I repeat: same spectra. I saw your spectra and commented multiple times that they suffer from window leakage. the signals are strictly periodic so their spectra only have discrete peaks without ‘skirts’. Please look at the Matlab code and show us how your FFT is done - the difference is in the processing. I assume that a computer OS is leaving a WAV file in intact ti the LSB level. Noted that the signals have non zero mean values that are different and the difference in the order of 2dB but D.C. is not reproduced nor perceived.
When it comes to resistor differences, we found problems in our initial research and development in the 1970's, where even some very expensive, military grade parts sometimes had problems, especially volume pots. The initial Mark Levinson JC-2 potentiometer was very expensive for us to buy, AND in certain locations on the volume pot, gave audible distortion. How do I know, because my associates heard it when we were in a listening test, and I verified it by putting an IM analyzer on it with the same setting back in 1974. It was awful! I had to show Mark, because it would ultimately effect our reputation if found by others, but I had to PROVE it to him by taking a pot off the assembly line and putting it on a THD analyzer and finding the 'bad spot' just by adjusting it. Bingo! Now that was a revelation at the time (48 years ago). Now, was it the manufacturer's fault? Partly, was it Mark's fault? Partly. But it was the principle fault of Richard Burwen, who recommended that an ideal dual volume control could be made with a relatively high value (100K) LINEAR TAPER dual pot, loaded with 50K on each wiper to ground, made an excellent tracking, semi-log volume control, without regard for potential current through the wiper, even a little. This cost Mark tens of thousands of dollars ultimately. This is reality folks. Now back to fixed resistors. In the early days we just bought just about any type of carbon resistor, with only accuracy as a criterion. 5% was a very good resistor, back in the 1960's, 1% was too expensive and saved for expensive products and attenuators. We did OK in the old days with that. Then Mark Levinson, probably because of Richard Burwen's suggestion went 100% to 1% metal film resistors in the early 1970's as they dropped in price barely affordable to a hi end audio product. This was a good thing, but we used just about every available brand and the audio circuit board would be covered by many colors of 1% film resistors because each manufacturer made different colors from each other most of the time. We were happy enough for about 10 years, when there was a listening test on fixed resistors published in 'Hi Fi News' and then we all woke up to the differences and we got more discriminating. That's how we learn to make better audio products, we cut/try and listen.
OK, lets assume for the moment that the spectra are the same, even though I haven't been able to confirm it. So, is your amazement still centered on "same spectra but different sound"? How do you explain, then, the completely audible 43Hz envelope modulation that is plainly audible? What exactly is that? Is it part of the real spectrum, waveform, or not?
IMO, it's a change in waveform over time, something an FFT may or may not see. But what exactly does that imply? That FFTs are bad? Or that, (OMG not again) we can't measure everything we hear?
Of course, significant measured distortion in FIXED RESISTORS is more difficult to measure. That is why I leave that up to Ed Simon. Still, there appear to be audible differences between resistor brands and types no matter that the measured distortion is really low. That is just something that I live with in order to make Class A products.
the two waveforms are drastically different as it can be seen by the naked eye. Nothing wrong with the FFT: both spectra have the same magnitude but differ in the phases of the side bands. The FFT resolves that perfectly (returns complex numbers ).
Actually the AM signal was generated by taking an inverse FFT of the FFT of the FM signal - just rotating the phases in the iFFT domain before transforming back. this is why the magnitude spectra are identical guaranteed by math
My unqualified impression is that this discussion heads toward credibility of “how many angels can dance on the tip of needle”, because those two signals are physically absolutely different and question why that is not obvious in the FFT is a different matter.🙂
How on earth could be possible not to distinguish by ear, between warble of 1 kHz signal amplitude modulated with 43 Hz and constant amplitude 1 kHz signal shifting its frequency up and down by 43 Hz.
It’s the same as asking why cube and ball of the same volume have different appearance.
To paraphrase late Steve Jobs “You are measuring it wrong”.
How on earth could be possible not to distinguish by ear, between warble of 1 kHz signal amplitude modulated with 43 Hz and constant amplitude 1 kHz signal shifting its frequency up and down by 43 Hz.
It’s the same as asking why cube and ball of the same volume have different appearance.
To paraphrase late Steve Jobs “You are measuring it wrong”.
that changing only the phases of modulation side bands seem to have such a large effect on the perception. Not explained by masking theory at least
Indeed.
So now lets consider the case of John Curl resistor sound. What do people try to measure? HD is what. However, it is known that resistor excess noise can vary a lot depending on construction, especially end cap connections. Has anyone tried to correlate audiophile resistor excess noise with audibility?
ESS makes very clear what they think in their slide deck. They say the ear is "exquisitely" sensitive to signal-correlated noise. In JC case presumably that could be resistor current noise.
Oh gawd not the ESS throw away comments on noise modulation again. Mark, you must stop believing everything you read in marketing slides!
Its like this: John and a lot of other people are suffering from mass hallucinations, or we are measuring it wrong. Its never the engineer that's wrong?
No, you are only proving that two signals with the same FFT magnitude can be vastly different. Which is obvious for everybody with a half brain. To have a strictly real result from a FFT, so the magnitude alone is relevant, the signal must have an even symmetry (i.e. in your matlab language x[n]=conj(x[N-n])), which obviously a FM signal doesn’t. You conveniently left out the phase in your matlab calculations and declared the two signals “FFT identical”. I tried to call you about earlier, but you choose to ignore my comment.
You are confusing the less knowledgeable, and feeding those with an agenda, by conflating phase shift and phase modulation, by calling them generic ”phase”. These are vastly different things and while an absolute phase shift is questionable audible, phase modulation can lead to completely different signals.
I don’t see what acoustic masking theory has anything to do with all these.
Last edited:
your comment is sonsensical. you get the magnitude (aka modulus) by using the abs() function in Matlab as shown in the script. Same magnitude spectrum means that abs(Ha)=abs(Hf) as the Matlab script tests for. The spectra do not need to be all real. sorry if I missed your earlier comment - I did not realise that you meant real as in its mathematical meaning.
For the millionths time: the two signals are vastly different ( as observed when looking through a scope) but identical seen through a spectrum analyzer, ie same magnitude spectra.
Masking theory: the question of if the side bands are masked by the carrier (so that no
modulation is heard) or not.
now that the definitions are in place we can leave it to others to draw their conclusions irrespective of their brain sizes