Extreme filtering software upscaling

[Ittaku]

4 minutes ago, eltech said:

Just one question.

Do you listen to your filters or just measure them?

I did mention that I abx tested them earlier in the thread to scientifically confirm my sighted listening. Hence why I couldn't hear anything different by 500m. 15m was the cutoff. So now yes I listen to tons of ripped CD music that I've upscaled at 30m taps...

[eltech]

1 minute ago, Ittaku said:

I did mention that I abx tested them earlier in the thread to scientifically confirm my sighted listening. Hence why I couldn't hear anything different by 500m. 15m was the cutoff. So now yes I listen to tons of ripped CD music that I've upscaled at 30m taps...

That's great. So your ABX tests confirm that you prefer a certain sort of sound? It sounds very scientific.

[Ittaku]

8 minutes ago, eltech said:

That's great. So your ABX tests confirm that you prefer a certain sort of sound? It sounds very scientific.

Yes it is. I do sighted listening first to see if I think I can hear a difference. If I can't I leave it there. Many accessories fall into that category. If I think I can tell a difference, then I use a (linux application) called abx which does a classic abx test where you can listen to files a and b separately and then x is a random one and you have to pick if it's a or b. Similar applications exist for windows I'm sure.

Basic principles:

https://en.wikipedia.org/wiki/ABX_test

The basic application info for linux (though it's just a package for ubuntu and easy to install):

http://phintsan.kapsi.fi/abx.html

[eltech]

You must be very happy to discover what you like. I'm happy for you.

[davewantsmoore]

On 09/02/2019 at 8:05 AM, Ittaku said:

Here is a blind test comparing linear vs minimum phase filters

The results say that people cannot tell the difference between the default minimum/linear SoX.

 

The more experienced listeners (assuming musicians/producers are more experienced) were more aware that they could not tell the difference.

 

Thoughts?!

[davewantsmoore]

20 hours ago, eltech said:

Actually, this test has been done and the improved sound quality persists.

It is not possible.

Something else is responsible for what they are hearing.

 

When we copies CDs using different equipment .... the 1 and 0 stored on the different CDs are the same .... BUT, when we examine the CDs.... Yes, each CD is different.    The pits/lands length, depth, clarity, etc... are different.   So there is a mechanism by which the CDs could sound different.

 

... but when putting the information inside files .... we can then compare the files and see that they are literally identical.    So unlike the CD, no mechanism exists which could make the files playback differently from each other.

 

[legend]

I have been trying to see if I can do the extreme processing myself as I used to do quite a lot of computer programming in the 1960s and 1970s – but things have moved on a bit since then!  However it is an interesting learning exercise to try – and to see if I can save the $7500 on an M-Scaler.

 

Slightly off-topic (it does involve high levels of upsampling, though not 'extreme' ones and has the complication of DSD) I have also been trying upsampled DSD files into my Chord Qutest DAC.  I noticed the other day quite by accident that my Amarra Luxe program automatically upsamples dsf files to 352k - rather than ‘just’ 176k as with PCM files.  I have no idea of the upsampling regime of Amarra Plus (whether it is frequency or time based and the number of taps/coefficients etc used as I can find nothing on the ‘net) but the sound from the 352k DSD files seems very like that gotten from the extreme upsampling and from the M-Scaler ie a sense of ‘ease’ with detail and speed/foot-tapping plus real sense of instruments in space.  These are only general impressions as I can’t do a direct comparison because I don’t have access to an M-Scaler and  I don’t have a DSD file of the Telarc Kamen track I have been using for the extreme upsampling listening – in retrospect I should have chosen the CD layer version of an SACD for which I do have a dsf file to send to @Ittaku for upsampling.

 

 I don't want to get into a DSD vs PCM war - I am not even sure whether Amarra converts DSD to PCM as its output says "352800 DSD(PCM)".  And I realise the DSD/dsf files have a head-start of 88k rather than 44k as with CD/Redbook ones before upsampling.  So it maybe just straws in the wind!

[eltech]

24 minutes ago, davewantsmoore said:

It is not possible.

Something else is responsible for what they are hearing.

 

When we copies CDs using different equipment .... the 1 and 0 stored on the different CDs are the same .... BUT, when we examine the CDs.... Yes, each CD is different.    The pits/lands length, depth, clarity, etc... are different.   So there is a mechanism by which the CDs could sound different.

 

... but when putting the information inside files .... we can then compare the files and see that they are literally identical.    So unlike the CD, no mechanism exists which could make the files playback differently from each other.

 

The thing is Dave out of respect for the OP and his topic, 

I repeatedly ask for any discussion to go to PM, but for the record I know, and you should know that I know what digital theory says. I'm well and truly aware of it. I'm pretty sure anybody here in reading this thread would be aware of it. I repeatet I am aware of it.

You you have not provided me with an education. All you've done is created further distraction. so I feel that you didn't actually want to discuss it with me, you just wanted to belittle me, or make yourself look more intelligent. If I'm wrong I'm very sorry, but from where I sit that's what it looks like. For the benefit of anyone else reading this if you've got something on this particular topic to say to me please take it to PM. And for the record I will not be replying to any further comments, or reply to anybody quoting me in this thread, and if anyone wants to set out to give me an education, all it's really saying is that they don't want learn something and they don't want to have an open mind. If you think the world is black and white and everything's been figured out, then that's great for you.  

But don't bother sending me a p.m. to give me an education about something I already know. There are plenty of things in this world that people don't understand. We can see the pyramids of Egypt, people have theorized about how they were built, but nobody actually knows how they were built. There are plenty of other unexplained mysteries in this world. The phenomena described in the article and observed by myself and other people I know including well respected international mastering engineers, is one of those mysteries. I don't care one iota if you believe it or not. I don't want to debate it this with people who haven't experienced it and people who want to just blindly dismiss it. I have better things to do. For the last time I know what digital theory says.

Cheers

All the best,

[rand129678]

15 minutes ago, legend said:

I have also been trying upsampled DSD files into my Chord Qutest DAC

Not sure if you're aware (many people aren't) but all of Rob's DAC's except Chord Dave decimate DSD... i.e the FPGA converts DSD to PCM internally...

 

https://www.head-fi.org/threads/chord-electronics-hugo-2-the-official-thread.831345/page-651#post-13950774

 

With Chord Dave, there is DSD+ mode which plays DSD without decimation.

Edited February 10, 2019 by Music2496

[davewantsmoore]

2 hours ago, eltech said:

You you have not provided me with an education.

There are many other people observing the discussion, who may derive benefit.

2 hours ago, eltech said:

All you've done is created further distraction.

I like to think I "ended" one, as opposed to created one.....  I may be moderately delusional about that, of course.

2 hours ago, eltech said:

you just wanted to belittle me

Not at all.   It isn't personal... or about bringing anyone down..... It is about providing context/balance to others reading along.

2 hours ago, eltech said:

If you think the world is black and white and everything's been figured out, then that's great for you.

This works in both directions.   If we think everything is black/white, when it's not (and it's true, lots and lots, if not most, things are not that simple)  .... then you'll struggle .....  BUT if you ignore things that ARE actually black/white .... then, you'll also struggle.

2 hours ago, eltech said:

But don't bother sending me a p.m.

No, I wouldn't do that.... as nobody else would see it, avoiding all the potential benefit of posting it.

2 hours ago, eltech said:

I don't care one iota if you believe it or not.

Good.   My "belief" or otherwise is a very poor indicator.   Facts exist no matter how we feel about them.

[davewantsmoore]

2 hours ago, Music2496 said:

With Chord Dave, there is DSD+ mode which plays DSD without decimation.

I wasn't aware of this.   Do you have a link which explains this clearly?    I assume that means the digital signal runs at the original rate?

[Nada]

15 hours ago, Ittaku said:

Yes it is. I do sighted listening first to see if I think I can hear a difference. If I can't I leave it there. Many accessories fall into that category. If I think I can tell a difference, then I use a (linux application) called abx which does a classic abx test where you can listen to files a and b separately and then x is a random one and you have to pick if it's a or b. Similar applications exist for windows I'm sure.

Basic principles:

https://en.wikipedia.org/wiki/ABX_test

The basic application info for linux (though it's just a package for ubuntu and easy to install):

http://phintsan.kapsi.fi/abx.html

I wonder if you might have the capacity to load up some short AB samples on line please? For me 30 second samples of something acoustic eg female vocal, strings and percussion would be best for discriminating. For fun don't identify the 15M tapped version and we can guess if you suggest what to listen for.

 

Id  really like to have a listen to your work and  use your cunning ABX method.

[rand129678]

19 minutes ago, davewantsmoore said:

I wasn't aware of this.   Do you have a link which explains this clearly?    I assume that means the digital signal runs at the original rate?

Sure, happy to trade sources

 

https://www.stereo.net.au/forums/topic/261591-extreme-filtering-software-upscaling/?do=findComment&comment=4060641

 

and
 

https://www.stereo.net.au/forums/topic/260051-chord-hugo-m-scaler-discussion-and-impressions-thread/?do=findComment&comment=4053223

 

[davewantsmoore]

2 hours ago, Music2496 said:

Sure, happy to trade sources

For the AV8805, the discussion around there actually being an analogue only path, is quite common.  The AVS thread on th AV8805 is full of it (as lots of people want to confirm they are avoiding A>D>A conversion).

 

As for the actual details, I can't find the Japanese site I had found wth the details  (I actually think I can find the site in my history, but the blog page is gone).   It had photos of the circuit board(s).   I want to say it was an 8 channel Cirrus Logic resistor IC..... but I might be thinking of something else.

 

 

For the 2L quote.... It's gone (it was on www.2l.no/hires) ..... but I see it's been replaced by this:  ttp://www.2l.no/hires/documentation/2L-MQA_Comparisons.pdf  .... which actually talks about 18-bits for an example, although that is only on high frequencies (so it's particularly significant, IMVHO).     I'd like to actually understand what this was, that they recorded so far down, and what microphone they used to do that.

[Ittaku]

7 hours ago, davewantsmoore said:

The results say that people cannot tell the difference between the default minimum/linear SoX.

 

The more experienced listeners (assuming musicians/producers are more experienced) were more aware that they could not tell the difference.

 

Thoughts?!

The results showed trends and statistically significant results in precisely one combination only - headphones and the piano playback. There were many many other variables added in addition to just the choice of filter, but it serves to test what general application of different filters means across the board with massively different hardware. To me it confirms a number of things that I've suspected.

• The choice of filter has only a subtle effect at best and concentrating on it as a big part of the quality of a DAC is not really purposeful, though may give some indication of the sort of sonic signature it might have.
• Sox's filters are a cut above the rest even with their default values.
• Wild variation in hardware playback means that the subtle effects of filtering differences may be irrelevant compared to other hardware limitations.
• There is no "correct" filter for everything - at least with the normal length filters we've been using till now.
• Perhaps approaching the infinite length filters is the way to make the most of these filters and we've been avoiding doing so till now for many many practical reasons and limitations.
• Further research is always required, appreciated, and... unlikely

[rand129678]

13 minutes ago, Ittaku said:

The choice of filter has only a subtle effect at best and concentrating on it as a big part of the quality of a DAC is not really purposeful, though may give some indication of the sort of sonic signature it might have.

 

13 minutes ago, Ittaku said:

Wild variation in hardware playback means that the subtle effects of filtering differences may be irrelevant com

I shared this earlier - from one of the most respected DAC designers who obviously has a very good understanding of both the digital side AND the analogue side (he's designed DACs and ADCs and amps for the pro world and HiFi gear of course), the late Charles Hansen of Ayre Acoustics:

 

https://www.computeraudiophile.com/forums/topic/35106-how-does-a-perfect-dac-analog-signal-look-different-than-a-cheap-dac/?page=7&tab=comments#comment-713189

 

Digital filters is down the list... not that the topic (like this thread) is not interesting.

 

But for example legend has been comparing Qutest+M-Scaler against MQA on the Pro-Ject S2 DAC but it's not a fair comparison, comparing the different analogue section and power supply section designs of these DACs... even putting aside the price differences...

 

As Charles Hansen described elsewhere but hints in the above post, these differences between DACs (different analogue section and power supply section designs) can swamp any differences in digital filter designs between the same DACs...

 

 

[Ittaku]

10 minutes ago, Music2496 said:

 

I shared this earlier - from one of the most respected DAC designers who obviously has a very good understanding of both the digital side AND the analogue side (he's designed DACs and ADCs and amps for the pro world and HiFi gear of course), the late Charles Hansen of Ayre Acoustics:

 

https://www.computeraudiophile.com/forums/topic/35106-how-does-a-perfect-dac-analog-signal-look-different-than-a-cheap-dac/?page=7&tab=comments#comment-713189

 

Digital filters is down the list... not that the topic (like this thread) is not interesting.

 

But for example legend has been comparing Qutest+M-Scaler against MQA on the Pro-Ject S2 DAC but it's not a fair comparison, comparing the different analogue section and power supply section designs of these DACs... even putting aside the price differences...

 

As Charles Hansen described elsewhere but hints in the above post, these differences between DACs (different analogue section and power supply section designs) can swamp any differences in digital filter designs between the same DACs...

Indeed, it's a very good summary and consistent with what I believe. I said in the opening post that the effect is subtle, and it's purely a simple way to optimise/tweak something we have control over. Bear in mind that in all my auditioning here, I'm listening to changes on a rather expensive system and DAC that has put a lot of effort and expense into addressing the "more important" things in his list.

[Ittaku]

4 hours ago, Nada said:

I wonder if you might have the capacity to load up some short AB samples on line please? For me 30 second samples of something acoustic eg female vocal, strings and percussion would be best for discriminating. For fun don't identify the 15M tapped version and we can guess if you suggest what to listen for.

 

Id  really like to have a listen to your work and  use your cunning ABX method.

See your private messages.

[rand129678]

18 minutes ago, Ittaku said:

Indeed, it's a very good summary and consistent with what I believe. I said in the opening post that the effect is subtle, and it's purely a simple way to optimise/tweak something we have control over. Bear in mind that in all my auditioning here, I'm listening to changes on a rather expensive system and DAC that has put a lot of effort and expense into addressing the "more important" things in his list.

The way you have been comparing differences makes good sense technically because the digital filtering is the only variable. Your analogue electronics remains constant in these comparisons and that's very important if you want to subjectively conclude anything in comparing digital filters.

 

But legend comparing MQA on the Pro-Ject S2 DAC against Qutest+M-Scaler throws in too many of the variables Charles Hansen describes above. Not to discount your experimenting @legend! I have an S2 DAC too. 

 

An interesting exercise with the excellent valued S2 DAC is to compare MQA on it vs feeding it PCM768kHz using different filters, thereby by-passing the internal oversampling...

 

With HQPlayer I can feed the S2 DAC with a loooong linear phase 1 million tap poly phase sinc filter and compare with MQA which is on the other end of the spectrum (so to speak), as well as a bunch of filters in between these extremes... all 'on the fly'. Doing it this way keeps the analogue electronics constant and that's very important if you want to subjectively conclude anything in comparing digital filters.

Edited February 10, 2019 by Music2496

[davewantsmoore]

6 minutes ago, Music2496 said:

An interesting exercise with the excellent valued S2 DAC is to compare MQA on it vs feeding it PCM768kHz using different filters, thereby by-passing the internal oversampling...

That doesn't bypass the internal oversampling of your DAC ... inside the DAC chip it resamples everything to a rate approaching 1.536mhz.

 

 

Designers using the chip, can program it so the internal resampling is bypassed (but they have to resample the audio themselves) ..... or, the designer can reprogram the oversampling filters ..... but my understanding is neither of these things are commonly done using the ESS chips.

[rand129678]

7 minutes ago, davewantsmoore said:

inside the DAC chip it resamples everything to a rate approaching 1.536mhz.

Again... a source? For the Pro-Ject S2 DAC specifically...

 

I'm not talking about an Auralic DAC here...

 

 

 

Edited February 10, 2019 by Music2496

[davewantsmoore]

16 minutes ago, Music2496 said:

Again... a source? For the Pro-Ject S2 DAC and the ESS9038Q2M John Westlake chose to use specifically...

The DAC is a SD modulator type of converter.  That's how they work.

 

 

http://file2.dzsc.com/product/18/05/25/829029_170233543.pdf 

 

You can see in the block diagram there are two resamplers.   Oversampling Filter, and "ASRC/Jitter Reduction".

 

You will read that the first can be bypassed (although not by the end-user).  Even if the designer did that,  my understanding is they would need oversample the data themselves outside the chip.

 

The second one ASRC/Jitter .... is where signal is converted up to ~1.5mhz and then fed into their SD modulator, which operates again, much faster.

[frednork]

@Ittaku @Music2496 @eltech @davewantsmoore  and others not on this page, just wanted to say I am really enjoying this thread,

Appreciate all your inputs.

[rand129678]

1 hour ago, davewantsmoore said:

The DAC is a SD modulator type of converter.  That's how they work.

This datasheet doesn't say everything about how things are implemented in the Pro-Ject S2 DAC and in every DAC using this chip. 

 

I've had discussions with John Westlake the designer of this DAC 

 

Just like Benchmark, Auralic and Ayre Acoustics do things differently using ESS Sabre chips...

 

ESS Sabre chips have always been a bit of a black box in some aspects.

 

Anyway, more important is that feeding this DAC PCM768kHz moves all images very very high above audio band and removes a lot of the DSP to outside the chip (a source of RF and IMD). As Jussi explains Laako (HQPlayer) explains: "there are always images at multiples of the sampling rate. Higher the rate, higher those images go in frequency and further apart from each other."

 

Moving the heavy lifting of WTA1 (which outputs PCM705kHz/768kHz) to outside  Hugo2/Qutest/Hugo TT2/Dave is one of the biggest benefits the M-Scaler brings to these DACs. Connecting these dots (Rob Watts quotes):

 

1. M-Scaler outputs 16FS (PCM705/768kHz), by-passing Dave/HTT2/Hugo2/Qutest's WTA 1 filtering stage.

 

2. Rob Watts on his DACs WTA 1 filtering stage:  "WTA 1 (getting to 16FS) uses 90% of the FPGA DSP, and most of the FPGA fabric."

 

3. Rob Watts on M-Scaler: "The RF noise that the FPGA generates is a nightmare; it's 12A of correlated current with large amounts of 2GHz noise. In the long term I would like to integrate an M scaler with a DAC; but I have not been able to figure out how to do it without it compromising sound quality."

 

Similar thinking to feeding DACs maximum input sample rate, in this case PCM768kHz for the Pro-Ject S2 DAC. Jussi Laako (HQPlayer developer) has measured this DAC and it objectively measures best fed PCM768/DSD512, which both measure practically the same... 

 

 

As initially mentioned, it's a nice DAC, when fed PCM768k for comparing MQA with, say, very long, million tap, linear poly phase sinc filters. And a bunch of filters that fit 'in between' these two extremes, as Jussi explains.

 

 

Edited February 10, 2019 by Music2496

[legend]

11 hours ago, Music2496 said:

The way you have been comparing differences makes good sense technically because the digital filtering is the only variable. Your analogue electronics remains constant in these comparisons and that's very important if you want to subjectively conclude anything in comparing digital filters.

 

But legend comparing MQA on the Pro-Ject S2 DAC against Qutest+M-Scaler throws in too many of the variables Charles Hansen describes above. Not to discount your experimenting @legend! I have an S2 DAC too. 

 

An interesting exercise with the excellent valued S2 DAC is to compare MQA on it vs feeding it PCM768kHz using different filters, thereby by-passing the internal oversampling...

 

With HQPlayer I can feed the S2 DAC with a loooong linear phase 1 million tap poly phase sinc filter and compare with MQA which is on the other end of the spectrum (so to speak), as well as a bunch of filters in between these extremes... all 'on the fly'. Doing it this way keeps the analogue electronics constant and that's very important if you want to subjectively conclude anything in comparing digital filters.

I am a physicist by training so am EXTREMELY aware of the reductionist approach of changing one variable at a time!

 

However on the Briggs-Myer test I am also a INTJ who uses intuition and helicoptering rather than the typical ISTJ of engineer/scientist that likes to obsess about detail.

 

Hence I like to explore the edges of topics where often the more interesting connections occur, even if it does conflict with the first point.

 

And having been involved in audio design for 25 years or more I can also get the feel of whether something is better or not without DBT testing changing one variable at a time  - though of course the latter is necessary to be scientifically rigorous/certain!

 

I think the main point I was trying to make in the comparison of the Project DAC S2 DAC plus MQA and the Chord Qutest plus M-Scaler/extreme filtering was that for me they pointed in the same direction/type of improved sound quality (but of course varied in the extent due not least to cost differences which obviously affect technical implementation) and that this MAY have been due to their common emphasis on the importance of timing/impulse responses.