is stiffer for forming shapes.
DUG said:
Actually, I have some soldering alloy that allows to solder aluminium at a pretty low temperature (around 200-230°C), so usable with a soldering iron. This alloy won a price in the "Concours Lépine" a few years ago, and I got it from a consumers show. Not sure how you can find it now via retail...
darkfenriz said:
That's not really the point, actually.
120 dB of resolution is pretty nice to have if you're serious about high-end audio electronics.
As for distortion, don't forget that even if you think one peak here and there at -100 dB are not a big deal, all cumulated, that can add up to unacceptable distortion over the whole spectrum. And anyway, this kind of resolution can allow you to make relevant THD measurements.
That being said, I really doubt you'll get anything useful below -90 dB when using a sound card in a PC. Even the Yamaha one (I myself own a DS2416). So I wouldn't really trust what I see below this threshold... Just a thought.
AudioWizard said:
Thats whats for notch filter was used by
Mr evil.
---
Aluminium solders are nothing new, multicore for example has Alu-sol. Quite easy to solder with this and ordinary soldering iron. Elfa and Farnell used to have this stuff at stock when i ordered last time, rather expensive but real multi-use solder, solders everything that is solderable by any means.
Problem with aluminium soldering is long-term stability, depending on solder and aluminium alloy how stable and corrosion resistant it is going to be. ALUSOL is supposed to be better than many others in this area.
http://www.farnell.com/datasheets/12269.pdf
Analog appnote AN347 has nice graphs for magnetic shielding, showing that at 1khz steel is superior compared to copper but at higher freq not so...
http://www.analog.com/UploadedFiles...76897654237552722724253930119076254AN-347.pdf
http://www.analog.com/UploadedFiles...76897654237552722724253930119076254AN-347.pdf
One reason for this is the skin effect, which you have noticeably at higher freuqencies.
This can be partly compensated by using copper plated steel, even better silver plated steel. Well, which you have to cover with lacquer or your shielding will suffer due to silver --> silver_oxide.
hth,
Andreas
Mr Evil, did you play around with different notch filters or BP filters ?
twin-T is pretty classical but pain in *** to tune. Bainter notch and gyrator based designs should archieve deep notch with much less sensitivity to component tolerances, but i dont know their suitability for this task?
Just did matching for 1khz twin-t caps, 0.2% tolerance in caps should get rid of most of the tuning, too bad that these are mylars, not exactly famous for temp stability.
twin-T is pretty classical but pain in *** to tune. Bainter notch and gyrator based designs should archieve deep notch with much less sensitivity to component tolerances, but i dont know their suitability for this task?
Just did matching for 1khz twin-t caps, 0.2% tolerance in caps should get rid of most of the tuning, too bad that these are mylars, not exactly famous for temp stability.
I did try a few other types, but the twin-t gave the best overall combination of distortion and Q. I tuned it by hand-picking wide-tolerance parts, e.g. all the 4.7nF capacitors are actually 4.8nF. It's not that much trouble to do by watching the results in real time - input white noise and do an FFT on the output and you can see the notch change as components are changed.mzzj said:
twin-t attenuator sensitivity to opamp distortion
twin-t attenuator sensitivity to opamp distortion?
I simulated twin-t attenuator with feedback(to raise its Q) and found rather unhappy results, ie sensitivity to feedback opamp generated distortion.
IE distortion generated by X1B in this pic (C) Mr Evil
I inserted ac source between X1B output and "centerpoint" of twin-t and it shows 3db-9db gain for 2x twin-t tune frequency, ie multiplying 2nd harmonic generated by X1B (and X1A) by a factor of 9db
(actual value depending on feedback factor, being 3db for 0.7 and 9db for 0.9x feedback. 0.7 feedback is about minimum required, notch response is already -3db down at 2khz.)
Maybe it would be better to leave twin-t to be a passive, but then it is going to attenuate 2nd harmonic 9db. Good side is that it is easyer to archieve deeper notch as it is not so damn narrow.
If decided to stick with feedback I think it would be usefull to take feedback from unity gain buffer instead of 40x amplified and >40attenuated signal as in Mr evils schematic
twin-t attenuator sensitivity to opamp distortion?
I simulated twin-t attenuator with feedback(to raise its Q) and found rather unhappy results, ie sensitivity to feedback opamp generated distortion.
An externally hosted image should be here but it was not working when we last tested it.
IE distortion generated by X1B in this pic (C) Mr Evil
I inserted ac source between X1B output and "centerpoint" of twin-t and it shows 3db-9db gain for 2x twin-t tune frequency, ie multiplying 2nd harmonic generated by X1B (and X1A) by a factor of 9db
(actual value depending on feedback factor, being 3db for 0.7 and 9db for 0.9x feedback. 0.7 feedback is about minimum required, notch response is already -3db down at 2khz.)
Maybe it would be better to leave twin-t to be a passive, but then it is going to attenuate 2nd harmonic 9db. Good side is that it is easyer to archieve deeper notch as it is not so damn narrow.
If decided to stick with feedback I think it would be usefull to take feedback from unity gain buffer instead of 40x amplified and >40attenuated signal as in Mr evils schematic
Forget my mumblings, thinking it twice it doesnt seem an issue, my assumptions were just little faulty. As if second harmonic (2khz) of measurement signal would be -100db and X1B would add 2nd harmonic at -100db level to this 2khz, it would appear at 4khz -200db below fundamental. Clearly not an issue. Or case of fundamental distortion happening in X1B, assuming 40db notch and if X1B adds -100db 2nd harmonic back to feedback point. Now my 3-9 db distortion increase comes in handy, as 2nd harmonic generated by X1B is multiplied thanks to feedback. This would result -133db...-137db lower measurement limit for 2nd harmonic.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.