D/A and A/D | Digital Show and Tell (Monty Montgomery @ xiph.org) 

But misleading.

​@@soloperformer5598how is it misleading?

And distracting

@@SamsungTshirt Its not, he just had to come in here and make sure you knew.

Not a chance, it's styled like a tutorial for elementary schools. I don't see how that's a problem.

It already is.

It always will be...

Still is! 😂

Another one conned.

what is your problem?@@soloperformer5598

Kinda!

But limited.

@@soloperformer5598 "BuT LiMiTeD" Bet you are one of those people that use the "Your audio isn't distorted enough" argument. 🫵🏻🤡🤡🤡

It is. Dithering is technique for preserving detail when reducing bit depth. Think of a bit crusher, it gets sharp noise artifacts when you reduce bit depth. The classing bit crusher just introduces aliasing artifacts to your sound. Dithering on the other hand is adding white noise then bit crushing. This white noise randomly increases or decreases the amplitude of your signal so when it rounded to the nearest small bit depth value, it might be rounded to a bigger or smaller number then it would have without the noise. So the dither IS just noise.

There is no "apparent integrity". Below the Nyquist limit there is only total integrity of any signal -- no matter what the content of that signal is. It doesn't matter if the signal is a pure sine wave, which in a Fourier transform would be the least information dense signal possible, or white noise, which would be the most information dense signal possible (same as picking a true random value for every sample). This is all covered by the Nyquist-Shannon sampling theorem.

It's a linear system. Since all band limited signals can be separated into a spectrum of frequencies, you can analyse what happens to each frequency to understand what happens to the entire signal.

@@srpenguinbr Excellent and concise explanation. Bravo.

All "real world" sounds are nothing more than single frequencies added together. In fact Monty already showed a square wave, which is infinite frequencies added together. But as we are all band limited (our ears) we can only hear the first 19 or so, and even then you'll have to be a child. You'll find it very inconvenient to demonstrate these examples if you were not using a single frequency, you might be able to handle a few sine waves but what is the point? Monty already showed you 19 sines added together.

@@FM-kl7oc Exactly, and since we don’t have infinite data bandwidth, the integrity would surely suffer.

It's an analogue reconstruction filter after the DAC, whether it is OS or NOS does not matter here. If it was without reconstruction filter, the signal output would indeed look different, but the ear would do the analogue filtering so the acoustic result would be the same.

@@nicksterjNot sure if RU-vid remove my comments or it was removed manually. Let me do it one more time here. Please search for the page "how-to-pick-the-best-filter-setting-for-your-dac". On that page, it shows a real stair step waveform output from a modern DAC, Topping E30, using NOS mode (implemented by using zero-order hold as you mentioned). The output for a perfect 1kHz digitized sine wave still look like a sine wave with many steps. For the perfect 10kHz digitized sine wave, you don't want to see its audio output. The very, IMHO, simply misleading by suggesting that you won't get the stair step output. 🤨

@@user-yk6uy7su9i Not sure if RU-vid remove my comments or it was removed manually. Let me do it one more time here. Please search for the page "how-to-pick-the-best-filter-setting-for-your-dac". On that page, it shows a real stair step waveform output from a modern DAC, Topping E30, using NOS mode. Topping E30 does have analogue reconstruction filter but it is not good enough for NOS mode. The output for a perfect 1kHz digitized sine wave still look like a sine wave with many steps. For the perfect 10kHz digitized sine wave, you don't want to see its audio output. Yes, I agreed that you would still be able to somehow filter the stair step waveform with your ear but the point is that this video is misleading as you could still get stair step waveform from a even modern NOS DAC (or modern DAC with NOS mode)

No you wouldn't, and that's why you aren't an engineer with a degree who has suffered through linear alg like the rest of us did.

It’s not that it’s represented by less than a bit, it’s that its amplitude is less than that corresponding to the interval between two whole values. So you can have a digitised signal that alternates between 0 and 1 (or between [-1, 0, 1]), consisting of a “true” signal with a lower amplitude than that + some broadband noise.

9:38 It's the conversion of a continuous value (analog) into a discrete value (digital). You can imagine it as the rounding of a decimal number (3.14159) to the closest integer (3). Hope this helps.

It is avoidable with infinite sine waves. But only at infinite sine waves. As long as you add up a finite amount of sines, they will be there. (Source: My calculus textbook)

If I understand what you mean by overshoot then no. You need infinite bandwidth, nothing has infinite bandwidth. You need infinite bandwidth everywhere, including in the ear which is impossible.

No.

Compression issue? What compression issue? Clearly, you haven’t listened to a word of the original poster’s excellent tutorial.

you say no thong compression could be native.

Digital data compression dosn't really work that way, if that's what you're referring to.

I mean for lossy formats

That's why you need the MISSING distortion analyser.

No, your ears wouldn't be able to hear them.

Jimmy, you are simply insisting on things that do not exist and using some very limited knowledge to support your completely erroneous theories. Nobody should even read your comment. It is a set of stair steps into an oblivion of ignorance on the topic, obscured by irrelevant technical language of topics you don’t understand. Just sentence after sentence of complete nonsense.

You don’t need to know any electical engineering to answer the question about “why no stair steps”. It’s sufficient to know that there is only one correct mathematical solution when you go from digital samples back to analog, which reproduces the original signal perfectly. It’s not a quirk of how electrical components work or how some DAC implementations work like you seem to believe. It has nothing to do with filtering or smoothing. The filtering of half sample frequency before sampling is used to avoid distortion from alising and has nothing to do with human hearing or ”background noise”.

Sorry for the confusion, this reply is directed to those interested in making their own DAC from scratch. I noticed my reply isn't under someone's question as it was supposed to... That being said, thanks for taking your time to reply, I have learned alot from your kind feedbacks.

You didn't watch the video. 17:38

​@subliminalvibes so it is because audio DACs are different to general purpose ones, or that they all approximate fixed voltage levels?

I'm no expert, I'm not even a talented amateur, but my understanding is: A DAC is an _analogue_ device. You can model an electric signal as a weight on a spring. The DAC output voltage pulls on the spring, the weight movement is change in output voltage (the velocity of the weight is current, it allows down at a rate dependent on capacitance) You can "jump" from one voltage to another all you like, but the spring weight is analogue, it can only move through every voltage to get from A to B.

Or, a DAC does not have discreet _output_ voltages, it has discrete _input_ voltages. And you can't listen to the input side, because that's just data, it's imaginary. You listen to the output, which is analogue, because _you_ are analogue.

@@subliminalvibes I did watch the video, twice, I'm just confused on how a DAC creates a smooth signal is all, I may have misinterpreted the video and if so I'd be happy to hear where I went wrong. I'm not trying to spark an argument, I'm just confused about something is all.

@@nicksterj Which is fine, though I don't really care for dither, but this is completely beside the point I am trying to make: The entire video blatantly ignores the fact that most 16 bit DAC's have a THD+N around 90dB or worse, which is why I am saying you need to have a very good quality 24 bit DAC to get PROPER 16 bit in your signal chain.

@@nicksterj Agree 100% about the actual bitrate of quality soundcards, exactly as I wrote in my 1st post.I have absolutely no idea what kind of players you are talking about, my personal experience is limited to various external soundcards and balanced signal chains. Only thing I will say is that there is a significant difference between a 90dB THD+N and a 105dB THD+N signal chain. There is a slowly increasing number of DAC's able to perform better than 110 SINAD, biggest advantage is lower noise floor IMO, especially with high sensitivity speakers.

I'm not sure it's fair to say "A cheap DAC will just stay at that level til the next number arrives". The filter is a necessary part of the digital to analog conversion, and holding a number until the next number arrives, know as a zero order hold, is just a consequence of the fact that the output contains a capacitor that gets charged to that voltage level._(and even if we don't put a capacitor explicitly there is always parasitic capacitance of wires, traces, parts, hell even between the output point and the earth itself)_ We can't really get a series of dots into our analog filter as an ideal DAC would because there can be no device that can output a voltage which is a dot in time.

​@@thewhitedragon4184 Perhaps I should have said simple instead of cheap. Sorry. My point is that it is impossible to recreate an instantanious voltage in a DAC. You say you don't know where the step signal comes from. It comes from the DAC trying to recreate the original signal.

@@stevenmsaxe Yeah the stairstep "effect" comes from the fact the we charge the output capacitor and it retains that voltage until another charging cycle comes. This essentially is a another sin(x)/x looking lowpass filter around the signal with some phase shift at the output

@@thewhitedragon4184

Yeah no that's just wrong... Digital signals (serial, ethernet, hdmi etc) usually switch between two states. They'd like to stairstep between the two voltages. But if you actually look at the voltage you will discover why no matter how simple or cheap a DAC is the output will not look like a stairstep. DAC's can be higher or lower quality, but they will never produce a stairstep.

P.S. My not-fully-thought-out theory is that phase information would fundamentally be higher-frequency information than the nyquist frequency, and yet if the phase information between different base frequencies somehow matters to how the human ear and brain perceive things, then indeed doing the band limiting needed for a 44.1KHz (or 48KHz or maybe even 192KHz) sample rate might still be perceivable with certain sounds -- that are combinations of certain base frequencies but at different phase angles. But, I haven't run the numbers and I don't trust that I'm right. It's just how I'm trying to make sense of what he's saying, so far.

There's no difference

If you don't understand the fact that music is built up of multiple sinewaves of varying frequency, amplitude and phase, then it's end of story: you will never grasp it!

The DAC does matter. His demo would be a bit different if he use non oversampling DAC

There's nothing "fishy" about it. If you think he's, somehow, cheating, then show how he is: build your own setup to demonstrate where he is wrong and what the effect of doing the experiment correctly is. It's called peer review and repeatability, and it's what Science is all about. This is a relative cheap and easy setup to replicate. Debunking the above video should be easy if Monty is gaming the system in some way. The fact that, after 10 years, no-one has, speaks volumes. I'm sorry, but making claims about this looking "fishy" just comes across as someone butt-hurt by the video who can't accept the facts.

@@karmac0ma The lowpass filter at the DAC's output is doing that. I'm not sure there are dacs that don't include that. You literally need at least a single resistor and capacitor minimum.

@@thewhitedragon4184 youd be right

That is literally what a DAC does. Takes digital signals (samples) and converts to analog (not samples). The voltage is not "held constant" at a speaker. By the time the audio signal reaches the speaker cables, it's already been transformed into an analog (smooth) signal.

@@tz4601 Ah. How is the smoothing done? With capacitors and inductors?

Also, remember that any speaker is itself a filter, both electrically and mechanically. So even if we were to produce a “perfect” electrical square wave, once it is fed into whatever physical speaker you choose it would be filtered first by the coil, piezo element, or whatever you choose as an actuator and then as well by the transduction into sound waves. However, due to the infinite series of odd harmonics needed to construct a mathematically fully perfect square wave, the energy content of such a signal would also of course be infinite. That’s why an ideal square wave is unphysical.

@@josephgleespen1433 You’ve summed it up perfectly. Scanning the comments reveals a stunning number of people who continue to insist there’s still some hidden magic resolving their cherished stair step theory. They mention Fourier transforms, which have nothing to do with the topic and are guaranteed to be even more mysterious to them. I don’t even know where to start with these people. That perfect sine wave coming out of the D/A converter is somehow not convincing enough that the very bright people who tackled the issues of digital recording of music 60 years ago knew what they were doing. A tiny bit of knowledge is a scary thing.

He's talking about tape hiss. When recording to a tape, you're physically blasting it with magnetic radiation. There is also other sources of magnetic radiation in the wild, namely the earth's magnetic core so you get noise on your recording added other then the noise already present at the signal. Your analog synth app honestly has very little to do with this other then the fact that if you plug a real synth to speakers and digital synth to speakers they can just make different voltage levels. Edit: Synth to speaker has no signal processing path, other then the frequency response of the speaker, potentially amp but let's assume that's flat, and like the impedance of the wire acting as and attenuator and like MHz filter. Your app has a DAC and it's noise and distortion to account for.

He's talking about people who used their bookshelf stereo system to record music to tape from vinyl, FM, or CD, which were universally terrible. Automatic level control meant that the signal was always too low. No Dolby noise reduction. And people buy the cheapest tapes they can find, not chrome or metal. The next step up is using a dedicated tape deck recorder, which did sound better, but most people didn't use that.

@@thewhitedragon4184 Saw now. Thanks for the good info!

@@jonah1976 Ah.

He demonstrated that bit depth effects where the noise floor is. A CD with 16 bits has a noise floor down around -96dB or so, which with dither can push that down further to around -120dB. However tapes have a noise floor way way higher than that, typically if you were to look at the dynamic range of audio cassette in terms of digital bits then you'd have only 5 bits of depth. He was trying to highlight that the *quantisation* *noise*, noise added to a digital signal due to rounding the sample up or down to the nearest sample value, was so low as to be inaudible. This of course ignores any other noise added, say from a noisy microphone or preamp. Now on cassette you can go a long way to reducing the noise by using an erase head that records a high frequency tone to the tape before the audio is recorded. This makes the noise much less noticeable, plus with Dolby it can get even better. But still if expressed in bits, it's far short of what digital sampling can do. Assuming you eliminate other noise sources as I said.

the only way to be sure is not scientifically accurate objective measurements, but a subjective test, uh huh, ok

@@gundabalf OK, Mister Sarcasm. There are imperfections in audio you can measure with instruments but can't hear, and those you can hear but may not measure with the instruments you happen to be using. Properly designed double-blind listening tests make the subjective become objective. What we perceive is, in the end, all that matters in sound reproduction

I disagree with most of what you said. The lollipop trace is not just meaningful theoretical, but actually is just how one should view and treat digital data when processing it. It's a series of dots when it's digital. The point of an ADC is to give out our set of dots and the DAC is to take that set of dots and turn it back into the exact continuous line for which an ADC would give that set of dots, meaning a DAC doesn't just interpolate them. Here are some other points of contention I have with what you wrote from a technical point of view: 1) An ideal ADC would do the exact thing you said a good ADC wouldn't do. An ideal ADC would take and infinitely small time sample each period. That would just be a multiplication with a dirac comb. And an ideal ADC wouldn't need a quantizer. The problem is we can't make circuits which are just multiplication by dirac combs, we can make circuits that take a very small time to sample but not infinitely small, and we can't store a number with infinite decimal points, we'd need a bit depth of infinity, so we round. That rounding to our bit resolution of 8, 16, 24 or 32 adds noise to our signal that is equal to the difference of the unrounded signal and rounded signal. That's because that rounding error still has some energy so it manifests as noise in our quantized signal's frequency content. This noise is added on top of any noise the quantization circuit itself adds to the signal path. 2) Analog filters aren't used to hide the stairstep effect, they are an integral part of digital to analog conversion. Sampling a signal in the time domain is achieved, ideally, by multiplying the signal with dirac comb signal. That gives us a series of dots. In the frequency domain, this multiplication becomes convolution with a dirac comb _(the fourier transform of a dirac comb is a still a dirac comb)_ . This convolution makes copies of the original signals frequency content centered around the sampling frequency. So if you could have a circuit output a perfect set of dots that go to a lowpass filter, you'd get a perfect reconstruction of your signal. 3) I don't know what you mean by artefacts here. Real adcs and dacs introduce distortion from finite sampling times, quantization and the zero order hold sure but so does recording to an analog medium as the needle can shit or ambient noise can make on your tape. When you reproduce sound from a recorded medium it is always slightly distorted I don't get the digitization part. 4) Dithering is the only point I agree with you on but even then not fully. Dithering is deliberate noise added to the signal to preserve detail when reducing bit depth while sampling noise is inherent to sampling, I didn't choose to add it. Dithering is used because you record and mix with 24 bit audio and then export to 16 bits so to make sure your song doesn't become bit crushed you dither. This applies to sine waves too.

@@thewhitedragon4184 while the lollipop trace is theoretically interesting, it doesn't represent what the hardware is actually doing, which is all that matters in the end.

@@thewhitedragon4184 damn spitting fatcs here

If noise is below the threshold of hearing, or below the level that is generated by the analog stages, then there is no value to increasing bitrate, as all it does is reduce quantization error, and thus noise. If frequencies are higher than can be heard (ultrasonics) or reproduced by the analog stages, then there is also no reason to increase the sampling frequency or change the bandlimiting. 192/96KHz and 32-bit sound is very useful in the recording, mastering and processing stages, but for audio delivery to the listener, 16-bit/48kHz are for all practical purposes at the limit of human hearing.

@@qmanol - With all due respect to his sine-wave machines and what they're worth, this is a gross oversimplification of what increased resolution actually means to sound. Bloody accountants thinking they have a clue about nuance.

@@nicksterj I'm no engineer but I call BS on that. It is absolutely related to resolution. Even if it's also related to things like THD and dynamic range. And no, it's bullisht to suggest that CD is 'good enough'. You guys just want to sound like you have some sort of inside scoop. You're basically reverse snobs. You're being ignorant.

@@nicksterj - Yes, my technical objection is the gross oversimplification of sound resolution to what his meters show in graphic form. He claims that if visible sine waves on a screen are smooth, then greater resolution has no effect except with noise floor. Rubbish. And the ground breaking statement that 'the staircase is actually a series of points'. Duh, really?

SonicScoop is another know-it-all talking out of his ass for clicks. I've been listening to hi-res vs standard res for months and I could pick out the hi-res every time. Similar to the audible improvement between shitty mp3 vs CD.

The spectrum analyzer does show harmonic distortion and noise, and was used quite extensively in his demonstration of bit depth and dither. I'm not really sure what else you want here.

@@sabiro2315 I wanted to see the output on the instrument I trust, the HP spectrum analyser.

​@@soloperformer5598 it IS an HP spectrum analyzer

@@maxrainer804 I know it is which is why I said it was the instrument I trust. There wasn't much output displayed on the HP spectrum analyser even thought the signal flow was from left to right.

​@@soloperformer5598what will a distortion analyzer show that a spectrum analyzer won't?