Master Thread: Are Measurements Everything or Nothing?

[j_j]

This kind of captures a lot of the dialogue in this thread (not aimed at the last few posts).
View attachment 514665

So, what is the corner case? That's part of the dialog, too. Of course, I'm not suggesting you're holding that position.

Absolute difference contains ALL physically existing errors. Note "physically" there, I've actually, in the past, tell me that I didn't credit the "spirit of the music".

 

[Geert]

This kind of captures a lot of the dialogue in this thread (not aimed at the last few posts).

90% of people prefer speakers with a flat response. All audiophiles: ”I’m part of the 10%”.

1% of people can distinguish high res audio. All audiophiles (incl. +70 years old), ”that’s me".

Less than 1% of DAC’s have an issue with jitter. All audiophiles, ”that's my DAC”.

Some locations have an extremely poor mains grid. All audiophiles, ”that’s where I live”.

A few amplifier show artifacts that aren't observed with the common test protocol. All audiophiles, ”that's exactly what I hear with my amp”.

Vey exceptionaly room acoustics and the speaker setup are so good that DRC doesn't add much. All audiophiles, ”I’m so lucky with my living room”.

...

 

[antcollinet]

From a signal processing perspective, digital audio and digital images are the same; the latter is just sampled in two (spatial) dimensions instead of one (temporal) dimension. Sampling rate is then samples per degree of the visual field rather than samples per second. Thus, the operation in digital audio which is directly analogous to spatial image scaling would be scaling playback speed (without using a time stretching algorithm).
If one ensures standard viewing conditions such that the spatial sampling rate from the viewer's perspective is known—like temporal sampling rate is in digital audio—maximum perceptually useful image resolution is more clearly defined.

There is a massive difference between digital imaging and digital audio.

In that in imageing - the individual samples (pixels) is all you get. The display also only has individual pixels. If you view an image 1:1 on screen then each image pixel (sample) is mapped one to one with a display pixel. There is no reconstruction of the image between samples. There is no "analogue" image waveform. Even when you print you have a DPI based spatial resolution.

Conversely with audio, the reconstruction filter combined with a band limited signal at the time of recording accurately fills in the voltage/time in between the samples. you can zoom in as far as you like, and you won't (in a correctly filtered signal) see the individual samples.

You will of course be able to see the (hopefully dithered) quantisation noise if you are zooming in with an oscilloscope. But:
A - That is bit depth (amplitude resolution) : the equivalent of colour accuracy in imaging - not time resolution (the eqivalent of pixel resolution in imageing)
B - It is well below the level of audibility. (you can't zoom in with your ears except by turning up the volume - and before noise reaches an audible level - if the music is playing at the same time - you won't have any ears left to zoom in with.


EDIT - and that is ignoring all the perceptual stuff as identified below by @j_j

 

[j_j]

From a signal processing perspective, digital audio and digital images are the same; the latter is just sampled in two (spatial) dimensions instead of one (temporal) dimension. Sampling rate is then samples per degree of the visual field rather than samples per second. Thus, the operation in digital audio which is directly analogous to spatial image scaling would be scaling playback speed (without using a time stretching algorithm).
If one ensures standard viewing conditions such that the spatial sampling rate from the viewer's perspective is known—like temporal sampling rate is in digital audio—maximum perceptually useful image resolution is more clearly defined.

Absolutely not. From a signal processing point of view wherein the end product is relevant, they are nothing close to alike. Vision does not care, to a very great extent, on things like harmonic distortion, and frequency response only matters as it blurs the spatial resolution. The eye is a spatial receptor at its most basic, and things like MTF and edge sharpening are part of a suite of effects that aid perception of image content. Things like nonlinear sharpening, changes in gamma, etc, can all enhance (or destroy, of course) any image, and very often some not so subtle changes in gamma and saturation can make a blah image into a pretty good one (but, yes, it's often overdone, as well). Ironically, although the eye is much more linear in short-term capture (before adaptation), the eye does not really depend on linear rendition very much at all, within very broad parameters. What's more, the eye is considerably "slower" in many fashions than the ear, due to the ear's innate frequency analysis.

The ear is at its base a frequency analyzer, hence frequency response changes can cause obvious perceptual differences, and any addition of distortions (i.e. gamma changes) introduce NEW FREQUENCIES at places far removed in frequency that can create very unpleasant distortions. The ear's short-term perception is much less linear (level adaptation happens in 1 millisecond on the basilar membrane), but the rendering of the sound to the ear must be much, much more linear, and the bandwidth is substantially wider.

So, no, images and sound are not the same even to any kind of signal processing past the sampling theorem. Sorry, but no, just no. I have had endless arguments with video people who want to impose their processing methods on audio, only to find that they don't do anything even slightly like what they expected. As always, video wins arguments, UNTIL THERE IS NO SOUND, and you can be pretty damn sure I'm tired of that, because then they want the doctor to come and fix a system that failed basic design issues AT THE LAST DAMN MINUTE.

 

[bmc0]

There is a massive difference between digital imaging and digital audio. [...]

Absolutely not. From a signal processing point of view wherein the end product is relevant, they are nothing close to alike. [...]

Sorry, I guess I should have worded things differently to make it clearer that the comparison I was drawing was extremely limited in scope. The point I was trying to convey was the one in my last sentence. To restate: Zooming in/out or changing viewing distance changes the spatial sampling rate. If spatial sampling rate is not controlled, maximum useful resolution is not well defined. In contrast, one doesn't tend to freely vary playback speed in audio, so maximum useful temporal sampling rate is fairly well defined.

"Not the same past the sampling theorem" is certainly correct, and that's as far as I intended my comment to go. I was certainly not meaning to imply that the processing one might want to do to an image (or video) is similar to audio in general. The eye likes things that the ear hates, and vice versa.

There is no "analogue" image waveform.

There is, but you're correct that monitors don't display it with "proper" reconstruction. In cameras, optical lowpass (blur) filters are often used to reduce moiré patterns (spatial aliasing). Image resampling (scaling) algorithms use lowpass filters for the same reason. And before I get yelled at again —a filter which is good for image resampling is generally a terrible choice for audio resampling.

 

[j_j]

And before I get yelled at again —a filter which is good for image resampling is generally a terrible choice for audio resampling.

Oh I'm not going to yell at that. A gaussian and a high-rejection FIR are indeed very different. Ditto (yes, somebody proposed this once) using splines for audio resampling. At least there is something to 'limiting discontinuity sharpness" in that.

 

[Dan Lavry]

Sorry, I guess I should have worded things differently to make it clearer that the comparison I was drawing was extremely limited in scope. The point I was trying to convey was the one in my last sentence. To restate: Zooming in/out or changing viewing distance changes the spatial sampling rate. If spatial sampling rate is not controlled, maximum useful resolution is not well defined. In contrast, one doesn't tend to freely vary playback speed in audio, so maximum useful temporal sampling rate is fairly well defined.

"Not the same past the sampling theorem" is certainly correct, and that's as far as I intended my comment to go. I was certainly not meaning to imply that the processing one might want to do to an image (or video) is similar to audio in general. The eye likes things that the ear hates, and vice versa.


There is, but you're correct that monitors don't display it with "proper" reconstruction. In cameras, optical lowpass (blur) filters are often used to reduce moiré patterns (spatial aliasing). Image resampling (scaling) algorithms use lowpass filters for the same reason. And before I get yelled at again —a filter which is good for image resampling is generally a terrible choice for audio resampling.

For Voltage change in time to accurately represent the music air vibration:

(Good for Audio, scopes, weighing scales, MRI…), I know very little thing about video signals:

Theory first:

Gaussian is the perfect interpolator. It connects the sample value to the precise shape (curvature) of the original signal.

You also need a “brick wall” filter. Immediate transition band pass to “band reject” From “gain of one” to total attenuation. Adjust the width of the gaussian to filter above Nyquist (half sample rate). The outcome would be perfect signal as in theory.

Problem:

Gaussian lasts forever. The longer you wait, the more accurate the result is. But there is a time limit, and practical considerations, time delay (latency). A sudden abrupt end to a Gaussian (rectangle distribution) result in less than desirable result This is not just an audio issue. So there are various fixes (adjustments) to the truncated Gaussian, they are the windows. Different windows are used for different applications (focus on different signal properties).

Such FIR filter – interpolator try to approximate the theoretically filter and be the interpolator, which is the same thing! They have the advantage of linear phase, and disadvantage of long latency.

It turns out that if you can use any great filter, it will interpolate well. Filtering means interpolation. So for low latency you need different kind of a filter. Simple IIR (emulating analog filters) have phase non linearity. But there are good solutions. I design for linear phase and low latency. The technology and theory advanced from the good old days.

 

[sonitus mirus]

I just have to pause and make a mostly irrelevant observation.

Having Amir, JJ, and Dan Lavry participating here is pretty remarkable.

Amir brought Microsoft-level engineering experience and modern lab-grade measurements into the public discussion. (though I initially debated this over at Head-Fi before I realized his goal).

Dan Lavry designed converters used in professional mastering environments and has consistently pushed for engineering grounded in psychoacoustics rather than marketing.

JJ’s work at Bell Labs and his role in perceptual audio coding and MPEG standards like MP3 and AAC was foundational to how digital audio is distributed today (which makes him a bit of a hero to me). I suspect a lot of people do not fully appreciate how central that contribution was. That entire line of work reshaped how music is stored and transmitted. It made large scale digital libraries and streaming practical, which is why most of us now have instant access to millions of tracks. That shift did not happen by accident.

And beyond them, there are several other serious engineers and researchers active here. Having that level of technical depth in one forum is rare. It kind of blows my tiny mind.

 

[Chrispy]

I just have to pause and make a mostly irrelevant observation.

Having Amir, JJ, and Dan Lavry participating here is pretty remarkable.

Amir brought Microsoft-level engineering experience and modern lab-grade measurements into the public discussion. (though I initially debated this over at Head-Fi before I realized his goal).

Dan Lavry designed converters used in professional mastering environments and has consistently pushed for engineering grounded in psychoacoustics rather than marketing.

JJ’s work at Bell Labs and his role in perceptual audio coding and MPEG standards like MP3 and AAC was foundational to how digital audio is distributed today (which makes him a bit of a hero to me). I suspect a lot of people do not fully appreciate how central that contribution was. That entire line of work reshaped how music is stored and transmitted. It made large scale digital libraries and streaming practical, which is why most of us now have instant access to millions of tracks. That shift did not happen by accident.

And beyond them, there are several other serious engineers and researchers active here. Having that level of technical depth in one forum is rare. It kind of blows my tiny mind.

Many others, too. Always impressed and try and keep my simple minded observations out of their way

 

[j_j]

I just have to pause and make a mostly irrelevant observation.

Having Amir, JJ, and Dan Lavry participating here is pretty remarkable.

...

And beyond them, there are several other serious engineers and researchers active here. Having that level of technical depth in one forum is rare. It kind of blows my tiny mind.

What's entertaining is that all 3 of us have been thrown out of many "audio forums". You have to love it. Or something.

 

[RexrothPigeon]

What's entertaining is that all 3 of us have been thrown out of many "audio forums". You have to love it. Or something.

badge of honor for sure

 

[antcollinet]

There is

There is not - You are correct that the antialiasing filters before sampling exist - but that is before sampling. After sampling there is no conversion back to analogue. You always have discrete samples on output : pixels per inch (discrete samples) on display devices) or DPI in printing processes.

It is true that there is a form of dithering in printing (dot size can be adjusted a little bit), and the resolution is generally good enough, on photographic printers, for the naked eye to see a smooth image - but it is still discrete samples on output.

 

[j_j]

There is not - You are correct that the antialiasing filters before sampling exist - but that is before sampling. After sampling there is no conversion back to analogue. You always have discrete samples on output : pixels per inch (discrete samples) on display devices) or DPI in printing processes.

In modern systems this is mostly true, however, I did many years ago use a system that did a "reconstruction" on CRT's, which were then carefully converged and run through an optical filter before being written to film. Of course, then you have "film grain". But that was in the "analog" area.

I suspect one could also do this on "photo prints" that aren't done by inkjet, giclee, etc. I rather suspect it's not the case. I am somewhat certain that some printers have high enough DPI to properly antialias digitally to a point that a close-up eye couldn't find the pixels, though, so that's helpful.

One of the interesting things about film grain is that it creates a kind of "dither" in a way.

 

[ahofer]

So, what is the corner case? That's part of the dialog, too. Of course, I'm not suggesting you're holding that position.

Absolute difference contains ALL physically existing errors. Note "physically" there, I've actually, in the past, tell me that I didn't credit the "spirit of the music".

I’m not even qualified to say for sure if they are right, but I’ve certainly seen folks suggest that the ‘standard’ measurement suite doesn’t cover everything music can throw at the signal chain. Sometimes that is just averred without evidence.

Thinking of *credible* assertions (as opposed to the “oh music is so much more complex than a sinewave” Fourier-denier camp), I think @pma has offered some possibilities with Class-D amplifiers.

NC252MP (class D) vs. A250W4R (classAB) burst measurements into 4ohm//2.2uF load

NC252MP (class D) vs. A250W4R (classAB) burst measurements into 4ohm//2.2uF load As I have been doing, quite intensively, wide range of power amplifier measurements, I have decided to compare 2 different power amplifier design philosophies, class AB and class D. The amplifiers chosen have equal...

www.audiosciencereview.com

My broader point would be that you could always define some weird corner case that the standard suite doesn’t really contemplate. Whether that represents something that might occur in normal use is another matter. But that does not stop people from sticking their foot in that door and damning the whole enterprise on those shaky grounds. Measurements are no good because they don’t starch my shirts.

 

[jareinertson]

It's a hobby, not a sacred vow. Buying stuff and trying it out for the fun of it is a pursuit of perfection, given a whole host of constraints, even when those improvements may be inaudible. Liking something that measured poorly may be caused by several factors: 1.) it was the best one could do at the time for whatever reason, 2.) we had trained our ears to favor the particular distortions provided by the supplanted equipment, or 3.) we were exploring a unique buying opportunity, nostalgia, aesthetics, price, brand curiosity, or some other feature unrelated to measured performance.

I still like and use my Advent speakers in a second system, even though the Revel speakers I'm using in the primary system work a lot better in that arrangement. I have been loyal to those Advents since I bought them new, which has included extensive repairs over the years. At the time they were introduced, and for their price, they were revolutionary. Lots of impecunious college kids got to hear decent sound because of them. Just because I replaced them in my primary system doesn't mean I don't still like them, or that they did (and still do) what I wanted them to do. I think you'd have a hard time finding a post where I said I no longer liked them, and I don't think I'm an outlier on this forum.

It is a bit funny to see folks chasing a few more dB of SINAD without any hope (or, generally, claim) of audible improvement, but that's what hobbyists do.

The quest to "understand everything" is also what hobbyists do, if they believe that performance can be measured. Those who don't are the ones who claim every tweak resulted in an earth-shattering improvement, removing the final veil (until the next one), impressing the disinterested wife all the way into the kitchen, etc.

Rick "it's a hobby, not a marriage" Denney

I sold many Advent speakers back in the day. Enjoyed them a lot as did my customers. Ahh… Double Advents!

 

[bmc0]

There is not - You are correct that the antialiasing filters before sampling exist - but that is before sampling.

I don't quite understand your point. Antialiasing filters also exist when resampling. The image is conceptually a spatially continuous signal and treated as such.

After sampling there is no conversion back to analogue. You always have discrete samples on output : pixels per inch (discrete samples) on display devices) or DPI in printing processes.

Ignoring subpixels, an LCD is a bit like a NOS DAC without an output filter. Each pixel is a zero-order hold "stairstep" in the output waveform. Would you argue that digital audio suddenly does not conceptually represent a continuous waveform if you end up putting it through a filterless ZOH DAC? Technically, even an oversampling sigma-delta DAC produces "discrete" pulses (ahead of the output filter), but at a very high rate. You could add an optical lowpass filter to an LCD or a printed image to more properly reconstruct the signal. Would that make it analog? Consider as well that there's a limit to the angular resolution of the eye—a lowpass filter, in other words.

 

[voodooless]

You could add an optical lowpass filter to an LCD or a printed image to more properly reconstruct the signal. Would that make it analog? Consider as well that there's a limit to the angular resolution of the eye—a lowpass filter, in other words.

Camera sensors often add a low-pass filter to account for moire effects, which are en effect similar to imaging issues in audio sampling.

Most LCD screens do use some kind of dithering btw, to create more than the default 8 bit per pixel colors.

 

[Dan Lavry]

What's entertaining is that all 3 of us have been thrown out of many "audio forums". You have to love it. Or something.

I don’t know yours or Amir’s interaction with audio forums. It is fun to recall and share with you what happened to me 20 years ago. I was a technical moderator. It was about technical issues, but we had some threads dealing with ridicules claims. I made sure to stay neutral, never mentioned my company or products.

Company X was advertising their new clock. They claimed that their clock would make the music sound better, some secrete magic built in that signal. I stated that the concept of converters is based on equal time between adjacent samples, a very simple "equal time between adjacent samples" concept. Any deviation or clock modulation is a bad idea…,

Company X fought back insisting on their magic. Some people fell for the new crock. I felt compelled to hold my ground, so they sent in their chief engineer. He knew the facts but tried to hold the company’s line. After a few back-and-forth posts, he conceded in a post that their clock can’t improve music, and it is just a “good clock”. What happens when chief engineer position contradicting his company’s expensive marketing campaign?

The next day I was removed from the forum. One last post by the engineer, talked about their clock but without the crock. The thread was no longer accepting posts. I felt really sorry for the guy, an EE pressured by bosses to push a crock.

Snake oil and “politics” behind the scene won. So I stay away from forums. I am not mad or bitter. It was good to put my time to better use. It helped me decide to avoid social media completely.

 

[jareinertson]

We all have our opinions on Kool.
I believe the VTL Compact Monoblocks I used for 14 years were Kool, and also quite good measuring. for dtubers
View attachment 502987

For SS amps, it's hard to beat my Phase Linear 700B for good looks!
View attachment 502990
Then for just plain ole solid industrial design, my stack of Adcom GFA 545 IIs
View attachment 502993
All JMHO LOL

Got to love that 700B!

 

[ahofer]

I don’t know yours or Amir’s interaction with audio forums. It is fun to recall and share with you what happened to me 20 years ago. I was a technical moderator. It was about technical issues, but we had some threads dealing with ridicules claims. I made sure to stay neutral, never mentioned my company or products.

Company X was advertising their new clock. They claimed that their clock would make the music sound better, some secrete magic built in that signal. I stated that the concept of converters is based on equal time between adjacent samples, a very simple "equal time between adjacent samples" concept. Any deviation or clock modulation is a bad idea…,

Company X fought back insisting on their magic. Some people fell for the new crock. I felt compelled to hold my ground, so they sent in their chief engineer. He knew the facts but tried to hold the company’s line. After a few back-and-forth posts, he conceded in a post that their clock can’t improve music, and it is just a “good clock”. What happens when chief engineer position contradicting his company’s expensive marketing campaign?

The next day I was removed from the forum. One last post by the engineer, talked about their clock but without the crock. The thread was no longer accepting posts. I felt really sorry for the guy, an EE pressured by bosses to push a crock.

Snake oil and “politics” behind the scene won. So I stay away from forums. I am not mad or bitter. It was good to put my time to better use. It helped me decide to avoid social media completely.

This sounds like digital clocking, but the Tice Clock is the GOAT among ‘crocks’:

Tice R-4 TPT & Coherence ElectroTec EP-C "Clocks" Thomas J. Norton March 1991 | Stereophile.com

Thomas J. Norton wrote a review of the Tice Clock in March 1991 (Vol.14 No.3): When I submitted my piece for the October 1990 "Industry Update" on the digital alarm clock controversy, I was only aware of the $500 Coherence Industries version and the $25 Radio Shack standard model. In case you're...

www.stereophile.com