Jump to content
IGNORED

DSD Frustrations With Manufacturers


Recommended Posts

3 minutes ago, jabbr said:

 

I gather than when converting from say 24/96 PCM to say 44.1x256 DSD that certain filters need less processing when staying within rate family, yet I have not heard that the final result depends on staying within rate family. For chip families such as ESS which operate internally at ?100Mhz we aren't staying with a multiple of either 44.1 nor 48Khz regardless.

Some filters are also 2x, 4x, etc... for example and there's only one way to do that :~)

Founder of Audiophile Style | My Audio Systems AudiophileStyleStickerWhite2.0.png AudiophileStyleStickerWhite7.1.4.png

Link to comment
18 minutes ago, The Computer Audiophile said:

Some filters are also 2x, 4x, etc... for example and there's only one way to do that :~)

 

Well with all the ones I've tried with HQPlayer, it works, albeit with different processing, and I presume that ESS also uses a method that doesn't depend on SDM being an integer multiple rate of input (i.e. its all 100Mhz, right?)

Custom room treatments for headphone users.

Link to comment
49 minutes ago, The Computer Audiophile said:

Thanks Ryan.

 

This is more about using apps like HQPlayer to upsample content, using different filters and modulators, to DSD256 and keeping the sample rate as a multiple of the original. I assume the QB-9 upsamples internally to a multiple of the base rate. This is very similar just done outside the DAC and enables the user to select what he prefers sonically.

 

Hey Chris, I see.  To be honest, we've never been big proponents of upsampling outside of the FPGA, where we can do everything in a single pass to minimize rounding errors as well as ensure that the original data is preserved and not rounded out by some software post-process.  We only can guarantee what WE do internally, so this gives us a bit more control on making sure the end product is as true to the original as possible.  I'm sure you already know how we tend to think of DSD based on posts from Charley over the years, but I still think that converting a 192kHz sample to DSD is a fair argument for a 48kHz-based DSD rate being available if you're really set on doing it.  Let me discuss that aspect with Ariel.  It's just a matter of adding a line to the code and figuring out if there's a good way to display what it's doing with the characters available on the display.  

 

41 minutes ago, jabbr said:

I gather than when converting from say 24/96 PCM to say 44.1x256 DSD that certain filters need less processing when staying within rate family, yet I have not heard that the final result depends on staying within rate family. For chip families such as ESS which operate internally at ?100Mhz we aren't staying with a multiple of either 44.1 nor 48Khz regardless.

 

Well, it CAN operate at 100MHz.  I wouldn't ever recommend using it that way.

President

Ayre Acoustics, Inc.

Link to comment
1 minute ago, Ryan Berry said:

Hey Chris, I see.  To be honest, we've never been big proponents of upsampling outside of the FPGA, where we can do everything in a single pass to minimize rounding errors as well as ensure that the original data is preserved and not rounded out by some software post-process.  We only can guarantee what WE do internally, so this gives us a bit more control on making sure the end product is as true to the original as possible.  I'm sure you already know how we tend to think of DSD based on posts from Charley over the years, but I still think that converting a 192kHz sample to DSD is a fair argument for a 48kHz-based DSD rate being available if you're really set on doing it.  Let me discuss that aspect with Ariel.  It's just a matter of adding a line to the code and figuring out if there's a good way to display what it's doing with the characters available on the display.  

Thanks for the clear reply. I certainly hear you Ryan. 

 

I love the ability to make my own choice, in essence to have control over what I do when I want to do it, and to turn control over to you when I want. Thus, using HQPlayer when I want to and using the Ayre FPGA when I want. Choice is wonderful :~)

 

I highly recommend checking out @Miska's HQPlayer. I won't speak for him but I'm guessing that anything that's done in an FPGA can also be done with his software. The cool thing is that its got endless power because it runs on full blown PCs. 

 

I'd say leaving the display at 256 no matter if the playback is based on 44.1 or 48k is cool. 

 

Thanks again!

 

 

 

Founder of Audiophile Style | My Audio Systems AudiophileStyleStickerWhite2.0.png AudiophileStyleStickerWhite7.1.4.png

Link to comment
10 minutes ago, Ryan Berry said:

Hey Chris, I see.  To be honest, we've never been big proponents of upsampling outside of the FPGA, where we can do everything in a single pass to minimize rounding errors as well as ensure that the original data is preserved and not rounded out by some software post-process.

 

HQPlayer can do 1024x upsampling in a single pass, and on top of that run digital room correction filters and such with millions of taps.

 

Nice thing also is that since software processing runs asynchronously from any sample clocks, it can monitor the output and and also upcoming future input data, and re-process data based on decisions, while still meeting delivery deadlines. This is possible because modern CPUs can run at 5 GHz clock speeds. And modern GPUs, like my Nvidia RTX2080Ti with it's 18.6 billion transistors can do massive amount of DSP operations as well. All at very reasonable cost.

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
Just now, The Computer Audiophile said:

Thanks for the clear reply. I certainly hear you Ryan. 

 

I love the ability to make my own choice, in essence to have control over what I do when I want to do it, and to turn control over to you when I want. Thus, using HQPlayer when I want to and using the Ayre FPGA when I want. Choice is wonderful :~)

 

I highly recommend checking out @Miska's HQPlayer. I won't speak for him but I'm guessing that anything that's done in an FPGA can also be done with his software. The cool thing is that its got endless power because it runs on full blown PCs. 

 

I'd say leaving the display at 256 no matter if the playback is based on 44.1 or 48k is cool. 

 

Thanks again!

 

 

 

 

I know HQPlayer pretty well, we've used it in the past for testing and agree that it's a nice piece of software.  However, you will still run into a second pass of oversampling being done at the FPGA to get to 16X and apply our Minimum Phase filter, so you'll run into what I mentioned before.  

President

Ayre Acoustics, Inc.

Link to comment
5 minutes ago, Ryan Berry said:

 

I know HQPlayer pretty well, we've used it in the past for testing and agree that it's a nice piece of software.  However, you will still run into a second pass of oversampling being done at the FPGA to get to 16X and apply our Minimum Phase filter, so you'll run into what I mentioned before.  

HQ Player (HQP) is best suited for DACs that either explicitly allow the user to bypass further internal processing or DACs where input at the highest allowed frequency automatically bypasses further internal processing otherwise there will be some double processing and double filtering. At that point it gets difficult to tell what is contributing what and the end result may even be worse. This is especially a problem with PCM where double filtering can create some weird results.

 

Also, as a manufacturer, I would be suspicious of something like HQP as it would allow the user to mess with the "house sound" which is mostly in proprietary filtering and processing. This is why so few DACs allow users to bypass processing. 

 

I have a NOS capable DAC that I use with HQP and another that I can't use with HQP because of these considerations.

Link to comment
48 minutes ago, Ryan Berry said:

 

Well, it CAN operate at 100MHz.  I wouldn't ever recommend using it that way.

 

Thanks, I don't claim to know how the ESS chips work internally. Regarding DSD @ 48k multiples, I'm really just trying to understand if there is a significant benefit with 48k x based PCM source conversion. I can do it either way on my system. My brief testing seemed to indicate that there was no benefit for me regsrding "SQ" ... but I am interested in learning.

Custom room treatments for headphone users.

Link to comment
47 minutes ago, Ryan Berry said:

I know HQPlayer pretty well, we've used it in the past for testing and agree that it's a nice piece of software.  However, you will still run into a second pass of oversampling being done at the FPGA to get to 16X and apply our Minimum Phase filter, so you'll run into what I mentioned before.  

 

With DSD inputs too? Or does DSD inputs bypass your processing and pass through to the DAC chip?

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
10 minutes ago, Ryan Berry said:

 

That's a big topic full of a lot of opinions.  Some groups studied could tell no difference while a study in Tokyo suggested people could hear the differences between DSD and PCM with a preference toward DSD (though no differences heard between different rates of DSD, interestingly).  I've not heard much benefit of doing a conversion to DSD myself and I know that Charley really didn't care for the format, but that was probably more on principle than anything.  Either way, I tell people to trust their own ears and just enjoy the music.  The argument on what is right or wrong would go on endlessly and we'd all be sitting here debating on forums instead of listening to music.

Interesting studies. I would love to know if for the PCM playback multi-bit hardware was used. For true PCM rendering, the DAC must be able to process a 24 bit word and assign it a specific impulse value. R2R ladder DACs are the classic example but there are others. If PCM material is fed to any one bit chip DAC, it first has to be converted to a bitstream (DSD) and then modulated. So what one is really hearing is the chip manufacturers version of DSD playback not true PCM playback. Any difference heard when using a one bit chip DAC for both PCM and DSD material is by definition likely to be slight.

Link to comment
1 hour ago, Miska said:

Nice thing also is that since software processing runs asynchronously from any sample clocks, it can monitor the output and and also upcoming future input data, and re-process data based on decisions, while still meeting delivery deadlines. This is possible because modern CPUs can run at 5 GHz clock speeds. And modern GPUs, like my Nvidia RTX2080Ti with it's 18.6 billion transistors can do massive amount of DSP operations as well. All at very reasonable cost.

 

Being sure that I am interpreting this correctly ... if software processing runs asynchronously from any sample clocks, then what is the dependence of the final output rate family on the input rate family? (aside from some additional processing which is offset by a slightly different output rate)

 

Custom room treatments for headphone users.

Link to comment
4 minutes ago, jabbr said:

Being sure that I am interpreting this correctly ... if software processing runs asynchronously from any sample clocks, then what is the dependence of the final output rate family on the input rate family? (aside from some additional processing which is offset by a slightly different output rate)

 

Dependency is mathematical. Processing is clocked by the CPU and GPU clocks which are not related to input or output clocks.

 

This is totally different from typical synchronous process you have in a DAC for example.

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
23 minutes ago, Miska said:

 

With DSD inputs too? Or does DSD inputs bypass your processing and pass through to the DAC chip?

 

 

That's a fair point that I wasn't thinking about in DSD terms, Miska.  Good catch.  DSD passes directly through to the ESS untouched, so it wouldn't be subject to another round of oversampling like pre-upsampling PCM would.  It doesn't eliminate some of our other issues with the format as a whole, but you're right that it wouldn't be subject to double-oversampling.

President

Ayre Acoustics, Inc.

Link to comment

An easy way that I’ve always looked at the sample rate family issue is to consider how it works when downsampling. 
 

A recording done at 88.2 can easily be made ready for CD at 44.1 by removing every other sample. No math involved. 
 

Granted this is going the opposite way (down vs up sampling), but it’s easy to comprehend. 

Founder of Audiophile Style | My Audio Systems AudiophileStyleStickerWhite2.0.png AudiophileStyleStickerWhite7.1.4.png

Link to comment
52 minutes ago, bobflood said:

Interesting studies. I would love to know if for the PCM playback multi-bit hardware was used. For true PCM rendering, the DAC must be able to process a 24 bit word and assign it a specific impulse value. R2R ladder DACs are the classic example but there are others. If PCM material is fed to any one bit chip DAC, it first has to be converted to a bitstream (DSD) and then modulated. So what one is really hearing is the chip manufacturers version of DSD playback not true PCM playback. Any difference heard when using a one bit chip DAC for both PCM and DSD material is by definition likely to be slight.

 

The more I read, the more questions I end up with.  Are the files being tested conversions of a PCM file or rerecorded with a DSD ADC?  What if the ADC being used for PCM had a filter like the QA-9 instead of a brick wall filter?  How is the DSD vs. PCM data handled inside the DAC?  The list goes on and on and every test is likely flawed in some way.

President

Ayre Acoustics, Inc.

Link to comment
17 minutes ago, Miska said:

 

Without proper decimation filter that would give you plenty of aliasing, as all the content between 22.05 and 44.1 kHz would appear in the 0 - 22.05 kHz band with inverse spectrum.

 

Same way, you could upsample by copying every sample twice, but without proper interpolation filter this would again produce images of content from 0 - 22.05 kHz band between 22.05 and 44.1 kHz again with inverse spectrum. Note that most DAC chips use this method at and above 8x rates.

 

Absolutely. Mine was an errant oversimplification. 

Founder of Audiophile Style | My Audio Systems AudiophileStyleStickerWhite2.0.png AudiophileStyleStickerWhite7.1.4.png

Link to comment
1 hour ago, The Computer Audiophile said:

An easy way that I’ve always looked at the sample rate family issue is to consider how it works when downsampling. 
 

A recording done at 88.2 can easily be made ready for CD at 44.1 by removing every other sample. No math involved. 
 

Granted this is going the opposite way (down vs up sampling), but it’s easy to comprehend. 


 

Not taking into account aliasing nor filters, and what I’m offering below is purely my way of thinking about the issue as I’ve never written conversion software:

 

Consider that a PCM sample is a number, that relates to the intended ouput voltage, and consider that the DSD is a pattern of bits that average out to the intended voltage: at each sample in time a stream of bits could be inserted in the output stream. The number of bits inserted in the stream is determined by the output rate conversion factor and need not be an even multiple nor even be the same from sample to sample. In this case 24/96 is converted to 44.1x256, for example — of course this isn’t how it actually works but conceptually the input and output rate families are crossed without a (big) problem :handwave:

Custom room treatments for headphone users.

Link to comment

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now



×
×
  • Create New...