Continuing on from here: https://amigaworld.net/modules/newbb/viewtopic.php?topic_id=44789&forum=2#858763

So, regarding the high end digital reverb simulation I'm still skeptical that it's actually implemented in a directly physically modelled way in a manner akin to light transport in ray tracing. Light transport has come a long way, we now have almost realtime under RTX for example but invariably that requires very sophisticated denoising techniques because in the time budget you have, you can only sample a few rays per pixel.

Unless I can see some source that proves otherwise, I'm still going with the idea that these super high end reverb plugins rely on filters and delay lines. Why? Because mathematically that is the correct model for what is being simulated and software engineers don't generally go out of their way to find new models for things that have existing good solutions. They tend to refine those slightly instead.

So imagine a physical reverb simulation.

For example, temperature and humidity affect the transport of sound in air and that will invariably act in a way that can be quantified and directly modelled by a filter, mapping those individual controls to the effect they would have on a filter for the direct audio path from the emitter. Why? Because however you want to quantify how the air transport works, in the end, it comes down to different frequencies attenuating at different rates. Frequency doesn't have any significant effect on the propagation rate (in our dense atmosphere, Mars is different for example), so there's no meaningful physical effect on delay or phase. This brings you back to a filter function. As long as you got your parameters right, you can account for the apparent effects on temperature and humidity.

A simple LPF is probably too crude though, what you can do is to make your filter much more parameterised and have many more taps, more like an EQ than a simple cut-off.

As for the effect of the space itself, you set up a large number of separate delay paths, each of which timed based on the distances of the observer from the reflecting planes. You use another filter, controlled by other parameters based on hardness, etc, but still a filter nonetheless, to simulate the reflection. The return signal goes to the observer but also back into the input path to the emitter where it's re-attenuated again by the air properties. The fact it's mixed in this way means that it's then re-echoed ad infinitum, not only from the plane it was initially reflected from but from every other plane that the direct path hits. Moving the observer or emitter affects all the multiple delay lines, as does changing the room size or shape. Changing the temperature, humidity affects the air transport filter and changing any of the materials of any of the planes affects the filter simulating the reflectance. Maybe temperature and humidity have some quantifiable effects on the surfaces too, but once quantified become contours to their filter function.

I really don't think it's more than this because there's simply no way it needs to be, to capture the nuance. The fact that every return signal goes back into the input path to the emitter means that the complexity of the mixing of an exponentially increasing number of "echoes" emanating from the simulation doesn't increase the computational overhead at all. It's completely limited to how many surfaces there are, how many paths per surface you use and how finely tuned the filters are and the precision (bit depth and temporal) you are processing at. But fundamentally, in a mathematical and implementation sense it's equivalent to the simplest digital reverb, using filters and delay.

Last edited by Karlos on 22-Dec-2022 at 04:47 PM.

_________________
Doing stupid things for fun...

@Karlos

Interesting stuff. I agree with your trouncing of audiophile things. CD quality was and is good enough for me. The only reason I miss vinyl is for the art and the ritual not the scratches and the inconvenience.
I’m only just catching onto the whole streaming concept that this sub literally opens up the whole world of music I’ve not yet listened to and putting it into practice.
Early digital reverb and effects sound bad though. The art can only improve.

Last edited by golem on 22-Dec-2022 at 04:26 PM.

_________________
1200 desktop Blizzard 1260 with 2Gb Jaz SCSI drive, 32Mb, OS3.9 BB2
1200 desktop Blizzard 1230, 32Mb, OS3.2
Trusty 1.3 A500 underneath the TV with SCART

@golem

So I'm not averse to analogue sound. Vinyl does have a lovely warmth to it. However that's a subjective, personal interpretation of a wide range of artefacts Ans distortions introduced into both the capture and playback. It sounds fine, but when someone tells you it's "more accurate" than a digital recording simply because it's not quantized, you know the chances are they can be sold a solid silver, actively biased directional USB cable.

That's not to say digital sounds better, because appreciation of sound is subjective. However provided that the capture and playback are at least double the highest analogue frequency you intend to capture (Nyquist), you have filters on the recording that cut frequencies above that and finally an appropriate reconstruction filter on the output side you are golden. The only other parameters left is bit depth. Given the logarithmic perception of sound intensity, vanilla 16-bit actually has ample dynamic range. Put all these together and 48kHz 16 bit is as good as any normal person needs for audio reproduction. 44.1 kHz was chosen for a number of technical reasons around digital video but that cuts the highest frequency to 22kHz. Still higher than probably 98% of people can appreciate but 24kHz is the realm of canines.

Where digital audio gets a really bad rep is when lossy compression is introduced. However, this is generally based on good scientific understanding of how sound perception works. Most people can't differentiate two frequencies in the audio range that are less than a few Hz apart. Play them together and they can perceive the beating introduced by the difference but played sequentially as pure tones they won't notice. Unless they are Jacob Collier. Convolving the time based singal into discrete windows of frequency based data is how most modern audio compression (not FLAC) work because you can then throw away a staggering amount of data and yet reconstructed the audio seems identical. There's more to most audio compression still, such as head transfer functions that take advantage of how we can't properly hear quiet things after loud things.

Early MP3 encoding and playback was terrible for doing a bad job for most of this, but there are many better formats for different purposes that use more or less all the same tricks.

_________________
Doing stupid things for fun...

@Karlos

Quote:

So, regarding the high end digital reverb simulation I'm still skeptical that it's actually implemented in a directly physically modelled way in a manner akin to light transport in ray tracing. Light transport has come a long way, we now have almost realtime under RTX for example but invariably that requires very sophisticated denoising techniques because in the time budget you have, you can only sample a few rays per pixel.

There are some Ai modelled reverbs you can get. Sonible make one.

That being said, using analogue ( a real physical reverb) like a full metal plate, has zero CPU hit, faithfully reproduces the same effect to sound every time (taking into account room variance) however, adding in noise of the analogue domain, each time.

I brought this up originally because of your analogue can't reproduce the sound comment.

Mind you, being a single instance (a real device) you have to apply it more than once if doing tracks individually, by bouncing tracks to it and re-recording. Then adding them together after in mixing.

The other way is using sends. Where each track is sent to the physical reverb and added back into the mix after, as effect, at the same time.

So first way, each track is sent or played one at a time into the physical reverb chamber and re-recorded and then all those tracks mixed.

Second way, each track is sent to a physical reverb chamber in a mix and the sum is added back. ( a lot less time consuming, but blended.)

All reverb reflections and harmonics however, are applied at sample rates that have less artefacts created in the digital domain at higher resolution, that occur in lower resolution.
The other added benefit is a much lower noise floor on sample. Multiple tracks with analogue noise, sum together to create more noise!. (Digital calculated reverb, wins here at the cost of CPU)

As an example, if you were to dub/copy an analogue tape, the copy will have more "hiss" as that was also amplified form the original. (unlike a digitally copied one)

It was why I still very much use analogue devices in digital mixing and why sample rate matters.
CPU usage is limited, analogue doesn't care but has noise and mechanical limitations. 2 very different domains that are still very much required. An analogue reverb should be treated like it is a musical instrument.
Then there is the debate on dithering/noise sounding good or not.

My take on all of this stuff is that mixing should be done in the highest domain you can afford digitally, for the most faithful reproduction of the end result audio with all its harmonic colourisation effects that create the overall product. Downsampled to 48K for mix-out.
Playback of the end result is fine at 44.1 or 48KHz sample rate, sure, 96k for the ultra quiet stuff maybe better but I doubt you will hear it.

Last edited by SHADES on 24-Dec-2022 at 12:07 AM.
Last edited by SHADES on 24-Dec-2022 at 12:06 AM.
Last edited by SHADES on 24-Dec-2022 at 12:03 AM.
Last edited by SHADES on 23-Dec-2022 at 11:57 PM.
Last edited by SHADES on 23-Dec-2022 at 11:42 PM.

_________________
It's not the question that's the problem, it's the problem that's the question.

@golem

Quote:

CD quality was and is good enough for me. The only reason I miss vinyl is for the art and the ritual not the scratches and the inconvenience.

Arguably the best analogue recordings can be of "higher quality" than their CD equivalents, due to being in the analogue domain.
The frequencies of analogue are not "filtered" out by the electronic shelves used to remove anything over 21KHz or below 20 Hz.
The limitation in vinyl is the rotation speed, cutting head vibration force and substrate it is etched to.
All of these do contribute to colourisation of the recorded sound however, instead of removal of information (via filtering - also changing the original) that loss is dithered out in amplitude instead.

Each method has its own pros and cons.
Vinyl can have a frequency response from 7Hz to 50kHz and beyond, along with more than 75dB of dynamic range. That will be lost in CD via the filtering process for 20-20KHz and affect how all the sounds mix together when played back. (they aren't there to mix in) This can be a good thing as, arguably, it is mainly dither/noise of analogue gear. (Some PREFER this)

On the other hand, you get the rub of the needle on a physical medium, the rumble of the turntable rotating, noise of the tone arm's resonate frequency, how good the magnetic coil or other method is and environmental factors. There are laser vinyl players that remove all of this, but you still have some rumble from the creation process of the cutting head, turntable of the cutting machine etc used at source. It's just not amplified by the consumer playback machine.

CD however, has a much higher dynamic range when looking at calculation of amplitude in waveforms and doesn't suffer from analogue dither and noise. That rumble and hiss isn't created by the medium in either etching or playback so, you can get much louder and hear things that are much softer.

Last edited by SHADES on 23-Dec-2022 at 11:39 PM.
Last edited by SHADES on 23-Dec-2022 at 11:36 PM.
Last edited by SHADES on 23-Dec-2022 at 11:33 PM.

_________________
It's not the question that's the problem, it's the problem that's the question.

@SHADES

The loudness wars ruined CD more than anything else, IMO. Totally wasted the dynamic range.

As for Vinyl being able to reproduce up to 50kHz that assumes the whole mastering and copying process is able to capture and faithfully record it too. Even if there are no explicit signal filters in the process, the amplification circuitry and physical groove cutting will certainly have their own limits. There are resistors and capacitors in the amplification either end and you'll always have something acting as a filter, even if it's unintentional (even bare wires exhibit capacitance, inductance and resistance). Given that almost nobody can hear much above 22kHz, it's unlikely that much optimisation goes into preserving that content.

_________________
Doing stupid things for fun...

@Karlos

Quote:

Given that almost nobody can hear much above 22kHz, it's unlikely that much optimisation goes into preserving that content.

Ahhh, but how do those unheard frequencies "impact" the sounds that are audible!
Everything impacts.

I agree, you don't hear them, but their effects on the hearing range, that you certainly do.
That noise background does make impacts on how the other waves transverse in space/time.

"Noise" matters.

Last edited by SHADES on 24-Dec-2022 at 12:14 AM.
Last edited by SHADES on 24-Dec-2022 at 12:12 AM.

_________________
It's not the question that's the problem, it's the problem that's the question.

@Shades

Quote:

I brought this up originally because of your analogue can't reproduce the sound comment

Did I say that? I'm pretty sure what I said was that the differences between digital and analogue are minimised in the real world due to filtering. This was in response to the trope that digital can't faithfully reproduce an analogue signal that audiophiles love to serve up, because in their mind, the digital audio they hear is at some microscopic scale all quantized into chair step changes. Which of course it isn't because the whole process of reconstruction requires the raw DAC output is put through a reconstruction filter that has the same (or as similar as possible) characteristics as the one applied to the input during ADC. I mean the maths on this is pretty solid.

And so, if you choose a sampling rate that satisfies Nyquist for the highest frequency anyone can realistically hear, filtering out content above that, provided you have enough bit depth (16 is generally good enough, 24 is overkill 9000) then the differences between the (filtered) analogue signal and the reconstructed signal are basically nothing. Of course there will be differences between the unfiltered original and the reconstructed one, but it's only your pet dog and bats in your loft that will be let down.

_________________
Doing stupid things for fun...

@Karlos

Quote:

Did I say that? "As for valve amps, don't get me started. They are excellent for a range of applications (especially high power were most semiconductors can't cope) but purity of audio reproduction is not one of them."

Quote:

And so, if you choose a sampling rate that satisfies Nyquist for the highest frequency anyone can realistically hear

Not when mixing audio. You don't have to hear the individual frequency to have it effect the frequencies mixing in lower frequency domains, sometimes even low frequencies can have a time/phasing effect when applied to higher ones. As explained in the video I posted earlier. Which, you can reproduce, and hear it. As those errors/effects, come into the audible range.

Analogue works differently. It just becomes more or less attenuated but doesn't reflect back to add in new frequencies or, chop off high/low ones.

Last edited by SHADES on 24-Dec-2022 at 12:27 AM.
Last edited by SHADES on 24-Dec-2022 at 12:26 AM.
Last edited by SHADES on 24-Dec-2022 at 12:25 AM.
Last edited by SHADES on 24-Dec-2022 at 12:24 AM.
Last edited by SHADES on 24-Dec-2022 at 12:23 AM.
Last edited by SHADES on 24-Dec-2022 at 12:22 AM.

_________________
It's not the question that's the problem, it's the problem that's the question.

@SHADES

Quote:

Ahhh, but how do those unheard frequencies "impact" the sounds that are audible

Exactly the same way audible frequencies do. They mix together additively. Except, unlike with audible frequencies, the resulting contribution is as inaudible as the original high frequency. You could probably test this in a professional recording environment by playing some audio and recapturing in the recording booth, once dry and once in the presence of an ultrasonic source.

The only obvious way it could have any other impact is if the resulting signal had enough peak intensity to saturate some part of the process.

Having said all that, it's possible to encode audio frequencies into the difference of interfering ultrasonic waves and pretty much direct it at a target listener. The effect, when properly applied is that they can hear the encoded audio but someone a few inches away is oblivious to it.

_________________
Doing stupid things for fun...

@Karlos

Quote:

Exactly the same way audible frequencies do. They mix together additively.

Nope. As explained in the video. In digital, they(frequencies) can sweep up to the sample rate, then reflect back down again, into the hearing range.
In analogue, they sweep up and keep going getting quieter and quieter till it's just noise.

If it's audible range, summing works the same for what you hear but you don't get digital reflection so, it's a moot point.

Last edited by SHADES on 24-Dec-2022 at 12:45 AM.
Last edited by SHADES on 24-Dec-2022 at 12:44 AM.
Last edited by SHADES on 24-Dec-2022 at 12:31 AM.

_________________
It's not the question that's the problem, it's the problem that's the question.

@SHADES

Under what specific circumstances does mixing any pair of digital signals, at the same sample rate, produce anything other than the strict sum of those signals? The operation being performed is, assuming that you have a volume control for each input, a multiply accumulate: out = in1*vol1 + in2*vol2 for each sample point. Otherwise you aren't mixing them, you're doing ... I dunno something else. The process is exactly analogous to summing a pair of analogue voltage signals using op amps.

If any part of that wasn't true, I'd still be trying in vain to synthesise sound in code, since mixing signals is one of the most common steps, whether it's the output voices of a unit, or the outputs of a set of oscillators for a voice or the input phase, pitch or amplitude modulators. They all use multiply accumulate based mixing.

I'm wondering if we are taking about the same thing. Maybe our use of terminology has some, er, impedance mismatch.

As an audio producer your definition of mixing may be more expensive. Certainly mixing digital streams together at different sample rates requires a whole bunch of resampling to the mix rate that will introduce all kinds of weirdness (which may or may not be desirable depending on what you are making) that analogue signals have no equivalent problem with. Higher mix rates than your usual 44/48kHz definitely help with that.

_________________
Doing stupid things for fun...

@Karlos

Quote:

I'm wondering if we are taking about the same thing. Maybe our use of terminology has some, er, impedance mismatch. As an audio producer your definition of mixing may be more expensive. Certainly mixing digital streams together at different sample rates requires a whole bunch of resampling to the mix rate that will introduce all kinds of weirdness (which may or may not be desirable depending on what you are making) that analogue signals have no equivalent problem with. Higher mix rates than your usual 44/48kHz definitely help with that.

Agreed. Digital audio has unique requirements when looking at how audio mixes together, especially when creating effects like reverb and saturation and compress etc that you don’t have in analogue/valve.
In analogue, you get noise, hiss, colouration etc.
Summing of frequencies doesn’t change, unless you’re in digital and out of your frequency limits, then you get weird artefacts. It’s why I need both doing what I do.

Those artefacts are subjective, as you quite rightly pointed out, depending on what you’re doing. If I was making 8 bit glitch sounding ailising music, it wouldn’t use analog to generate it. Doesn’t work that way. You need to generate those artefacts and that’s due to how digital sampling works.

Last edited by SHADES on 24-Dec-2022 at 01:13 AM.
Last edited by SHADES on 24-Dec-2022 at 01:11 AM.

_________________
It's not the question that's the problem, it's the problem that's the question.

@SHADES

Quote:

Nope. As explained in the video. In digital, they(frequencies) can sweep up to the sample rate, then reflect back down again, into the hearing range

We were talking about mixing. However now in think we have the actual subject..Producing a frequency sweep like this does not produce such reflection artefacts if the signal generator behaves in the same way as a real capture device would: putting a filter in line.

I've experienced this when working on my synth code. If I program an oscillator naïvely to emit the instantaneous value at each sample point then as I ramp the pitch I get all kinds of distortions, subharmonics etc. What did I do wrong? I forgot the fact that my oscillator albeit entirely digital and operating at the mixing frequency at all times, is no different than any other input from the real world. It has to be filtered, just the same, because as the number of discrete sample points on the duty cycle gets less and less, there are more and more alias frequencies that pass through them. The effect isn't that pronounced on sine or triangle oscillators but on saw and square, good god almighty. I mean you know the harmonic series for saw and square so you can see that even low notes have very high harmonics. You get problems even at musically useful pitches, let alone anything even approaching Nyquist. And heaven forfend if I allow signals higher than Nyquist, which is of course possible if I don't include an antialias filter!

However, as soon as I remembered this basic necessity of signal processing, those issues went away. I didn't even need to use a full LPF cutoff, a travelling 5 sample window was more than adequate.

I would suggest that sweeping up to the sample rate to illustrate the problems with it indicates a lack of understanding of the basic principles of digital audio in the first place - there should be no frequency components in any stream that are higher than half the sample rate. It's not ultrasound suddenly coming in and messing up your audio range, it's straight up aliasing: multiple subharmonic waveforms that have periods coinciding with your unfiltered sample points. If you try to introduce any frequencies above that, you're breaking the (mathematical) rules of the game and you should expect weirdness.

Last edited by Karlos on 24-Dec-2022 at 01:28 AM.

_________________
Doing stupid things for fun...

@Karlos

Quote:

I've experienced this when working on my synth code. If I program an oscillator naïvely to emit the instantaneous value at each sample point then as I ramp the pitch I get all kinds of distortions, subharmonics etc. What did I do wrong? I forgot the fact that my oscillator albeit entirely digital and operating at the mixing frequency at all times, is no different than any other input from the real world. It has to be filtered, just the same, because as the number of discrete sample points on the duty cycle gets less and less, there are more and more alias frequencies that pass through them. The effect isn't that pronounced on sine or triangle oscillators but on saw and square, good god almighty. I mean you know the harmonic series for saw and square so you can see that even low notes have very high harmonics. You get problems even at musically useful pitches, let alone anything even approaching Nyquist. And heaven forfend if I allow signals higher than Nyquist, which is of course possible if I don't include an antialias filter!

It depends on what you are doing.
When using saturation, or reverb harmonic frequencies are generated as interactions with input. Not possible to filter on the in. Distortion is another. Reverb adds harmonics that are required to smash in and manipulate phase and frequency alike. Analogue has no need for any filter.

The metal plate on a real reverb is a metal plate. The limit is how quick you can vibrate and what the diaphragm can capture, sound wise.
Filtering after that effect to stop anything over 20k may help things calm down in digital (whole idea in oversampling plus filter) after but, it will affect the sound as well. Valve stuff doesn't do that.
It just gets more "noisy" and scrambles the signal into infinity. (no sample rate limit, no filter)
It's no like it's not done in digital but it's very very hard to do the processing. Very similar to ray tracing i would assume. and even then, is it correct? Analogue adds to it with signal pathway design, voltage, transformer hum, resistors etc etc
You can try your sweeps on analogue amplification without any filtering. Not hard to do, as long as they don't exceed and alias prior to amplification.
Come to think about it, I've never tried a true digital amp. I hear they are oscillated in the megahertz so, way out of the Khz range. That would be interesting. I know they are used in subwoofers a lot. Maybe they do have a filter on them and low freq wouldn't be so much of a problem because it's slow.

Well, i'll be.
The Tripath digital audio amps are considered to sound more like top $ valve audio gear in the 100kmar but for $20 lol
I'm out of my depth now. I'll have to read up more on Tripath.
https://en.wikipedia.org/wiki/Class-T_amplifier

Last edited by SHADES on 24-Dec-2022 at 02:24 AM.
Last edited by SHADES on 24-Dec-2022 at 02:23 AM.
Last edited by SHADES on 24-Dec-2022 at 02:20 AM.
Last edited by SHADES on 24-Dec-2022 at 02:20 AM.
Last edited by SHADES on 24-Dec-2022 at 02:13 AM.
Last edited by SHADES on 24-Dec-2022 at 02:12 AM.
Last edited by SHADES on 24-Dec-2022 at 02:09 AM.

_________________
It's not the question that's the problem, it's the problem that's the question.

@SHADES

Going back to basics on digital reverb, you take your dry input and put some attenuated copy on a delay line. The output of the delay line is then filtered and added back to the input. You have to do that filtering because you want to model the frequency attenuation specifics (damping) of the air and reflective medium.

If there's a problem with this design it's that you need enough of these delay paths with a spread of delay to get the desired overall reverb effect, each path having a discrete delay, whereas a real physical reflective surface produces a continuum, every point on the surface contributing to the reflection each with a slightly different delay. There's also the vibration of the surface itself as it absorbs energy from the incident wave and that vibration will have some effect, I'm sure. These interactions generally aren't reproduced in the delay/filter model. It's an approximation, not a simulation. I think this is the true limit of digital reverb. You can't model continuum delay because you don't have enough discrete delays to work with.

However given that there's no resampling going on in any of this there's not really any obvious way to generate accidental frequencies above the Nyquist because we are just mixing. Which is direct summing. Also, to be really technical, "frequencies above Nyquist" is a bit of misnomer because it implies the presence of a signal above the frequency that it can be represented digitally at the selected processing rate. Which means there is no signal - it's the illusion of a signal caused by our friend aliasing. You can't have anything above a 24kHz signal in a 48kHz stream because you literally can't plot it. At 24kHz, you only have, at best, two sample points pet duty cycle. If you try to produce something higher than that, you will get some repeating cycle of points but they won't correspond to your desired signal, they'll produce a set of alias frequencies, somewhere lower in the audio spectrum.

I can definitely see how constructive interference could produce unwanted lower frequencies, but I can't see how higher frequencies are generated in the process. Let's say you take a signal and invert mix it with itself offset by one sample. Almost everything will be cancelled out except for the small adjacent changes. The highest frequency here is still any repeated set of pairs of points, which is still half the sampling rate. There just isn't a way that any additive or subtractive process of mixing two digital signals can produce a frequency above that, because there's no way for the stream to represent it. Unless you are resampling to a higher rate first, but then your input signals didn't contain anything above half the original sample rate either. You might then get some artefacts above that rate in your new higher sample rate output but only as a result of aliasing problems during the resample, not from the mixing step itself.

_________________
Doing stupid things for fun...

@Karlos

Quote:

I can definitely see how constructive interference could produce unwanted lower frequencies, but I can't see how higher frequencies are generated in the process. Let's say you take a signal and invert mix it with itself offset by one sample. Almost everything will be cancelled out except for the small adjacent changes.

Harmonics get generated. Digital reverb is hard on CPU. Analogue has no hit but you still create harmonics coming back in from noise, natural plate resonates etc. Then there's saturation and other things you need a lot of the time. All adding in more colour, can amplify frequencies already present in mixing.

When you talk about inverting frequency, sound is rarely a single frequency. As i explained earlier, the top and bottom of a snare drum, have different textures, 180 degrees out of phase. As the stick hits the top it pushes down the skin, on the bottom of the drum, the skin is pushed out. That tone of the overall drum has undertones, natural harmonics and skin tones. The heads are never tuned the same. Offsetting by a few milliseconds can make a huge difference, depending on the tone of the overall added correction is. you may want a particular tone to pop out in sound, or be pushed back. The natural harmonics can be controlled this way, without touching EQ. The overall combination when including room microphones adds depth but even more harmonics. The Snare of that drum has a lot of very high frequency so, adding in reverb (very common) in come the potential problems.

https://www.youtube.com/watch?v=qrXXNKhjBE8 around 15 mins in. Lots of people use phase change to stop frequencies from affecting each other or, to accentuate. Just depends on what you want.

Anyway, all mixing should be in high rate due to all these reasons. mix down, doesn't matter as much, as the "soup" can be calculated properly of output by that stage.

Last edited by SHADES on 26-Dec-2022 at 10:50 AM.
Last edited by SHADES on 26-Dec-2022 at 10:49 AM.
Last edited by SHADES on 26-Dec-2022 at 10:46 AM.
Last edited by SHADES on 26-Dec-2022 at 10:46 AM.
Last edited by SHADES on 26-Dec-2022 at 10:37 AM.

_________________
It's not the question that's the problem, it's the problem that's the question.

@SHADES

I didn't say inverting frequency I said inverting the signal, e.g flipping sign of the samples.

If you'll pardon the broken record pun, to restate it, there are no frequencies in any digital signal that are above half the sample rate. This is not due to filtering or any kind of constructive or destructive adding just a hard mathematical limit. You can't represent any periodic wave shape with a frequency above this limit. At exactly half the sample rate, any signal would be equivalent to a perfect square wave and be reproduced as a triangle wave on any hardware implementation (digital or analogue) that does linear interpolation. A good filter might reconstruct it as a sine wave.

By the same token any process running at a same rate of x is incapable of producing any harnonic with a frequency above x/2. You can of course write a process that performs resampling of the stream to a higher sample rate first, process that resampled signal and end up with harmonics above x/2 that need to be filtered out again if you want to output at the original sample rate and not sound awful. But that's literally the only way it can happen. It's not software implementation choices or anything, it's just maths.

For a visual analogy, draw me a perfectly black square exactly 2.5 pixels to a side on a white background...

Last edited by Karlos on 26-Dec-2022 at 02:02 PM.

_________________
Doing stupid things for fun...

Anyway, I think we are in agreement, as much as it doesn't look like it: You want the process things at the highest quality you need resolution. For sound, resolution is the bit depth and the samplimg rate.

For some algorithms it's just easier (engineering wise) to go to much higher sampling rates. You can ignore a lot of computational noise and aliasing if it's going to be a over the final audio frequency range.

Consider the venerable Yamaha DX7. It produces 12 bit digital samples at 50kHz, per voice. It doesn't even mix voices digitally. The DAC just switches between them in fixed time slices, meaning the DAC output rate is 16*50kHz, switching between the instantaneous 12 bit values of each voice in turn. You can consider this an 800kHz output, but no processing happens higher than 50. The reason this doesn't sound awful is that the immediate output goes through a reconstruction filter, in the analogue domain, that effectively produces the analogue sum of the 16 discrete inputs, as well as performing the LPF cutoff to get rid of unwanted artefacts above 22 kHz or so. The end result is a smooth continuous output

Last edited by Karlos on 26-Dec-2022 at 04:37 PM.

_________________
Doing stupid things for fun...

@Karlos

Quote:

I didn't say inverting frequency I said inverting the signal, e.g flipping sign of the samples

Right, so, instead of tracking 0 to 180degrees you go 0 to -180 degrees. The sign wave is inverted. or am i missing something here?
If you have two frequencies and one is inverted, in the same time domain, they cancel out right?

Quote:

no frequencies in any digital signal that are above half the sample rate

Not that aren't damaged or aliased, correct.
In analogue, there are, the medium and mechanics are limitations, up to infinity, amplitude over noise or, attenuation. There is no sample rate or frequency limit as such. All of that added "colour' still mixes in with the rest of the sound. Analogue EQ's cannot "cramp" because of this and you can use bell curves right up to the 21khz without issue. The trade off, is noise and what you can hear. That being said, a lot of people, myself included, add it in via sends to real reverb gear, prior to mix down, if that's what we are after.
Quote:

By the same token any process running at a same rate of x is incapable of producing any harmonic with a frequency above x/2.

Which is why plugins like ProQ use oversampling and why higher resolution prior to mix down is used.
It very much has a purpose. Mathematically proven. Nothing is breaking here.

If curves at mix down are plotting in the right direction, at mix down. The filters at lower resolution that cut off any plots above the sample rate don't matter as much because the effects of colliding have mixed into whatever got cancelled or amplified, prior so, the overall plots of the final mix (without aliasing or much less) now get converted to 44.1 and you get a much nicer sounding mix at 44.1.

You can use the same techniques prior to going to 8 bit as well and hear the difference.
The video i gave you, showed the maths plotted and how aliasing occurs when mixing in 44.1. You can reproduce this, yourself. All of it is "math" based, how the aliasing occurs, all of it. and can be proven.

Last edited by SHADES on 28-Dec-2022 at 05:45 AM.

_________________
It's not the question that's the problem, it's the problem that's the question.