• 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). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

Minimum Phase vs Linear Phase

scott wurcer

Active Member
Audio Luminary
Technical Expert
Joined
Apr 24, 2019
Messages
142
Likes
174
The confusing thing is that the chip makers never clearly state how many bits a DAC is effectively working with, in terms of resolution. It could actually well be that even the lowly AKM4490 already puts out more than 24bits (noise-removed) resolution when fed with 32-bit data (only few, if any at all, DAC frontends supply 32bit data to the DAC chip proper, via I²S. Does anyone know a DAC that does?).
I was thinking of asking the same question, does feeding 32bit integer data to any DAC give a performance enhancement? As for Windows, in normal use, I see no evidence that there is ever anymore than 24bits transferred (8 LSB's are always 0), that is if even 24bits are used.

The heavily averaged synchronous measurements are good for test instrumentation but as you said not for real time audio.
 

DonH56

Technical Expert
Technical Expert
Patreon Donor
Joined
Mar 15, 2016
Messages
2,990
Likes
3,559
Location
Monument, CO
IME averaging can greatly reduce the noise floor but does not help fundamental linearity. That is, the noise floor can be reduced, but nonlinearity-induced spurs (from myriad sources) are not amenable to averaging. The distortion is correlated so averaging does not help. SNR should improve, SINAD might, THD nay, SFDR nay. Another thing often overlooked is that the front end of a delta-sigma ADC must still have full linearity, as must the output threshold and all output buffers after a D-S DAC. There is still an analog piece at the very end of a data converter (one end or the other). When you are talking 120+ dB everything matters, including things like resistor voltage and temperature coefficients otherwise neglected. I spent a lot of time working with Vishay years ago to make a special "0-TC" resistor load for a 16-bit, 100 MS/s DAC that needed essentially "ideal" linearity and settling with <10 ppm tempco. My six-week test schedule ballooned to six months but I did meet spec after inventing a few test schemes. Hope I don't have to do that again anytime soon...

That said I don't know anything about how Windows etc. handles the data. In the past for instrumentation we had to get special drivers to handle wide data as Windows would truncate in some cases and in others zero-pad to a fixed bit length; neither were desirable for the application.
 

maty

Major Contributor
Joined
Dec 12, 2017
Messages
1,360
Likes
553
Location
Tarragona (Spain)
http://help.izotope.com/docs/ozone/pages/modules_paragraphic_eq.htm

[ Selecting Phase

Ozone now offers filter phase control when in digital mode with three selectable phase modes:

Minimum phase filters that have a transient response similar to analog EQs in that most of the ringing is concentrated after the transient.

Linear phase filters that maintain symmetric response, meaning equal pre- and post-ringing, as is often characteristic of digital EQs.

Mixed phase mode allows each individual band to have a phase response varying between minimum phase (-1), linear phase (0), and maximum phase (+1). Maximum phase filters concentrate all ringing before the transient where it is most audible.



In mixed phase mode the phase response of a given node can be adjusted with vertical handles that appear on a selected node when Mixed phase mode is selected. These behave much like the horizontal handles which affect Q/bandwidth. Alternatively, phase response values may be entered manually in the Show Info table. Different phase responses will often produce quite subtle sonic differences unless you are doing very steep filtering. ]

https://splice.com/blog/mastering-with-ozone-8-equalization/

1. Try not cut or boost over 3dB

The general rule here is that if you have to cut or boost a frequency band over 3dB, it means there’s most likely a problem with the mix. Remember that mastering is all about subtle touches that eventually add up to a polished, professional sound...
There are very interesting info in the site.
 

MRC01

Senior Member
Joined
Feb 5, 2019
Messages
463
Likes
340
The WM8741 has several different filters, each with documentation. In the linked doc, look at pages 37, then 50 with graphs starting on page 54. Is this typical of other DAC chips? With so many choices built in, how do designers decide which to use?

For example, with this chip I'd imagine you'd want OSR high (max oversampling rate), then filter 3 would be the standard "sharp" filter and filter 4 would be min phase. But there are other options. My DAC has a "sharp" vs. "slow" setting, but I don't know exactly which of these 5 different filters it selects.
 

MRC01

Senior Member
Joined
Feb 5, 2019
Messages
463
Likes
340
I read some detailed info there but didn't see an actual answer to Maty's question why 24/192 sounds different. I'll offer one: re-released high-res recordings are often remastered. This guarantees they will sound different. But those differences aren't necessarily coming from the higher bit rate.
EDIT: he does mention this in the middle of the linked doc. But it can be hard to find, buried in all the other information. So I've highlighted it here.

One simple/easy test you can make: take the 24/192 and downsample to 16/44 using Audacity or similar software. Make sure you use the highest quality settings, dither, etc. Now you have a file where you know the ONLY difference is the bit rate. Then ABX test this file against the 24/192.
 
Last edited:

maty

Major Contributor
Joined
Dec 12, 2017
Messages
1,360
Likes
553
Location
Tarragona (Spain)
Almost everything I hear in my second system (the first one is usually kidnapped) are rips of vinyl, CD and SACD made from analog master. And the FM radio (classical music).

I know many years ago about the scams, you have to be very careful and inform yourself well before clicking. Many times they are upsamplers from the CD, not even from the master. HDtracks is a site to avoid.

It is the easiest test because true 24/192 sounds unreal. More difficult test is 24/48 vs 24/96. 16/44 and 16/48 are worse than 24/96, at least from very good vinyl recordings. I save a few 24/192 but I play them on my second system using SoX 24/96, but my problem is I do not like the sound of their implementation in JRMC :(

I am reading the long article... until 16 vs 24. I have not found anything that explains why I do not like to play / listen to 24/192 FLAC.
 
Last edited:

maty

Major Contributor
Joined
Dec 12, 2017
Messages
1,360
Likes
553
Location
Tarragona (Spain)
One simple/easy test you can make: take the 24/192 and downsample to 16/44 using Audacity or similar software. Make sure you use the highest quality settings, dither, etc. Now you have a file where you know the ONLY difference is the bit rate. Then ABX test this file against the 24/192.
Many years ago I opened a thread and I upload a lot of files from the original 24/192 WAV, but is not in the open Internet. With SoX from foobar2000, bandwith at 95%.

https://www.google.com/search?q=foobar2000+sox+maty+nauscopio -> Google Images

Old picture

 
Last edited:

MRC01

Senior Member
Joined
Feb 5, 2019
Messages
463
Likes
340
... I have not found anything that explains why I do not like to play / listen to 24/192 FLAC.
When you say unreal, do you mean "good", or "bad". That is, figuratively, or literally?

Some of this is getting confusing. If recordings are 16 and 24 bit, what is the benefit of a 32 bit DAC?
AFAIK, a 32-bit DAC is not a benefit just a different format for storing the same data. 24-bit implies integer, but it can be encoded as a 32-bit floating point, which typically uses 23 bits for the value and 8 bits for the exponent (plus the sign bit makes 32). Most computers align data to words, not bytes, so a 24-bit value takes up 32 bits in memory when it's word aligned.
The most linear DAC we've seen here (that Amir has measured) has been to about 120 dB which is 20 bits. So I've never seen a DAC having 32 bit resolution. I'm not an EE but it seems impossible to achieve that resolution in a DAC, since a DAC by definition has an analog output and this goes below the thermal noise of metal film resistors.
 

maty

Major Contributor
Joined
Dec 12, 2017
Messages
1,360
Likes
553
Location
Tarragona (Spain)
Unreal = bad.

It is like I am inside an empty cathedral (by the way, the acoustics of the Tarragona's cathedral are horrible: my memory of how bad the Haendel's Messiah sounded, with the cathedral full, still lingers).

It is too late, tomorrow will not be festive as it has been today in Catalonia (Spain, EU). Bye.
 
Last edited:

scott wurcer

Active Member
Audio Luminary
Technical Expert
Joined
Apr 24, 2019
Messages
142
Likes
174
So I've never seen a DAC having 32 bit resolution. I'm not an EE but it seems impossible to achieve that resolution in a DAC, since a DAC by definition has an analog output and this goes below the thermal noise of metal film resistors.
I lost the graph but someone presented the quantum mechanical limit for sampling frequency vs bits. We are not there but I don't think 32 bits at 96k is provably impossible.
 

Julf

Addicted to Fun and Learning
Joined
Mar 1, 2016
Messages
549
Likes
386
Location
Amsterdam, The Netherlands
Almost everything I hear in my second system (the first one is usually kidnapped) are rips of vinyl, CD and SACD made from analog master. And the FM radio (classical music).
So all recorded more than 30 years ago?
 

MRC01

Senior Member
Joined
Feb 5, 2019
Messages
463
Likes
340
I lost the graph but someone presented the quantum mechanical limit for sampling frequency vs bits. We are not there but I don't think 32 bits at 96k is provably impossible.
Sure, 32-bit is certainly possible on the DSP side. I'm questioning whether it makes any improvement to the analog output. Analog audio electronics don't get much better than about 120 dB, and achieving even that is a stretch. And that's already close to the theoretical noise limits of the opamps, resistors and other components. Amir's fancy AP rig can do what: -130 dB? So even the best analog signal chain is not even 24-bit resolution, let alone 32.
 

scott wurcer

Active Member
Audio Luminary
Technical Expert
Joined
Apr 24, 2019
Messages
142
Likes
174
Sure, 32-bit is certainly possible on the DSP side. I'm questioning whether it makes any improvement to the analog output.
Not the point, I'm talking about the purely intellectual exercise of what is possible here. There is a limit based on first principles for what can be done.
 

MRC01

Senior Member
Joined
Feb 5, 2019
Messages
463
Likes
340
I understand that wasn't your point, but it was my point ;)
I agree the digital/information/math limits can be interesting. But as long as we're recording sounds using microphones, we bump into analog limits long before we reach digital limits.
 

mansr

Addicted to Fun and Learning
Joined
Oct 5, 2018
Messages
508
Likes
1,191
Location
Hampshire
AFAIK, a 32-bit DAC is not a benefit just a different format for storing the same data. 24-bit implies integer, but it can be encoded as a 32-bit floating point, which typically uses 23 bits for the value and 8 bits for the exponent (plus the sign bit makes 32). Most computers align data to words, not bytes, so a 24-bit value takes up 32 bits in memory when it's word aligned.
Quite a few DAC chips will accept 32-bit integer samples and do their best to reproduce them. Almost all chips will accept 32-bit input and ignore the low 8-16 bits. I am not aware of any DAC chip supporting floating-point input.

Software, on the other hand, often uses 32-bit floating-point with 23-bit (24 if counting the implicit leading 1) precision. The advantage over integer formats is that the dynamic range is vastly expanded. A signal can be attenuated by several hundred dB and amplified back without loss of precision. For storage or distribution this doesn't matter, but it simplifies processing in a DAW. Similarly, sample values can exceed the 1.0 reference level without clipping provided a suitable overall adjustment is applied before exporting to an integer format. Again, useful in production, not as a final format.
 

maty

Major Contributor
Joined
Dec 12, 2017
Messages
1,360
Likes
553
Location
Tarragona (Spain)
Fast post

About the bits, in the very great SACD plugin from foobar2000 I ALWAYS choose Direct 64fp, 30 kHz LPF -> 64-bit floating point and low pass at 30 kHz to cut the distortion.

foobar2000-sacd-dsd2pcm-direct-64fp-30khz.png


It was a pain, the SACD ISO sounded better since foobar2000. Many months ago I got or something changed and since JRMC they sound better.

At 88200 because the cheap/tweaked ODAC USB works up to 24/88-96.

And the limit, for now, is the analog section. And, probably, that the last chip needs to be heated to reach its optimum performance, as seems to happen with the ESS ES9038Pro in the Sabaj D5, about 30 minutes.
 

maty

Major Contributor
Joined
Dec 12, 2017
Messages
1,360
Likes
553
Location
Tarragona (Spain)
Top Bottom