• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

What am I missing - up-sampling and DSD conversion?

MaxwellsEq

Major Contributor
Forum Donor
Joined
Aug 18, 2020
Messages
2,001
Likes
3,049
I see people up-sampling 44k/16 PCM CD quality source files to, e.g. 96k or 192k. Or even converting 44k/16 PCM files to DSD. But I don't get it, what's the benefit?

PCM up sampling
I can see one possible benefit from up-sampling to 192k, that any in-band artefacts from the playback DAC reconstruction filter will be well above the audible range, but I'm not convinced these artefacts are audible when no up-sampling has taken place. But otherwise, there's no content benefit, since the ADC brickwall is part of the content.

The downsides seems obvious to me as well - larger files containing no new content and a risk that the conversion process actually adds artefacts!

Converting to DSD
PCM to DSD conversion is not the same as writing an analogue audio feed directly to DSD format. So I don't see how the PCM takes on some magical DSD characteristic. I can see that changing the noise shape may have some impact, but the source material is constrained by 16 bits. There's also the reconstruction filter justification, I suppose.

I've also read that some DACs are measurably better at DSD than PCM, but are these differences audible, given the source is 44k/16?

Again, the downsides are bigger files and the risk that the conversion process adds artefacts.

What am I missing? Should I start converting to DSD right away? (BTW my DACs are ADI-2 Pro FS (AKM) and Sabaj A10d 2022(ESS)).
 
software up-sampling could in theory do a better job than what your dac does internally.
 
software up-sampling could in theory do a better job than what your dac does internally.
Thanks - that's the sort of thing I'm missing.

Since both of my DACs measurably perform at, or close to perfection when playing 44k/16 PCM, what measurably is better after software conversion?

Also, there's quite a lot of evidence of SRC being done badly (e.g. issues with Windows), so is there not a balancing risk that the up-sampling or DSD conversion measures worse or has new artefacts? Presumably I have to be very careful about which tools to use?

But if I do pick a perfect up-sampler, there will be a benefit, presumably because the reconstruction filter artefacts are now ultrasonic?
 
Thanks - that's the sort of thing I'm missing.

Since both of my DACs measurably perform at, or close to perfection when playing 44k/16 PCM, what measurably is better after software conversion?

Also, there's quite a lot of evidence of SRC being done badly (e.g. issues with Windows), so is there not a balancing risk that the up-sampling or DSD conversion measures worse or has new artefacts? Presumably I have to be very careful about which tools to use?

But if I do pick a perfect up-sampler, there will be a benefit, presumably because the reconstruction filter artefacts are now ultrasonic?
Just a random results fro Amirs' measurements:

1711794036057.png



If you do your own up-sampling you can do better.
More stop band attenuation, less ripple in the pass band, your choice how fast/slow the filter is etc..
Question indeed is: audible or not?
 
But if I do pick a perfect up-sampler, there will be a benefit, presumably because the reconstruction filter artefacts are now ultrasonic?
Amir himself showed a quick small example where software upsampling improved DAC output.

He made a typo here writing 92kHz - I believe he upsampled to 96kHz or 192kHz

index.php


Poor filter past Nyquist:

index.php
 
Amir himself showed a quick small example where software upsampling improved DAC output.

He made a typo here writing 92kHz - I believe he upsampled to 96kHz or 192kHz

index.php


Poor filter past Nyquist:

index.php
Thanks!

What I'm not sure with this is if Amir is using native 192kHz source signals from his AP to demonstrate that this DAC behaves better with 192kHz or if he is up-sampling tests from 44kHz/16 to 192kHz and demonstrating a benefit. I have to assume the former.

But, of course that doesn't mean that the benefit is not there for up-sampled content as well.

Both of my DACs have SOTA filters, so I'm not sure yet if I'd get a benefit, assuming I found a flawless up-sampler.
 
What I'm not sure with this is if Amir is using native 192kHz source signals from his AP to demonstrate that this DAC behaves better with 192kHz or if he is up-sampling tests from 44kHz/16 to 192kHz and demonstrating a benefit. I have to assume the former
No in the review of that DAC he says he used Roon to upsample.
 
There is plenty of discussion in this recent topic:


Essentially what you do is what the DAC already does for you (upsample, convert to delta sigma). Yes, in some cases, PC algorithms may show a slight improvement in objective performance, but audibility has never been proven. In some cases, differences are more than slight. In all of those cases, the DAC fails to properly apply filtering. Just don't buy those products... @Music1969 showed one of those, Here is another, Marantz SA10:
index.php

Reaching 1% THD at 7 kHz due to its poor filtering:
index.php

:facepalm:
The Fosi is curious since the ES9038Q2M usually performs quite a bit better than this.
 
Last edited:
What are the 2 DACs exactly ?
As per my 1st post, the main system is an RME ADI-2 Pro FS R Be, which has an AKM AK4493; my computer DAC is a Sabaj A10d with 2 x ESS ES9038Q2M. Both have measurements that exceed human audibility and feature default filters with minimal in-band artefacts and excellent stop-band rejection.
 
There is plenty of discussion in this recent topic:


Essentially what you do is what the DAC already does for you (upsample, convert to delta sigma). Yes, in some cases, PC algorithms may show a slight improvement in objective performance, but audibility has never been proven. In some cases, differences are more than slight. In all of those cases, the DAC fails to properly apply filtering. Just don't buy those products... @Music1969 showed one of those, Here is another, Marantz SA10:
index.php

Reaching 1% THD at 7 kHz due to its poor filtering:
index.php

:facepalm:
The Fosi is curious since the ES9038Q2M usually performs quite a bit better than this.
Thanks for your response and the link to the other thread. I hadn't realised that DSD1024 was a thing (albeit not very useful).

The graphs you show are genuinely flawed. I'm not sure I would hear any benefit of effectively offline conversation using PC algorithms to higher sample rates or DSD given I have an ADI-2 Pro FS R Be and Sabaj A10d
 
Here's 25 pages. If you find something, let us know :)
Excellent, I'll have a good read. Before I start does it cover DSD?
 
I'm not sure I would hear any benefit of effectively offline conversation using PC algorithms to higher sample rates or DSD given I have an ADI-2 Pro FS R Be and Sabaj A10d
I’m sure you would not ;)
 
I’m sure you would not ;)
It's not that simple and there are even some benefits that could be even hear able (if it fails under let's say 80 dB SINAD) in case of very low (50 mV and lower) outputs (for sensitive IEM's and/or headphones and listening very loud while they have very low THD in mids/uper mids) and it's not about sample rate but because you lose in most cases 6 dB output (half the V) in case of DSD while SINAD remains almost the same in case of DSD 128 and in audible range or you get a higher bit depth converting to 32 bit integer on today's DAC's so you lose less lowering the output. DSP manipulation benefits from 64 bit FP which then afterwards transparently scale to 32 (16, 24) bit integer. You don't need more than 21 bit integer SINAD (120 dB) in any case as that's the limit of our hearing sensitivity in mids and uper mids where it's most sensible and most data is anyway (main tones, and most of 2nd and 3rd order harmonics). Needless to say THD of analog transcivers (drivers) will be much higher.
There is only one PCM to DSD conversion implementation writen in current use which is SMP 4 optimised and very CPU intensive, much more than let's say complex 64 bit integer tool chain (PEQ's, effects, FIR, resampling and all together from which you can benefit greatly disreging of use case) and back to 32 bit integer so I won't recommend it in any case.
 
Last edited:
Have someone else rename these files to different colors or animals and hide the key and then give them a listen on headphones. It's free. Warning these are big files so download can be slow.
 
Here's 25 pages. If you find something, let us know :)
OK, I've read that. Quite a lot of noise and heat and limited amount of light. Some good points raised, though.
 
Answering my own question (naughty I know):

Up-sampling my 44kHz/16 bit files using a PC and software e.g. to 192kHz cannot add any new data unless there is some interpolation algorithm applied. But it may allow me to use complicated algorithms unavailable to a real-time hardware DAC which is trying to add as little delay as possible.

Such a 192kHz file has no new information, but for DACs with poor/no filtering, it may have a positive audible and measurable impact, since some artefacts are higher in frequency. If I used some terribly dumb NOS DAC that can process 192kHz, I may hear a difference. I've no interest in owning such a ridiculous design!

BUT, instead, I have two top-class DeltaSigma DACs which already use oversampling internally. So I can NOT see any point where there will be a measurable or audible difference to turning my Red Book files in 192kHz. I'm therefore confused as to why people with good DACs bother. OK, so there may be an interpolation algorithm which fixes some consequences of a what I think is too low an fs (I'd have insisted on at least 48kHz, but that's because I was in pro audio!), but nobody seems to have pointed me to what this magic PCM-cleany-uppy software is!

What I don't fully grok is converting Red Book to DSD. An analogue source feeding a DSD ADC I get, but converting PCM to DSD gives none of the theoretical DSD benefits as far as I can see. BUT - there does seem to be some view that some DeltaSigma DACs have an architecture better at handling DSD than PCM. Given the extraordinary measurements for my ADI-2 Pro FS R Be and Sabaj A10d, I'm still not yet convinced this will make a measurable difference or one I can hear.

Time is money and so is storage space. As far as I can tell, leaving my Red Book files as 44kHz/16 FLAC is the most pragmatic approach

Thanks for everyone's input!
 
To simply it for you if DAC has 65~70 dB SINAD at 30~40 mV output PCM which is 5~6 dB higher at 2x mV and you use DSD 128 conversion which cuts the mV output by half and retains SINAD on very good high sensitivity (SPL per mV) low impedance IEM's/headphones that are also very low THD you will be able alright to hear the difference. If it's a great DAC retaining that in high 80"s to 90 on that output level you won't. Spoiler there are a lot of such higher end IEM's and headphones but not that much DAC's that can hold so good there.
Will hopefully tomorrow order a pair of such (92 dB SPL @ 35 mV 23 Ohms) which I desired for a very long time as I finally catched them on discount and from a dealer that I trust enough (Denon AH-D5200).
 
Back
Top Bottom