Not sure if 'getting rid of image bands' is a very good way to describe what upsampling does. If done properly the process moves image bands farther apart from each other in the frequency domain. Why? Because if we band limit audio going into an A/D from 0Hz to 20kHz, then upsample to, say, to 96kHz, we will have created an empty band from 20kHz to 48kHz. That empty band occupies frequency space that once had images in it.
In terms of what happens if we upsample without filtering, it depends on how we interpret the resulting digital signal. If the original ADC acquired data sequence represents impulses multiplied by instantaneous analog signal amplitude, and if zero stuffing is taken to mean a set of original impulse samples separated by filled-in null data points, then the original digital signal meaning would appear to remain unchanged. If the stuffed zeros are taken to mean there are samples that represent zero-crossings that never occurred in the original analog signal, then it looks more like maybe we have produced inconsistent data.
Now, if we take ADC acquired samples and apply a zero-order hold to each one then we have altered the original sampled data. The result in the frequency domain is HF droop. However if we zero stuff without filtering and upclock it out to a dac, we will have reduced the average analog signal output amplitude according to the number of zeroes added. (That signal amplitude reduction can be corrected for in a digital filter of course, if one were to be applied).
In terms of what happens if we upsample without filtering, it depends on how we interpret the resulting digital signal. If the original ADC acquired data sequence represents impulses multiplied by instantaneous analog signal amplitude, and if zero stuffing is taken to mean a set of original impulse samples separated by filled-in null data points, then the original digital signal meaning would appear to remain unchanged. If the stuffed zeros are taken to mean there are samples that represent zero-crossings that never occurred in the original analog signal, then it looks more like maybe we have produced inconsistent data.
Now, if we take ADC acquired samples and apply a zero-order hold to each one then we have altered the original sampled data. The result in the frequency domain is HF droop. However if we zero stuff without filtering and upclock it out to a dac, we will have reduced the average analog signal output amplitude according to the number of zeroes added. (That signal amplitude reduction can be corrected for in a digital filter of course, if one were to be applied).
Last edited:
That's what I meant to say - makes it easier to remove them by analog filtering. Should have been more precise.
This is changing sample-hold operation to a diferent kind. The same effect is achieved without oversampling, but time limiting of the sample-hold output, dropping to zero when time expires. By shortening time limit, the average amplitude get lower, but HF drop is less significant. Also mirror images get a lower amplitude. It creates a filtering effect without FIR.
Last edited:
VHDL is more of a handache than Verilog due to all of the extra junk you have to add to each module to make it usable. Functionally, they're pretty much the same, with similar quirks, but no one really chooses to use VHDL if they have a choice.
Verilog is generally easier to approach if you forget everything you learned about normal software programming (C/Java/etc.) and approach it as defining hardware, more like Spice.
Everything in the digital chain is noise-immune, by design - as long as your 0's and 1's make it out of your computer, there's nothing to be gained, unless you couple noise from the computer to your device (looking at you RME). It's only the ADC/DAC itself that is susceptible to PSU noise, unless something is really wrong with your computer, but you'd see this in a benchmark with crashing/freezes/hanging, etc.
There are some ADCs/DACs that perform better at higher sample rates too (TDA1541A comes to mind). Higher sample rates for Nyquist-rate converters generally incurs more switching (charge injection and glitch issues), the need for higher settling time, and higher overall power consumption with regards to thermal changes and that noise/performance shift.
Higher sample rates for sigma-delta converters need a change in the filter to increase in-band bandwidth, which adds more of the shaped noise to the desired in-band signal. This effective reduction of the OSR reduces the SNR/DR. The modulator clocking rate is generally fixed (save for low-power modes or something), so it doesn't have the same issues as Nyquist-rate converters with regards to that.
This is what I referred to in the following posts:
What do you think makes NOS sound different?
What do you think makes NOS sound different?
@Extreme_Boky:
Hi ... I actually was just mentioning this as I consider the "phenomenon" that DACs don't perform equally well at the various sampling frequencies to be an important parameter that is not often addressed. As it is I personally believe in the sound benefits of higher sampling rates, however, most - if not all - DACs, due to various circuitry compromises, likely perform better at some sampling frequencies than at others. I just think it would be great to know at what sampling frequency any given DAC is optimized so as to really enjoy the benefits of this DAC.
To this end I have also been looking at DSD DACs (which IMHO may very well be a superb conversion approach) and I have actually been put off by the energy- and computer requirements of HQPlayer in DSD mode. I would, as fedde touches upon, appreciate if HQPlayer (consumer) also included the feature of performing upsampling in non real-time mode ...
However, as I understand it upsampling PCM signals requires a much less powerful computer - basically most all computers should be able to do this without being overly challenged - which - in my perspective - is good news. It means that I may then e.g. optimize my DAC for 384 kHz or 768 kHz playback and then enjoy the benefits of the upsampling - or, when really high res recordings become available (like 384 kHz - 1.536 MHz - or similar DSD formats) enjoy the native files at these frequencies. Threading carefully here - and not wishing to initiate a discussion about this - IMHO 44.1 kHz can be improved upon ...
@Ken Newton:
A fine sentence I think
... However, as you might also have observed, lurking around the audio realm are deep lakes of relatively unknowns which only occasionally are touched upon. In my audio perspective one of these "deep lakes" has been, & may to some extent still be, the reasons why NOS and OS perceptively may sound differently. Yet, having read parts of this thread the depths of this particular lake has become more transparent - which is a fine accomplishment ...
Well, will end here ... Have a good day ;-)
Jesper
Hi ... I actually was just mentioning this as I consider the "phenomenon" that DACs don't perform equally well at the various sampling frequencies to be an important parameter that is not often addressed. As it is I personally believe in the sound benefits of higher sampling rates, however, most - if not all - DACs, due to various circuitry compromises, likely perform better at some sampling frequencies than at others. I just think it would be great to know at what sampling frequency any given DAC is optimized so as to really enjoy the benefits of this DAC.
To this end I have also been looking at DSD DACs (which IMHO may very well be a superb conversion approach) and I have actually been put off by the energy- and computer requirements of HQPlayer in DSD mode. I would, as fedde touches upon, appreciate if HQPlayer (consumer) also included the feature of performing upsampling in non real-time mode ...
However, as I understand it upsampling PCM signals requires a much less powerful computer - basically most all computers should be able to do this without being overly challenged - which - in my perspective - is good news. It means that I may then e.g. optimize my DAC for 384 kHz or 768 kHz playback and then enjoy the benefits of the upsampling - or, when really high res recordings become available (like 384 kHz - 1.536 MHz - or similar DSD formats) enjoy the native files at these frequencies. Threading carefully here - and not wishing to initiate a discussion about this - IMHO 44.1 kHz can be improved upon ...
@Ken Newton:
A fine sentence I think
... However, as you might also have observed, lurking around the audio realm are deep lakes of relatively unknowns which only occasionally are touched upon. In my audio perspective one of these "deep lakes" has been, & may to some extent still be, the reasons why NOS and OS perceptively may sound differently. Yet, having read parts of this thread the depths of this particular lake has become more transparent - which is a fine accomplishment ... Well, will end here ... Have a good day ;-)
Jesper
This kind of response doesn't bring anything to discussion, it just shows arrogance coupled with inferior superiority syndrome.
You have been participating positively, and with restraint of late. No need to take it over the top now. Maybe, attempt to better explain your point, instead.
@Extreme_Boky:
Hi ... I actually was just mentioning this as I consider the "phenomenon" that DACs don't perform equally well at the various sampling frequencies to be an important parameter that is not often addressed. As it is I personally believe in the sound benefits of higher sampling rates, however, most - if not all - DACs, due to various circuitry compromises, likely perform better at some sampling frequencies than at others. I just think it would be great to know at what sampling frequency any given DAC is optimized so as to really enjoy the benefits of this DAC.
To this end I have also been looking at DSD DACs (which IMHO may very well be a superb conversion approach) and I have actually been put off by the energy- and computer requirements of HQPlayer in DSD mode. I would, as fedde touches upon, appreciate if HQPlayer (consumer) also included the feature of performing upsampling in non real-time mode ...
However, as I understand it upsampling PCM signals requires a much less powerful computer - basically most all computers should be able to do this without being overly challenged - which - in my perspective - is good news. It means that I may then e.g. optimize my DAC for 384 kHz or 768 kHz playback and then enjoy the benefits of the upsampling - or, when really high res recordings become available (like 384 kHz - 1.536 MHz - or similar DSD formats) enjoy the native files at these frequencies. Threading carefully here - and not wishing to initiate a discussion about this - IMHO 44.1 kHz can be improved upon ...
@Ken Newton:
A fine sentence I think
Well, will end here ... Have a good day ;-)
Jesper
Arriving at a very low impedance over a wide bandwidth is art. Jeff Zhu (Holo Audio DAC) mentioned that he spent years playing with decoupling capacitors, layouts and tracks' lengths/routes - to get this right.
Marcel,
I saw a sentence in an old ADC App. Note that caught my eye. It seemed to suggest that dither can transform jitter spurs. to noise, just as it does to distortion harmonics. I don’t believe that I’ve read that notion anywhere before. Perhaps, I’ve misunderstood what was being suggested. Do you know anything about whether that’s true or not?
@Extreme_Boky:
Hi ... I actually was just mentioning this as I consider the "phenomenon" that DACs don't perform equally well at the various sampling frequencies to be an important parameter that is not often addressed. As it is I personally believe in the sound benefits of higher sampling rates, however, most - if not all - DACs, due to various circuitry compromises, likely perform better at some sampling frequencies than at others. I just think it would be great to know at what sampling frequency any given DAC is optimized so as to really enjoy the benefits of this DAC.
To this end I have also been looking at DSD DACs (which IMHO may very well be a superb conversion approach) and I have actually been put off by the energy- and computer requirements of HQPlayer in DSD mode. I would, as fedde touches upon, appreciate if HQPlayer (consumer) also included the feature of performing upsampling in non real-time mode ...
However, as I understand it upsampling PCM signals requires a much less powerful computer - basically most all computers should be able to do this without being overly challenged - which - in my perspective - is good news. It means that I may then e.g. optimize my DAC for 384 kHz or 768 kHz playback and then enjoy the benefits of the upsampling - or, when really high res recordings become available (like 384 kHz - 1.536 MHz - or similar DSD formats) enjoy the native files at these frequencies. Threading carefully here - and not wishing to initiate a discussion about this - IMHO 44.1 kHz can be improved upon ...
@Ken Newton:
A fine sentence I think
Well, will end here ... Have a good day ;-)
Jesper
Any reasonably modern computer (Raspberry Pi 4 or stronger, 64-bit FPU) can give you real-time resampling with a 160dB+ noise floor, but you need to actually use the CPU. You just have to provide the taps in an unrestricted floating point algorithm. DSD is largely unnecessary in the modern world. Perhaps it had benefits when 25MHz was considered significant, but these days it's easily traded with equivalent PCM, e.g. DSD512 should be no better than your current PCM audio on an ESS DAC. Modern sigma-delta DACs work similar to the DSD principle with generally a few more bits, maybe 4 or 5. The age of single-bit sigma-delta converters, which would be true to DSD, and their spurs/harmonics, are over. Their inherent linearity is not worth the downsides when a 2nd-3rd order sigma delta ADC is easy to design, and a 4th-5th order sigma delta DAC is also easy to design thanks to modern digital simulation.
With sigma delta, and oversampled SAR/R2R/etc. converters, the trouble always comes back to filtering. How well is the input data filtered during the conversion from PCM to "DSD-like" data or, for both, interpolated oversampled data? How many taps in the filter? Is it a simple digital sinc-in-frequency (SincX) filter? Well, that's not nearly good enough because an ideal filter is sinc-in-time, which is much more computationally expensive to approach, and perhaps impossible to reach (causality issues).
This is the same problem that early CD player designers had. Referencing the SAA7220 - it was a great achievement of the era! But 120 taps of an FIR filter is absolutely pathetic these days. I could easily drop a 10000 or 100000 tap FIR filter on a digital audio file in real-time (obviously delayed, but still streaming) now. 120 taps is nowhere -near- being sinc-in-time, and likely sounds bad. Note that chip's datasheet has -zero- information about its added THD/SNR/etc. There's no wonder so many people replaced or removed it and had sonic improvements - running without its significant distortion was better, even with the increased distortion from running filterless (speaker/amp/preamp-filtered) or low-order filtered NOS.
A DSD DAC is simple. It is a low-pass filter - ideal pulse density modulation can perfectly recreate a signal with an adequate x-order analog filter. It can be good or it can be bad, but DSD audio itself (on the ADC side) suffers from the problems of single-bit sigma delta modulation. The only cure is faster and faster sample rates and the hopes that spurs don't occur in the passband. Oversampled PCM or at least multi-bit sigma delta is a much better approach to solve the issue of the simple fact that a sinc-in-time analog antialiasing filter is impossible, which would be necessary for 44.1kHz audio to reproduce a 22.05kHz bandwidth and not fold distortion.
Last edited:
I think it is better to start with R2R DAC in real NOS mode (avoiding those using error scrambling techniques). Then read this post about natural sound properties in test files: https://www.diyaudio.com/forums/digital-line-level/371931-makes-nos-sound-108.html#post6731260
If you attend live performances, you will want these properties to be present in your living room. In result of my post #1075 further samples were used deliberately to not carry these properties in source material, I also heard a comment that this particular sample was defective. .LOL.
On my R2R-11 there are no benefits of upsampling, even after upgrading to ultra-low jitter oscilators. There is more difference in sound quality between night and day (i.e. due to the power overload) than between these files. On one side it is a proof that resampling was a good quality, on the other it is not worth of increased storage and wasting Internet bandwith.
In summary, the outcome of these tests is to play in a native format and don't bother with resampling. If you have a Delta-Sigma DAC, different analogue chain, your experience can be different.
I just came across a posting with this unconventional Denon Alpha Dac with a continuously switching sharp/slow FIR filter, depending on the incoming signal.
No idea what problem should be solved with this solution, and obviously they didn't conquer the market with it.
But always nice to have a glimpse on what's being tried, possibly leading to new ideas.
The Black Hole......
Hans
P.S. Alpha stands for Adaptive Line Pattern Harmonized Algorithm.
No idea what problem should be solved with this solution, and obviously they didn't conquer the market with it.
But always nice to have a glimpse on what's being tried, possibly leading to new ideas.
The Black Hole......
Hans
P.S. Alpha stands for Adaptive Line Pattern Harmonized Algorithm.
Last edited: