Discrete A/D converters

[Rochey]

Guys, think of it like this:

All ADC's are made up of 2 parts... the modulator and digital decimation filter.
The modulator oversamples the signal at 512 times the input. Most Modulators are either 1bit, or 5 or 6 bit.
The output of a 1bit modulator is actually the same as the DSD standard.
After the modulator, we use a digital decimation filter that takes the oversampled 1bit (or multibit data) and creates PCM.

Whenever you filter anything, you lose some data. That's the whole point of a filter, to chop out what you don't want, and keep what you really want (at that point in time).

So, the message is, the purest form of AD there is, is DSD, because it's the converted data without any digitally filtered content.

I think Korg have nailed it.. for directly recording analog, DSD is perfect. With the Korg portable recorders, you can bring home the audio and create PCM data at any sampling rate at any bit depth.

Imagine recording Hendrix or the Who in DSD, and mixing down for CD in the 80's... and doing a DVD remix of it now (at 24bit 96kHz, or 192kHz).

or worse yet.. a direct mp3 from teh DSD

 

[rafafredd]

OK, so as I see DSD is better than PCM converters that use 1bit modulators then filters, but what about PCM converters that uses 5 and 6 bit modulators, as you mentioned? I would think they are somewhat better than 1bit DSD... Also, there´s no raw 5bit or 6bit converters, as far as I know... Maybe the next thing will be DSD 5bit, or DSD 6bit?

Care to explain how 1bit raw compares favorably with 5 or 6 bit filtered??

 

[Rochey]

PCM4222 will be the first ADC with a 6bit modulator output.

I believe that there are certain undocumented modes on other converters that allow a 5 bit output.

At the end of the day, PCM is still filtered from an oversampled source. It's filtered down to whatever sample rate and bit depth you want.
As I said before, as soon as you filter, you're losing data. (whether it's a tiny bit, or a lot)

To achieve grater than 120dB of dynamic range, most say that you need a multi-bit modulator.

Digital decimation filter design is a complex topic, way beyond my understanding it enough to explain here. However, I can share one or two things -- if anyone wants to correct me, please do - i don't claim to be a master at this:

Filter design has 2 main specifications: phase linearity and group delay. Typically, one works in the opposite direction of the other... i.e. if you want low group delay (low latency) then typically the phase linearity and stop band attenuation suffer.
However, in applications like live sound, low group delay is more important than phase linearity and stop band attenuation. Typically, converters sit in either the low group delay, or flat phase linearity.

The 120dB Cirrus Logic CS5381 converters typically have a 12 sample group delay.
The 123dB AKM AK5394A have a 63 sample group delay.
The 118dB TI PCM4202 have 37 samples up to 96kHz, and 9 samples at 192kHz.
The new 124dB PCM4222 from TI has 3 different options: a reduced group delay of 21 samples (still quite linear), a "classic" response (VERY linear) 39 samples.

Or... as a final option, designers can take the 6bit modulator output from the PCM4222 directly into their own custom designed digital filter in FPGA.

I suspect most people will take the PCM out chosing the Classic or reduced filter design.


There's a great paper from Digital Audio Denmark that i'm currently reading, that discusses DSD and DXD (384kHz PCM -- generated from a 5bit modulator) as a medium for aquiring audio and storing audio.
http://www.digitalaudio.dk/technical_papers/axion/dxd%20Resolution%20v3.5.pdf

cheers

R

 

[jdbakker]

[quote author="Rochey"]All ADC's are made up of 2 parts... the modulator and digital decimation filter.[/quote]
Nitpick: all current audio ADCs (and DACs) are based on the Delta-sigma principle. There are numerous other ADC architectures (Flash, successive approximation etc), but in current process technologies delta-sigma gives best dynamic range for medium-speed applications, without needing extensive trimming or temperature compensation.

Some audiophiles(/phools) still swear by older monolithic Flash converters.

[quote author="Rochey"]So, the message is, the purest form of AD there is, is DSD, because it's the converted data without any digitally filtered content.[/quote]
...iff your ADC uses a 1-bit modulator. If I understand the math right, there is no trivial way to go from a multibit modulator to DSD (the conversion is possible, but requires digital filtering). Most manufacturers are migrating to multibit modulators.

In my humble opinion, the Korg paper is more marketing than anything else. I find it particularly misleading that they use a 20kHz square wave to make their case, as the magnitude of its third harmonic (60kHz) and fifth harmonic (100kHz) have far less effect on audio performance than general transient response (overshoot, ringing etc) which is not illustrated at all by these "tests". As far as I can tell, transient response is as much a function of the modulator design as it is of the digital filter, so it may not be a great testcase for DSD...

Similarly, the DXD paper saying that "the pulse response [...] is a measure of the timing precision of the digital representation of the signal" is at best an oversimplification. The DXD paper is also confusing various issues:

[quote author="The DXD paper"]The advantage of DXD is that the signal is not limited to a one bit representation like the DSD format. A-D convertors can therefore be used having a higher order delta sigma modulator of 5 bit, for example.[/quote]
Modulator order and bit depth are orthogonal features; the link in the first post of this thread discusses a 6th order 1-bit ADC. Sounds much like a journalist(/PR person?) has been reading a few tech briefs and has cooked up something nice-sounding.

From what I've seen when our group was doing tests on digital audio watermarking schemes, the ear is much more sensitive to phase oddities in the audio band than to absolute system bandwidth. It is true that digital filtering may impact this, but that's not what Korg et al are claiming.

For those who care, there's an extended argument going on between Philips/Sony, claiming that 1-bit DSD for SACD is the best thing since sliced bread, and several others who say that 1-bit delta-sigma converters are fundamentally flawed wrt instability and idle tones. Google for "dsd lipshitz" and prepare for some .... interesting math.

[quote author="Rochey"]Filter design has 2 main specifications: phase linearity and group delay. Typically, one works in the opposite direction of the other... i.e. if you want low group delay (low latency) then typically the phase linearity and stop band attenuation suffer.[/quote]
Yup. To a certain extent, processing power plays a role too. The number of FIR taps / IIR poles you can get on your chip determines how steep a filter you can have, which may limit or broaden your options wrt filter topology.

[quote author="Rochey"]The new 124dB PCM4222 from TI has 3 different options: a reduced group delay of 21 samples (still quite linear), a "classic" response (VERY linear) 39 samples.[/quote]
That's my favourite feature of this part. It's great to have this user-tunable. I particularly like it because it makes it easier to use this part for baseband sampling in radio systems. Hey, noone said you could only use 'em for audio.

[quote author="Rochey"]Or... as a final option, designers can take the 6bit modulator output from the PCM4222 directly into their own custom designed digital filter in FPGA.[/quote]
Very nice indeed; more converters should do this. I wish I had several more hours in a day, or I'd plonk one down on a board connected to a FPGA to toy with custom filtering profiles. May still do so if I can get myself a client...

Thanks for the ADC delay data, very instructive !

JDB

 

[Samuel Groner]

In my humble opinion, the Korg paper is more marketing than anything else.

Full agreement on my side.

I find it particularly misleading that they use a 20 kHz square wave to make their case.

That square wave testing is a never ending story. I remember a well-known german magazine writing an article on "why is analogue better than digital" and showing "impulse responses" of various systems. The impulse response of the "analogue" system was perfectly symmetrical... Well... And the conclusion was: analogue tape recorders are better because of this. :roll:

Some audiophiles(/phools) still swear by older monolithic Flash converters.

A few month back we had a call at work asking where to get these old "true 16 bit" converters. :green:

Samuel

 

[chrissugar]

[quote author="Samuel Groner"]
That square wave testing is a never ending story. I remember a well-known german magazine writing an article on "why is analogue better than digital" and showing "impulse responses" of various systems. The impulse response of the "analogue" system was perfectly symmetrical... Well... And the conclusion was: analogue tape recorders are better because of this. :roll:
[/quote]

Problem is people do not understand the premises of digital audio and they draw the wrong conclussions. They do not understand that those harmonics above nyquist have no reason to be there, and that would not make any difference in audio if well implemented.

[quote author="Samuel Groner"]A few month back we had a call at work asking where to get these old "true 16 bit" converters. :green: [/quote]

Yeah, I know how it is. :grin: I know some audiophools very excited by their Philips multibit TDA chips and nonoversampling DACs. I listened to some, (also some DIY variations) and they invariably sound like crap.
Sometimes I'm wondering about the sonic references of these guys. I think they never listen to some real acoustic music.

chrissugar

 

[mikep]

[quote author="Rochey"]PCM4222 will be the first ADC with a 6bit modulator output.
[/quote]

6bit=5.1bit=surround sound-bit

but seriously, every time this part comes up it looks better and better. I like the idea of custom FPGA filtering, adds a bit of future upgradability. I think I'm going to have to grab one of these and hack something together when it is available. I need a new A/D for my mastering rig (to replace the OLD PSX-100) and I havent done any real digital foolery since I was in grad school. Any idea when the EVM for the 4222 is going to be ready for sale? I wonder what the subjective difference of the filter choices is.

mp

 

[Rochey]

Gents,

EVM's in April/May along with the release. Prelim datasheets are up.

I'd love to see a demo board with an FPGA onboard.
mmmm 124dB

 

[JohnRoberts]

[quote author="chrissugar"][quote author="Samuel Groner"]
That square wave testing is a never ending story. I remember a well-known german magazine writing an article on "why is analogue better than digital" and showing "impulse responses" of various systems. The impulse response of the "analogue" system was perfectly symmetrical... Well... And the conclusion was: analogue tape recorders are better because of this. :roll:
[/quote]

Problem is people do not understand the premises of digital audio and they draw the wrong conclussions. They do not understand that those harmonics above nyquist have no reason to be there, and that would not make any difference in audio if well implemented.

[quote author="Samuel Groner"]A few month back we had a call at work asking where to get these old "true 16 bit" converters. :green: [/quote]

Yeah, I know how it is. :grin: I know some audiophools very excited by their Philips multibit TDA chips and nonoversampling DACs. I listened to some, (also some DIY variations) and they invariably sound like crap.
Sometimes I'm wondering about the sonic references of these guys. I think they never listen to some real acoustic music.

chrissugar[/quote]

While there will always be potential for differences of opinion about judgement calls at the cutting edge of technology any arguement that old technology is better is not supported by results.

Another often misinterpreted characteristic is "Gibb's phenomenon", the apparent ringing from proper subtraction of overtones from a wideband square wave.

JR

 

[Quince]

I was turned off by DSD when I saw the detailed noise analysis in one of Hawksford's presentations (it can be found on the Web I think).