In datasheets for some op-amps, Gdelay is indicated as time Propagation Delay: OPA633, OPA603, LT1363-LT1365, LT1227, LT1357, LT1469 and many others.
I have indicated the optimal value for power amplifiers many times; it should be no more than 100 ns with as long a linear section as possible (not less than 1 MHz).
For those who do not understand, I repeat: a musical signal is an impulse signal and each of its "hooks", every moment is the first period. Therefore, to say that Gdelay is just a signal delay and nothing more, it means nothing to understand how an amplifier works ...
Yes music consists of both "impulses" and steady tones. But thats not the point, the point is that both impulses as well as steady tones are.. wait for it.... continuous.
This means that continuous proportionality is maintained for everything on a CD/LP/Tape except for the start of each track on said CD/LP/tape. And out of these three it is probably only CD that have that problem due to that the two latter ones has so much noise that the "continuous" criteria is already met before the musicians start to play. And for CD there is always a little time before music start playing that cares for the "startup" "problem"... so not to hurt the fire strike of a string or gasp for air.
If you still are confused about why the "impulses" in the music is not a valid case to support you theory, you should realise that the BW is so low in music that even a cymbal crash is perfectly continuous so nit a case for the "startup" "problem".
You have noting - sorry.
//
This means that continuous proportionality is maintained for everything on a CD/LP/Tape except for the start of each track on said CD/LP/tape. And out of these three it is probably only CD that have that problem due to that the two latter ones has so much noise that the "continuous" criteria is already met before the musicians start to play. And for CD there is always a little time before music start playing that cares for the "startup" "problem"... so not to hurt the fire strike of a string or gasp for air.
If you still are confused about why the "impulses" in the music is not a valid case to support you theory, you should realise that the BW is so low in music that even a cymbal crash is perfectly continuous so nit a case for the "startup" "problem".
You have noting - sorry.
//
Last edited:
Because you do not understand what people say to you. You keep repeating the exact same arguments others find to be wrong or irrelevant.
Prop delay translates directly to phase as frequency increases, additive to whatever capacitive phase lag exists. The only problem with excessive delay is it affects the stability of the amplifier. 100ns phase delay is 1 full period at 10MHz, so the loop gain had better be below 0 dB at 5 MHz, due to delay alone. Add the 90 degrees of phase due to capacitive lag, and you have to be below zero dB by 2.5 MHz. That's if you have only a single pole.
That, to me, is the only argument for low delay through the amp: higher GBW.
Your argument for constant delay through the amp is stronger, but I think most amplifiers are fairly good, in this regard. Only phase lag or lead of EQ causes delay to wander, but the ear expects it in that case.
Think of a square wave from a digital FIR output filter, like on a CD player output. It is band-limited to 20 kHz and perfectly time coherent (constant group delay). Now, bandlimit a square wave to 20 kHz with an RC filter. It no longer has constant group delay, but it is still correct to the ear, because the phase corresponds to the band-limiting. Studies have shown that people can't hear phase, which makes sense if you consider how the ear works.
That, to me, is the only argument for low delay through the amp: higher GBW.
Your argument for constant delay through the amp is stronger, but I think most amplifiers are fairly good, in this regard. Only phase lag or lead of EQ causes delay to wander, but the ear expects it in that case.
Think of a square wave from a digital FIR output filter, like on a CD player output. It is band-limited to 20 kHz and perfectly time coherent (constant group delay). Now, bandlimit a square wave to 20 kHz with an RC filter. It no longer has constant group delay, but it is still correct to the ear, because the phase corresponds to the band-limiting. Studies have shown that people can't hear phase, which makes sense if you consider how the ear works.
This might be appreciated here, burst recorded with a microphone from 13mm tweeter.
This is clearly a memory type distortion. That will be hidden by THD measurements. You can't say it's an IMD - where is the other sine? You can't say it's a TID - where is the square wave?
That could be a memory type ...
- electrical distortion of the D/A converter IC
- electrical distortion of the post-D/A circuit
- electrical distortion of the power amplifier
- electro->mechanical distortion of the tweeter
- mechanical distortion of the tweeter dome (Polymer Dynamics and Relaxation ???)
- mechanical distortion of the microphone diaphragm
- mechanical->electro distortion of the microphone
- electrical distortion of the microphone preamp
- electrical distortion of the pre-A/D circuit
- or even the A/D converter IC has inherently some of these distortions as well ????
TID and IMD are types of distortion. They exist, whether you measure them or not.
You may use sines or square waves or anything to measure, but the distortion does not change, because it is caused by the nonlinearity of the amplifier or whatever device you are discussing.
By measuring in different ways you get a different manifestation of that underlying non-linearity.
Jan
You may use sines or square waves or anything to measure, but the distortion does not change, because it is caused by the nonlinearity of the amplifier or whatever device you are discussing.
By measuring in different ways you get a different manifestation of that underlying non-linearity.
Jan
Agreed.
IIUC your point there would seem to be an implied requirement of time-invariance for it to be true. Since no physical system is perfectly time-invariant, that might be something to keep in mind when measuring.
Agree. I did not question the existence of IMD or TID [or rather TIM (?)]. I have measured IMD myself many times.
Disagree. Not just the distortion, the non-linearity causing it is changing by changing of TBE, VCE etc... These easily changing on the order of milliseconds for a transistor like the BC550 as shown on the left photo here: https://www.diyaudio.com/forums/solid-state/5928-comments-thermal-distortion-2.html#post6339797
(The one on the right has a much lower thermal cut-off frequency) You can see loops there on the left because thermal distortion is not time-invariant. Please note: VCE, IB, are the same during up part and the down part of the individual thermal 'loops'.
Disagree. That underlying non-linearity is not a fixed non-linearity. It's changing as the audio signal changing because it's linked to bias conditions. The bias conditions are not time invariant because of the nature of memory type distortions like thermal distortion. So the non-linearity is not time-invariant. It is easy to see by understanding the exponential characteristic of thermal changes. It also has an effect after excitation, not just during excitation.
My previous comment was for those who question the existence of memory-type distortions. Pavel's measurement shows the existence of memory type distortions in audio.
A rather complex model would explain everything, but I've not seen anything modelling exactly the distortions in audio so far.
Regards,
Krisztian
I don't see any photo on the page that link takes me to. What post # are you trying to link to? Direct links only work correctly if everyone has the same number for the posts-per-page forum display setting, which not everybody does.
Ye, I agree, the test signal itself changes bias conditions and thus non-linearities. But that should only be valid for the thermal consequences of the signal, no?
The signal itself always causes excursions positive and negative around the bias point.
Jan
Last edited:
Thanks for info! It's #68 post in that thread: comments on thermal distortion?
Is there any way for a shared link to be good for all, regardless of the number of pages displayed at the viewing?
Please excuse the poor quality sketching below. It was produced on a "reMarkable 2" pad that my daughter gave me as a gift. Nice girl... more than I deserve.
There are non-linearities generated from “square law” and “S-type” transfer curves that generate harmonics along with non-harmonic artifacts that track with the exponential rate of those harmonics. These artifacts seem generally unknown, ignored or considered inconsequential in THD measurement. The discussion below is specific to an “S-type” transfer curve rather than the “square law” curve as likely more relevant to this thread.
In 1984 I presented a paper at the Audio Engineering Society Convention in New York entitled “On the Replacement of T.H.D. as the Standard Figure of Merit”. This was accepted for presentation at the convention by John Vanderkooy, hence its substance can be considered having acceptable technical accuracy, though not necessarily worthy of replacing T.H.D. as the title suggests.
The paper was motivated by the discovery of results by testing non-linearities observable on an oscilloscope. This involved creating a device having an S-type transfer function with a means to subtract the fundamental frequency from non-linearities on the output. The objective was to isolate the non-linearity from the fundamental and compare the shape and phase of the non-linearity as a whole to harmonics produced in THD measurements. The device performed as expected, in that the fundamental could be completely removed by variance of an adjustable control to remove the fundamental from the output, thus yielding a near pure 3rd harmonic.
What was unexpected was that the gain control required constant re-adjustment of the level of the fundamental needed to take out its content from the non-linearity under circumstances the input level was changed. This was under circumstances that the device generating the distortion was of fixed gain and that the cancellation signal was also of fixed proportion to the input. The conclusion was that the amplitude of the fundamental passing through the “S-type” curve was non-linear. In subsequent testing it was discovered that the shape of the non-linearity had to take on the shape of that in Figure 1 to be repeatable without gain adjustment.
It should be noted that 3rd harmonic distortions changes exponentially with linear changes of the input. This means that for the non-linearity as a whole to remain of constant shape requires that all non-linearities, as to include the 3rd harmonic, must also change at that exponential rate. It turns out that there are two fundamentals that exist in the output when input levels are fixed, one changing at a linear rate and one that would otherwise change exponentially. The nature of the combination of the 3rd harmonic and the non-linear fundamental is shown in Figure 2. The totality of the non-linear shape in Figure 1, and as identified as (c) in figure 2, is created from the incremental summation of (a) and (b).
This means that T.H.D. measurements having fixed input stimulus do not accurately represent the totality of distortion, as it combines linear and non-linear fundamentals. Perhaps most importantly is that the totality of the wave shape in Figure 1 reveals that the RMS power of the non-linearity is strongly aligned with the peak of fundamental, therefore it is hard to imagine that the 3rd harmonic is responsible on its own as the singular cause of psycho-acoustic experiences.
Last edited:
It's excellent! I've been thinking about buying one for myself for a year. Could be that pressure sensitivity is switched off.
I have to read it a few more times to understand. So does that mean that in the case of an S-curve, after the error subtraction does the fundamental harmonic change as a function of the signal level non-proportionally?
I didn't really understand the second sentence. I think even in one sine cycle especially for lower frequencies the positive and negative excursions pull back and forth the bias point a little because of thermal variations like on my OnSemi BC550 curve tracer photo: https://www.diyaudio.com/forums/att...-comments-thermal-distortion-bc560c-semi_-jpg
On the photo, VCE varies with the absolute value of a 50Hz sine. So each loop starting from the bottom left corner is a half sine along the horizontal axis. Each loop up and down is 5ms. The BC550 clearly easily tracks the self-dissipation induced thermal variations. VCE varies and so does the current gain. (The vertical axis is IC).
Krisztian
As I wrote: VCE varies with the absolute value of a 50Hz sine. So yes a half sine cycle of 50Hz is 10ms hence 5+5ms for a loop up and down. You have 100 half sine cycle in a second when its frequency is 50Hz.
If you look at a THD spectrum containing a 3rd harmonic known created from an S-curve, the peak value of the non-linear fundamental would correspond on the fundamental spectral line about 9.5 dB higher than the peak of the 3rd, given that the non-linear fundamental tracks the 3rd proportionately at about a 3 to 1 ratio.
To expand somewhat. If the 3rd harmonic changes at a cubic rate, a 2x increase of input amplitude would correspond to an 8x change in the 3rd harmonic spectrum line. Given that the non-linear fundamental is tracking this exponential rate it still manifests as being 9.5dB higher than the 3rd.