• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). There are daily reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

DacMagicPlus Filter Mirror Images - Short Study

chris719

Senior Member
Joined
Mar 22, 2019
Messages
371
Likes
414
I agree, resampling from 44.1kHz to 96kHz with a proper re-sampler and resampling filters is the way to get better results and usually better sound as well ;).

The challenge here is that there is no off-the-shelf ASRC with a stopband that begins at Fs/2. Every single chip I looked at (SRC4192/SRC4392, AD1896, CS8421, AK4137) has a stopband starting at 0.5465*FsIn except the AK4137 which is 0.5417*FsIn. This means that it can only be fixed using an external DSP or FPGA or by software in a PC.

Nearly every single IC from major manufacturers have this magic 0.5465*Fs stopband and it's been that way for a very long time.
 

bennetng

Major Contributor
Joined
Nov 15, 2017
Messages
1,562
Likes
1,570
The challenge here is that there is no off-the-shelf ASRC with a stopband that begins at Fs/2. Every single chip I looked at (SRC4192/SRC4392, AD1896, CS8421, AK4137) has a stopband starting at 0.5465*FsIn except the AK4137 which is 0.5417*FsIn. This means that it can only be fixed using an external DSP or FPGA or by software in a PC.

Nearly every single IC from major manufacturers have this magic 0.5465*Fs stopband and it's been that way for a very long time.
I have a Roland interface purchased in 2001 with a CS8420 and the SRC quality/jitter suppression is not all that satisfactory. Even the DSP chip on my Creative gaming soundcard has better performance, and it can do far more than 2 channels of SRC as well.
https://www.audiosciencereview.com/forum/index.php?threads/digital-distortion.2059/post-55543
https://www.audiosciencereview.com/forum/index.php?threads/digital-distortion.2059/post-55659
The plots above are partially limited by my old Audition's FFT display accuracy, but still clearly demonstrated the differences.

The different amount of skirting is clearly measurable in the Roland's analog output as well, among all other nasty stuff.
https://www.audiosciencereview.com/...r-s-pdif-digital-audio-filter.2189/post-59502

All tests were done using the same Toslink cable.

Given the result that CS43198 has great SPDIF jitter performance and 10dB better attenuation than AK449x when both of them are using the sharpest filters I don't know what I can further complain. The amount of noise shaping is comparable to DSD256 and one can see much stronger shaping in 2L's DXD demo tracks.
https://www.audiosciencereview.com/...es-dsd-sound-better-than-pcm.5700/post-128714
 

chris719

Senior Member
Joined
Mar 22, 2019
Messages
371
Likes
414
I have a Roland interface purchased in 2001 with a CS8420 and the SRC quality/jitter suppression is not all that satisfactory. Even the DSP chip on my Creative gaming soundcard has better performance, and it can do far more than 2 channels of SRC as well.
https://www.audiosciencereview.com/forum/index.php?threads/digital-distortion.2059/post-55543
https://www.audiosciencereview.com/forum/index.php?threads/digital-distortion.2059/post-55659
The plots above are partially limited by my old Audition's FFT display accuracy, but still clearly demonstrated the differences.

The different amount of skirting is clearly measurable in the Roland's analog output as well, among all other nasty stuff.
https://www.audiosciencereview.com/...r-s-pdif-digital-audio-filter.2189/post-59502

All tests were done using the same Toslink cable.

Given the result that CS43198 has great SPDIF jitter performance and 10dB better attenuation than AK449x when both of them are using the sharpest filters I don't know what I can further complain. The amount of noise shaping is comparable to DSD256 and one can see much stronger shaping in 2L's DXD demo tracks.
https://www.audiosciencereview.com/...es-dsd-sound-better-than-pcm.5700/post-128714

Yeah, CS8420 was one of the earlier generation devices, it's not very good. It's blown away by SRC4392.

I wouldn't necessarily say CS43198 has great SPDIF jitter performance technically, since it doesn't have an SPDIF interface. Whatever is giving Topping good numbers in the D30Pro, it's the SPDIF receiver and/or an ASRC.

I brought those into the discussion because they would be a quick, low-power fix for a designer if they had filters that entered the stopband at Fs/2.
 

bennetng

Major Contributor
Joined
Nov 15, 2017
Messages
1,562
Likes
1,570
Redo the Realtek tests with max C-state set to C3 (UEFI auto default) rather than forcing C7 to show cleaner results. Disabling other C-states won't help and the CPU will be much hotter, so don't do that. Newer motherboards should have this fixed (mine is 6 years old with $60 power supply).
48.png


Additional test. The same file played through Firefox with 2x sample rate to activate Windows SRC. No one cares about this anyway LOL.
48-96.png


A harder one. 44.1k to 96k with Windows SRC. The tone is scaled proportionally (21.13125kHz) to have an apple to apple comparison.
44-96.png


Related crosspost:
https://www.audiosciencereview.com/...u-m2-review-audio-interface.19911/post-674665
 
Last edited:

AudioSceptic

Major Contributor
Joined
Jul 31, 2019
Messages
2,094
Likes
1,935
Location
Northampton, UK
Not because I see it, but because it is the right way to do so. To do so in order to achieve proper signal reconstruction. And it is not my problem what someone hears or not. I want the technical things to be done properly and I do not like excuses like “it does not matter”.

For example in case of these filters all of them are wrong.

View attachment 112098
Quite. Why do none of them even start rolling off until at least 21k? Again, these don't look like filters for 44.1k sampling.
 

AudioSceptic

Major Contributor
Joined
Jul 31, 2019
Messages
2,094
Likes
1,935
Location
Northampton, UK
Is there any source that would show D30 filters response at @48kHz sampling rate? Amir's measurement is at @44.1kHz, so we cannot tell the filter response @48kHz based on Amir's measurement at @44.1kHz.

I agree, resampling from 44.1kHz to 96kHz with a proper re-sampler and resampling filters is the way to get better results and usually better sound as well ;).
I've never seen the need for 96k (and 192 is just crazy and wasteful), but wouldn't 64k make a lot of sense? Not hugely wasteful, but would allow complete suppression by Fs/2 (32k) with a slow, easy rolloff from 20k.
 

Mnyb

Major Contributor
Forum Donor
Joined
Aug 14, 2019
Messages
1,484
Likes
1,946
Location
Sweden, Västerås
I've never seen the need for 96k (and 192 is just crazy and wasteful), but wouldn't 64k make a lot of sense? Not hugely wasteful, but would allow complete suppression by Fs/2 (32k) with a slow, easy rolloff from 20k.

We are talking resampling on your computer or other device on the fly ? the files may still be 16/44.1 thats what I do ? it wont tax usb or ethernet much to send 24/96 maybe you should not use 24/192 with an older wifi endpoint
 

bennetng

Major Contributor
Joined
Nov 15, 2017
Messages
1,562
Likes
1,570

KSTR

Major Contributor
Joined
Sep 6, 2018
Messages
2,190
Likes
4,809
Location
Berlin, Germany

AudioSceptic

Major Contributor
Joined
Jul 31, 2019
Messages
2,094
Likes
1,935
Location
Northampton, UK
This is a great thread (all your threads are awesome, btw).



I draw the opposite conclusions from your results here. Your results show that, at worst, the amount of IMD caused by aliases in the passband with the "slow" filters is more than about -125 dB down -- i.e., definitely inaudible. You don't show the time domain measurements, but the steep filter necessarily has about double the amount of pre-ringing. There is debate about how much pre-ringing is audible, but to the extent it may be audible, it's more likely to be audible than IMD at -125 dB down. (When you hear increased presence with the sharp filter, I'd argue what you're hearing is a psychoacoustic effect of the pre-ringing.)

A lot of people on this forum (not you) don't seem to understand that you can't have good attenuation at Fs/2 on 44.1kHz sampled data while preserving 20kHz bandwidth without also having considerably more pre-ringing. You can't have your cake and eat it too because of the way the math works. The reason DAC designers from all of the major IC vendors have generally settled on filters with a 0.54Fs stopband isn't because they're incompetent -- it's because they realize the inherent tradeoff in time-domain performance. For telecom applications or higher sample rates, Fs/2 stopband is obviously ideal, but at 44.1Khz for audio applications there is no ideal solution, and I'd even go so far as to say that Fs/2 is more "wrong".
Fs/2 is correct. The problem is the small "window" between that and the assumed 20k upper limit of our hearing, a legacy of Red Book. But why do the lazy filters start rolling off as late as 21-22k instead of 20?

You also get extreme "solutions" like Ayre's where the rolloff starts before 10k, all in the name of improved time domain performance, but which are you hearing, the rolled off HF or the time domain improvements? <https://www.stereophile.com/content/ayre-acoustics-qb-9-usb-dac-measurements>
 

AudioSceptic

Major Contributor
Joined
Jul 31, 2019
Messages
2,094
Likes
1,935
Location
Northampton, UK
We are talking resampling on your computer or other device on the fly ? the files may still be 16/44.1 thats what I do ? it wont tax usb or ethernet much to send 24/96 maybe you should not use 24/192 with an older wifi endpoint
Sorry, I wasn't clear. I'm not interested in resampling when I play something. I just think that Red Book was a great attempt with limited tech at the time, but should have been replaced long ago with 64k sampling (or maybe even 48k as that is the standard for audio in digital video). 48k makes filtering above 20k a bit easier (!), but 64 would a *lot* easier.
 

chris719

Senior Member
Joined
Mar 22, 2019
Messages
371
Likes
414
Sorry, I wasn't clear. I'm not interested in resampling when I play something. I just think that Red Book was a great attempt with limited tech at the time, but should have been replaced long ago with 64k sampling (or maybe even 48k as that is the standard for audio in digital video). 48k makes filtering above 20k a bit easier (!), but 64 would a *lot* easier.

The problem might still exist I think. Marketing would insist on using the full Nyquist bandwidth and designing to be flat to 32k, someone would give in, and the rest of the industry would cave to match it.
 

AudioSceptic

Major Contributor
Joined
Jul 31, 2019
Messages
2,094
Likes
1,935
Location
Northampton, UK
The problem might still exist I think. Marketing would insist on using the full Nyquist bandwidth and designing to be flat to 32k, someone would give in, and the rest of the industry would cave to match it.
Yes, that's always a possibility. I'd like to think that sanity would prevail, but perhaps hi-res already proves me wrong.
 

restorer-john

Master Contributor
Joined
Mar 1, 2018
Messages
9,574
Likes
27,199
Location
Gold Coast, Queensland, Australia
Not because I see it, but because it is the right way to do so. To do so in order to achieve proper signal reconstruction. And it is not my problem what someone hears or not. I want the technical things to be done properly and I do not like excuses like “it does not matter”.

This, from the man who designed the first digital tape recorder for NHK and wrote the (actual) entire book on digital. Heitaro Nakajima.

Sure, IIR filters, hand adjusted and expensive.
scan519.jpg


scan521.jpg


scan520.jpg
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
15,885
Likes
26,661
Sorry, I wasn't clear. I'm not interested in resampling when I play something. I just think that Red Book was a great attempt with limited tech at the time, but should have been replaced long ago with 64k sampling (or maybe even 48k as that is the standard for audio in digital video). 48k makes filtering above 20k a bit easier (!), but 64 would a *lot* easier.
Yeah, J_J wanted 64 khz sampling with bandwidth to 28 khz or so. Lavry made/make their gear to 96 khz, but filter at I think 30 or 35 khz.
 

pos

Addicted to Fun and Learning
Forum Donor
Joined
Feb 13, 2018
Messages
550
Likes
673
I know and have designed a proper filter 2 days ago. This is out of the question/topic.
I assume you are talking about the new filter for the D30Pro here?
Any curve (measured or simulated) you could share, at 44.1 as well as maybe 48, 96, and 192?
 

chris719

Senior Member
Joined
Mar 22, 2019
Messages
371
Likes
414
I assume you are talking about the new filter for the D30Pro here?
Any curve (measured or simulated) you could share, at 44.1 as well as maybe 48, 96, and 192?

I though he was speaking globally but maybe not. There’s no free lunch with the integrated filters; if it’s just custom coefficients then I’m pretty sure any earlier attenuation is going to trade-off other parameters given a fixed number of taps.
 

chris719

Senior Member
Joined
Mar 22, 2019
Messages
371
Likes
414
This, from the man who designed the first digital tape recorder for NHK and wrote the (actual) entire book on digital. Heitaro Nakajima.

Sure, IIR filters, hand adjusted and expensive.
View attachment 113020

View attachment 113021

View attachment 113022

A 13th order passive Chebyshev filter with ferrite core inductors is probably more damaging to the audio than allowing some imaging. Just because he wrote “the book” doesn’t really mean much. It’s not like he invented sampling theorem.

It is very clear to me that given that 99.9% of all DACs have filters allowing some images through, there is no real issue here. The few products that do it correctly aren’t even universally preferred among subjectivists.
 
Top Bottom