Sampling rate vs output question

[davewantsmoore]

8 hours ago, Grant Slack said:

isn't the full-range, single-driver loudspeaker an entire thing in the world of hifi? And therefore worth discussing in terms other than, "I can't think of any actual ones"? 

8 hours ago, Grant Slack said:

A quick browse on the internet will reveal countless examples of this type of loudspeaker.

Ones which outputs both 40hz and 3khz at an equal SPL ??!?      (not including any advantage via box loading, eg, vented or 'BLH')

 

In short, yes you are correct (single driver speakers are a thing) ... it's just none closely match the parameters of the simulation.... and as soon as you realistically model it ... the distortion products drop to a point where they are much more likely masked.

 

[davewantsmoore]

4 hours ago, March Audio said:

A slight aside on the main topic but single driver speakers have a fundamental flaw.  The size of the driver is inappropriate for much of the frequency range and they will inevitably beam higher frequencies creating a poor match with on and off axis sound reaching the listener.  Dispersion is related to source size and sound wavelength.  Ie there is a very good reason tweeters are smaller

1000%

 

... but, even if we chose to disregard this and use a single driver....   Then ... it requires an extreme driver to make sufficient SPL at 40Hz  (from only piston action) .... ie very large (increases the directivity problem you rightly pointed out) ..... or very high excursion (increases the phase modulation distortion problem) .....  It's just generally unworkable.... so not really worthy of serious consideration.

[davewantsmoore]

3 hours ago, MLXXX said:

The fact that Gibbs was able to provide a mathematical explanation as to how a  squarewave will appear after being bandlimited, does not take away from the fact that after the bandlimiting the waveform shape is no longer square.

You are correct

3 hours ago, MLXXX said:

Necessarily the waveshape is not preserved if the original waveform included out of band components.

.... but how is that relevant?      (Hint:  It isn't.  You can't hear it)

[davewantsmoore]

On 10/09/2019 at 11:52 AM, MLXXX said:

The figure of 10μs is sometimes mentioned for human perception of ITD and that would indeed be a very short period (corresponding to the period of a full cycle of a wave with a frequency of 100kHz).

 

Don't mix up the difference between being able to perceive a frequency ... between being able to perceive a time difference.

 

They aren't the same thing

[MLXXX]

1 hour ago, davewantsmoore said:

 

Don't mix up the difference between being able to perceive a frequency ... between being able to perceive a time difference.

 

They aren't the same thing

Who suggested they were the same thing? Certainly I didn't.

 

I don't understand a culture of continually assuming that others misunderstand.

[March Audio]

 because your comments seem to be conflating several different things

Edited September 11, 2019 by March Audio

[MLXXX]

2 minutes ago, March Audio said:

 because your comments seem to be conflating several different things

What comment of mine appears to you to conflate several different things? 

 

The fact that I mentioned that a 10 microsecond time disparity corresponds to the period of a full cycle of a 100kHz waveform was not a conflation. It was intended as a helpful yardstick for appreciating how short a period 10 microseconds is. 

 

There is no suggestion (at least not from me) that the ability to perceive a particular high audio frequency could be equated with the ability to hear an interaural timing difference of the period corresponding to that high frequency.

[March Audio]

3 hours ago, MLXXX said:

What comment of mine appears to you to conflate several different things? 

 

The fact that I mentioned that a 10 microsecond time disparity corresponds to the period of a full cycle of a 100kHz waveform was not a conflation. It was intended as a helpful yardstick for appreciating how short a period 10 microseconds is. 

 

There is no suggestion (at least not from me) that the ability to perceive a particular high audio frequency could be equated with the ability to hear an interaural timing difference of the period corresponding to that high frequency.

It may well be a misunderstanding but I'm clearly not the only reader who has drawn the conclusion.  You have been talking across waveform timing, shape and frequency. 

 

You have talked about a time domain issue in terms frequency. 

 

So what is your wider point? We know 10uS is a short duration. 

Edited September 11, 2019 by March Audio

[davewantsmoore]

4 hours ago, MLXXX said:

Who suggested they were the same thing? Certainly I didn't.

LOL

 

57 minutes ago, March Audio said:

So what is your wider point? We know 10uS is a short duration. 

About 57 pages back it was that if ITD was circa 10us ..... then we surely need to consider 'doppler distortion'.

 

... and now that 44.1khz could never be enough to capture these differences, because the waveform changes shape when it is captured  (ie. it doesn't contain all the frequencies from the original signal, due to the need for it to be bandlimited to <half the sampling rate).

 

I would given up 83 pages ago, but I always worry that someone will find it in the future  (perhaps I shouldn't).    I wasted too much time around the turn of the millenium mining the internet for what turned out to be very poor info/advice.

[Steffen]

11 hours ago, MLXXX said:

much of the mathematical analysis of digital sampling is based on the assumption that the waveforms continue indefinitely. Provided there are many cycles of a waveform, that assumption can be reasonable and practical to adopt.

 

On the question of phase accuracy, this becomes a messy business with high audio frequency waveforms that last only one or two cycles.

 

Any signal can be expressed as superposition of harmonic (sine wave) oscillations. It doesn't have to be a periodic waveform. With periodic waveforms, the harmonics required to create the signal are integer multiples of the fundamental. With non-periodic signals, an infinite number of frequencies in between those integer multiples may be required. The Dirac impulse, for example, contains all frequencies from 0 to infinity. In other words, your assumption above is always true, regardless of the signal.

 

To see how well a non-periodic signal – say, a single triangular impulse – can be sampled you just have to bandwidth-limit the signal and see what the result looks like. A digital sampling that samples fast enough for the chosen bandwidth limit will be able to perfectly capture this result.

Edited September 11, 2019 by Steffen

[MLXXX]

3 hours ago, Steffen said:

To see how well a non-periodic signal – say, a single triangular impulse – can be sampled you just have to bandwidth-limit the signal and see what the result looks like. A digital sampling that samples fast enough for the chosen bandwidth limit will be able to perfectly capture this result.

If you bandwidth limit a continuous triangular wave you get something rounded, that may resemble a continuous sinewave.

 

If you bandwidth limit a single cycle of a triangular wave you may end up with what looks likes a sine wave plus a damped oscillation.

 

__________ 

 

In the screen capture below of the Audacity audio editing software GUI, the top waveform is one cycle of a 15360 Hz sine wave generated at a sample rate of 384kHz, with the left channel delayed by 3 samples relative to the right channel.

 

The bottom waveform is the conversion to 44.1kHz, which would invoke bandwidth limiting to avoid aliasing.

 

Non-zero values for the left and right channels in the converted waveform start before non-zero values for the original waveform channels start, and end after non-zero values for the original waveform channels end.  The reconstructed waveform (not shown here, merely the calculated resampled values) will differ in shape from the original unprocessed, non-bandwidth limited waveform.

 

 

 

[MLXXX]

Here's an interesting article that reports on interaural time delays of  around 10μs being detected by test subjects wearing headphones: Human interaural time difference thresholds for sine tones: The high-frequency limit

 

Interestingly the waveforms were presented with quite slow rise times (100mS).

 

The audio frequencies at which this extraordinary acuity was found to exist were below 1500Hz. 

[MLXXX]

6 hours ago, March Audio said:

So what is your wider point? We know 10uS is a short duration. 

You yourself may have a precise appreciation of how long 10uS is. Others might not be able to distinguish it readily from 10mS, To those people, both might be just considered simply as "short periods of time".

 

My wider point is that if some human beings can detect an interaural time difference (ITD) as low as 10uS in some circumstances then perhaps the conventional wisdom that 44.1kHz is a sufficient sample rate could be doubted, given that a very rapid onset (commensurate with 10uS) may be impossible to generate or replicate crisply and cleanly at that sample rate.

 

I see that in a test where an ITD figure of around 10uS was established (see my post immediately above) the onset of the sinewave tone for test subjects was performed as a fade-in over a 100mS period.  And there was  a fade-out, also  over 100mS. The test frequencies were below 2kHz.  These characteristics are not at all demanding to generate digitally and are obviously far removed from being transients. A sample rate of 44.1kHz would have been more than sufficient to generate the test signals.

 

I have not seen any test results involving transients that show a human capacity to detect remarkably small ITDs of around 10uS.  Has anyone else seen such results?

Edited September 11, 2019 by MLXXX

[MLXXX]

On 11/09/2019 at 5:47 AM, Grant Slack said:

isn't the full-range, single-driver loudspeaker an entire thing in the world of hifi?

Yes of course it is.  @davewantsmoore could peruse the following webpage for some current examples, if he has any doubt about the matter: https://www.songsimian.com/best-full-range-speakers-driver-review/

 

_________

 

A technically advanced website that looks at different sources of loudspeaker output nonlinearities is the Klippel website. One of its many technical pages points out the significance (as a function of frequency) of the inductance of the voice coil and of other factors (including the Doppler effect) in causing IMD in the output of the loudspeaker:  https://www.klippel.de/know-how/measurements/nonlinear-distortion/intermodulation-distortion.html 

[March Audio]

17 hours ago, MLXXX said:

What comment of mine appears to you to conflate several different things? 

 

The fact that I mentioned that a 10 microsecond time disparity corresponds to the period of a full cycle of a 100kHz waveform was not a conflation. It was intended as a helpful yardstick for appreciating how short a period 10 microseconds is. 

 

There is no suggestion (at least not 

 

So all that can be said, again, is that timing resolution is *not* related to sample rate. 

Edited September 12, 2019 by March Audio

[March Audio]

9 hours ago, MLXXX said:

If you bandwidth limit a continuous triangular wave you get something rounded, that may resemble a continuous sinewave.

 

If you bandwidth limit a single cycle of a triangular wave you may end up with what looks likes a sine wave plus a damped oscillation.

 

__________ 

 

In the screen capture below of the Audacity audio editing software GUI, the top waveform is one cycle of a 15360 Hz sine wave generated at a sample rate of 384kHz, with the left channel delayed by 3 samples relative to the right channel.

 

The bottom waveform is the conversion to 44.1kHz, which would invoke bandwidth limiting to avoid aliasing.

 

Non-zero values for the left and right channels in the converted waveform start before non-zero values for the original waveform channels start, and end after non-zero values for the original waveform channels end.  The reconstructed waveform (not shown here, merely the calculated resampled values) will differ in shape from the original unprocessed, non-bandwidth limited waveform.

 

 

 

 

 

 

You have missed a fundamental point here.  You are just looking at the sample points.  The waveform that fits those sample points will still have correct timing. Its counter intuitive I know.

 

see the video at  20:50  and before you say it, this has nothing to do with continuous and non continuous wave forms.

 

 

 

 

 

 

 

 

your pre ripples are caused due to the fact that your sine wave was not edited to start at a zero value, hence its actually a high frequency change to the first sample, hence Gibbs effect when you bandwidth limit

 

192kHz  sample rate 15kHz sine

 

 

 

 

 

 

 

Edited September 12, 2019 by March Audio

[davewantsmoore]

21 minutes ago, MLXXX said:

Yes of course it is.  @davewantsmoore could peruse the following webpage for some current examples, if he has any doubt about the matter: https://www.songsimian.com/best-full-range-speakers-driver-review/

You just posted 4 examples which (in not a small way) do not match the simulation paper you posted.

 

ie. they do not make equal SPL (in a sealed box) to 40hz and 3khz.

 

If you EQed them at the low end.... and took the big distortion vs SPL penalty, then they still wouldn't get within an octave (4x the excursion!!!) of 40Hz.

 

It's ok that you don't realise your mistake.... but when you do so in such a cute and smug way, you shouldn't wonder why people are frustrated.   

 

 

 

 

21 minutes ago, MLXXX said:

A technically advanced website that looks at different sources of loudspeaker output nonlinearities is the Klippel website. One of its many technical pages points out the significance (as a function of frequency) of the inductance of the voice coil and of other factors (including the Doppler effect) in causing IMD in the output of the loudspeaker:  https://www.klippel.de/know-how/measurements/nonlinear-distortion/intermodulation-distortion.html 

Indeed.   (and I'm surprised that Klippel call it "doppler effect", but that's beside the point)

 

... but there's two telling things buried in Klippes papers

 

* Phase modulation is less significant(ly audible) than amplitude modulation

* "Doppler distortion" can be avoided by using a sufficiently low low pass filter frequency, vs the low frequency (more accurately, vs the amount of displacement of the cone, by the low frequency).

 

What this means... is that it's just not significant in almost every real world speaker.   ie. one which employs a low pass filter.... and/or one which doesn't undergo large excursion.

 

[MLXXX]

12 minutes ago, March Audio said:

 

So all that can be said, again, is that timing resolution is *not* related to sample rate. 

Yes the above has been mentioned a number of times in this thread. I don't think anyone has disputed it.

 

An issue I have been referring to is the lack of crispness in reproducing a transient. 

 

You have stated: 

9 minutes ago, March Audio said:

because your ears are also band limited.  You wouldn't hear the additional high frequency harmonics

That also is not in dispute; that a bandlimited (upper band limit a little above 20kHz) square wave can sound the same to human ears as a (relative;y) non-bandlimited square wave.

 

The issue I have sought to highlight is that if human hearing can detect an ITD as low as 10uS in some circumstances, that ability might be hampered if the onset of the waveforms presented to the human ears are not rendered crisply and cleanly. This is not a question of listening in mono to a bandlimited squarewave but using both ears to perceive small differences in the timing as between the sound waves presented to one ear and the sound waves presented to the other ear.

 

You could respond by saying that human hearing is "bandlimited" and would necessarily blur rapid onset transients anyway making it unnecessary to reproduce them crisply and cleanly; that it would be sufficient to slow the rise and fall times, and it wouldn't matter if pre-ringing and post-ringing were present. You could well be right. However I haven't seen any research results from an investigation into this.  Could someone point me to such an investigation?

[davewantsmoore]

9 hours ago, MLXXX said:

In the screen capture below of the Audacity

... and you've just highlighted the problem with resampling.

 

Did you intentionally mis-use the sampling theorem to try to make a point .... or do you just not understand it very well ?!   (Don't worry, I know the answer).

 

 

This is why, for example... you will see people making converters with very very long filters.... this is why you will see people trying to "avoid" resampling all together (or control it in a way which tries to avoid the issues, eg. MQA).

 

This is why, 346 pages ago, I said something to the effect of don't confuse what real world devices and resamplers will do.... with theory, and what is right or wrong about what we do not cannot hear.

[March Audio]

36 minutes ago, MLXXX said:

Yes the above has been mentioned a number of times in this thread. I don't think anyone has disputed it.

 

An issue I have been referring to is the lack of crispness in reproducing a transient. 

 

You have stated: 

That also is not in dispute; that a bandlimited (upper band limit a little above 20kHz) square wave can sound the same to human ears as a (relative;y) non-bandlimited square wave.

 

The issue I have sought to highlight is that if human hearing can detect an ITD as low as 10uS in some circumstances, that ability might be hampered if the onset of the waveforms presented to the human ears are not rendered crisply and cleanly. This is not a question of listening in mono to a bandlimited squarewave but using both ears to perceive small differences in the timing as between the sound waves presented to one ear and the sound waves presented to the other ear.

 

You could respond by saying that human hearing is "bandlimited" and would necessarily blur rapid onset transients anyway making it unnecessary to reproduce them crisply and cleanly; that it would be sufficient to slow the rise and fall times, and it wouldn't matter if pre-ringing and post-ringing were present. You could well be right. However I haven't seen any research results from an investigation into this.  Could someone point me to such an investigation?

 

I made a few edits, please re read.

 

The signals *are* rendered correctly and as crisply (sharp) as the bandwidth (and your hearing) allows.

 

Look at it a different way.

 

You can hear an ITD of two sine waves 10 uS apart.  the pre ringing you saw was a function of a rise time greater than 20kHz which you cant hear.  Its not an artifact, its what happens. The audible sine waves are still 10uS apart.

Edited September 12, 2019 by March Audio

[davewantsmoore]

8 hours ago, MLXXX said:

Interestingly the waveforms were presented with quite slow rise times (100mS).

Yes, that is what I've been saying all this thread.

 

Timing differences that you can perceive, don't rely on high frequency content.....  44.1 can keep represent enough timing differences .... BUT, the allure to using higher rates, is that can make it easier for you to not to accidentally damage these timing differences..... eg. by using limited/realworld reasmplers (and may ther things).

 

[MLXXX]

17 minutes ago, davewantsmoore said:

It's ok that you don't realise your mistake.... but when you do so in such a cute and smug way, you shouldn't wonder why people are frustrated.   

You do read a lot into posts! As for "not realising [my] mistake" I am fully aware that you are trying to focus on the exact parameters of the 2017 paper. I think that's quite unnecessary, and not helpful. It strikes me as a bit like saying, "your cited mathematical simulation of the expected earthquake was for a Richter scale 6 event, but real events will probably only be at Richter scale 4". 

 

A more constructive approach, in my view, would be to estimate the IMD of speaker systems including wide-range single driver systems. And if you did that you would find the effects, including the often cited Doppler effect, significant, particularly at high listening levels corresponding to large cone excursions.

Edited September 12, 2019 by MLXXX

[March Audio]

Another thing to point out about the ripples.  Their frequency is close to nyquist.  Thats above audible range even at 44.1.

 

Dirac Pulse (note a dirac pulse is essentially an illegal signal you cant have a 0 - 1 - 0 sample value you have to filter at 1/2 sample rate)

 

 

Actual measured output from dac

 

 

note the ripple timebase.  The frequency of the ripple is above 20kHz, in fact eyeballing it looks like about 45uS or 22kHz (which is eaxactly what you would expect).   So you simple dont hear them.

Edited September 12, 2019 by March Audio

[March Audio]

I think its fair to say that 44.1 sample rate is a bit "close for comfort"  but the reality is that if 48kHz was chosen it would be a pretty much "blameless" system.  Higher sample rates just let in ultrasonic garbage and do nothing to help sound quality.

Edited September 12, 2019 by March Audio

[MLXXX]

14 minutes ago, March Audio said:

I think its fair to say that 44.1 sample rate is a bit "close for comfort"  but the reality is that if 48kHz was chosen it would be a "blameless" system.  ...

I've often expressed that view myself. The exact design of the antialias filtering is much less critical for a 48kHz SR than for 44.1kHz.