MQA technical analysis

[Shadders]

Hi mansr,

Thanks - i was going to write to Hifi News to correct.

 

As such - can you confirm that the information in the 24kHz to 48kHz is aliased ? (or not).

 

I thought the leaky filters allowed aliasing - and if so, if i have to direct Hifi News to the relevant information on this forum, which posting provides the evidence ?. Thanks.

 

Regards,

Shadders.

[mansr]

12 minutes ago, Shadders said:

Thanks - i was going to write to Hifi News to correct.

 

As such - can you confirm that the information in the 24kHz to 48kHz is aliased ? (or not).

 

I thought the leaky filters allowed aliasing - and if so, if i have to direct Hifi News to the relevant information on this forum, which posting provides the evidence ?

The encoder downsampling to 96 kHz introduces some aliasing below 48 kHz as shown in this post:

 

[Shadders]

6 minutes ago, mansr said:

The encoder downsampling to 96 kHz introduces some aliasing below 48 kHz as shown in this post:

 

Hi mansr,

Thanks - what i am not sure about the graphs etc., is whether they show significant aliasing from 24kHz upwards.

 

The quadrature mirror filters join the 0-24kHz and 24kHz-48kHz together, which i assume is accepted.

 

So, what is the reason for using leaky filters ?

 

Do these leaky filters operate across the 48kHz bandwidth and hence leak the image from 96kHz downwards into the top end of the 48kHz ?

 

I am unsure about this aspect.

 

Thanks and regards,

Shadders.

[mansr]

3 minutes ago, Shadders said:

Hi mansr,

Thanks - what i am not sure about the graphs etc., is whether they show significant aliasing from 24kHz upwards.

 

The quadrature mirror filters join the 0-24kHz and 24kHz-48kHz together, which i assume is accepted.

 

So, what is the reason for using leaky filters ?

 

Do these leaky filters operate across the 48kHz bandwidth and hence leak the image from 96kHz downwards into the top end of the 48kHz ?

The filter I'm talking about here is the one reducing the input file to 96 kHz from some higher rate. It introduces a small amount of aliasing from frequencies above 48 kHz as seen in the graphs above. This aliasing is low enough that it's difficult to detect with most files.

 

The band splitting is a separate thing. The obvious method is to use quadrature mirror filters, but we don't know for sure that that's what MQA does.

[Shadders]

Hi mansr,

Thanks.

OK - so we do not know if there is aliasing in the band 24kHz to 48kHz.

 

As an aside - does anyone know what the filter is used on the front end of a studio ADC ?.

 

I looked at the Pacifics Microsonic offerings, and no mention of the ADC filter.

Thanks and regards,

Shadders.

[Miska]

46 minutes ago, Shadders said:

I looked at the Pacifics Microsonic offerings, and no mention of the ADC filter.

 

PMD didn't have digital filter for 176.4 kHz ADC rate. But when output was set to 44.1 kHz, it uses normal FIR anti-alias filter for the rate conversion. I don't remember the FIR in more detail...

 

[mansr]

40 minutes ago, Shadders said:

OK - so we do not know if there is aliasing in the band 24kHz to 48kHz.

Frequencies in the 48-72 kHz range are aliased onto 24-48 kHz, though not to a huge extent.

 

40 minutes ago, Shadders said:

As an aside - does anyone know what the filter is used on the front end of a studio ADC ?

Here's the rising edge of a square wave (~5 ns rise time) recorded on a Tascam UH-7000 at 192 kHz:

 

I could test it with some sine waves around 96 kHz and see what the roll-off looks like...

[Shadders]

20 minutes ago, Miska said:

 

PMD didn't have digital filter for 176.4 kHz ADC rate. But when output was set to 44.1 kHz, it uses normal FIR anti-alias filter for the rate conversion. I don't remember the FIR in more detail...

 

Hi Miska,

Thanks - i was referring to the analogue filter on the front end of the ADC - do you know anything about that ?

Thanks and regards,

Shadders.

[Shadders]

21 minutes ago, mansr said:

Frequencies in the 48-72 kHz range are aliased onto 24-48 kHz, though not to a huge extent.

 

Here's the rising edge of a square wave (~5 ns rise time) recorded on a Tascam UH-7000 at 192 kHz:

 

I could test it with some sine waves around 96 kHz and see what the roll-off looks like...

Hi mansr,

Thanks - i will send to Hifi news if when i read further any other aspects.

Thanks for the filter picture. I would expect it to be an 8th order - but not sure about the type -bessel, butterworth, elliptical etc. So, should be linear phase - constant group delay.

Do you know what ADC is prevalent in the studio in the 1980's amd 1990's, and currently ?

Thanks and regards,

Shadders.

 

[mansr]

4 minutes ago, Shadders said:

Thanks - i was referring to the analogue filter on the front end of the ADC - do you know anything about that ?

With a sigma-delta architecture operating at several MHz, there isn't really any need for much of an analogue filter. Realistic inputs don't have anything for the filter to remove anyway.

[Miska]

2 minutes ago, Shadders said:

Hi Miska,

Thanks - i was referring to the analogue filter on the front end of the ADC - do you know anything about that ?

Thanks and regards,

Shadders.

 

I don't really remember much... It is likely not delta-sigma ADC, so it would need some more aa-filtering in analog domain. But of course it depends how much designer expect there to be content up to Nyquist frequency and how much they are ready to sacrifice bandwidth for the filter transition band. I think PMD was optimized for CD pass-band, so I think it started rolling off somewhere near above CD's Nyquist.

 

MQA seems to be working in a similar way, by sacrificing band above 22.05/24 kHz for the filter roll-off to keep the filter very short. So instead of more ringing it makes the rise-time longer. So while winning transient domain from ringing perspective it loses on rise time.

 

Normal studio gear uses typically second order low-pass with -3 dB point around 100 kHz, since they use delta-sigma converters and there's no need to do a lot of analog pre-filtering since most will be done in digital domain. Then you may have bunch of options for the digital filter.

 

[Shadders]

10 minutes ago, mansr said:

With a sigma-delta architecture operating at several MHz, there isn't really any need for much of an analogue filter. Realistic inputs don't have anything for the filter to remove anyway.

Hi mansr,

I was thinking of the 1980's and 1990's before sigma-delta - when the album was recorded direct to digital, or analogue tapes transferred during this time.

For the Tascam - is there in information on the filters (i assume sigma-delta, then filtered) ? Thanks.

Regards,

Shadders.

[mansr]

5 minutes ago, Shadders said:

Thanks for the filter picture. I would expect it to be an 8th order - but not sure about the type -bessel, butterworth, elliptical etc. So, should be linear phase - constant group delay.

Since it is symmetrical around the edge, it is indeed linear phase.

 

5 minutes ago, Shadders said:

Do you know what ADC is prevalent in the studio in the 1980's amd 1990's, and currently ?

Current gear seems to mostly use ADC chips from TI, AKM, and Cirrus Logic.

[Shadders]

1 minute ago, mansr said:

Since it is symmetrical around the edge, it is indeed linear phase.

 

Current gear seems to mostly use ADC chips from TI, AKM, and Cirrus Logic.

Hi mansr,

Thanks.

It is a shame Brian Lucey is not here - i could ask what Pacific Microsonics ADC is used and then contact then to determine the filter type and order.

I will try Tascam to see if they will provide any information on the filters - basic details should suffice.

Regards,

Shadders.

[mansr]

4 minutes ago, Shadders said:

I was thinking of the 1980's and 1990's before sigma-delta - when the album was recorded direct to digital, or analogue tapes transferred during this time.

Sorry, you'll have to ask someone older than me.

 

4 minutes ago, Shadders said:

For the Tascam - is there in information on the filters (i assume sigma-delta, then filtered) ?

The ADC chip is the PCM4220 from TI. The datasheet has some details about the architecture including filters.

[Miska]

22 minutes ago, Shadders said:

I was thinking of the 1980's and 1990's before sigma-delta - when the album was recorded direct to digital, or analogue tapes transferred during this time.

For the Tascam - is there in information on the filters (i assume sigma-delta, then filtered) ? Thanks.

 

90's was already SDM converters. Chips from same vendors as these days, TI, AKM and Cirrus Logic.

 

For pro-audio gear, AKM and Cirrus Logic seem to be most popular.

 

I still have some tens of these older DSD-capable true 1-bit ADC chips on tape, originally from 2001:

http://www.ti.com/product/pcm1804

 

[Shadders]

5 minutes ago, Miska said:

 

90's was already SDM converters. Chips from same vendors as these days, TI, AKM and Cirrus Logic.

 

For pro-audio gear, AKM and Cirrus Logic seem to be most popular.

 

Hi Miska,

Thanks. I checked - late 90's seems to be the earliest - but in any case - i can still progress with assumed FIR filters.

Do you know how many taps they will have had? (256, 512 - data sheet as per mansr indicated did not state - as per the DAC;s too.. ???)

Thanks and regards,

Shadders.

[Miska]

1 minute ago, Shadders said:

Thanks. I checked - late 90's seems to be the earliest - but in any case - i can still progress with assumed FIR filters.

Do you know how many taps they will have had? (256. 512 ???)

 

Those are cascade FIR designs, just like DAC chips. Last stage to 44.1 kHz less than 200 taps, preceding stage usually 25-50 taps (to 88.2/96k) and the stage before that less than 25 taps (to 176.4/192k).

 

[Shadders]

1 minute ago, Miska said:

 

Those are cascade FIR designs, just like DAC chips. Last stage to 44.1 kHz less than 200 taps, preceding stage usually 25-50 taps (to 88.2/96k) and the stage before that less than 25 taps (to 176.4/192k).

 

Hi Miska,

Thanks - Ok - will assume a 256 tap FIR - and use that as an example.

Regards,

Shadders.

[Don Hills]

7 hours ago, mansr said:

...

 

Thanks, Mans. That's cleared some of the fog (and created some more... )

Plenty for me to think about.

 

Regards...

[Fokus]

9 hours ago, Shadders said:

Do you know what ADC is prevalent in the studio in the 1980's and 1990's, and currently ?

 

 

As others said, from the early-mid 90s on all sigma-delta stuff, with first or second order analogue pre-filtering and then digital half-band anti-aliasing filtering in the ADC's downsamplers.

 

And before that linear ADCs sampling at baseband, and thus with analogue AA filters, something like 8th order elliptic.

 

The SDM ADCs with their HB filters are easy to recognise: do a spectral analysis of a CD, when the spectrum runs right up to Fs/2 you are looking at HB: there is always a bit of aliasing.

 

The older ADC class is typefied by the Sony PCM1610/1630 systems. I have quite a few older CDs where the spectrum reaches a lot of attenuation at Fs/2, i.e. free of aliasing. Remarkable.

There is not much data on these old ADCs available on the net, but here is some

Sony DAT review.

 

 

If I remember correctly the Pacific Microsonics ADC (or rather, the entire system including HDCD encoding) used 2x oversampling, so part of the CD-production AA filter was done in the digital domain.

 

Oh, and Audio Note had a prototype non-filtered linear ADC at around 2000. I still have a CD with test recordings PQ gave me  ;-)

 

 

[Miska]

1 hour ago, Fokus said:

As others said, from the early-mid 90s on all sigma-delta stuff, with first or second order analogue pre-filtering and then digital half-band anti-aliasing filtering in the ADC's downsamplers.

 

And before that linear ADCs sampling at baseband, and thus with analogue AA filters, something like 8th order elliptic.

 

The SDM ADCs with their HB filters are easy to recognise: do a spectral analysis of a CD, when the spectrum runs right up to Fs/2 you are looking at HB: there is always a bit of aliasing.

 

The older ADC class is typefied by the Sony PCM1610/1630 systems. I have quite a few older CDs where the spectrum reaches a lot of attenuation at Fs/2, i.e. free of aliasing. Remarkable.

There is not much data on these old ADCs available on the net, but here is some

Sony DAT review.

 

Yeah, although with recent material the recording is quite typically done at 96 kHz sampling rate and then mastered for CD with software conversion from 96k to 44.1k. That's what also brought us the comparison site: http://src.infinitewave.ca

 

I have good example of this too...

 

Here's early (first?) CD version of Pink Floyd's DSOTM, "Time":

 

Here's 30th Anniversary SACD's RedBook-layer:

 

And here's the latest remaster:

 

And for comparison, 30th Anniversary SACD's DSD-layer:

 

 

1 hour ago, Fokus said:

If I remember correctly the Pacific Microsonics ADC (or rather, the entire system including HDCD encoding) used 2x oversampling, so part of the CD-production AA filter was done in the digital domain.

 

It runs at 176.4 kHz, so 4x oversampling.

[mansr]

This filter talk is all very interesting, but I feel we're drifting from the topic of MQA. If you want to continue on this tangent, please start a new thread.

[asdf1000]

Hello, did anyone have a .pdf copy of Ayre's whitepaper that used to be here:

 

https://www.ayre.com/pdf/Ayre_MP_White_Paper.pdf

 

 

[shadowlight]

34 minutes ago, Em2016 said:

Hello, did anyone have a .pdf copy of Ayre's whitepaper that used to be here:

 

https://www.ayre.com/pdf/Ayre_MP_White_Paper.pdf

 

 

https://web.archive.org/web/20101119144911/https://www.ayre.com/pdf/Ayre_MP_White_Paper.pdf