Hartono said:if we have an ADC with 2Hz sampling period
we can only put 1Hz signal in first sampling period or next sampling period (next 0.5 second).
but if we have 1Mhz sampling, the 1Hz signal can start from many different point within 1 second.
Cheers
Yes, but I don't see why that makes a difference. As I said, even with 44.1kHz sampling, a 20kHz wave can be exactly captured and reproduced, and that includes the phase of the wave, so, including at which point the wave started.
I told you it is counter-intuitive...
Jan Didden
yes it can be reproduced
but with much higher sampling..............................................................................
...........................................................................................~~~~~~
~~~~~~~~~~~~~~~~~~~~~~~~
ok in my last example the 1 hz sinewave with 1 mhz sampling will have 1 million possible group delay.
where as in 2 hz sampling only 2 possible group delay. (0.5 second and 1 second)
but with much higher sampling..............................................................................
...........................................................................................~~~~~~
~~~~~~~~~~~~~~~~~~~~~~~~
ok in my last example the 1 hz sinewave with 1 mhz sampling will have 1 million possible group delay.
where as in 2 hz sampling only 2 possible group delay. (0.5 second and 1 second)
janneman said:Yes, but I don't see why that makes a difference. As I said, even with 44.1kHz sampling, a 20kHz wave can be exactly captured and reproduced, and that includes the phase of the wave, so, including at which point the wave started.
I told you it is counter-intuitive...
Jan Didden
Hi Jan,
Even in case of a (digital) recursive reconstruction filter?
Cheers,
Triangle wave, or sawtooth is nice in my opinion to view signal from an amp, sine says absolutely nothing on an oscilloscope..
hmm Now you guys are going to hang me for this thought, one can maybe use a kind of core that is active at higher frequencies and use fewer wounds.. that way it's more actve at rf oscillations and almost inactive at lf, just how we want it huh..
Btw, This thread is amazing.. I read to page 25 or so then I couldnt stand it any more hehe..
Soon we will be replacing the guitars on the artists that made the recording due to too much thd or something..
Anyway I'm sure it's possible to not use output coil in an amp, but remember,.It not if it burns or not that is the question, it's getting rid of parasitic (bubble) oscillations that is the problem..
Baah making the amp not burn with capacitive load, even I can handle THAT..
ps/ Dire Straits still ROCKS on my 800ohm vintage tube lo-fi stero system..
hmm Now you guys are going to hang me for this thought, one can maybe use a kind of core that is active at higher frequencies and use fewer wounds.. that way it's more actve at rf oscillations and almost inactive at lf, just how we want it huh..
Btw, This thread is amazing.. I read to page 25 or so then I couldnt stand it any more hehe..
Soon we will be replacing the guitars on the artists that made the recording due to too much thd or something..
Anyway I'm sure it's possible to not use output coil in an amp, but remember,.It not if it burns or not that is the question, it's getting rid of parasitic (bubble) oscillations that is the problem..
Baah making the amp not burn with capacitive load, even I can handle THAT..
ps/ Dire Straits still ROCKS on my 800ohm vintage tube lo-fi stero system..
Hi Jan,
The thread has whizzed on rightly since this morning.
I won't do what you say I must !!!
You won't try what I suggest you should !!!
Hence no meeting of minds.
I OBJECT to you suggesting that it is necessary to limit amplifier input testing to 50kHz bandwidth.
What on earth for.
That would give you a comfortable pipe and slippers performance, not the definition currently available from digital audio.
Yes I do have an RF filter on my own amplifier.
I just sim checked it to -3dB at 2MHz.
And NO do not want to change it !!!!!!!!
Why, because the amplifier's response is so much better because of it.
Remember Jan, this input filter (on most amplifiers too) is also part of the NFB sensing network.
And NO (again) I do not suffer other "hf noise, junk etc" you think I might. Or at least if I do the amplifier remains unaffected because it to is very fast. During testing I use it to amplify AM signals and it was nicely stable with antenna input.
I just developed ( with help ) a buffer / pre-amplifier. It is reported as being ultra transparent.
Its bandwidth ?
>2MHz, giving a phase response within +/-1deg from 10Hz to 100kHz at +12dB.
Jan, how many degrees of hf phase shift would be introduced by your suggested 50kHz bandwidth, and how on earth could you test amplifiers with such a poor test source ?????
There are still 3 pages of posts for me to cover.
Cheers .......... Graham.
The thread has whizzed on rightly since this morning.
I won't do what you say I must !!!
You won't try what I suggest you should !!!
Hence no meeting of minds.
I OBJECT to you suggesting that it is necessary to limit amplifier input testing to 50kHz bandwidth.
What on earth for.
That would give you a comfortable pipe and slippers performance, not the definition currently available from digital audio.
Yes I do have an RF filter on my own amplifier.
I just sim checked it to -3dB at 2MHz.
And NO do not want to change it !!!!!!!!
Why, because the amplifier's response is so much better because of it.
Remember Jan, this input filter (on most amplifiers too) is also part of the NFB sensing network.
And NO (again) I do not suffer other "hf noise, junk etc" you think I might. Or at least if I do the amplifier remains unaffected because it to is very fast. During testing I use it to amplify AM signals and it was nicely stable with antenna input.
I just developed ( with help ) a buffer / pre-amplifier. It is reported as being ultra transparent.
Its bandwidth ?
>2MHz, giving a phase response within +/-1deg from 10Hz to 100kHz at +12dB.
Jan, how many degrees of hf phase shift would be introduced by your suggested 50kHz bandwidth, and how on earth could you test amplifiers with such a poor test source ?????
There are still 3 pages of posts for me to cover.
Cheers .......... Graham.
Graham Maynard said:
Graham,
Pray tell how 'the definition curently available from digital audio' will somehow cause music information to appear above, say, 20kHz as we had with the LP and CD?
Graham Maynard said:
No, its not.
Graham Maynard said:
Uhh... a couple of degrees, I guess. About a couple of orders of magnitude better than the best speakers that neither I nor you can afford. Why do you ask?
Jan Didden
Hi Johan,
Forgive me not answering your post#384 at the moment.
I simply cannot cope with all the time wasting noise arising in this thread.
If this were an amplifier I would bin it.
Tests need to be done. .
Properly. I suggest with 5uH
And NOT with a simple bandwidth limiting input filter causing prior modification of an AF signal !!!
Cheers ......... Graham.
Forgive me not answering your post#384 at the moment.
I simply cannot cope with all the time wasting noise arising in this thread.
If this were an amplifier I would bin it.
Tests need to be done. .
Properly. I suggest with 5uH
And NOT with a simple bandwidth limiting input filter causing prior modification of an AF signal !!!
Cheers ......... Graham.
From Jan - quote:
Originally posted by Graham Maynard
[snip]Remember Jan, this input filter (on most amplifiers too) is also part of the NFB sensing network.[snip]Cheers .......... Graham.
No, its not.
____________________________________________________
Jan, I shall not be responding to you......... Graham.
Originally posted by Graham Maynard
[snip]Remember Jan, this input filter (on most amplifiers too) is also part of the NFB sensing network.[snip]Cheers .......... Graham.
No, its not.
____________________________________________________
Jan, I shall not be responding to you......... Graham.
janneman said:
Yes, but I don't see why that makes a difference. As I said, even with 44.1kHz sampling, a 20kHz wave can be exactly captured and reproduced, and that includes the phase of the wave, so, including at which point the wave started.
I told you it is counter-intuitive...
Jan Didden
Yes, counterintuitive. A 20 KHz tone sampled at exactly the Nyquist rate (40 KHz) gets 2 samples per period and they may fall anywhere. Where is the trick for perferct recovery?
Apart from the fact the maths says so (belive if you want or don't, it is up to you) a hint to grasp intuitively what happens is to look at the function of the reconstruction filter, which plays a central role.
The spectral signature of a sampled signal includes the so called base band - the original continous signal - plus harmonics of the sampling frequency adorned by sidebands reproducing the baseband replacing DC with the corresponding harmonic either stright or mirrored right and left respectively.
From this it is clear an ideal rectangular filter which spans just up to the top of the base band (20 KHz) and rejects completely what lies futher up, will have at its output a perfect copy of the base band, that is, the original continous signal.
So far goes the theory though as logical as may look, still lets us far from intuitively grasping how is it that a train of pulses sampling a 20 KHz wave may reproduce the tone as if nothing had happened.
As said, the crux is in the reconstruction filter. An ideal rectangular filter has an inpulse response in the form of a sinc function. This is a symetrical waveform centered at the nominal pulse position (remember a sample is ideally a very narrow pulse modulated in amplitude by the original signal), that ripples by ever decreasing positive and negative excursions behind and ahead of the pulse position in time.
Now, think about the composition of multiple waverforms of this kind - which are continous in time - in a train corresponding to the sampled signal. A characteristic of this functions, the reconstruction filter, and the sampling rate, is that the zero crossings of the upstream and downstream ripples for each "wavelet", correspond to the nominal pulse positions of the previous and following samples, meaning the reconstructed amplitude is not perturbed by previous and following pulses at those points in time. Yet, within the interval between pulses, you get as recovered signal the composite contribution - with ever decreasing weight - of the previous and following samples. (*)
Hope this at least hints to a visualization. The world is more complex though, an the actual implementation requires both feasible reconstruction filters and compromises. Final accuracy depends as allways on how well these compromises are handled.
It is the task of industry technical working groups to establish these compromises in terms of accuracy, data volumes and manufacturability.
Rodolfo
(*) It is this composite contribution from preceding and following samples what fills up the interval between samples in such a way as to recover the original waveform unscathed.
I've been running some sims to explore this using an "ideal" amplifier (VCVS) driving the TS equivalent circuit, for a mid range unit I've measured, via a series 5uH inductor. The input is a 1kHz rectified sine wave (single rectification, not bridge) passed through a 20kHz 1st order low pass filter. This waveform (start of) is not uncommon in music sequences - a drum hit at the end of a quiet passage generates a similar starting signal.
A picture of the input signal is included below. I'll post other pics in subsequent replies.
I then differenced the signal at the output of the coil with a suitably delayed version of the input signal to obtain an error signal. Input signal peak value 1V, error signal peak about 3.3mV. I then summed the error signal with the original (before rectification) sine wave in order to get an FFT plot of the result (this is just to get the results in some recognisable format - not intended to imply any sustained error generation). The spectrum of this was very broad with dominantly even harmonics: 2nd harmonic at -62dB, 4th at -67dB, ...., 20th at -80dB. Similar results were obtained comparing the coil output signal to an RC delayed version of the input. I also tried the same test with a 50uH coil and got about 20dB higher harmonics.
The conclusions I've drawn from this are.
1) Stored energy in line to the speaker is bad and generates errors in response to changing amplitude signals - starting from zero is probably worst case but audio has constantly changing levels of various amplitudes.
2) The errors generated appear proportional to the amount of inductance in line: so a passive crossover containing series inductors is also bad and will likely swamp the effects due to a (much) smaller output coil.
Of course, none of this addresses the issue of audibility particularly since the errors generated by a single event are of fairly short duration. Don (AMV8) did make some enlightening observations on this in post #333 however.
Cheers,
Tony
A picture of the input signal is included below. I'll post other pics in subsequent replies.
I then differenced the signal at the output of the coil with a suitably delayed version of the input signal to obtain an error signal. Input signal peak value 1V, error signal peak about 3.3mV. I then summed the error signal with the original (before rectification) sine wave in order to get an FFT plot of the result (this is just to get the results in some recognisable format - not intended to imply any sustained error generation). The spectrum of this was very broad with dominantly even harmonics: 2nd harmonic at -62dB, 4th at -67dB, ...., 20th at -80dB. Similar results were obtained comparing the coil output signal to an RC delayed version of the input. I also tried the same test with a 50uH coil and got about 20dB higher harmonics.
The conclusions I've drawn from this are.
1) Stored energy in line to the speaker is bad and generates errors in response to changing amplitude signals - starting from zero is probably worst case but audio has constantly changing levels of various amplitudes.
2) The errors generated appear proportional to the amount of inductance in line: so a passive crossover containing series inductors is also bad and will likely swamp the effects due to a (much) smaller output coil.
Of course, none of this addresses the issue of audibility particularly since the errors generated by a single event are of fairly short duration. Don (AMV8) did make some enlightening observations on this in post #333 however.
Cheers,
Tony
Attachments
Graham Maynard said:
No problem, Graham.
I am also reeling under the 30 odd additions since then, then with 5 other forums, and also trying to keep a moderately normal day within the constraints of 24 hours....
There are lots I would also like to post, but impossible. I often cannot avoid the basic feeling that much argument here looses track (more or less) of the fact that we only need to deal with response up to give/take 20 KHz. Any effects generated must still be reflected there, one way or another.
But there; one learns new things and read interesting trains of thought, even if some can have very little influence on the subject matter.
So fine with me.
Regards.
In another thread I talk about my stability woes (amplifier, not mental, for those who jump to such conclusions), and these have been directly connected to the output inductor, and to effects that can happen on a single or few cycles. I still don't think the value of 2uH can have any *direct* effect on audio, but I can also state with confidence that it can interact with the amp to have a significant effect that's probably audible, and probably bad 
Let me describe two inductors:
The first consists of 15.5 turns of AWG19 enameled wire on a standard old fashioned 10 ohm 2W carbon composition resistor. This is pretty typical of the output coils used by many people. The combination of coil and resistor has a Q of 1.73 and has a phase shift of 60 degrees, both at 1MHz.
The second consists of 14 turns of a similar size wire on a 1W film resistor. 10 ohms again. Don't know the exact type, but it has magnetic metal end caps and is likely carbon film. It's about half the diameter of the big carbon comp. It's also typical of the coils many people use. This assembly has a Q of 1.73 and has a phase shift of 60 degrees, both at 1MHz.
From what I've told you, would you be surprised if the first network caused a power amp to be unstable, oscillating at about 2-5 MHz on signal peaks, and the second worked just fine? No other changes, just the output coil and its resistor.
The last piece of info is the impedance. Z=5 ohms for the first coil, and 2 ohms for the second, both at 1MHz. That's 0.92 uHp for the first, 0.37 uHp for the second. It appears that due to odd wiring, feedback compensation, or the phase of the moon, some amps can be very fussy about the output network, and bigger is no guarantee of stability. I ruled out magnetic interference and pickup, and ultimately changed the amplifier compensation to be stable with almost any output network.
My point with all this verbiage is that one can't change the output network willy nilly, even slightly, and say anything about how the amp sounds, without completely reevaluating the stability and overall performance of the amp on the test bench. My guess is that simulation wouldn't have predicted this either, unless every important parasitic was accounted for.
So now the house is in silence, my amp is still apart on the bench, and it's all the fault of this rather long thread.
Let me describe two inductors:
The first consists of 15.5 turns of AWG19 enameled wire on a standard old fashioned 10 ohm 2W carbon composition resistor. This is pretty typical of the output coils used by many people. The combination of coil and resistor has a Q of 1.73 and has a phase shift of 60 degrees, both at 1MHz.
The second consists of 14 turns of a similar size wire on a 1W film resistor. 10 ohms again. Don't know the exact type, but it has magnetic metal end caps and is likely carbon film. It's about half the diameter of the big carbon comp. It's also typical of the coils many people use. This assembly has a Q of 1.73 and has a phase shift of 60 degrees, both at 1MHz.
From what I've told you, would you be surprised if the first network caused a power amp to be unstable, oscillating at about 2-5 MHz on signal peaks, and the second worked just fine? No other changes, just the output coil and its resistor.
The last piece of info is the impedance. Z=5 ohms for the first coil, and 2 ohms for the second, both at 1MHz. That's 0.92 uHp for the first, 0.37 uHp for the second. It appears that due to odd wiring, feedback compensation, or the phase of the moon, some amps can be very fussy about the output network, and bigger is no guarantee of stability. I ruled out magnetic interference and pickup, and ultimately changed the amplifier compensation to be stable with almost any output network.
My point with all this verbiage is that one can't change the output network willy nilly, even slightly, and say anything about how the amp sounds, without completely reevaluating the stability and overall performance of the amp on the test bench. My guess is that simulation wouldn't have predicted this either, unless every important parasitic was accounted for.
So now the house is in silence, my amp is still apart on the bench, and it's all the fault of this rather long thread.
Hi John & all,
I just found this thread and did not read it exhaustively so please forgive me if this has been covered already.
When designing in spice I find that an o/p coil appears to help stability.
I normally use 0.5uH
However the ringing that results from the coil with square waves I find is best damped ( with 1uF across 8 ohms ) with a 1 or 2 ohm resistor in parallel with the coil.
putting aside the situation with a 8uF capacitive load which is not my concern - do you think 0.5uH with 1 or 2 ohm in parallel will also be noticeably detrimental to the sound ?
mike
I just found this thread and did not read it exhaustively so please forgive me if this has been covered already.
When designing in spice I find that an o/p coil appears to help stability.
I normally use 0.5uH
However the ringing that results from the coil with square waves I find is best damped ( with 1uF across 8 ohms ) with a 1 or 2 ohm resistor in parallel with the coil.
putting aside the situation with a 8uF capacitive load which is not my concern - do you think 0.5uH with 1 or 2 ohm in parallel will also be noticeably detrimental to the sound ?
mike
Cellardoor said:
Hello Tony,
Quite interesting. I understand this last picture is actually the output signal with the coil? Would it be possible to have the same pic of the output signal *without* the coil? I guess that signal will also have substantial harmonics due to the asymmetric waveform. How much did the coil actually add?
What was your output signal level? Would it be possible to post a diagram of the speaker equivalent circuit?
Jan Didden
Hi Mike,
I do not believe that 0.5uH//2.2R would be audible, so I suggest you go ahead with this (with necessary Zobel ahead of the choke of course).
Hi Conrad,
Output chokes should be air-core. When you wind them outside of a resistor that is not non-inductive you are likely to have some coupled interaction which could lead to some form of a tuned peak.
Do 10 ohm carbon composition resistors not have a spiral manufacture too?
You wrote >>My point with all this verbiage is that one can't change the output network willy nilly, even slightly, and say anything about how the amp sounds, without completely reevaluating the stability and overall performance of the amp on the test bench. My guess is that simulation wouldn't have predicted this either, unless every important parasitic was accounted for. <<
Spot on Conrad, so my choice for resistors in an output stage ie. power emitter resistors, Zobel and in parallel with a choke if fitted, would be 'thick' film types.
Also, if I was using an output filter I would not have it near the PCB but close to the output sockets.
(Quiet house?
I find I can think through problems better when there are no distractions anyway.)
http://www.crystalradio.net/cal/indcal2.shtml
Cheers ........... Graham.
I do not believe that 0.5uH//2.2R would be audible, so I suggest you go ahead with this (with necessary Zobel ahead of the choke of course).
Hi Conrad,
Output chokes should be air-core. When you wind them outside of a resistor that is not non-inductive you are likely to have some coupled interaction which could lead to some form of a tuned peak.
Do 10 ohm carbon composition resistors not have a spiral manufacture too?
You wrote >>My point with all this verbiage is that one can't change the output network willy nilly, even slightly, and say anything about how the amp sounds, without completely reevaluating the stability and overall performance of the amp on the test bench. My guess is that simulation wouldn't have predicted this either, unless every important parasitic was accounted for. <<
Spot on Conrad, so my choice for resistors in an output stage ie. power emitter resistors, Zobel and in parallel with a choke if fitted, would be 'thick' film types.
Also, if I was using an output filter I would not have it near the PCB but close to the output sockets.
(Quiet house?
I find I can think through problems better when there are no distractions anyway.)
http://www.crystalradio.net/cal/indcal2.shtml
Cheers ........... Graham.
- Status
- Not open for further replies.