John Curl's Blowtorch preamplifier part II

Status
Not open for further replies.
Dick

As Fourier states all continuous functions may be expressed as the sum of sinewaves. The value shown is what the multiplier for that sinewaves. What is not shown often is the phase. Now it is common practice to display the 20 x log of the multiplier values referenced to a standard. On my AP sys2 I can choose what to call zero. As the display is log the rms to peak difference would just show up as 3 db everywhere.
 
Disagree. It is the matter of "How the world Looks" from outside, without knowing how that "particles" look and what animates them. But if to accept it as "How the world works" the science ends. Finita la comedia.

Not even vaguely. Science actually had a big leap forward with the understanding of statistical mechanics and quantum theory, and the non-existence of hidden variables is well-established.
 
SY and Scott

Most of the audio oriented FFT analyzers show the amplitude either only or as the prompted screen. That is because most users only look at frequency response. Of you use a general purpose analyzer it usually offers phase, window selection etc. But all of the basic field use models don't.

Of course you could do a market survey to see what percentage do or don't.
 
Irrespective of whether noise is random (it certainly is by definition), any finite length time domain noise signal can be treated as periodic and Fourier transformed to give the frequency representation. And back again to generate the original noise signal. That's really basic stuff which has been explained to you again and again.

and

No. Frequency and time are conjugate variables. One representation is exactly equivalent to the other.

Two worlds here, the mathematical one and the one we live in with all its limitations.

Mathematically, frequency and time may be conjugate variables, but this presupposes that the exact mathematical representation on either frequency or time side is known to begin with. I am sure that if you could find a mathematical expression of noise in the time domain, you could go to the F-side and back again without loss of information. Describe random noise as a function of time and the next Fields medal is all yours. In the case of music, what is the time domain function of Lou Reed's "Walk on the Wild side"?

So, mathematically, between the time and frequency domains there may be unlimited re-entry visas, but this is not how we use FFT's in daily life. There are no well defined functions. We do it on mathematically undescribable signals, so take measurements, window them in time frames and split up into frequency bands what is inside those windows.

What can we miss by doing so? Quite a lot actually, but I need a small diversion first to make what comes next more real. FFT's have been used extensively in the search to unravel speech for automated speech recognition, but with limited succes. Apparently, the human ear is able to distinguish markers in sound that cannot be identified by FFT. Otherwise, speech would have been cracked.

Now, to the point. Take a realistic situation where you want to look at the distortion products of a 100 Hz tone. You need a window larger than 10 ms for that. That will capture at least two periods of the second harmonic. Now, asume that the second harmonic has the character of a pulse, with a high first period and a low second. What you will see on your screen is the average of the second harmonic, not the peak. The ear most likely is able to maintain resolution where the FFT is not capable of doing so.

This could be the explanation why two amps that measure identical can still have a different acoustic signature.

Anyways, there is theory and application, and I hope to have pointed out a shortcoming of FFT in its application for audio analysis. Important information gets lost.

vac
 
Last edited:
Disabled Account
Joined 2012
Ok. Now we are getting somewhere close to what i have thrown out.... statisical studies of metadata has pointed the way to correlations and then real solutions in many fields.
Waveform shape cannot be seen on an FFt or THd et al. The phase, when known is not in a form that helps you get a quick take on its resulting sound qualitites.... yes you can make enough tests and reconstruct the original input but still need to know how sensitive the ear is to that waveform compared to other waveforms. etc etc etc.

On the other hand, -100db elements are not usually found in tube audio equipment. Thus, my stab at tubes... distortion and group-delay are sometimes severe enough to be able to hear.

Especially, if you look at the whole system from front to end... that might as well include the recording chain as well. What does the whole thing look like? Not one piece at a time connected to nothing else.

With waveform shapes as the hearing focus, for the moment, we are at a big disadvantage with FFT's that are not telling us anything about the shape... all by itself.

So what new test can better correlate with listening that we havent tried or can we make something new? At least for a few key waveforms that are sensitive ones related to hearing sensitivities. Maybe Dianna Deuche (sp?) can help us out with waveform selection.

Thx, RNMarsh
 
Last edited:
In the case of music, what is the time domain function of Lou Reed's "Walk on the Wild side"?

It's a well-defined sequence of numbers that you can read off the CD. Same thing with a finite acquired noise sequence. One can transform between the domains practically to any arbitrary accuracy, depending on what your hardware and computing budget are.
 
Member
Joined 2005
Paid Member
The bass in " Walk on the Wild Side " is unique. Herbie Flowers, the session bass player, played both an acoustic bass and electric bass track in the original multitracks. Unable to decide which they liked better, in the end it was decided to use both - so each bass note is played twice on an acoustic bass and an electric bass giving the bass an unusual sound on the track. Herbie was pleased, he got twice the usual 30 pound session fee.

Was there a meter, analyser, FFT, or other scientific marvel which would have told you that?

Spent an evening drinking with Lou Reed in the late 70's, he was abrasive and obnoxious, but aren't we all?
 
Last edited:
... Now, to the point. Take a realistic situation where you want to look at the distortion products of a 100 Hz tone. You need a window larger than 10 ms for that. That will capture at least two periods of the second harmonic. Now, asume that the second harmonic has the character of a pulse, with a high first period and a low second. What you will see on your screen is the average of the second harmonic, not the peak. The ear most likely is able to maintain resolution where the FFT is not capable of doing so.

This could be the explanation why two amps that measure identical can still have a different acoustic signature.

Anyways, there is theory and application, and I hope to have pointed out a shortcoming of FFT in its application for audio analysis. Important information gets lost. vac

There is no doubt in this conclusion, and I am just curious, why not to compare precisely the input and scaled output pulse wave forms?
Is there a rule approved by God precluding to make comparisons in time representation?
 
Last edited:
Now, to the point. Take a realistic situation where you want to look at the distortion products of a 100 Hz tone. You need a window larger than 10 ms for that.

A window length of precisely 10mS would do it rather nicely.

That will capture at least two periods of the second harmonic. Now, asume that the second harmonic has the character of a pulse, with a high first period and a low second.

If we assume that then its no longer the second harmonic. The second harmonic is by definition a 200Hz sinewave in this instance.

What you will see on your screen is the average of the second harmonic, not the peak.

Yes - DF96 has already pointed out that if you want to peer inside the window to see what's changing within the window time then probably wavelets are your best bet.
 
Here is the realistic picture.

I feed 400 Hz to the input and see on half power that 2'nd harmonic is -80 dB below fundamental, 3'rd harmonic is -135 dB, and the rest is invisible below noise level. Then I decrease level 10 dB and see that the 2'nd harmonic is below 100 dB, the rest is invisible. Increasing the level up to the maximum power I see the tail goes up, and at almost full power distortions are quite high, but when with decreased power they go down, their order gets lower, it is fine. It looks like high-end amp.
Then I feed mixed 1 KHz and 20 KHz and see how they intermodulate. Decrease first frequency gradually and see that at 40 Hz intermodulation is almost the same, below that it goes higher. I increase time constant of servo and see that intermodulation decreases. It is good. For even better result I would need better, more expensive output transformer, and now I can decide, do I want to go this road, or not.

FFT plots are useful tools, when used realistically. You have to remember that harmonics can not have character of pulses. They are steady sinewave parts of the complex signal. If you see on oscilloscope some pulse this pulse on Fourier plot is a mix of strictly sinewave harmonics.
 
Status
Not open for further replies.