Ok I want to point out why just using a sine wave isn’t interesting.
I stole these graphs from another site since they show what oversampling filters do to signals
This is just as specious in isolation as the sine wave example, measurements from a spring2 I think.
NOS square wave
OS square wave with reconstruction filter
Now which reproduces the signal better?
Was that the graph from torque? Regardless, love it!
It can be correct depending on how many electronic components you decide to include in the definition of the term “DAC” for the purpose of this or that discussion. In this case I meant the output filter of the DAC, and without which the digital-to-analog conversion function remains incomplete, unfinished. (In other discussions “DAC” means a fully assembled market-ready device. That doesn’t make them incorrect, it’s just a different definition used for a different purpose.)
But yes, my whole technical point was about the output filter of the DAC, because its absence is what causes most of the signal defects in the typical NOS DACs being promoted by NOS fans, namely filterless NOS DACs.
I don’t know enough about the oversampling/non-oversampling distinction to say how that impacts the output by itself (assuming for example that a sharp or “brick wall” filter is always used and the only difference is in the oversampling part of the design). I would assume it makes less of a difference than the filtered/filterless distinction, so I’m not too interested to delve into that part of it too much.
It’s from the SBAF KTE spring2 measurements thread, I’ve seen better examples of filters and impulse response, but it get’s the point across.
When you apply convolutions to an upsampled signal you introduce ultrasonic ringing, it’s unavoidable, you can move it around a bit and change the amplitude/area of effect, but it will always be there.
You can make arguments it’s at frequencies that are inaudible, you can make the same argument for the stair steps in the sine wave.
That is perfectly irrelevant, as square waves do not correspond to any acoustic signal that can be heard with human ears. A perfect square wave requires the presence of an infinite spectrum of frequencies, but humans can only hear up to 20 kHz. The artifacts you’re showing on your “non-acoustic” square signal cannot be heard by humans, but the stair-step “sine” waves I showed you at 100% audible frequencies can be heard as being clearly distorted vs. clean sine waves. You have a similar example with sawtooth waves here: Aliasing - Wikipedia
It’s just as relevant as reproducing a sine wave.
Were picking arbitrary signals and looking at how they are reproduced.
Neither is indicative of what we listen to.
Looking at sine wave in isolation really does not give you a picture of how an amplifier or DAC responds to a more complex signal. It can be useful, to understand, some things about an amplifier, but you also have to look at time domain effects, impulse response etc.
Surprisingly I do understand the math behind basic signal processing, and basic information theory.
Any invertable filter adds no information, all a filter can do is remove information, so whether the sine wave looks smoother or not i doesn’t fundamentally make the DAC sound better.
It might be subjectively better sounding, or it might not.
It gives me part of the information, and crucially it shows me if there are glaring issues with the reproduction of individual audible frequencies, which is what complex music is made of anyway, there’s just usually a lot of them. For simpler instruments like a triangle in a symphonic orchestra you have just a few sine waves that are the harmonics produced by that instrument. The analysis result you saw for one separate sine wave will absolutely affect every one of those harmonic sine waves produced by that triangle. Immediately the defects in the sine wave reproduction tell you that a lot of sine waves that are part of a complex signal will likely be affected in the same way, and that you have here a device that is not reproducing all signals equally well. And then you measure a filtered DAC in the same way and you see the output signal keeps the same shape the input signal had near-perfectly, whether it’s one pure sine or 10000 sines at different frequencies.
This isn’t about “smoother” or not, it has to be smooth because I know the experiment started with a smooth signal before the ADC. The requirement is fidelity or accuracy of reconstruction, not “smoothness” per se.
Out of curiosity have you spent some time listening to some NOS DACs? Reading back through the comments, seeing the graphs and the case being closed for you, so wasn’t sure. Which was my point, why does it seem to be preferred by some, that’s the science explanation I’m curious about.
Is that a hill you want to die on?
So you are saying humans cant hear spike pulses (like gun shots) either because to represent them, you need an infinite number of sine waves?
Just wanted to say using minumum phase is more modern, and sounds better.
Minimum phase puts the whole artifacts after the impulse.
So when the bass kicks, it kicks as it should and the echo is after the kick (just like real life). Linear phase has some echo before the signal which is weird and unnatural.
Linear-phase is only suitable for very limited occasions where you have two signals of the same things and want to change them separately. Like when you want to EQ the left channel and right channel differently. That’s where you need the phase to be linear.
That’s why iFi nano has minimum phase for “listen” option. Because that’s how things sound in real life. First the sound, then the echo.
Yes. More specifically, “instant” pulses that rise and fall extremely fast are not heard as such because that would require ultrasonic hearing. We hear them as slower rises and falls, as gentler slopes at least 50 microseconds in duration, because that’s all our 20 kHz hearing allows. It’s useless to try to reproduce them as 5 microsecond transients because they will still be heard as 50 microsecond transients anyway (in fact this is how some NOS fans argue that nothing bad will happen because of the lack of a reconstruction filter: human hearing has its own low-pass filter at 20 kHz that will be applied to all incoming signals anyway).
Haven’t seen any evidence that pre- or post-ringing can be heard, so I don’t see the relevance if there’s some of it before the pulse. But again I’ve seen what the minimum-phase filter does to the shape of the signal in the time domain (especially the higher frequencies) and I didn’t like it:
I don’t think science has humans modelled and explained at such a level as to say why people prefer things. It involves too many things, from biology to neurology to life experiences to personality types etc. You’d have to get together a multi-disciplinary team from many scientific fields to get such a question answered, and I don’t know that this has been done. Closest thing I can think of is the book “How Music Works” by David Byrne, where he follows the history of audio production and reproduction and how people’s listening preferences changed along with the live setups and the technology. Even deals with placebo effects briefly, showing there are real changes in the brain when liking something because of placebo that are the same ones as when liking something because it’s measurably different etc. So it seems like a good source to get a general understanding of how complicated the topic of “why people like stuff” is.
It sure changes the phase, but since the phase is changed for all the channels equally, there’s no worry.
If you have a multi-channel EQ and decide to change one channel’s frequencies, that’s where things get bad.
Whether the whole filters are audible is another discussion. But what linear phase does to the signal is worse compared to the phase shift (which shoudn’t be audible at all since all the channels are changing equally).
Question. What ladder DACs have you actually heard or is your life purely about numbers and not whether or not your music sounds good?
Plus I’m calling this out, starting out with a big post from ASR on this forum is a bit of a troll move. As someone that notices ladder DACs reproducing sound more accurately to my ears than delta sigma DACs I honestly don’t give a damn about the numbers when my sound is neutered. A song should not be messed with tonally for the sake of numbers.
Already dealt with this above, maybe read the thread before accusing others of trolling. 
I use a DSD1793 which is similar to ladder dacs. Not the same, but still similar.
To my ears, the PCM section has problems with transition. The DSD section which acts as a low pass filter and nothing else, sounds better.
As someone who worked at Bell Labs on ASICs primarily focused on DACs and ADCs and how interconnects and pre-amps factor into the entire solution for customers on in the GHz range I understand. But the fact nobody has a done transient analysis and base everything of single sine wave measurements means someone is picking a choosing the datasets.
If you can hear something substantially different in terms of accuracy to pure analog, but the numbers aren’t backing it up (in terms of information), then there is a clear issue with the data taken in the tests or how the conclusions are being drawn out.
My Spring 2 has a terrible DSD implementation where on that unit I cannot compare DSD fairly to NOS modes fairly. DSD, PCM, and NOS kind of doesn’t really matter a terrible amount seeing how that describes how the data is related to the actual DAC portion of a DAC to go digital bitstream → DAC → analog filters/analog back end. I will say that a DAC that implements DSD properly will be technically superior and will have a higher global maxima inherently in its design than PCM for the DAC portion. However, implementation of DSD with delta sigma may still suffer transients that a ladder DAC may not experience.
Delta Sigma designs rely on 4-6 (chord uses 5 in the mojo) 1 bit integrating amplifiers that are summed together. Keyword here is integrating… Which means off the bat there is some design challenges for transients and inherent phase constraints. However, at low frequencies and removing phase questions aside will give the smooth and clean lower frequency sign wave, which is where we see all the measurements for THD. This figure has an impact on the sound but it’s only a piece of the puzzle, I can flat out look at that number and say some of my gear will not respond well and others will be fine.
Now this is were things get interesting between DSD and PCM with delta sigma approaches. For PCM the pulse train into the digital part measures in the normal 44.1k for regular CDs, based on that pulse train internally (usually up sampled to the highest sample supported to keep the digital to analog simpler) the pulse stream in converted to those 4-6 1-bit integrating amps. With PCM I think the max is like 768kHz samples I’ve seen but no clue if the D to A actually implaments that. Now with DSD, bump that up to like 22MHz for the pulse train coming in, MUCH better transient response!
As for ladder DACs, it’s simple. It’s a parallel 1-bit network of resistors without any need for capacitors. Assuming the amplifier can ensure a buffer with minimal phase to be introduced and a capacitor for the transients, it just works.
That’s the beauty of DSD. It’s inherently linear. A lowpass filter gives the signal. No need for data-weighted averaging or anything else.
Multi-bit delta-sigma modulators bring back the problems of ladder dacs.
The bad thing is that it needs quite a bit of process in digital form if the source is PCM. But nowadays, a smartphone can do the job.
Not exactly. That is an extreme over simplification.
Uhm… Either you are phrasing something improperly or I am missing some type of reference that is missing here because the last thing I would say is a low pass filter.
With my 1 piece of equipment I can definitely say I experience VHS level compression with recessed mids. But one does not purchase a Spring 2 for DSD exclusively.
Ladder dacs have linearity problem which is one of the problems of multi-bit DS modulators, too.
This document offers DEM solution for this problem.
A lowpass filter is enough for DSD (1-bit DS modulator) to get converted to analog. While I couldn’t find what Spring II does exactly, I know Airist DAC which is another ladder dac, converts DSD to PCM.
Whatever Spring II is doing to the DSD signal is not necessary and actually damages it.
DSD1793 has a different path for DSD which is only a lowpass filter.
