HQ player settings with iFI Micro DSD

[Dyson]

This is what the data sheet has:

[ATTACH=CONFIG]19333[/ATTACH]

I believe the data sheet is correct...

 

What iFi says applies only to PCM1795...

 

In any case, that 6-bit part has only marginal effect. It makes up 0.02% of full-scale signal values if input is 24-bit and 0.0001% of full-scale signal if input is 32-bit. And the DAC becomes pure delta sigma for levels below -36 dB. I think main reason for the 6-bit part is to make up for the low order modulator to achieve wanted SNR. Maybe they didn't have DSP resources to put higher order modulator there.

 

Funny, from the figure and the explanation that follows in the data sheet I understand the opposite, the 6 upper bits are the MSB and contribute for 99.8% to the output signal (true multi bit conversion, not converted). That's why they can call it "true" PCM conversion even if a part is done with sigma delta conversion (but counting for only 0.02% in the final signal). Or am I wrong on how I read it?

[Miska]

Funny, from the figure and the explanation that follows in the data sheet I understand the opposite, the 6 upper bits are the MSB and contribute for 99.8% to the output signal (true multi bit conversion, not converted). That's why they can call it "true" PCM conversion even if a part is done with sigma delta conversion (but counting for only 0.02% in the final signal). Or am I wrong on how I read it?

 

If you go back to the data sheet picture I posted you can see that the SDM covers MSB and the 18 LSBs. ICOB covers 6 bits below the MSB. Those 6 bits cover top 36 dB of value space while rest 108 dB value space is covered by the rest. If you replace 18 LSB bits with 26 LSB bits then the ratio becomes even bigger as the remaining part becomes 157 dB.

 

You need to remember that a full scale signal covers full value space and the upper MSB bits become active only at waveform peaks, but not anywhere near the zero-crossing region. For waveforms with peak levels below -36 dB those six bits never become active.

 

Effectively what the solution does, is to lower the SDM noise by 36 dB, since otherwise the 3rd order modulator running at 64x fs would have inadequate SNR in wanted pass band.

 

Side effect of the architecture is that with PCM inputs you get strong image frequencies at multiples of 352.8/384k as I have shown in my measurements - that are absent if you use DSD input...

[Dyson]

If you go back to the data sheet picture I posted you can see that the SDM covers MSB and the 18 LSBs. ICOB covers 6 bits below the MSB. Those 6 bits cover top 36 dB of value space while rest 108 dB value space is covered by the rest. If you replace 18 LSB bits with 26 LSB bits then the ratio becomes even bigger as the remaining part becomes 157 dB.

 

You need to remember that a full scale signal covers full value space and the upper MSB bits become active only at waveform peaks, but not anywhere near the zero-crossing region. For waveforms with peak levels below -36 dB those six bits never become active.

 

Effectively what the solution does, is to lower the SDM noise by 36 dB, since otherwise the 3rd order modulator running at 64x fs would have inadequate SNR in wanted pass band.

 

Side effect of the architecture is that with PCM inputs you get strong image frequencies at multiples of 352.8/384k as I have shown in my measurements - that are absent if you use DSD input...

 

I asked about it at AMR/iFi support and they confirmed that what they list as specifications is indeed what is used. They discovered (in fact long and costly engineering) quite a few of undocumented features (or Easter Eggs as they wrote) in one of the last chips of Burr-Brown Japan, but of course will not tell more than what is publicly available. So we know at least as differences to the official DSD1793 data sheet that it is used @ 32 bits, the modulator for the PCM delta sigma conversion is running at 256x and that it can do 2xDXD and 512DSD. How the DACs are used in the micro iDSD, is clear way out the original specs.

[MG32]

so is HQ Player able to confirm/deny the burrbrown the micro dsd works on dsd256 and dsd512 as this is not possible according to the datasheet in the 1793. to me this is more significant than the 24 or 32 bit.

[Miska]

so is HQ Player able to confirm/deny the burrbrown the micro dsd works on dsd256 and dsd512 as this is not possible according to the datasheet in the 1793. to me this is more significant than the 24 or 32 bit.

 

DSD256 is possible based on the data sheet. DSD512 is possible (undocumented) with "mono mode" when two channels are combined. However, it is not clear which one gives better results, DSD256 or DSD512.

 

I measured following figures...

 

Built-in "Standard" filter, 44.1k input: noise=-95.9dBr, THD=0.00743%, IMD=0.00101%

HQPlayer poly-sinc, 768k input: noise=-103.9dBr, THD=0.00277%, IMD=0.00084%

HQPlayer poly-sinc-2s, ASDM5, DSD256 input: noise=-106.7dBr, THD=0.00247%, IMD=0.00078%

HQPlayer poly-sinc-2s, ASDM5, DSD512 input: noise=-105.0dBr, THD=0.00293%, IMD=0.00063%

[Miska]

So we know at least as differences to the official DSD1793 data sheet that it is used @ 32 bits, the modulator for the PCM delta sigma conversion is running at 256x and that it can do 2xDXD and 512DSD. How the DACs are used in the micro iDSD, is clear way out the original specs.

 

Even if it supports 32-bit input data, the output SNR with PCM input doesn't go anywhere near resolution of 24-bit data. If you'd like to get decent performance with PCM input you need to run it at 705.6/768k input rates (and still it does better at DSD256/DSD512).

[Lionelh2]

DSD256 is possible based on the data sheet. DSD512 is possible (undocumented) with "mono mode" when two channels are combined. However, it is not clear which one gives better results, DSD256 or DSD512.

 

I measured following figures...

 

Built-in "Standard" filter, 44.1k input: noise=-95.9dBr, THD=0.00743%, IMD=0.00101%

HQPlayer poly-sinc, 768k input: noise=-103.9dBr, THD=0.00277%, IMD=0.00084%

HQPlayer poly-sinc-2s, ASDM5, DSD256 input: noise=-106.7dBr, THD=0.00247%, IMD=0.00078%

HQPlayer poly-sinc-2s, ASDM5, DSD512 input: noise=-105.0dBr, THD=0.00293%, IMD=0.00063%

 

Sorry if this been asked many times before, but what should HQ Player be set for if doing DSD up convert into iFi iDSD Micro? 24bit/32bit/default? I an on Win10 using the iFi ASIO driver.

[kurb1980]

Bits has no affect on DSD when ASIO is concerned w/Micro with that said set it to 24 bit. As outlined above by Miska the DAC chip is capable of "32bit" but in his tests the SNR doesn't go anywhere near that.

[skypickle]

forgive my naivete but I can not hear anything when I try to play dsf files. I can succesfully add and play flac files.

However dsf files do not play. Specifically,

these two sets of files downloaded as 5 .6 mhz dsd:

The Jazz Side Of The Moon | HDtracks - The World's Greatest-Sounding Music Downloads

The Raven 176/24 | HDtracks - The World's Greatest-Sounding Music Downloads

Help with config?

[skypickle]

Well I kept playing around and it started to play after i unchecked 'pipeline SDM'.

 

Just to make sure i was not crazy, I changed the driver selection in settings to default (realtek speaker out put). Of course now it would not play dsd audio, but would play flac. I went back into settings and switched it back to the iFI drivers.

 

No sound still from dsd files. I broke it again. or maybe the ifi drivers are broken. I disconnected and reconnected the ifi. No improvement. I dont know what I did to get it to play the first time.

 

Is there a noob guide somewhere?

What is SDM?

When do I use WASAPI vs ASIO ?

Why cant I add mp3?

[skypickle]

I changed to ASIO and now it plays DSD. But flac files sound terrible-sound is staticky and tempo is slow. I wonder why WASAPI was so finicky before. I didnt have any other audio apps open, just a few browser tabs. How can I troubleshoot this?

[Miska]

What is SDM?

 

Sigma-Delta Modulation, of which subset is also known as DSD (trademark and marketing term).

 

When do I use WASAPI vs ASIO ?

 

Apart from few exceptions (Chord DACs). Use ASIO whenever DAC manufacturer provides ASIO drivers. When not available, fall back to WASAPI.

 

Why cant I add mp3?

 

Because it is lossy format and HQPlayer is about quality. In addition, it would increase license price of HQPlayer significantly due to patent license fees, thanks to minimum annual fee of $15000.

[skypickle]

Thank you for your replies. I fumbled around and finally got HQplayer to play with wasapi by enabling SDM PACK DoP. I really need to spend time with the manual to understand all the little settings.

Which oversampling to use?

Which modulator?

Why is there even a bit rate limiter under SDM, etc,etc.

 

 

$15K for mp3 decoding?! WOW

[Miska]

Which oversampling to use?

 

Get started with poly-sinc family (see manual for descriptions). The ones that have -2s at the end are a variant that is lighter on the CPU.

 

Which modulator?

 

I recommend ASDM7, but rest is up to your choice. I use Micro with it's filter switch set to "Standard" (below the unit).

 

Why is there even a bit rate limiter under SDM, etc,etc.

 

When you use for example automatic rate selection and if you want to limit the maximum used rate to lower that what the DAC (seems to) support. For example there are some DACs that claim to support 6.1 MHz DSD under some circumstances, but only actually work at 5.6 MHz. For such cases one can set the limit to 5.6 to avoid HQPlayer from picking up the 6.1 rate. But in normal circumstances, just set it to maximum rate your DAC supports. In case of iFi iDSD Micro, 768 kHz for PCM and 24.576 MHz for DSD.

 

$15K for mp3 decoding?! WOW

 

Yes, it is $0.75 per software license, but $15000 annual minimum. So if I'd sell less than 20000 licenses per year (which is the case), my yearly fee would be that flat $15000 instead of the $0.75 per license sold. So for me, it would mean I'd instead need to divide the $15000 with number of licenses sold per year and add that on top of my net price.

 

As usual, all these kind of things are tailored for big mass market corporations (selling in millions of units) and screwing small vendors. There are lot of similar examples across the board.

[craighartley]

Get started with poly-sinc family (see manual for descriptions). The ones that have -2s at the end are a variant that is lighter on the CPU.

 

 

 

I recommend ASDM7, but rest is up to your choice. I use Micro with it's filter switch set to "Standard" (below the unit).

 

ASDM7 in preference to AMSDM7?

[Miska]

ASDM7 in preference to AMSDM7?

 

Both are fine, but in case someone is limited to DSD256 with Micro (Mac, etc), ASDM7 works nicely at that rate too.

[bogi]

In addition, it would increase license price of HQPlayer significantly due to patent license fees, thanks to minimum annual fee of $15000.

 

I am really curious, if authors/owners of all free SW players with mp3 support pay annual $15000. Doesn't seem to be probable.

 

To be clear, I'm not asking you to support MP3 in HQPlayer. I don't need that.

[Jud]

Both are fine, but in case someone is limited to DSD256 with Micro (Mac, etc), ASDM7 works nicely at that rate too.

 

I listened briefly and found I preferred ASDM7 at the approx DSD256 rates my old desktop will support (which is quite good for a 2009 CPU, thank you, Miska).

 

Also, I don't know whether it should make any difference and therefore whether I'm imagining it, but regardless of whether the source was 44.1 or 48KHz based, I felt the sound was better at 12.28 MHz than 11.28.

[Miska]

I am really curious, if authors/owners of all free SW players with mp3 support pay annual $15000. Doesn't seem to be probable.

 

No they probably don't. I'm not a lawyer, so I really cannot claim to know anything. But I'd assume if they don't make any money, it is hard to sue them for "damages" and get anything out of their "revenue".

[Miska]

Also, I don't know whether it should make any difference and therefore whether I'm imagining it, but regardless of whether the source was 44.1 or 48KHz based, I felt the sound was better at 12.28 MHz than 11.28.

 

This may be related to how sampling rate fits the USB packet rate. At 44.1k the packet size varies because the packet interval is not in sync with the sampling rate, while 48k is in sync with the packet interval and thus the sizes vary less. This causes different patterns on the USB interface interference pattern.

[mansr]

No they probably don't. I'm not a lawyer, so I really cannot claim to know anything. But I'd assume if they don't make any money, it is hard to sue them for "damages" and get anything out of their "revenue".

That's pretty much it. I'm not a lawyer, but I have been involved with such software projects. The patent holders mostly ignore you if you're not making significant money.

[YashN]

I am really curious, if authors/owners of all free SW players with mp3 support pay annual $15000. Doesn't seem to be probable.

 

Nowadays, it strikes me as odd to pay a license annually to create crappy audio.

[bogi]

Some SW requires user to download MP3 encoder/decoder, so "the thing" may be solved also this way.

 

Many people are not able to distinguish 320kbps MP3 from WAV. With ASDM7 modulator they play well in HQPlayer. On other side playing 128k MP3 in HQPlayer would be like using 91 octane gasoline in Ferrari.

[bogi]

Nowadays, it strikes me as odd to pay a license annually to create crappy audio.

 

I would like to see the list of paying hi-end companies

[lmitche]

This may be related to how sampling rate fits the USB packet rate. At 44.1k the packet size varies because the packet interval is not in sync with the sampling rate, while 48k is in sync with the packet interval and thus the sizes vary less. This causes different patterns on the USB interface interference pattern.

Thanks, this is interesting. I'll try up-sampling everything to the higher dsd512 rate with the T+A DAC.