EEVblog Electronics Community Forum
Electronics => Projects, Designs, and Technical Stuff => Topic started by: Kokoriantz on November 03, 2022, 06:13:57 am
-
Half a century ago I encountered first time from the refrain of a pop song by the Turtles Happy Together. At that time, solid state was vs tube was the subject and this refrain sounded distorted by SS amps while no problem with simple tube radios. I found many such music pieces that was also bad with SS amps with most difficult ones, Sing, Sing of Goodman, requiems of Verdi and Mozart.
It has nothing to do with harmonic or IM distortions, as very low THD modern amps do same. What I did notice that the NFB at 20khz has an effect, as above 40db the pop song becomes good and 60db for the requiems.
I post bellow in wav format the refrain, stripped from CD of unknown group.
I simulated the end part of this refrain on Tina with 9 ideal amps each 20db NFB, the result was exactly that of the input. So no transient distortion. I simulated with 3 lm3886 models, also excellent, no slew rate limitation problem. The LM3886 in real world with gain of 26db reproduces the refrain with distortion, but with 180/100 ohm feedback resistors with gain of 2.8, the reproduction of the refrain is very good.
I simulated with Rod Elliott's P3A that I know it can't, on simulator it could with no problem.
What non linear element I can try to simulate this type of distortion?
-
So you say the song sounds "correct" when played from a simple tube amp radio but "wrong" when played through a solid state amp (and probably bigger speakers)?
If you want to re-create the distortion of tube amplifiers then try looking up "class A tube transfer curve". This will be a nonlinear curve, you can approximate it with modern transistors in a class A configuration OR you can implement it in software.
There are audio filter plugins for programs like audacity that try to recreate "old radio" style distortion. Unfortunately there are many unique flavours of this, so you might not find one that makes it sound like you remember.
You will find that the tube amplifier is only going to be one part of the equation. Much audio distortion comes from the choice of speaker and cabinet.
Another option is to buy an old tube radio and modify it to take a 3.5mm jack input, then record the output with a microphone. This new recoding can then be played through any modern audio system with reasonable speakers and give you a reasonably correct rendition of what you want. Be warned that this can be dangerous -- many old tube radios do not operate on isolated power supplies and some even connect the metal internal chassis to mains voltages! You will have to find one that is suitable.
When I was young I used to play games on my brother's Dreamcast. The bootup sound (https://www.youtube.com/watch?v=sOsw-Au9JCo) is very different in my memory compared to what it now sounds like (every time I hear it now I think it sounds completely "wrong"). I think the speakers and acoustic properties of the old family CRT television added quite a lot of distortion, especially on the low end, which dramatically changed how it sounded (and I never knew this was a "distorted" form of the sound).
EDIT: How on earth to I link to youtube without it embedding and replacing my words? >:(
-
You got it all wrong. Tubes have nothing to do in this story. It is due to feedback that these types of music gets distorted. As I said you need more than 40db NFB @20khz to get it right. Or, if the phase shift is lower than 90° lower NFB is needed which is the case with tube amps. Open loop amps or wide band amps as JLH do not have this kind of distortion.
-
https://youtu.be/Hlo1I3rx6n4?t=2305
-
Ah yes, that 'feedback distortion' that audiophiles keep claiming exist but nobody can demonstrate in waveforms or simulation.
-
What makes you think NFB creates distortion? Is it because the output of an audio amp with lots of NFB sounds different to an amp that has no or little NFB?
EDIT: If your negative feedback goes out of phase then it becomes positive feedback, which definitely creates distortion and is well understood, but most equipment is designed to avoid this. I presume you're not talking about this?
-
I can show it. Tomorrow I will wire a LM3886 with 26db gain and record the output with the refrain. You can hear it by yourself if you pass the wav file or the youtube on post 4 on any standard amplifier, 90% chance it will sound garbage. It is to repeat that the amount of NFB required depends upon the phase shift in high frequencies, that is higher is the NFB better it is.
-
It is not the NFB that creates the distortion, from experience I think it is the integrator function that needed for stability is the cause. With simulation, I understood that there is a kind of non linearity is involved along the phase shift that the simulator doesn't see it, this is the problem that I am trying to understand. The F3A of R Elliott, does distort in real but not on simulator, what it needs to do?
-
It was shown long ago by writers such as N Crowhurst that negative feedback (properly applied) reduces overall distortion, but can shift the relative amplitudes of the harmonic spectrum in that distortion that remains.
Some people like the sound of even-order distortion (as made by single-ended amplifiers without feedback), but almost everyone agrees that odd-order distortion is annoying.
As a real audiophile, concert-goer, and record collector, I might suggest that the recording process on those original Turtles' recordings was designed to sound "good" on a teen-ager's radio, not a proper high-fidelity system.
I remember an anecdote in Stereophile magazine (under its original ownership, not the current one) around 1968 where a pop group was recording in a professional studio, told the engineers that they didn't like the resulting sound, and used a previous recording as an example of what they wanted.
The engineers had to remove transistors from their professional Langevin amplifiers to satisfy the request.
-
Do like the I.F. amplifiers in radar. Limit the gain-per-stage and use negative feedback (such as un-bypassed emitter resistor or cathode resistor) in each stage and do away with the 40+Db 'reach-around' feedback. The latency of all the stages doesn't change within the usable frequency design range. The phase shift from input to output may appear to change when looking at in vs. out but all signals arrive with the same delay and no reach-around negative feedback distortion or anti-distortion is introduced when the reach-around is no longer needed or incorporated.
-
Sing,Sing,Sing of B Goodman origina.
https://www.youtube.com/watch?v=r2S1I_ien6A (https://www.youtube.com/watch?v=r2S1I_ien6A)
Turtles remastered
https://youtu.be/pSw8an1u3rc?t=40 (https://youtu.be/pSw8an1u3rc?t=40)
The short refrain file I posted is a recent recording from an unknown group but it is not mp3, as many accuse to be reason why it sounds distorted. The Sing, Sing, Sing sounds bad because according to a sound engineer, it was recorded with 1940's technology, make it mp3 and you get garbage sound. No, if your amp has wide bandwidth in open loop or feedbackless SET, sounds no any distortion. With multiple horn instruments, a very pleasant beat occurs, you can hear from a marching band or simple three horn, trombone + saxo + trumpet, this pleasant beat I never heard it reproduced in any HIFI. If the amp has feedback distortion, the beat becomes dirty sound as you can hear from this 27 horns music of Goodmans orchestra.
-
amp has feedback distortion
I've done some mixed signal and analog work, and have never heard of this. Info on the exact source and definition online seems to be largely "but it sounds worse". Is there some kind of Bode plot or THD measurement or anything at all that shows what people mean by "negative feedback distortion"? Everything is sine waves so it should be easy to measure.
In general, a transfer function with negative feeback can always be converted to a transfer function without any feedback (become open loop, see attached), so the term itself seems to be either meaningless or at least misleading.
-
I recorded the total "Happy together" and the requiem of Verdi.
To avoid copyright problems I introduced errors on the original at the beginning.
The recordings are mono right channel.
-
Recording requiem of Verdi with lm3886 26db gain
-
Happy together original mp3
-
recording with lm3886 gain 20.8
-
You got it all wrong. Tubes have nothing to do in this story. It is due to feedback that these types of music gets distorted. As I said you need more than 40db NFB @20khz to get it right. Or, if the phase shift is lower than 90° lower NFB is needed which is the case with tube amps. Open loop amps or wide band amps as JLH do not have this kind of distortion.
Unless I am missing something, 40 dB at 20 kHz means an NE5532 (10 MHz GBW) can be used with a gain of 5 and still produce “good” audio in your opinion. Yes?
Can you be specific about the specs required for a “wideband” amp that does not suffer from this distortion that you cannot measure?
Jason
-
I think NE5532 is 60db@20khz. For perfect sound it can provide 0db gain. With NE5534 stable with gain of 3 has 70db, that is good to 10db gain. The same for lme49720 and lm4562.
There are a few 10khz bandwidth opamps, my favorite is AD826, the great CFA AD8008 has bandwidth of 100khz with 1k input impedance for unity gain stable, but cost 18$. https://www.analog.com/media/en/technical-documentation/data-sheets/ad8007_8008.pdf (https://www.analog.com/media/en/technical-documentation/data-sheets/ad8007_8008.pdf)
-
Bonjour a tous!
I have Just noticed this interesting but Very old topic.
Since 1970s in numerous AES papers and other industry/mfg app notes.
eg AES papers
TIM: Otala and John Curl: https://www.ka-electronics.com/images/pdf/Leinonen_Otala_Curl_TIM_Measurement.pdf (https://www.ka-electronics.com/images/pdf/Leinonen_Otala_Curl_TIM_Measurement.pdf)
SID: Slewing induced distortion Craig TODD (CTO/VP/RD Dolby Labs) Walt JUNG (audio/analog EE, author) for instance.
https://en.wikipedia.org/wiki/Slew-induced_distortion (https://en.wikipedia.org/wiki/Slew-induced_distortion)
Also Bob Adams at ADI and the usual gang at LT, Nationals Semi etc.
Golden standard of audio measurement is the fine Audio Precision machines, from Rich Cabot and Bruce Hoffer ex TEK and AP founders.
https://www.ap.com/ (https://www.ap.com/)
Many papers on every aspect of audio measurement and distortion.
We used AP SYS 1 and SYS 2 since 1980s..1990s.
Hope this is interesting to some!
Bon Weekend!
Jon
-
Salue Jean Paul.
Thank you for information. This type of distortion by experience is due to phase shift and NFB. The slew rate distortion is independent from NFB.
This type of distortion is independent from the signal level, but it occurs only when great number of voices is involved, as large jazz orchestras or chorals. It is the beat that is provoking some kind of distortion. There is a measurement using multi tones, I will try using very narrow spaced 16 notes in 1khz-3khz range and hear in real what I get.
-
Bon samedi cher Monsieur Kokoriantz:
The best test of distortion is two tone eg CCIR or SMPTE:
Your analyzer or gen generates two tones closely spaced in freg eg CCIR 10kHz and 10.4 kHz. equal level
Intermod will then give sum and difference.
SMPTE IM is a 6 kHz modulated by 60 Hz 4:1 level difference.
Read the lowpass filtered difference tone eg 400 Hz CCIR or 60 Hz SMPTE.
VERY sensitive and more relevant test than THD
AP and other audio analyzers may have an IM tone or option.
IM test relation to feedback is not clear to me.
I defer to Black, Nyquist, Blackman and the others at BTL that developed feedback theory in 1920s for long lines telephone transmission.
listening tests require fine program materials
Recommend:
Hig res Beethovens 9th
Solo piano sonatas
Folk music of the Andes
Opera/Jazz/etc with high res recording.
Philips, Sonly and others produced test CDs in 1980s to check CD payers and recorders.
The AV Forums have many test DVD/BR for home cinema, with excellent audio tests.
We used the EBU SQAM Subjective Quality Assessment Material, was avail from EBU decades ago, now rare.
DL a FLAC version here
https://tech.ebu.ch/publications/sqamcd
SQAM has many test tones, sweeps, noise, and very carefully recorded music, including voice and solo instruments.
SQAM is a REALLY great test of your system, speakers, ears.
Finally beware of auditory "recruitment" when certain tones and levels excite inner ear distortion, especially as one ages the effect can be worse.
Happy listening,
Bon Weekend
Jon
from a feedback free and ancient audio guy....
and a random double blind ABX tyoe relay box, from 1970s..1980s.
-
I formulated a 13 tone generator of 1khz spaced +/-3% or flat to sharp with 5hz interval.
I applied it on example amp in LTspice and here is the input and output spectrums. This way of measuring 1khz chorus subtracting from the output, the difference can tell me someting. Of course it is both harmonic and modulation distortions, may be I can see as Jonpaul suggest it can be intermodulation distortion emphasized by integrator.
The distortion of this amp with same input level but single 1khz is 0.07%
-
no Simulator is accurate for audio.
Just build it, apply two tones ( not 12) and listen, as you sweep in amplitude and frequency
j
j
-
I posted on 12&13 the original and recorded requiem of Verdi to show the difference, no one listened to it. I am listening such distortions since 5 decades.
Bellow is the multitone 1khz of the LTspice example amp and AD824 which has 40db@20khz in open loop with a gain of 26db.
-
Here is the evidence of IM due to feedback integrator.
The circuit on bench is AD844.
https://www.analog.com/media/en/technical-documentation/data-sheets/AD844.pdf (https://www.analog.com/media/en/technical-documentation/data-sheets/AD844.pdf)
This CFA opamp has an independent trans impedance followed by a diamond unity gain, the output of the trans impedance is available as Tz. If a capacitor is grounded, it will decrease its bandwidth, if resistor is grounded, it will decrease the gain and enlarge the bandwidth.
First I tested it with 330pF to get 70db open loop gain at 1khz. Then with 600kohm to have 70db gain flat to 60khz-3db. The last is 300kohm to get 64db to 100khz.
It is evident that the IM distortion is lower with wide bandwidth even with lower gain.
-
How to measure the total distortions with LTspice?
I started using two frequencies 500hz+1.5khz. I get the difference and sum 1khz&2khz. But to get a total signal to all others, may be called SINAD, I first run transient for 1s to have precise voltages of the fundamental signals. As they are equal, I measure on on the FFT and multiply by √2 to get the RMS value of what the output should be. Ltspice can measure the RMS value of the .raw graph by (ctrl+click) on the trace name and you get signals+IM+THD+N. By subtracting and deviding one gets the Total undesired/signals.
Ltspice can do more, on FFT graph if selected 20Hz-20khz range by (CTRL+click) on the name gives the Power bandwidth and same measure only one signal can have the signal component. I don't know yet how to add the two signals and make similar ratio, but it is possible to have only audible signal/total distortions ratio.
-
To get an idea of IM distortion, only first order is used with 500+1.5kHz signal. Need to run 1s and read the .four 500 4 v(out). You will get the peak and relative values of the signals and the side bands 1khz and 2khz. I don't know if it is as THD, to √ the squares of the sidebands relative values.
With Tina it is much faster, as it calculates only the Fourier function.
-
See the SMPTE and CCIF/ITU-R two tone IM.
Standards well accepted and used.
https://www.ranecommercial.com/legacy/note145.html (https://www.ranecommercial.com/legacy/note145.html)
http://personal.ee.surrey.ac.uk/Personal/P.Jackson/ee2.lab/S5_adm/AudioSpecifications.pdf (http://personal.ee.surrey.ac.uk/Personal/P.Jackson/ee2.lab/S5_adm/AudioSpecifications.pdf)
https://www.ap.com/technical-library/more-about-imd/ (https://www.ap.com/technical-library/more-about-imd/)
Jon
-
Salue Jean-Paul. Merci beaucoup for the links.
So the IMD is the rms sum of the sidebands ratio. There are many audio components as speakers, microphones, SET amps, pick-up cartridges that have over 3% IMD and sound nice. I am turning in circle to find a way to distinguish the good from the bad sounding sidebands.
I found from Bob Cordell mutitone IMD to replace TIM,
http://www.cordellaudio.com/papers/multitone_test.pdf (http://www.cordellaudio.com/papers/multitone_test.pdf)
I tried it and here is the result FFT. High frequencies above 3khz fundamental is not important as the problem to resolve is mutitone distortion in less than 4khz, highest musical note.
-
The IM is usually a pair of tones at a high freq eg 7, 10, 11 kHz with a low freq difference. 60 Hz or 1 kHz
Jon
-
The old fashion, was 1/4 60hz and 7khz so that the low frequency components could easily filtered out and am detector measures the 7khz modulation which gives idea about 2nd and 3rd order pairs sidebands. Such distortion you can hear very easily in jazz music where the voice of the singer or saxo, gets modulated by the double C as if the sound is reflected by a rotating fan. I use Fever of Peggy Lee to test this. Above 2% 2nd harmonic it becomes audible.
-
I wanted to see on spectrum the distortion of amp with complex sound. I used 1s extract from Verdi and passed it through AD844 distorted by diodes and a resistor for 1.3% even or 0.3% odd. In either case, I didn't hear or see on the spectrum any difference. I want to remind the shure V5, once revered pick-up cartridge used mostly in studios or radio stations, has 2% second and 0.5% 3rd with 3%IMD.
What I conclude that what makes a sound bad is another aspect.
-
I passed through the AD844 with 0.45% odd harmonic distortion
Fourier coefficients
k Amplitude (C) Phase (ø)
0. 917.22u 0
1. 447.88m -90.01
2. 10.93u -6.28
3. 1.76m -89.9
4. 9.13u -8.66
5. 920.15u -91.03
6. 8.18u -2.57
7. 565.38u -93.17
8. 8.42u 7.44
9. 363.85u -95.27
10. 8.98u 13.06
11. 234.49u -95.7
12. 8.44u 15.07
13. 150.87u -92.95
14. 6.3u 19.72
15. 99.6u -87.44
16. 3.73u 43.65
an extract from Pink Floyd's the Wall, where a hundred of teenagers shouting "Hey teacher we don't need the wall".
bellow is the input and output.
I don't hear any bad sound, if you do, please describe.
-
And this is with 1.33% mainly even order harmonics.
Fourier coefficients
k Amplitude (C) Phase (ø)
0. 6.3m 180
1. 459.68m -90.01
2. 5.95m -226.05m
3. 868.18u -89.78
4. 992.8u -1.12
5. 456.16u -90.71
6. 319.38u -2.47
7. 282.29u -92.66
8. 110.37u -2.2
9. 183.91u -94.63
10. 28.92u 10.53
11. 120.86u -95.1
12. 11.42u 110.78
13. 79.95u -92.8
14. 18.55u 145.81
15. 54.51u -88.34
16. 19.58u 161.69
Now I see the different distortion standards for Tube and transistor amplifiers. With tubes, the behavior is near theory as simulation, but transistors must have another type of distortion that makes the sound dirty with the same distortion numbers.
I will realize the AD844 circuit and record this same file to see reality vs simulation.
-
I found the second report of Otala for TIM description from Hifisonic site. As described I made a theory amp with only the input differential in old time transistors. I did see with 3.5khz square and 1/4 15khz sine, saturation of the differential error current +/-1ma with 1nF miller. I passed the "Wall" wav file I couldn't get the current saturated, the peaks are at high frequencies and not with choral. Here what the document says:
A low slew rate or poor power bandwidth may indicate the presence
of TIM, but the converse is not generally true. Equally, a
high slew rate or good power bandwidth does not automatically make TIM less likely to occur. The presence of TIM in an amplifier may also be suspected from an unrealistically low
THD specification, which may indicate the possibility of large
feedback and, consequently, heavy compensation [lo]
-
Listening carefully to bad sounding amp, I got the impression that the bad sound is due to lower amplitude definition, as if 16bit has is getting 8bit when numerous voices at same note is reproduced. Bellow is the graph of 16 voices linearly arranged +/-3% to be flat to sharp (bémol-dièse) of the same note. The mixed signal, as one can see, goes to very low levels and the 16 voices get reconstructed only if at very low level amplitude is faithfully amplified. There goes wrong, The AD844 with 1mv 1khz input generates 1.3% THD+N. May be the noise and the harmonics at low level are the cause of loss in definition.
I need a better measuring tool then I have, I read Behringer UMC202 to good one for low budget.
-
I added to AD844 two noise generators. A current noise generator parallel to the integrating capacitor representing the noise in the input and VAS stage and a series voltage noise generator representing the noise of the output EF stage.
I measured the 13 tone Signal to noise ratio with 100pF, 1nF and 10nF. Shown bellow.
The capacitor decreases the gain at 1khz, the input and VAS noise increase and decrease with the same ratio, so the signal to noise remains the same but the output noise does not decrease with the capacitor. By this, the output noise is the one will reduce the S/N by reducing the gain with higher capacitor. Strangely from 100pF to 1nF nothing happens, from 1nF to 10nF, instead of decreasing by 20db S/N ratio it got decreased by 30db.
May be it is the same with the distortion, That is the output stage distortion influences the mostly by the capacitor value.
How have thought to make low noise output stage?