So I assume you have some ABX results to share with us?
-
WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!
Hypex nCore vs Class A amps
- Thread starter ppataki
- Start date
So I assume you have some ABX results to share with us?
Nope. I considered such behavior a bug in my DSP code - long since fixed. One man's garbage ...
If you want to do something like this intentionally, you may play with parameters in any number of plugins removing/transforming transients (e.g. https://www.google.com/search?q=click+removal+plugin) on a prog rock or heavy metal record.
Nope. I considered such behavior a bug in my DSP code - long since fixed. One man's garbage ...
If you want to do something like this intentionally, you may play with parameters in any number of plugins removing/transforming transients (e.g. https://www.google.com/search?q=click+removal+plugin) on a prog rock or heavy metal record.
Those tools affect the audible range (on purpose).
This is interesting consideration indeed. Even though, there appears to be some confusion. WXA-50 uses ICEpower 125asx2, right? Which, according to https://icepower.dk/products/amplifier-power-modules/asx-series/, is "based on ICEpower’s HCOM class D topology". Which, according to the HCOM patent (https://patents.justia.com/patent/8013678), does use the feedback loop going from a point after the output filter.
The 125asx2 is supposed to be different from the older ICEpower module version described here: https://www.diyaudio.com/forums/cla...-rev-engin-complete-finished-post1957813.html, which does indeed only use the loop sourcing from a point before the filter. My impression was that when the WXA-50 was released in 2016, the need for the feedback from after the output filter was already well understood. Is that right?
It appears we have different definitions of "audible range". Let me try explaining once again what I mean, from another angle. Apologies if you already know all of it - what follows could be still interesting to others.
Imagine a PCM file at 192KHz with this sequence of four repeating values: [0,+Max,0,-Max, ...]. Let's say this sequence has 2,000 samples. What would you hear when it is played? Obviously, this is a representation of 192,000/4=48,000 KHz sinusoid. Even if the audio chain you use could reproduce it, you wouldn't be able to hear it. Clearly, this signal lasting slightly over 10 milliseconds is out of the human audible range.
Let's start removing samples one by one from the tail of the sequence, and play the signal again. A some point you'd start hearing clicks, and by the time there are only three samples left [0,+Max,0], we'd arrive at the signal that is very clearly in the human audible range, while being derived by a simple procedure from a signal out of the audible range.
Some people argued - and this is partially valid - that what we actually hear are oscillations in the transducer(s), excited by the pulse. To check for that, very accurate and highly damped transducers were introduced instead of regular ones, reproducing the pulse without subsequent ringing. The click was still heard.
Next argument was that the oscillating device we actually hear is the conglomeration of bones in the middle ear. Experiments were conducted (on rodents) when the bones were eliminated and the cochlea was excited with a pulse directly. The measured neuro-physiological effect was still consistent with the animal hearing a click.
The next argument after that was that it is the cochlea itself that vibrates after being excited with a pulse, and that's what we actually hear. This turned out to be true, yet only partially. The cochlea does get excited by the pulse, yet it doesn't settle into a vibration pattern prolonged enough to be detected by the regular frequency-sensitive neural machinery.
What was ultimately found is described in the peer-reviewed papers I referenced in this thread, and in many other papers: there is a different, anatomically distinct neuro-physiological mechanism, specifically tuned to detect very short transients, which are out of audible range of the frequency-sensitive neural machinery.
The evolutionary value of this mechanism is clear: an animal able to detect a very faint and very short transient - for instance generated by a dry leaf broken under a leg of an approaching predator still hidden from view - has better chance to survive, compared to an animal only capable of detecting relatively-long-running harmonic oscillations.
With the transients removed, or morphed into harmonic waveforms and/or noise, music subjectively changes - sometimes for the better actually, but in the context of some genres hugely for worse. For instance, removal of the sounds of nail hitting a close-miked acoustic guitar string is often appreciated. Removal of sharp attacks in rock music usually is not.
The misappropriation of the click-removing plugins can demonstrate this effect, which is dependent on your audio reproduction chain and the music piece, so it is not a universally applicable ABX test but rather a "slowly turn the knob until you suddenly hear a qualitative change" personal experience.
I trust you on the use of the 125asx2. And I agree with you that it makes use of post filter feedback. My bad.
The various patents around self oscillating class d with post filter feedback seem to have forced icepower to use a more "tempered" version with the consequence that the frequency response is not fully load independant. But we are now talking about nuances, not the caricatural behavior from other topologies.
That's helpful. Thanks!
[0,+max,0] contains all frequencies. Clearly you can hear that!
Yeah the precision, the clear effortless sound reproduction and the reduced crossover distortion make them sound bad.
Corrected
Well yeah,Corrected
but there is a small switching delay/dead time.
Yep, happening at the switching frequency, so completely different. But also source of distortion.
Hmm, still not a hint of experimental verification, just more and more handwaves and analogies.
I'm getting a wee bit suspicious that this is massive bullshitting.
I'm getting a wee bit suspicious that this is massive bullshitting.
- Joined
- Apr 24, 2019
- Messages
- 1,501
- Likes
- 2,822
You chose to ignore the very basic principles of Nyquist and the sampling theorem and make up meaningless experiments.
Could you please write here the definition of the the theorem you use.
- Joined
- Apr 24, 2019
- Messages
- 1,501
- Likes
- 2,822
The sampling theorem, simply removing or adding samples to a digital data stream is not BW limited in the analog domain. More to the point what does simply trimming digital bits off of a waveform have to do with how any actual signal chain or device operates (unless broken)?
Last edited:
Here is a figure of the starting waveform in your experiment. Here I am using fewer periods so we can see what's going on. Blue dots are the samples, and orange curve is the "analog" waveform.
When the waveform is changed into a tone burst, here is what happens. The "analog" waveform shown is the only band-limited waveform that will match the samples according to the sampling theorem. This "analog" waveform is no longer single tone.
Here are the spectral analyses of the full-length waveforms in your experiment (dB magnitudes are normalized to the peak value). With a relatively large duty cycle burst, the side lobes from the spectral leakage are pretty weak and therefore inaudible.
With a lower duty cycle, the side lobes spreading into the audible frequency range have become large enough to be audible.
Therefore, it does not require your hypothesis to explain what you said you hear.
When the waveform is changed into a tone burst, here is what happens. The "analog" waveform shown is the only band-limited waveform that will match the samples according to the sampling theorem. This "analog" waveform is no longer single tone.
Here are the spectral analyses of the full-length waveforms in your experiment (dB magnitudes are normalized to the peak value). With a relatively large duty cycle burst, the side lobes from the spectral leakage are pretty weak and therefore inaudible.
With a lower duty cycle, the side lobes spreading into the audible frequency range have become large enough to be audible.
Therefore, it does not require your hypothesis to explain what you said you hear.
Similar threads
