Only if by 'approximate reality' you mean make a noisy but otherwise accurate within the limits of real world eningeering, record of the voltage at the recorders input.
Here are a couple of files you may find interesting, they are 3 bit quantised piano, first with no dither, then with simple TPD dither then lastly with a modestly noise shaped dither.
The first one sounds **NASTY** but fades to silence as the signal drop below 1 LSB, the second and third ones sound very hissy as you would expect, but the piano clearly sounds like a piano and can be heard through the hiss and even as it fades down below the hiss.
Undithered: http://media.soundonsound.com/sos/feb08/audio/digitalaudio/piano_3.mp3
Dithered: http://media.soundonsound.com/sos/feb08/audio/digitalaudio/dithered_piano_3.mp3
Noise Shaped: http://media.soundonsound.com/sos/feb08/audio/digitalaudio/noiseshaped_piano_3.mp3
Regards, Dan (Who really thinks you should come back AFTER studying some discreet maths and some signal processing).
Here are a couple of files you may find interesting, they are 3 bit quantised piano, first with no dither, then with simple TPD dither then lastly with a modestly noise shaped dither.
The first one sounds **NASTY** but fades to silence as the signal drop below 1 LSB, the second and third ones sound very hissy as you would expect, but the piano clearly sounds like a piano and can be heard through the hiss and even as it fades down below the hiss.
Undithered: http://media.soundonsound.com/sos/feb08/audio/digitalaudio/piano_3.mp3
Dithered: http://media.soundonsound.com/sos/feb08/audio/digitalaudio/dithered_piano_3.mp3
Noise Shaped: http://media.soundonsound.com/sos/feb08/audio/digitalaudio/noiseshaped_piano_3.mp3
Regards, Dan (Who really thinks you should come back AFTER studying some discreet maths and some signal processing).
the innumeracy is breathtaking to engineers
64 bits is way more than Audio, "Sound" even exists - from vacuum/shock front to Brownian Noise, air molecules/s bouncing off your ear drum
64 bits time/distance resolution is enough to see Special and General Relativity effects in lifting your coffee cup
Platonic forms/ideals just aren't useful in reasoning about the real world, meausrements or human sensory perceptions
64 bits is way more than Audio, "Sound" even exists - from vacuum/shock front to Brownian Noise, air molecules/s bouncing off your ear drum
64 bits time/distance resolution is enough to see Special and General Relativity effects in lifting your coffee cup
Platonic forms/ideals just aren't useful in reasoning about the real world, meausrements or human sensory perceptions
Last edited:
Yep, 64 bits is massive overkill for measuring anything, and even 24 bits is a pretty much unobtainable ideal in a 20K bandwidth (Mostly because the analogue stuff in front of the converter is nowhere near that good).
Regards, Dan.
Regards, Dan.
That has been repeatedly suggested, to no apparent effect. DF96's prediction earlier was exactly correct.
I'm still trying to figure out how one can avoid dither (much less why in the world one would even want to!) at 16 bits or greater when real microphones, preamps, and consoles are used. Or perhaps I should say that I'm well aware that it can't be (and shouldn't be), but this has not occurred to Kastor.
Avoiding learning the fundamentals is a mindset that I just cannot fathom, I will admit.
Ah, but SY, if you don't know how anything works then everything is possible (I had a manager like that once).....
Love your sig, and actually cake sounds like a good idea.
Regards, Dan.
Love your sig, and actually cake sounds like a good idea.
Regards, Dan.
I suspect that you don't quite have quantizer dynamic range conceptualized correctly. Quantizer dynamic range essentially is the ratio between the smallest magnitude signal the quantizer can produce (meaning, control) and the greatest magnitude signal, usually expressed logarithmically in dB. This measurement does not include 'digital black' or the quantizer's null state. If we were to include the null state, then your 4-bit quantizer could be said to have an nearly unlimited dynamic range, going from whatever desired full scale magnitude down to an null state. Including this null state when expressing quantizer dynamic range would be like turning off the engine of your car and then marveling at how quiet it is.
It also seems like your conceptualization of quantization noise may not be quite solid. Quantization noise is the sample by sample ERROR in converting signals between analog and digital domains, and is due to finite quantizer resolution/bits. The fewer the quantizer bits, the greater is the sample by sample quantized error, with respect to the magnitude of analog signal being quantized. One of the many non-intuitive aspects of DSP is that quantization error does not by itself distort the quantized signal, which is why it is more descriptive to refer to the error as noise, albeit mathematical noise, rather than a manifestation of nature.
Last edited:
The HM-602 states that it uses a non-oversampling DAC, a TDA-1543. The wobbles on the Oscilloscope trace look to be about 40 kHz. It could be that the reconstruction filter is failing to attenuate the sample frequency. I suggest looking at the output waveform of an over-sampled DAC. I would hope that a PC Codec would be better than that.
Or that they didn't bother with a reconstruction filter since their target market laps up nonsense.
https://head-direct.com/Products/?act=detail&id=199
Some basic binary operations - calculations would be helpful for newbies (like me) in digital audio:
1) A 16bit binary number could get a max value:
0111 1111 1111 1111 b = 7FFF h = +32,767 d
and a min value:
1000 0000 0000 0000 b = 8000 h = -32,768 d
So the multitude of the possible different decimal values (quantities) that could represent a 16bit binary number is:
+32,767 - (-32,768) = 65535
If we translate this multitude to a dB scale we have:
20 X log 65535 = 96.329 dB
which i think is the named dynamic range of a 16bit, e.g. analog to digital converter.
Thus an analog signal with a reference level of 0dBV = 1Vrms = 1.4142Vp = 2.8284Vpp could be broken in quantities with a minimum level diference between them of:
2.8284 / 65535 = 43.15uV
From those i've tried so far in analog audio projects, it is amazingly difficult to obtain a so small noise floor.
2) A 24bit binary number could get a max value:
0111 1111 1111 1111 1111 1111 b = 7FFFFF h = +8,388,607 d
and a min value:
1000 0000 0000 0000 0000 0000 b = 800000 h = -8,388,608 d
So the multitude of the possible different decimal values (quantities) that could represent a 24bit binary number is:
+8,388,607 - (-8,388,608) = 16,777,215
wich if translated in a dB scale we have:
20 X log 16,777,215 = 144.49dB
which is the dynamic range of a 24bit analog to digital converter.
The same analog reference signal as above could be broken in quantities with a minimum level diference between them of:
2.8284 / 16,777,215 = 0.1685uV
In any analog system (audio or anything else) a such noise floor is impossible from those i know.
Thanks
1) A 16bit binary number could get a max value:
0111 1111 1111 1111 b = 7FFF h = +32,767 d
and a min value:
1000 0000 0000 0000 b = 8000 h = -32,768 d
So the multitude of the possible different decimal values (quantities) that could represent a 16bit binary number is:
+32,767 - (-32,768) = 65535
If we translate this multitude to a dB scale we have:
20 X log 65535 = 96.329 dB
which i think is the named dynamic range of a 16bit, e.g. analog to digital converter.
Thus an analog signal with a reference level of 0dBV = 1Vrms = 1.4142Vp = 2.8284Vpp could be broken in quantities with a minimum level diference between them of:
2.8284 / 65535 = 43.15uV
From those i've tried so far in analog audio projects, it is amazingly difficult to obtain a so small noise floor.
2) A 24bit binary number could get a max value:
0111 1111 1111 1111 1111 1111 b = 7FFFFF h = +8,388,607 d
and a min value:
1000 0000 0000 0000 0000 0000 b = 800000 h = -8,388,608 d
So the multitude of the possible different decimal values (quantities) that could represent a 24bit binary number is:
+8,388,607 - (-8,388,608) = 16,777,215
wich if translated in a dB scale we have:
20 X log 16,777,215 = 144.49dB
which is the dynamic range of a 24bit analog to digital converter.
The same analog reference signal as above could be broken in quantities with a minimum level diference between them of:
2.8284 / 16,777,215 = 0.1685uV
In any analog system (audio or anything else) a such noise floor is impossible from those i know.
Thanks
And the concept of a NOS (non-oversampling) DAC could be clearly described from the attached screen-shot from oscilloscope.
The upper waveform is the output of an oversampling DAC while the lower is the output of the NOS. With analog audio terms, the upper waveform is plenty from overshoot and ringing and i have seen it many times in tests with instruments. It is realy the enervating side of digital audio devices as it implies a very high TIM distortion (transient intermodulation distortion) which in analog audio circuits usualy is caused from non compensated feedback loops.
The lower waveform is the output of NOS and indeed is a clean and enviable square wave. But only a 1KHz test waveform does not say the whole story.
I would like to see the square wave response of NOS at 100Hz and at 10KHz which are the difficult test signals.
Thanks
The upper waveform is the output of an oversampling DAC while the lower is the output of the NOS. With analog audio terms, the upper waveform is plenty from overshoot and ringing and i have seen it many times in tests with instruments. It is realy the enervating side of digital audio devices as it implies a very high TIM distortion (transient intermodulation distortion) which in analog audio circuits usualy is caused from non compensated feedback loops.
The lower waveform is the output of NOS and indeed is a clean and enviable square wave. But only a 1KHz test waveform does not say the whole story.
I would like to see the square wave response of NOS at 100Hz and at 10KHz which are the difficult test signals.
Thanks
Attachments
Last edited:
So whether 'step' or 'stair case' what does it mean?
As far as my understanding of his DAC goes, it relies on interpolation in the PC, so its not non-interpolating. Also the sound of NOS DACs isn't a matter of my opinion, rather my observation that they sound more natural.
My current hypothesis is that its due to glitch yes. Hence my avoidance of R2R DACs and favouring of segmented current source designs more optimized for dynamic performance than for linearity.
Perhaps someone needs to invite him into this thread, so he can explain it for us?
He is the one that has spent XX years designing a top-flight DAC, not me, I'm just designing concepts in thin air, i.e. conceptualization, not with hard electronics.
That is not to say that hard electronics are excellant to thin air. If you've vested a certain amount of time, sweat and tears into something you may have a certain, err, vested interest in your... vindications!!!
I'll assume step distortion is this. Starting with a sequence of decisions, writing "I" here not referring to me personally, referring to the third-party analysis
1 - I don't like Delta-Sigma, I don't believe in it due to......
- A lack of natural sound
- A lack of perfection
- A lack of evidence
- Ultrasonic noise, which is not in the spec sheets
- Glitch error
- Time-domain
- An advancement in the Sigma-Delta technology, which is returning to R2R anyway
The last one is citing Altmann.
2 - I need to use an R2R DAC, see above.
3 - I don't believe in reconstruction filters, due to......
- A lack of perfection
- A lack of evidence
- Time-domain error
----- Let me pause here for a moment -----
Abraxalito, I saw you writing in a thread recently within this sub-forum discussing exactly this.
I advise everyone in this thread to look at this thread linked just to expand this topic into some more clarity!!
I.e., that some of the most advanced designers, achieving the lowest specs currently available, don't believe in our usual reconstruction filters
http://www.diyaudio.com/forums/digital-line-level/259397-new-chord-hugo-dac.html
-----
4 - Now I have arrived at an R2R DAC, without a reconstruction filter.
5 - This DAC seems to have "step distortion" in it, when I analyze it, there are steps everywhere instead of sines, are these steps audible? What do I do with them?
According to XXHighEnd, they are in-band error, thus we need to upsample them away into a much wider spectrum, let's say 10 Hertz to 705.6 kHz
----- Pause -----
I'm not saying this is true, but, is it? He says his Nos DAC comes in at 0.01% THD.
Thus, audio bit depth, is linked to step distortion, is linked to THD?
I'm neutral and the above isn't from him, I'm just assuming that's how this so called step distortion works, if it exists.
Is it your perceptual opinion or do you have some kind of hard theory / reasoning why they sound more natural?
Here is Audio Note's take on it......
ANK Audiokits - DAC Kit 2.1
"To complete the pure digital to analogue design of the DAC 2.1 the digital board uses the Audio Note Non Oversampling architecture for sonic purity and has No Analog Filtering (analog filtering has been totally removed). We have found Oversampling and having Analog Filtering in the signal path has a marked detrimental effect on sound quailty."
How does upsampling interpolate?
Such as which IC?
Last edited:
Cirrus Logic paper on reconstruction filters, as recommended by you earlier for me to study
http://www.cirrus.com/en/pubs/whitePaper/DS668WP1.pdf
Please tell me which filter is perfect in this paper, IIR or FIR?
Then, we can install it in all modern audio equipment and move forwards from confused discussions like these, where the topic starter has not read the whitepapers.
What does accurate within the limits of the real world mean?
I very much doubt it'll generate light, perhaps a fair amount of heat though.
Glitching is about the only foible that S-D DACs aren't prone to.
I don't have much of a clue what 'perceptual opinion' might mean. I do have a hypothesis for why they sound more natural - there's less glitching and a lower error due to settling.
In the XXHighEnd world, I have no idea. In the vast majority of digital filters its by FIR filtering, the superposition of multiple time-shifted sinc functions.
TDA1545A and TDA1387 I've played with, TDA1541A is a popular choice that I haven't built anything around.
"In the XXHighEnd world, I have no idea. In the vast majority of digital filters its by FIR filtering, the superposition of multiple time-shifted sinc functions."
How about in the Chord Hugo world then, it's FIR as well.
How about in the Chord Hugo world then, it's FIR as well.
Last edited:
In the Hugo its by FIR filter (a very long one, implemented in an FPGA). But then there's no DAC chip as such so I think it must use PCM-> PWM conversion or alternatively there's a 'discrete' low-bit DAC and its using an S-D modulator.
I am not personally aware of portable digital devices, using digital media as the format, which can produce analog sound, which do not have a D/AC, i.e. digital to analog conversion step in them.
Please shed some light on this new bizarro parameter you've introduced to us.
Sonove said:"Well. Although I wrote in the pages of the HM-602 and Gudaguta,
It is this short and non-oversampling. ↓
Sine wave regeneration of 1kHz"
An externally hosted image should be here but it was not working when we last tested it.
"The DAC of today, you have a smooth output by the filter and oversampling, digital data rickety originally,
"But You're a rickety originally Oke out as it is without tampering" in NOS DAC
I ... Is that I feel it? (Bearish)
I feel like somehow understand also is that feeling. It is interesting to somehow.
(Highs fall and will issue a staircase waveform as it is, it's a Damedame anymore numerical goes back up the noise floor is fact)
It Anyway,
It was interesting or rather surprisingly quite listening to the HM-602 this time,
Despite the "rattling Mutcha", without discomfort so much when I listen fact As you can see the output waveform.
None at all confident it can be determined by blind I mean.
----- Terms.
I did my best and keenly aware keenly ... I Is terrifically sloppy Nante ear Oira.
Add and RMAA measurement results to the page of HM-602.
Retro style (?) Measurement result is interesting.
However, aim ... What was
It had arrived by the time you forget completely
Add Page of HiFiMAN HM-602.
Response of the power ON is bad, or broken? I am in fear a little me."
Look at the white text in the picture, top-left, third line down.
Look at this link as well, I online translated his text from there ----- http://sonove.angry.jp/Log2.html
Last edited:
How is this statement pertinent to what I wrote? I wrote that the Hugo doesn't have a DAC chip (as far as I'm aware, please correct me if this is in error), not that it did not have a D/A conversion step in it.
Hello Fotis Anagnostou,
Thank you for your very interesting post.
You equate NOS to less TIM.
I have never seen that before and it is well noted.
I've not very informed when it comes to TIM, I will check the Matti Otala / Jan Lohstroh papers later.
For onlookers interested, the first paper seems to be called "An Audio Power Amplifier for Ultimate Quality Requirements".
http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=1162523&url=http%3A%2F%2Fieeexplore.ieee.org
Lohstroh, J. ; Philips Research Laboratories, Eindhoven, Netherlands ; Otala, M.
Edit ----- I hardly know what TIM is, my standpoint here is neutral.
Just expanding the list of "why to Nos", which, with a complete list, I can see the theories clearly.
Personally, I don't think Nos sounds very good, compared to my ES9018, LoL.
Last edited:
- Status
- Not open for further replies.