Author Topic: Maximum slew rate typically found in music/voice  (Read 7075 times)

0 Members and 1 Guest are viewing this topic.

Offline CatalinaWOW

  • Super Contributor
  • ***
  • Posts: 5571
  • Country: us
Re: Maximum slew rate typically found in music/voice
« Reply #25 on: September 17, 2023, 03:24:06 am »
All of this discussion is wrapped around amplifier design and the effect of slew rate on distortion etc.

I have had an interest in slew rate for a different reason.  I transcribe old LPs (and other recordings) to .mp3 for convenience in listening and to avoid the problems controlling dirt, wear and other things in mechanical media.  I am not an audiophoole, and in fact with my old ears don't need particularly hi-fi.  But would like to remove the ticks, pops and clicks from the vinyl.  The automatic click removal tool in Audacity actually does yeoman work in removing the highest amplitude clicks.  But if it's parameters are adjusted to get rid of all of the audible clicks it has a noticeable negative impact to the sound, even to my pedestrian ears.

You can manually go through the recording and eliminate the rest, but it is incredibly tedious and if the record is particularly noisy it can take many hours.  Hence I started to write my own automatic click detector.  Slew rate is an obvious metric, so many clicks have a slew rate easily within the recording equipment's capability, but out of family for the music on the record (and probably all music, since the stuff I am listening to has cymbal strikes and other similar sounds).  But I never came up with a meaningful result.  Partly because I found that the psychoacoustic stuff was too complex for me.  You could have two clicks with identical time traces and one would be painfully audible and another inaudible even with careful listening depending on the sounds surrounding them in time.  Many clicks that were objectionable had lower slew rates than "real" sounds.

Long diatribe, with the only point being that slew rate may have other impacts than on the purity of reproduction.
 
The following users thanked this post: edavid

Offline mzzj

  • Super Contributor
  • ***
  • Posts: 1298
  • Country: fi
Re: Maximum slew rate typically found in music/voice
« Reply #26 on: September 17, 2023, 07:06:08 am »

I really wasn't looking to get too complicated with this, I just thought that someone, somewhere, would have done this research and might have published it somewhere. 

Hal
See my link above earlier.. " The prize for greatest slew rate found goes to a single peak from 'Year 3000' by Busted, which hit 150mV, equivalent to the slew rate of a full level sine wave about 10kHz, i.e. five times greater than Baxandall's figure, and equivalent to a level of 2.5V/µsec for a 100W power level."

https://www.renardson-audio.com/slew-tid.html
 

Online David Hess

  • Super Contributor
  • ***
  • Posts: 17429
  • Country: us
  • DavidH
Re: Maximum slew rate typically found in music/voice
« Reply #27 on: September 17, 2023, 01:13:28 pm »
Sampling frequency above that is needed because 2x sampling frequency from Nyquist's theorem does not really work IRL because real low pass filters are not perfect and thus you want more margin. 44kHz sampling works good enough not because it works fine for 20kHz but because even though humans can hear high frequencies, they do it poorly. And neither actual sound contains much in that part of spectrum.

Before they used oversampling ADCs and DACs, 8th order elliptic filters sort of met the needed specifications, however anything modern uses oversampling ADCs and DACs and digital filtering so the antialiasing requirements are easily met now.  The aliasing filter problem does not apply to modern 44.1kHz equipment.

:-+ And I also don't see why the relation tr = 0.35 / BW would apply here, which is rather the 10%...90% risetime of the step response of a Gaussian filter with a given -3dB bandwidth. This waveform is not a sine wave, but the integral of a Gaussian bell.

It still mostly applies, just like the 0.35 rule works for oscilloscopes with their bandwidth filter engaged which is decidedly not a Gaussian response.  I tested this once with a Tektronix 7A22 set at various bandwidths and 0.35 matched exactly, as I suppose Tektronix intended it to.

The fastest slew rate will be from a combination of frequencies so a step response, which has a comb of high frequencies, into the bandwidth filter will give a good approximation of the highest slew rate.
 

Offline gf

  • Super Contributor
  • ***
  • Posts: 1425
  • Country: de
Re: Maximum slew rate typically found in music/voice
« Reply #28 on: September 17, 2023, 01:52:36 pm »
:-+ And I also don't see why the relation tr = 0.35 / BW would apply here, which is rather the 10%...90% risetime of the step response of a Gaussian filter with a given -3dB bandwidth. This waveform is not a sine wave, but the integral of a Gaussian bell.

It still mostly applies, just like the 0.35 rule works for oscilloscopes with their bandwidth filter engaged which is decidedly not a Gaussian response.  I tested this once with a Tektronix 7A22 set at various bandwidths and 0.35 matched exactly, as I suppose Tektronix intended it to.

The fastest slew rate will be from a combination of frequencies so a step response, which has a comb of high frequencies, into the bandwidth filter will give a good approximation of the highest slew rate.

Yes, 0.35 seems to be the tr*BW product of traditional Tek scopes. An ideal Gaussian step response has actually 0.332 (which is of course pretty close).
[ Edit: An ideal Gaussian filter is not realizable anyway. In the analog domain, approximations like "Gaussian to -6dB" seem to be common, and in the digital domain the infinite extent needs to be truncated/windowed, so that we end up with an approximation, too. ]

My point was that the calculation 1.6*Apeak/tr = 1.6*Apeak/(0.35/f) leads to 4.57*f*Apeak, which underestimates the maximum slew rate of a sine wave Apeak*sin(2*pi*f*t), whose maximum slew rate is actually 2*pi*f*Apeak = 6.28*f*Apeak.
« Last Edit: September 17, 2023, 02:44:31 pm by gf »
 

Offline gf

  • Super Contributor
  • ***
  • Posts: 1425
  • Country: de
Re: Maximum slew rate typically found in music/voice
« Reply #29 on: September 17, 2023, 03:04:33 pm »
Does the DAC have more stringent slew rate requirements?  Slewing from rail to rail in a single sample period is \$2 \, V_{pk}\$ in one 44100'ths of a second, or \$88200 \, V_{pk}\$ per second, which is less than the aforementioned reconstruction limit.  This means that with a theoretically perfect brick-wall low-pass filter, the \$0.1385 \, V_{pk}\$ slew rate suffices, but this slew rate is \$1.5707\$ times (\$\pi/2\$) as fast as just slewing from rail to rail in a single sample period.  This affects our choice of DACs, as just being able to slew from rail to rail in a single sample period is not sufficient; it needs to slew basically 1.5707 times rail-to-rail range, in a single sample period.

If we did a Fourier analysis of the error spectrum at different (higher than necessary) DAC slew rates, we'd find a faster slew rate does push the error noise somewhat higher in the output spectrum, which makes it easier to filter out this particular error using analog circuits.  Note that we're still assuming a brick-wall low-pass filter at 22050 Hz for reconstructing the highest-frequency components in the audio signal, though.

A DAC emits a pulse with a particular shape for each sample. In order to avoid non-linear distortion, the pulse shape must be the same for each sample (except for differently scaled amplitude, of course).

However, the only pulse shape which enables "perfect reconstruction" is a Dirac delta pulse, i.e. in order to obtain an ideal, undistorted reconstruction (after applying the brickwall filter), the DAC would need to output a Dirac delta pulse for each sample (which requires infinite slew rate, of course).

For any other pulse shape, the output of the DAC can be considered as if it were the above mentioned stream of delta pulses, convolved (in the continuous-time domain) with the DAC's pulse shape. The consequence is that any pulse shape other than Dirac delta leads to linear distortion, or in other words, it affects the frequency response. The convolution with the pulse shape acts as a filter.

The linear interpolation between the samples you are considering above corresponds to a (hypothetical) DAC which emits a triangular pulse with FWHM=1/fs for each sample. This results in a sinc²-shaped lowpass frequency response, whose roll-off approaches 20*log10((sin(pi/2)/(pi/2))^2) = -7.84dB at fs/2 (or for for CD audio at 44.1kSa/s, it would be about  -6.3dB at 20kHz).

If you want to compensate this frequency roll-off in the digital domain before sending the data to the DAC, then you need additional dynamic range headroom for up to 2.47x higher amplitudes being sent to the DAC (implying the need for a 2.47x higher slew rate for the triangular pulses as well).

BTW, I guess that many are not aware that even the commonly used rectangular pulse shape with a pulse width of 1/fs (a.k.a. first-order hold) introduces a sinc-shaped frequency response roll-off (in the amount of e.g. -3.1dB at 20kHz, for CD audio at 44.1kSa/s).
 
The following users thanked this post: newbrain

Offline Karel

  • Super Contributor
  • ***
  • Posts: 2275
  • Country: 00
Re: Maximum slew rate typically found in music/voice
« Reply #30 on: September 17, 2023, 03:43:29 pm »
Please read this article:

A Method for Measuring Transient Intermodulation Distortion TIM

https://www.ka-electronics.com/images/pdf/Leinonen_Otala_Curl_TIM_Measurement.pdf

 

Offline newbrain

  • Super Contributor
  • ***
  • Posts: 1801
  • Country: se
Re: Maximum slew rate typically found in music/voice
« Reply #31 on: September 19, 2023, 02:10:45 pm »
This means that at that frequency, the phase resolution of CD audio is 360°/12 = 30°.  If we produce two sinusoidal signals at 3675 Hz (each full wave taking about 272 µs), one for each ear, both starting at zero phase, but one minutely delayed, humans can generally discriminate at down to 10 µs difference.  That corresponds to a 360° × 10µs / 272µs ≃ 13° phase difference at 3675 Hz.  Thus, CD stereo audio does not have sufficient phase resolution at 3675 Hz to match human hearing.
Nope. Sampling and reconstruction simply do not work that way.
The theory says that any band limited signal can be reconstructed perfectly, and that include its phase; any reasonably implemented sampling system (CD) is able to reproduce the phase shift with inaudible accuracy.
Please check this very old but still extremely valid video, it's absolutely worth watching the whole thing, but the relevant part starts at 20.56.
https://youtu.be/UqiBJbREUgU?feature=shared&t=1256
Nandemo wa shiranai wa yo, shitteru koto dake.
 
The following users thanked this post: nctnico, Wolfram, gf

Offline Zero999

  • Super Contributor
  • ***
  • Posts: 20363
  • Country: gb
  • 0999
Re: Maximum slew rate typically found in music/voice
« Reply #32 on: September 19, 2023, 02:42:12 pm »
Wow, there's a lot of interesting, yet irrelevant information in this thread.

It's true, the original question isn't very well phrased, but it does have merit. An amplifier does not have to have its full power bandwidth extend all the way up to 20kHz. Look at a loudspeaker cabinet and note, the tweeter has around a tenth of the power rating, of the woofer. That should be enough to give you a clue. Most of the energy in music, can be found at lower frequencies. An LM741 run at +15/-15V will have more than sufficient slew rate to output an audio waveform with an amplitude of 20V peak-to-peak. There are other reasons not to use the 741.
 

Offline TimFox

  • Super Contributor
  • ***
  • Posts: 9003
  • Country: us
  • Retired, now restoring antique test equipment
Re: Maximum slew rate typically found in music/voice
« Reply #33 on: September 19, 2023, 03:27:02 pm »
Wow, there's a lot of interesting, yet irrelevant information in this thread.

It's true, the original question isn't very well phrased, but it does have merit. An amplifier does not have to have its full power bandwidth extend all the way up to 20kHz. Look at a loudspeaker cabinet and note, the tweeter has around a tenth of the power rating, of the woofer. That should be enough to give you a clue. Most of the energy in music, can be found at lower frequencies. An LM741 run at +15/-15V will have more than sufficient slew rate to output an audio waveform with an amplitude of 20V peak-to-peak. There are other reasons not to use the 741.

The 741 putting out 20 V pk-pk will push its slew rate limit.  Minimum 0.3 V/us rating corresponds to only 4.8 kHz at 20 V pk-pk.
The actual slew rate limit is like clipping:  you don't want to hit it, since the feedback around the circuit will fail and no good will come of it, and stuff happens as you approach it.
As I pointed out above, the input differential stage in a typical op amp is reasonably linear at small amplitudes, but at larger current outputs (the maximum current into the compensation capacitor defines the slew rate) it is less linear and distortion increases.
 

Offline Zero999

  • Super Contributor
  • ***
  • Posts: 20363
  • Country: gb
  • 0999
Re: Maximum slew rate typically found in music/voice
« Reply #34 on: September 19, 2023, 05:47:40 pm »
Wow, there's a lot of interesting, yet irrelevant information in this thread.

It's true, the original question isn't very well phrased, but it does have merit. An amplifier does not have to have its full power bandwidth extend all the way up to 20kHz. Look at a loudspeaker cabinet and note, the tweeter has around a tenth of the power rating, of the woofer. That should be enough to give you a clue. Most of the energy in music, can be found at lower frequencies. An LM741 run at +15/-15V will have more than sufficient slew rate to output an audio waveform with an amplitude of 20V peak-to-peak. There are other reasons not to use the 741.

The 741 putting out 20 V pk-pk will push its slew rate limit.  Minimum 0.3 V/us rating corresponds to only 4.8 kHz at 20 V pk-pk.
The actual slew rate limit is like clipping:  you don't want to hit it, since the feedback around the circuit will fail and no good will come of it, and stuff happens as you approach it.
That's more than good enough, considering the bandwidth of music. You'll only get 20V peak to peak, at low frequencies, up to a few hundreds of Hz, not at 4.8kHz.

As far as not hitting the slew rate vs clipping is concerned. The distortion from limited slew rate doesn't sound as bad as clipping. The effect is more similar to when the sound has travelled through a wall, or ceiling than clipping.

Having a wider bandwidth op-amp is good, but primarily because negative feedback is maintained at higher frequencies, which minimises the distortion due to the non-linearities in the output stage, rather than the slew rate.
 

Offline magic

  • Super Contributor
  • ***
  • Posts: 7454
  • Country: pl
Re: Maximum slew rate typically found in music/voice
« Reply #35 on: September 19, 2023, 06:11:10 pm »
Minimum 0.3 V/us rating corresponds to only 4.8 kHz at 20 V pk-pk.
That's more than good enough, considering the bandwidth of music.

See my link above earlier.. " The prize for greatest slew rate found goes to a single peak from 'Year 3000' by Busted, which hit 150mV, equivalent to the slew rate of a full level sine wave about 10kHz, i.e. five times greater than Baxandall's figure, and equivalent to a level of 2.5V/µsec for a 100W power level."

https://www.renardson-audio.com/slew-tid.html

100W into 8Ω being 40Vpp. Not so good for the old 741.


Having a wider bandwidth op-amp is good, but primarily because negative feedback is maintained at higher frequencies, which minimises the distortion due to the non-linearities in the output stage, rather than the slew rate.
Correction of output stage distortion has nothing to do with slew rate, it only depends on loop gain at given frequency.
 

Offline Zero999

  • Super Contributor
  • ***
  • Posts: 20363
  • Country: gb
  • 0999
Re: Maximum slew rate typically found in music/voice
« Reply #36 on: September 19, 2023, 09:40:04 pm »
Minimum 0.3 V/us rating corresponds to only 4.8 kHz at 20 V pk-pk.
That's more than good enough, considering the bandwidth of music.

See my link above earlier.. " The prize for greatest slew rate found goes to a single peak from 'Year 3000' by Busted, which hit 150mV, equivalent to the slew rate of a full level sine wave about 10kHz, i.e. five times greater than Baxandall's figure, and equivalent to a level of 2.5V/µsec for a 100W power level."

https://www.renardson-audio.com/slew-tid.html

100W into 8Ω being 40Vpp. Not so good for the old 741.
That's a moot point because the 741 is not a power amplifier and can't handle the current either.

Quote
Having a wider bandwidth op-amp is good, but primarily because negative feedback is maintained at higher frequencies, which minimises the distortion due to the non-linearities in the output stage, rather than the slew rate.
Correction of output stage distortion has nothing to do with slew rate, it only depends on loop gain at given frequency.
That's true.
 

Offline magic

  • Super Contributor
  • ***
  • Posts: 7454
  • Country: pl
Re: Maximum slew rate typically found in music/voice
« Reply #37 on: September 20, 2023, 05:46:18 am »
But it can handle the voltage swing. Or at least a quarter of it (sorry, 100W is 80Vpp, not 40Vpp).

Which means that over 0.6V/μs is needed to reproduce some real world CDs at 20Vpp full scale.

Which also means that 741 likely wasn't used to produced these CDs, maybe for good reason ;)
« Last Edit: September 20, 2023, 05:48:13 am by magic »
 

Offline Zero999

  • Super Contributor
  • ***
  • Posts: 20363
  • Country: gb
  • 0999
Re: Maximum slew rate typically found in music/voice
« Reply #38 on: September 20, 2023, 09:39:28 am »
But it can handle the voltage swing. Or at least a quarter of it (sorry, 100W is 80Vpp, not 40Vpp).

Which means that over 0.6V/μs is needed to reproduce some real world CDs at 20Vpp full scale.

Which also means that 741 likely wasn't used to produced these CDs, maybe for good reason ;)
I don't know where the figure of 0.3V/µs taken from, but the datasheet says 0.5µ/s.
https://www.ti.com/lit/ds/symlink/ua741.pdf

Alright I take the point, if can't quite be able to produce all real world CDs, just most of them. There are a couple of outliers.

I'm not advocating the 741 audio, just point out that an audio amplifier doesn't have to have sufficient slew rate to output a 20kHz sine wave at the maximum output voltage.
 

Offline TimFox

  • Super Contributor
  • ***
  • Posts: 9003
  • Country: us
  • Retired, now restoring antique test equipment
Re: Maximum slew rate typically found in music/voice
« Reply #39 on: September 20, 2023, 02:23:50 pm »
0.5 V/us is typical
The only minimum value I could find is 0.3 V/us, which I used and noted as minimum
  National LM741A datasheet  https://www.mit.edu/~6.301/LM741.pdf

Of course, no one seriously recommends the 741 for modern audio, since there are reasonably-priced devices with roughly 10x the slew rate.
TL071/2/4:  8 V/us min, 16 V/us typ
NE5532:  9 V/us typ
« Last Edit: September 20, 2023, 02:43:10 pm by TimFox »
 

Online David Hess

  • Super Contributor
  • ***
  • Posts: 17429
  • Country: us
  • DavidH
Re: Maximum slew rate typically found in music/voice
« Reply #40 on: September 20, 2023, 03:23:39 pm »
Of course, no one seriously recommends the 741 for modern audio, since there are reasonably-priced devices with roughly 10x the slew rate.
TL071/2/4:  8 V/us min, 16 V/us typ
NE5532:  9 V/us typ

Signetics designed a number of high slew rate bipolar parts at that time, but just a couple the NE series survived.  Designers who were desperate used the LM318 but it relies on emitter degeneration so has high noise.  National had the still popular LM833 at 5 min and 7 typ V/us ; I never understood how they achieved that and low noise but suspect the published schematics are simplified and do not show how the transconductance was reduced.  Motorola had a bunch of high slew rate 741 replacements which are still available.

I suspect the JFET parts were the most common 741 replacements for audio because their low input bias current and current noise allows higher impedance networks.
 

Offline TimFox

  • Super Contributor
  • ***
  • Posts: 9003
  • Country: us
  • Retired, now restoring antique test equipment
Re: Maximum slew rate typically found in music/voice
« Reply #41 on: September 20, 2023, 04:20:55 pm »
I went looking for the seminal article by W G Jung et al. in 1979 about slew-induced distortion, and found this in a free location:
https://web.archive.org/web/20190303175101/http://pdfs.semanticscholar.org/cdbd/3ffcb276aabaf6cccd9b6bf067ee665c82d9.pdf

Before the 5534 series came out, there were some good high slew-rate products from Harris (later Intersil), that I believe used dielectric isolation or some such construction technique.
e.g.  https://pdf1.alldatasheet.com/datasheet-pdf/view/66679/INTERSIL/HA-2505.html
« Last Edit: September 20, 2023, 04:25:45 pm by TimFox »
 

Offline Nominal Animal

  • Super Contributor
  • ***
  • Posts: 7198
  • Country: fi
    • My home page and email address
Re: Maximum slew rate typically found in music/voice
« Reply #42 on: September 20, 2023, 11:22:52 pm »
The theory says that any band limited signal can be reconstructed perfectly, and that include its phase; any reasonably implemented sampling system (CD) is able to reproduce the phase shift with inaudible accuracy.
1. Shannon-Nyquist only applies exactly if you have infinite precision, i.e. exact discrete samples.  Quantizing the samples limits the accuracy.   Discretized quantized signal is limited in both bandwidth and phase resolution, and is limited in the resolution at which it can represent the wave packet arrival time.

2. Human time-domain signal arrival time discrimination between each ear is not exactly a phase shift; the ear does not work like that.  You do need to model it as the arrival of a wave packet to understand exactly how it occurs.  Specifically, it requires the leading edge of the incoming wave packet to be sufficiently steep and sufficiently high maximum amplitude; in a Fourier transform, this is not a single-frequency event.

You can easily simulate this, if you actually wanted to.  Just generate a suitably shaped pulse as I described earlier at a higher sample rate.  Generate several variants, delaying each by a one additional leading zero sample, then pad with lots of zeros to a suitable size.  Decimate to CD audio sample rate and quantize to 16 bit samples, then resample back to original sample rate, and compare the initial and final waveforms.

For example, using the aforementioned waveform using \$f = 3675\text{ Hz}\$ at 176400 samples per second (4x CD audio rate), with four variants; containing one, two, and three leading zero samples compared to the first (corresponding to 5.7µs, 11.3µs, and 17µs delays), using Python 3 and SciPy:
Code: [Select]
import scipy
import scipy.signal
from math import sin, pi
from sys import stdout, stderr

# Original discrete waveform
wave = []
for i in range(0, 12001):
    x = i / 12000.0
    wave.append( (1.0 - x)*(1.0 - x)*sin(500.0 * pi * x) )

# Construct four samples, each delayed by one sample
orig1 = scipy.array(([0]*36000) + wave + ([0]*47999), dtype=scipy.float64)
orig2 = scipy.array(([0]*36001) + wave + ([0]*47998), dtype=scipy.float64)
orig3 = scipy.array(([0]*36002) + wave + ([0]*47997), dtype=scipy.float64)
orig4 = scipy.array(([0]*36003) + wave + ([0]*47996), dtype=scipy.float64)

# Resample by 1/4
audio1 = scipy.signal.resample(orig1, 24000)
audio2 = scipy.signal.resample(orig2, 24000)
audio3 = scipy.signal.resample(orig3, 24000)
audio4 = scipy.signal.resample(orig4, 24000)

# Quantize to 16 bit precision (but no clipping)
for i in range(0, 24000):
    audio1[i] = round(32768.0 * audio1[i]) / 32768.0
    audio2[i] = round(32768.0 * audio2[i]) / 32768.0
    audio3[i] = round(32768.0 * audio3[i]) / 32768.0
    audio4[i] = round(32768.0 * audio4[i]) / 32768.0
audio1 = scipy.array(audio1, dtype=scipy.float64)
audio2 = scipy.array(audio2, dtype=scipy.float64)
audio3 = scipy.array(audio3, dtype=scipy.float64)
audio4 = scipy.array(audio4, dtype=scipy.float64)

# Resample back to original rate
result1 = scipy.signal.resample(audio1, 96000)
result2 = scipy.signal.resample(audio2, 96000)
result3 = scipy.signal.resample(audio3, 96000)
result4 = scipy.signal.resample(audio4, 96000)

for i in range(0, 96000):
    stdout.write("%5d   %9.6f %9.6f   %9.6f %9.6f   %9.6f %9.6f   %9.6f %9.6f\n" % (i, orig1[i], result1[i], orig2[i], result2[i], orig3[i], result3[i], orig4[i], result4[i]))

Note that scipy.signal.resample() uses the Fourier method, i.e. it is the mathematically correct resampling method.  Since it assumes the signal is periodic, we pad it with large number of zeroes before and after.
Each output line contains the sample index, and original and final sample amplitudes (-1 to +1) for each of the four waveforms.

Observe the first wavelength prior to the pulse, and two full 3675 Hz wavelengths at the beginning of the pulse (at sample 36001/36002/36003/36004), i.e. samples 36050 - 36099:
Code: [Select]
              Original              Delay +1             Delay +2             Delay +3
Index     Initial     Final     Initial     Final     Initial     Final     Initial     Final
35950    0.000000 -0.001446    0.000000 -0.001074    0.000000 -0.000089    0.000000  0.000918
35951    0.000000 -0.000985    0.000000 -0.001470    0.000000 -0.001077    0.000000 -0.000077
35952    0.000000  0.000092    0.000000 -0.001007    0.000000 -0.001465    0.000000 -0.001068
35953    0.000000  0.001165    0.000000  0.000072    0.000000 -0.000994    0.000000 -0.001463
35954    0.000000  0.001591    0.000000  0.001147    0.000000  0.000090    0.000000 -0.001000
35955    0.000000  0.001085    0.000000  0.001573    0.000000  0.001165    0.000000  0.000081
35956    0.000000 -0.000092    0.000000  0.001068    0.000000  0.001587    0.000000  0.001160
35957    0.000000 -0.001265    0.000000 -0.000108    0.000000  0.001075    0.000000  0.001593
35958    0.000000 -0.001730    0.000000 -0.001281    0.000000 -0.000106    0.000000  0.001094
35959    0.000000 -0.001173    0.000000 -0.001746    0.000000 -0.001279    0.000000 -0.000078
35960    0.000000  0.000122    0.000000 -0.001190    0.000000 -0.001740    0.000000 -0.001251
35961    0.000000  0.001415    0.000000  0.000102    0.000000 -0.001176    0.000000 -0.001721
35962    0.000000  0.001929    0.000000  0.001391    0.000000  0.000123    0.000000 -0.001171
35963    0.000000  0.001313    0.000000  0.001900    0.000000  0.001415    0.000000  0.000116
35964    0.000000 -0.000122    0.000000  0.001282    0.000000  0.001923    0.000000  0.001404
35965    0.000000 -0.001556    0.000000 -0.000151    0.000000  0.001300    0.000000  0.001917
35966    0.000000 -0.002122    0.000000 -0.001579    0.000000 -0.000140    0.000000  0.001304
35967    0.000000 -0.001428    0.000000 -0.002137    0.000000 -0.001574    0.000000 -0.000126
35968    0.000000  0.000183    0.000000 -0.001434    0.000000 -0.002136    0.000000 -0.001556
35969    0.000000  0.001793    0.000000  0.000182    0.000000 -0.001437    0.000000 -0.002124
35970    0.000000  0.002424    0.000000  0.001792    0.000000  0.000177    0.000000 -0.001436
35971    0.000000  0.001627    0.000000  0.002419    0.000000  0.001785    0.000000  0.000167
35972    0.000000 -0.000214    0.000000  0.001617    0.000000  0.002411    0.000000  0.001770
35973    0.000000 -0.002052    0.000000 -0.000225    0.000000  0.001608    0.000000  0.002400
35974    0.000000 -0.002767    0.000000 -0.002061    0.000000 -0.000235    0.000000  0.001610
35975    0.000000 -0.001835    0.000000 -0.002770    0.000000 -0.002070    0.000000 -0.000219
35976    0.000000  0.000305    0.000000 -0.001831    0.000000 -0.002777    0.000000 -0.002045
35977    0.000000  0.002442    0.000000  0.000315    0.000000 -0.001836    0.000000 -0.002752
35978    0.000000  0.003260    0.000000  0.002453    0.000000  0.000311    0.000000 -0.001819
35979    0.000000  0.002134    0.000000  0.003269    0.000000  0.002450    0.000000  0.000315
35980    0.000000 -0.000427    0.000000  0.002136    0.000000  0.003265    0.000000  0.002441
35981    0.000000 -0.002981    0.000000 -0.000432    0.000000  0.002135    0.000000  0.003251
35982    0.000000 -0.003938    0.000000 -0.002992    0.000000 -0.000430    0.000000  0.002120
35983    0.000000 -0.002519    0.000000 -0.003952    0.000000 -0.002984    0.000000 -0.000441
35984    0.000000  0.000671    0.000000 -0.002533    0.000000 -0.003937    0.000000 -0.002991
35985    0.000000  0.003848    0.000000  0.000658    0.000000 -0.002512    0.000000 -0.003943
35986    0.000000  0.004997    0.000000  0.003835    0.000000  0.000681    0.000000 -0.002521
35987    0.000000  0.003091    0.000000  0.004986    0.000000  0.003858    0.000000  0.000669
35988    0.000000 -0.001129    0.000000  0.003082    0.000000  0.005005    0.000000  0.003845
35989    0.000000 -0.005328    0.000000 -0.001133    0.000000  0.003096    0.000000  0.004996
35990    0.000000 -0.006761    0.000000 -0.005326    0.000000 -0.001123    0.000000  0.003095
35991    0.000000 -0.003919    0.000000 -0.006752    0.000000 -0.005317    0.000000 -0.001118
35992    0.000000  0.002258    0.000000 -0.003906    0.000000 -0.006744    0.000000 -0.005310
35993    0.000000  0.008398    0.000000  0.002270    0.000000 -0.003898    0.000000 -0.006741
35994    0.000000  0.010195    0.000000  0.008405    0.000000  0.002277    0.000000 -0.003903
35995    0.000000  0.004890    0.000000  0.010194    0.000000  0.008408    0.000000  0.002265
35996    0.000000 -0.006378    0.000000  0.004883    0.000000  0.010193    0.000000  0.008392
35997    0.000000 -0.017539    0.000000 -0.006386    0.000000  0.004876    0.000000  0.010180
35998    0.000000 -0.018608    0.000000 -0.017542    0.000000 -0.006398    0.000000  0.004871
35999    0.000000  0.001413    0.000000 -0.018599    0.000000 -0.017557    0.000000 -0.006394
36000    0.000000  0.050781    0.000000  0.001434    0.000000 -0.018616    0.000000 -0.017548
36001    0.130504  0.131896    0.000000  0.050812    0.000000  0.001418    0.000000 -0.018608
36002    0.258733  0.240106    0.130504  0.131929    0.000000  0.050794    0.000000  0.001419
36003    0.382492  0.364944    0.258733  0.240133    0.130504  0.131910    0.000000  0.050787
36004    0.499667  0.493286    0.382492  0.364960    0.258733  0.240112    0.130504  0.131897
36005    0.608254  0.613143    0.499667  0.493290    0.382492  0.364939    0.258733  0.240100
36006    0.706400  0.716591    0.608254  0.613137    0.499667  0.493273    0.382492  0.364931
36007    0.792428  0.800821    0.706400  0.716581    0.608254  0.613128    0.499667  0.493271
36008    0.864871  0.867126    0.792428  0.800812    0.706400  0.716583    0.608254  0.613129
36009    0.922494  0.918578    0.864871  0.867121    0.792428  0.800824    0.706400  0.716581
36010    0.964317  0.957566    0.922494  0.918576    0.864871  0.867140    0.792428  0.800815
36011    0.989628  0.984315    0.964317  0.957565    0.922494  0.918597    0.864871  0.867125
36012    0.998001  0.996887    0.989628  0.984314    0.964317  0.957581    0.922494  0.918579
36013    0.989298  0.992400    0.998001  0.996885    0.989628  0.984320    0.964317  0.957568
36014    0.963673  0.968675    0.989298  0.992397    0.998001  0.996881    0.989628  0.984318
36015    0.921571  0.925416    0.963673  0.968672    0.989298  0.992386    0.998001  0.996890
36016    0.863718  0.864380    0.921571  0.925415    0.963673  0.968658    0.989298  0.992401
36017    0.791107  0.788576    0.863718  0.864382    0.921571  0.925404    0.963673  0.968674
36018    0.704987  0.701035    0.791107  0.788582    0.863718  0.864377    0.921571  0.925413
36019    0.606835  0.603840    0.704987  0.701044    0.791107  0.788583    0.863718  0.864376
36020    0.498335  0.497894    0.606835  0.603851    0.704987  0.701050    0.791107  0.788574
36021    0.381345  0.383468    0.498335  0.497908    0.606835  0.603858    0.704987  0.701038
36022    0.257871  0.261123    0.381345  0.383484    0.498335  0.497914    0.606835  0.603849
36023    0.130026  0.132465    0.257871  0.261138    0.381345  0.383487    0.498335  0.497909
36024    0.000000  0.000305    0.130026  0.132477    0.257871  0.261139    0.381345  0.383484
36025   -0.129983 -0.131816    0.000000  0.000312    0.130026  0.132477    0.257871  0.261134
36026   -0.257699 -0.260466   -0.129983 -0.131815    0.000000  0.000313    0.130026  0.132465
36027   -0.380963 -0.383020   -0.257699 -0.260470   -0.129983 -0.131813    0.000000  0.000294
36028   -0.497669 -0.497894   -0.380963 -0.383026   -0.257699 -0.260468   -0.129983 -0.131836
36029   -0.605823 -0.604212   -0.497669 -0.497899   -0.380963 -0.383026   -0.257699 -0.260488
36030   -0.703576 -0.701168   -0.605823 -0.604212   -0.497669 -0.497901   -0.380963 -0.383038
36031   -0.789260 -0.787480   -0.703576 -0.701165   -0.605823 -0.604218   -0.497669 -0.497906
36032   -0.861413 -0.861237   -0.789260 -0.787476   -0.703576 -0.701172   -0.605823 -0.604218
36033   -0.918805 -0.920235   -0.861413 -0.861236   -0.789260 -0.787481   -0.703576 -0.701173
36034   -0.960460 -0.962583   -0.918805 -0.920241   -0.861413 -0.861236   -0.789260 -0.787490
36035   -0.985670 -0.987229   -0.960460 -0.962595   -0.918805 -0.920235   -0.861413 -0.861254
36036   -0.994009 -0.994141   -0.985670 -0.987244   -0.960460 -0.962585   -0.918805 -0.920258
36037   -0.985340 -0.984042   -0.994009 -0.994152   -0.985670 -0.987235   -0.960460 -0.962604
36038   -0.959818 -0.957904   -0.985340 -0.984047   -0.994009 -0.994149   -0.985670 -0.987244
36039   -0.917884 -0.916478   -0.959818 -0.957903   -0.985340 -0.984053   -0.994009 -0.994145
36040   -0.860262 -0.860138   -0.917884 -0.916473   -0.959818 -0.957916   -0.985340 -0.984039
36041   -0.787941 -0.789103   -0.860262 -0.860137   -0.917884 -0.916491   -0.959818 -0.957900
36042   -0.702166 -0.703879   -0.787941 -0.789111   -0.860262 -0.860152   -0.917884 -0.916481
36043   -0.604406 -0.605658   -0.702166 -0.703897   -0.787941 -0.789116   -0.860262 -0.860151
36044   -0.496340 -0.496429   -0.604406 -0.605682   -0.702166 -0.703888   -0.787941 -0.789124
36045   -0.379819 -0.378746   -0.496340 -0.496452   -0.604406 -0.605660   -0.702166 -0.703899
36046   -0.256839 -0.255274   -0.379819 -0.378759   -0.496340 -0.496422   -0.604406 -0.605667
36047   -0.129506 -0.128374   -0.256839 -0.255271   -0.379819 -0.378729   -0.496340 -0.496422
36048   -0.000000  0.000061   -0.129506 -0.128357   -0.256839 -0.255249   -0.379819 -0.378723
36049    0.129462  0.128455   -0.000000  0.000085   -0.129506 -0.128347   -0.256839 -0.255243
36050    0.256667  0.255207    0.129462  0.128478   -0.000000  0.000084   -0.129506 -0.128347
36051    0.379438  0.378378    0.256667  0.255224    0.129462  0.128472   -0.000000  0.000073
36052    0.495676  0.495605    0.379438  0.378387    0.256667  0.255219    0.129462  0.128448
36053    0.603396  0.604311    0.495676  0.495612    0.379438  0.378389    0.256667  0.255188
36054    0.700757  0.702087    0.603396  0.604321    0.495676  0.495621    0.379438  0.378355
36055    0.786098  0.787055    0.700757  0.702105    0.603396  0.604334    0.495676  0.495590
36056    0.857961  0.858002    0.786098  0.787079    0.700757  0.702118    0.603396  0.604309
36057    0.915124  0.914252    0.857961  0.858025    0.786098  0.787086    0.700757  0.702097
36058    0.956611  0.955360    0.915124  0.914266    0.857961  0.858025    0.786098  0.787068
36059    0.981720  0.980823    0.956611  0.955358    0.915124  0.914261    0.857961  0.858008
36060    0.990025  0.989990    0.981720  0.980804    0.956611  0.955353    0.915124  0.914246
36061    0.981391  0.982211    0.990025  0.989961    0.981720  0.980806    0.956611  0.955342
36062    0.955970  0.957145    0.981391  0.982181    0.990025  0.989973    0.981720  0.980801
36063    0.914204  0.915044    0.955970  0.957125    0.981391  0.982204    0.990025  0.989974
36064    0.856812  0.856842    0.914204  0.915039    0.955970  0.957153    0.981391  0.982208
36065    0.784782  0.784007    0.856812  0.856850    0.914204  0.915065    0.955970  0.957158
36066    0.699350  0.698242    0.784782  0.784022    0.856812  0.856866    0.914204  0.915067
36067    0.601983  0.601191    0.699350  0.698255    0.784782  0.784024    0.856812  0.856867
36068    0.494349  0.494324    0.601983  0.601196    0.699350  0.698242    0.784782  0.784027
36069    0.378295  0.379030    0.494349  0.494319    0.601983  0.601174    0.699350  0.698253
36070    0.255808  0.256860    0.378295  0.379019    0.494349  0.494294    0.601983  0.601194
36071    0.128986  0.129740    0.255808  0.256848    0.378295  0.378998    0.494349  0.494322
36072    0.000000  0.000031    0.128986  0.129730    0.255808  0.256836    0.378295  0.379028
36073   -0.128943 -0.129632    0.000000  0.000024    0.128986  0.129729    0.255808  0.256859
36074   -0.255637 -0.256627   -0.128943 -0.129637    0.000000  0.000032    0.128986  0.129739
36075   -0.377915 -0.378625   -0.255637 -0.256632   -0.128943 -0.129625    0.000000  0.000027
36076   -0.493687 -0.493713   -0.377915 -0.378632   -0.255637 -0.256622   -0.128943 -0.129639
36077   -0.600974 -0.600319   -0.493687 -0.493719   -0.377915 -0.378628   -0.255637 -0.256635
36078   -0.697944 -0.697004   -0.600974 -0.600321   -0.493687 -0.493724   -0.377915 -0.378630
36079   -0.782942 -0.782268   -0.697944 -0.696999   -0.600974 -0.600336   -0.493687 -0.493713
36080   -0.854517 -0.854492   -0.782942 -0.782257   -0.697944 -0.697021   -0.600974 -0.600311
36081   -0.911449 -0.912073   -0.854517 -0.854479   -0.782942 -0.782282   -0.697944 -0.696992
36082   -0.952770 -0.953668   -0.911449 -0.912063   -0.854517 -0.854503   -0.782942 -0.782257
36083   -0.977777 -0.978426   -0.952770 -0.953662   -0.911449 -0.912081   -0.854517 -0.854488
36084   -0.986049 -0.986084   -0.977777 -0.978424   -0.952770 -0.953674   -0.911449 -0.912079
36085   -0.977449 -0.976871   -0.986049 -0.986081   -0.977777 -0.978432   -0.952770 -0.953680
36086   -0.952131 -0.951297   -0.977449 -0.976862   -0.986049 -0.986089   -0.977777 -0.978439
36087   -0.910532 -0.909939   -0.952131 -0.951277   -0.977449 -0.976873   -0.986049 -0.986091
36088   -0.853370 -0.853363   -0.910532 -0.909912   -0.952131 -0.951294   -0.977449 -0.976868
36089   -0.781629 -0.782204   -0.853370 -0.853337   -0.910532 -0.909932   -0.952131 -0.951283
36090   -0.696540 -0.697354   -0.781629 -0.782188   -0.853370 -0.853356   -0.910532 -0.909921
36091   -0.599564 -0.600140   -0.696540 -0.697355   -0.781629 -0.782200   -0.853370 -0.853348
36092   -0.492363 -0.492371   -0.599564 -0.600159   -0.696540 -0.697357   -0.781629 -0.782196
36093   -0.376775 -0.376217   -0.492363 -0.492399   -0.599564 -0.600150   -0.696540 -0.697354
36094   -0.254780 -0.253986   -0.376775 -0.376245   -0.492363 -0.492386   -0.599564 -0.600146
36095   -0.128468 -0.127896   -0.254780 -0.254004   -0.376775 -0.376233   -0.492363 -0.492377
36096   -0.000000  0.000031   -0.128468 -0.127899   -0.254780 -0.253998   -0.376775 -0.376221
36097    0.128425  0.127915   -0.000000  0.000037   -0.128468 -0.127901   -0.254780 -0.253988
36098    0.254609  0.253873    0.128425  0.127922   -0.000000  0.000030   -0.128468 -0.127899
36099    0.376395  0.375873    0.254609  0.253871    0.128425  0.127917   -0.000000  0.000022

The 0° and 180° phases are correctly positioned, but the very first half wave – that which triggers the stereocilia, and humans being able to discriminate down to 10µs differences in the triggering time of same frequency bands between ears – is deformed.  There is also quite a lot of quantization noise, and a spurious pre-pulse wave with significant amplitude (0.005, with quantization units of at 0.000030 or so).  Essentially, the pulse leading edge is "smeared".

If you examine the full output of the above Python script, you'll see that after the first wave or so, the output wave tracks very closely to the initial signal.  The error or difference between initial and final signals are concentrated around the beginning of the wave packet, and this is what affects the signal arrival time difference discrimination in human hearing.
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Maximum slew rate typically found in music/voice
« Reply #43 on: September 21, 2023, 01:14:15 am »
The theory says that any band limited signal can be reconstructed perfectly, and that include its phase; any reasonably implemented sampling system (CD) is able to reproduce the phase shift with inaudible accuracy.
1. Shannon-Nyquist only applies exactly if you have infinite precision, i.e. exact discrete samples.  Quantizing the samples limits the accuracy.   Discretized quantized signal is limited in both bandwidth and phase resolution, and is limited in the resolution at which it can represent the wave packet arrival time.

Sure, but suppose we have a quadrature sampled signal, i.e. 4 samples per cycle, so that Nyquist is happy enough.  If the signal is a sine wave, then the samples are A sin(wt + ph), A sin(wt + ph + pi/2), etc.  Depending on ph, either one pair of samples, or the other, is close to zero (ph ~ pi n / 2), and thus d(sample)/dph is large (~A) and the others are near +/-A and d(sample)/dph ~ 0; or both are similar (ph ~ pi (2n + 1) / 4) and all four slopes are comparable (d(sample)/dph ~ +/- A sqrt(2) / 2 in each slot).  In either case, we have a total of two samples' worth at full gain, and thus somewhere on the order of 0.5-1 LSB of phase resolution, depending on how you want to count it (assuming A is ~full scale).

I'm not sure what the complaint is here--?

There might be complications due to mixing of frequencies, and practical signal chains, but those are issues of any practical signal chain, and if we had ideal brick wall filters, the above analysis can be approached arbitrarily (indeed it doesn't need to be quadrature, it could be, er, trisature so to speak, or anything strictly less than Nyquist condition if we don't mind that quantization noise becomes a larger fraction of phase noise for some samples; but that too doesn't really matter since we'd then be considering arbitrarily long windows of samples, as is necessary to resolve a signal arbitrarily close to Fs/2, and then phase is still perfectly recoverable in the presence of quantization noise; the N=4 case above is simply where it's evident within a ~minimum of samples).  So in a real system with modest oversampling and whatever quantization level dictating overall bandwidth, things come through just fine.

Mind, I'm not interested in getting into psychoacoustics here, and probably neither was the OP.  At the very least, such a topic shouldn't be taken as a given, and must be explained and justified.

For what little I know of it, it's my understanding the ear is particularly well tuned to correlations of specific kinds, like impulsive sounds (edges, short wavelets of many frequencies), and to -- given some assumptions about the expected spectrum of familiar sounds -- the comb-filtering effect of delays between ears, diffraction around the head, and reflection/diffraction within the (external) ear shape itself; and by extension, the effect of ambient (room or object) reflections.  Which, I believe, the ears give a little front-back discrimination for applicable sounds, obviously mostly left-right, and very little up-down; which makes sense for a species expected to have evolved on a grassland to forest edge environment.

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline gf

  • Super Contributor
  • ***
  • Posts: 1425
  • Country: de
Re: Maximum slew rate typically found in music/voice
« Reply #44 on: September 21, 2023, 08:05:31 am »
You can easily simulate this, if you actually wanted to.  Just generate a suitably shaped pulse as I described earlier at a higher sample rate.  Generate several variants, delaying each by a one additional leading zero sample, then pad with lots of zeros to a suitable size.  Decimate to CD audio sample rate and quantize to 16 bit samples, then resample back to original sample rate, and compare the initial and final waveforms.

I think the main problem is that the the original signal violates the sampling theorem.
It does contain frequency contents >= 12kHz.
Then the reconstructed signal is not supposed to be equal to the original one.
However, when I compare the tabulated Delay+1 and Delay+2 values (with one sample offset), the differences seem to be in the order of magnitude expected with 16-bit quantization.

Quote
If you examine the full output of the above Python script, you'll see that after the first wave or so, the output wave tracks very closely to the initial signal.

I think this is mainly the difference between the original signal and the band-limited original signal (filtered with a zero-phase brickwall filter). If the original signal was already band-limited to < 12kHz per se, then I think you would not see such a big difference.
 

Offline Nominal Animal

  • Super Contributor
  • ***
  • Posts: 7198
  • Country: fi
    • My home page and email address
Re: Maximum slew rate typically found in music/voice
« Reply #45 on: September 21, 2023, 09:47:28 pm »
I think the main problem is that the the original signal violates the sampling theorem.
I don't see any way of low-pass filtering the original signal that would not "smear" the attack phase of the wave packet.

Another way to put it, I guess, is to say that while humans cannot typically sense sounds above 20 kHz or so, our time discrimination (arrival time difference between each ear) is such that components above 20 kHz are involved, with the time discrimination ability measured to be on the order of 10 µs.

The reason I keep talking about wave packets is because it much better matches the physical operation of the stereocilia and hair cells.  It is the time difference between activation of the corresponding hair cells in the ears that has the 10 µs time discrimination ability.

Now, how would you model that as a simple frequency response?  I don't know exactly how.  It is much easier to do using wave packet model, where if you do an infinite-width Fourier analysis of the waveform, you indeed do have frequencies above the range at which humans can hear a sinusoidal continuous signal.  It should not be a surprise that a pure frequency domain representation cannot capture a mixed frequency-domain (stereocilia and hair cells, that respond to a narrow band of frequencies) and time domain (band activation time difference in the auditory brainstem) sensors.



The uncontested facts are that while a typical human ear can sense acoustic waves within 20 Hz to 20 kHz or so, and differences as small as 10 µs between initial arrival time in each ear.  The peak sensitivity is a bit below 4000 Hz, which means that at that frequency, the time discrimination is on the order of one 25th of the wavelength.  Another sensitivity peak is around 1000 Hz, where the time discrimination is on the order of one hundredth of the wavelength.

The question seems to be whether CD audio can capture that or not.

My understanding is that in optimal conditions (i.e., volume set depending on background noise levels, background noise having no sharp spectral peaks) mono CD audio does convey basically all perceivable information to a typical human (if we exclude dynamic range changes, like an orchestra playing some parts very loud, and some parts very quietly; human hearing, effectively, has an automatic gain control relative to the surrounding noise floor, if you will).  For stereo, it does not convey all the information needed for the human auditory senses to perceive the exact direction of specific types of sounds, "smearing" their directionality.  The reason for this belief is the physical structure of the audio sensor, having stereocilia waving in a fluid that in groups trigger nerve cells, with groupwise arrival time differences from each ear processed in the auditory brainstem (via the cochlear nucleus).  It is neither purely time-domain, nor purely frequency-domain apparatus, but a mix.

I've probably messed up my explanations and my math above –– I do make errors often ––, but the above paragraph is basically what I've been trying to convey.

The end result, in my opinion, is that 24-bit quantization allows a larger dynamic range, and 192 kHz sample rate better captures the stereophonic information human hearing can detect, and is not just audiophoolery or done because it makes filtering and quantization noise shaping easier.
« Last Edit: September 21, 2023, 09:52:22 pm by Nominal Animal »
 
The following users thanked this post: CatalinaWOW

Offline CatalinaWOW

  • Super Contributor
  • ***
  • Posts: 5571
  • Country: us
Re: Maximum slew rate typically found in music/voice
« Reply #46 on: September 21, 2023, 11:52:11 pm »
Or saying the same thing as Nominal (I think), in somewhat different words.  Our brain is not just a simple RMS meter connected to a microphone and front end amplifier.  There are at least two types of sensors and literally hundreds or thousands of each type.  Non-linear responses are clearly evident and there is evidence of temporal variation on time scales ranging from under a second to decades.  Our experience of sound is the brain combining and interpreting all of those different inputs.

Reconstructing the pressure variations in the ear canal produced by the original sound source is the goal of all audio systems, but a simple measure of performance measured in the 20-20k band at some amplitude is a necessary but not necessarily sufficient criteria. 

Unfortunately at our current level of understanding we can't really define a better measure, and so how it sounds is all we can do.  And that is subjective and also subject to social pressures.  For most of us CD quality sound is more than good enough, and for a somewhat smaller subset of us streaming quality digital audio is just fine.  But I believe there are some who really do hear defects in CD quality sound, and many more who have convinced themselves they can for a variety of reasons.
 
The following users thanked this post: Someone, Nominal Animal

Online NiHaoMike

  • Super Contributor
  • ***
  • Posts: 9323
  • Country: us
  • "Don't turn it on - Take it apart!"
    • Facebook Page
Re: Maximum slew rate typically found in music/voice
« Reply #47 on: September 22, 2023, 04:42:24 am »
Those who use equalization/ "room correction" will want more than 16 bits playback capability, since by the very nature it operates, it will reduce some frequencies so that they'll end up with less than 16 bits of resolution if you start with just 16 bits. Starting with 24 bits (or even 20 "real bits"), there will be room for some pretty aggressive adjustments and still have at least 16 bits of resolution left over for every frequency in the range.
Cryptocurrency has taught me to love math and at the same time be baffled by it.

Cryptocurrency lesson 0: Altcoins and Bitcoin are not the same thing.
 

Offline metebalci

  • Frequent Contributor
  • **
  • Posts: 460
  • Country: ch
Re: Maximum slew rate typically found in music/voice
« Reply #48 on: September 22, 2023, 06:11:09 am »
When reading the posts, I was thinking there has to be some amount of studies done in this topic (I have read some amount of psycho/technical acoustics and auditory neuroscience papers many years ago). Here is the abstract of one I quickly found from 2007:

"Misalignment in timing between drivers in a speaker system and temporal smearing of signals in components and cables have long been alleged to cause degradation of fidelity in audio reproduction. It has also been noted that listeners prefer higher sampling rates (e.g., 96 kHz) than the 44.1 kHz of the digital compact disk, even though the 22 kHz Nyquist frequency of the latter already exceeds the nominal single-tone high-frequency hearing limit fmax∼18 kHz. These qualitative and anecdotal observations point to the possibility that human hearing may be sensitive to temporal errors, τ, that are shorter than the reciprocal of the limiting angular frequency [2πfmax] ~ 9us, thus necessitating bandwidths in audio equipment that are much higher than fmax in order to preserve fidelity. The blind trials of the present work provide quantitative proof of this by assessing the discernability of time misalignment between signals from spatially displaced speakers. The experiment found a displacement threshold of d≈2 mm corresponding to a delay discrimination of τ≈6 μs."

http://boson.physics.sc.edu/~kunchur/papers/Audibility-of-time-misalignment-of-acoustic-signals---Kunchur.pdf

So it looks like 20 kHz is obviously not enough to replicate natural listening experience.

I dont think enough is known about how this works (in terms of neural coding, early processing in the brain etc. at least it was the case ~10y ago) so the only way is doing experiments and this gives a starting point. It is not much ungrounded to move to 96 or 192 kHz.
 

Offline tggzzz

  • Super Contributor
  • ***
  • Posts: 21227
  • Country: gb
  • Numbers, not adjectives
    • Having fun doing more, with less
Re: Maximum slew rate typically found in music/voice
« Reply #49 on: September 22, 2023, 07:16:22 am »
The concept of "degregation of fidelity" misses the elephant in the room.

Consider recording an orchestra in a concert hall. Where exactly should the microphones be placed? Conductor's podium? First row of the audience? Centre, right left? Centre of the audience? With or without a complete audience?

The concept of a single place for the microphones is, of course, too simplistic. But it does highlight the point that there cannot be a single "correct" sound - it is all a choice made by the recording engineer.

Back when I had ears and CDs were new, I and a friend did A-B comparisons between a CD and vinyl. His setup was good, but not audiophool. We could tell a difference, but we could not tell which was which nor which was better.
There are lies, damned lies, statistics - and ADC/DAC specs.
Glider pilot's aphorism: "there is no substitute for span". Retort: "There is a substitute: skill+imagination. But you can buy span".
Having fun doing more, with less
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf