32 bit stereo ADC, wow

[martingriffith]

Just found this http://esstech.com/index.php?p=products_ADC

 

[Rochey]

Hello Martin,

their claim of "32-bit" doesn't really blow me away I'm afraid.
Nobody has a true 24bit converter right now - the analog normally disappoints, causing the least significant bits of the datastream to be noise. Even the 32bit DAC's on the market only stretch to the 25th bit through some very clever measuring. (FFT with very small bins, then take an average!)

having said that, there are a few headlines that impress... -120dB THD @ 50mW is impressive, but I have to call bshit on some things...

Without a decent datasheet (with performance graphs) then it's hard to believe they were able to get high performance with just 50mW power consumption. A 123dB SNR ADC, with -108dB THD (typ) consumes ~300mW. The competition before it was consuming 700mW+. So, the jump down to 50mW is very suprising. 200, maybe I could believe. 150, that's a "wow" -- 50mW??? I need to see data.

There's no dynamic range (SNR) data in their marketing sheet (I'm in marketing... I know marketing fluff when I see it  ) There's no details of how they've read the -120db THD. What about the self-noise in the device? The noise floor of any device is normally the THD+N (N means noise) at specified input voltages. (normally full scale input, and another input -60dB below full scale).

That marketing fluff has none of that.

If anyone here has really seen this IC, and thrown it under an Audio Precision tester, I'd be really interested to see what it's really like. If it's really the saviour of ADC technology, then so be it. (no hard feelings... I'll actually be really impressed!)

If one of their marketing folks reads this... "please, do what the rest of us do in the converter industry - help customers read about your parts!". The rest of us, AKM, Cirrus, Wolfson and TI actually have real performance details in their datasheets. We actually let potential customers read more than just fluffy marketing data before having to sign NDA's etc. It's a very strange engagement strategy.

by the way - some disclosure - I work for TI. I'm a systems definition engineer for home audio converters. However, the contents of this message are a personal opinion, not one of my employer.

Finally - can we get some inputs from some of the real digital guys? JDB? Svart?

 

[jdbakker]

Rochey made most major points, and I have to agree with his analysis.

The only reason you'd possibly want 32 bits out of an ADC rather than 24 is to dither the lower bits before truncating. On paper this is a valid concern, but Bruno Putzeys points out that ADCs are one of the few cases where device noise can be said to provide self-dithering. I don't see this as an issue until well after ADCs start breaking the 130dB dynamic range barrier.

The 50mW figure is fishy. While from a design POV I feel that most manufacturers could disclose more information than they're doing now, this 2-page hype sheet can be read any which way. With creative interpretation I could argue that they're saying it consumes 50mW per channel. Nowhere does it explicitly say that the 50mW figure applies to full performance; could be a low-power mode. Is it indeed -120dB THD, or -120dB THD+N? What measurement bandwidth?

(About converters and power consumption: most modern sigma/delta-converters have an analog and a digital side. The digital side runs a filter, and filter quality and chip area limits set a lower bound to the power consumption. On the analog side for a given architecture noise and distortion tend to go down when quiescent current goes up. Combine that with the fact that you need to select a semiconductor process that works well with both your analog and digital subsystems, so you can't just optimize for one or the other.)

It might be a nice chip, and some people have been happy with the earlier Sabre DACs. However without more data it's hard to say anything sensible.

JDB.

 

[Svart]

Well I won't call BS until someone actually tests the thing but I can say that I doubt that a small player in the market like ESS would be the first to come out with a true 32 bit part.

I need to see more than just a buzzsheet.  Datasheets would be a start, a demo setup would be best.. 

 

[JohnRoberts]

It sounds nice to use a full 32 bit decimation if you are working with 32 bit wide data paths.

While vague (spot, total, thd+n ?) the 120 dB distortion spec is only 20 bits down, and noise floor is not even specified.

I do find it a little odd that they publish a "Confidential" advance information sheet on the WWW 

Not enough info to call it good or bad and of course the price always matters...

I'd personally like to see spectrum analyzer plots for HF but in band multi-tone IMD test. Different people may have their personal favorite benchmarks.

It's clear that good A/D convertors will deliver quantization floors well below their analog noise floors, and analog audio paths are just not physically capable of 32 bit X 6dB per bit dynamic range, so digital paths this wide are just for convenience crunching data in the digital realm. Not good or bad, but not an analog consideration.

JR

 

[burdij]

Wow, now we can even get a high fidelity recording of the noise. 

 

[Samuel Groner]

Will the extra 8 bits help the musicians to play better?

Samuel

 

[olafmatt]

What is a 32bit ADC anyway? Many ADCs are 1-bit Delta-Sigma or Multibit-Delta-Sigma followed by some DSP to output 'regular' 24bit PCM. So if I just change the DSP code to output 32bit PCM instead of 24bit then I made a 32bit ADC without changing the actual conversion at all. So unless you make a 32bit ladder converter you can call a converter any bit-depth you like, just make your DSP output the word-length you want.  ;D

Olaf

 

[Nishmaster]

I don't see how it could be physically possible to approach -192dB noise floor offered by a 32 bit converter anyway. Is this some miraculous resistorless device operating in a bath of liquid nitrogen?

JR, I possibly see your point about offering a native 32 word to reduce overhead, but tacking on some zeroes doesn't really take much power, and that 32 bits incoming comes at the expense of bus bandwidth.

 

[Tubemooley]

I have one of the earlier Sabre DAC's on a twisted pear audio board. I haven't had time to wire it up. These guys at ESS are a pain-in-the-butt. Everything is all confidential. When the first Sabre DAC came out, there's only one distributor in USA and the IC was $50. each and you had to order a minimum of five. I signed an NDA and got the datasheet. It's all stamped with my company name on every page. Talk about paranoia. They appear to be trying to market to the larger high-end consumer audio makers for their DAC's. A few companies have bought in. I agree with all of the techno analysis you guys have presented here. They ain't gettin' no 32-bits. These guys blow. DW.

 

[Svart]

When dealing with a lot of companies that make ICs, I've experienced a lot of the NDA/company name on the datasheets/password protected Acrobat documents/etc. 

It's a pain in the ass but these companies think it helps somehow and I don't think they'll stop anytime soon.

There is a company that we deal with that will only talk to a single person at our site, the person whose name was on the original contract documents... talk about paranoia..

 

[Tubemooley]

I know the local factory guy for AKM here in Boston. He hates me at this point. Long story not worth repeating. But I just got free samples of their latest 32-bit DAC. I have a contact at TI who has sent me free eval boards. Cirrus is a bit more difficult. Analog Devices is relatively cool. TC Electronic and BridgeCo are all NDA'd up. Yaddee, yaddee, yaddee.

 

[Michael Tibes]

I couldn't work it out quickly, maybe someone can enlighten me: is this 32bit fixed point or a floating point format (as with the typical difference between a 24 and 32 bit wav file)?

Michael

 

[Andy Peters]

When dealing with a lot of companies that make ICs, I've experienced a lot of the NDA/company name on the datasheets/password protected Acrobat documents/etc. 

It's a pain in the ass but these companies think it helps somehow and I don't think they'll stop anytime soon.

There is a company that we deal with that will only talk to a single person at our site, the person whose name was on the original contract documents... talk about paranoia..

BroadCom, Marvell, 3dLabs, all of the big chip companies think their parts are so special that everything has to be locked down under NDA etc.

It wouldn't be so bad if their tech support didn't suck hard.

-a

 

[analag]

You all scared the poor guy...

 

[Robson Santos]

So for audio it wouldn't be viable to use 32-bit ADC/DAC and for the cost that is totally high, because I'm looking for something with 32-bit, something more modern to use in a studio rack that I'm building, so if 20, 24- little enough, now it's stuck in my head.
Anyone to help us there?

 

[living sounds]

There is no point in 32-bit converters in pro audio. Going from 24 bit to 32 bit is as useful as adding wings to your bycicle. It will probably hurt less, but it won't do any good. The theoretical SNR of 24 bit is already far below any reasonably attainable noise floor in the buffer stages, let alone the rest of your equipment.

 

[Robson Santos]

There is no point in 32-bit converters in pro audio. Going from 24 bit to 32 bit is as useful as adding wings to your bycicle. It will probably hurt less, but it won't do any good. The theoretical SNR of 24 bit is already far below any reasonably attainable noise floor in the buffer stages, let alone the rest of your equipment.

Yes, I'm so into this 32-bit ADC that I forgot about it, good to remember.

 

[Biasrocks]

There is no point in 32-bit converters in pro audio. Going from 24 bit to 32 bit is as useful as adding wings to your bycicle. It will probably hurt less, but it won't do any good. The theoretical SNR of 24 bit is already far below any reasonably attainable noise floor in the buffer stages, let alone the rest of your equipment.

The most useful function of 32 bit conversion that I have seen is the ability to recover clipped material by simply turning the audio file down in your DAW. Not possible with 24 bit conversion, it just stays clipped.

https://mediatech.edu/32-bit-audio-end-audio-clipping/

 

[living sounds]

The most useful function of 32 bit conversion that I have seen is the ability to recover clipped material by simply turning the audio file down in your DAW. Not possible with 24 bit conversion, it just stays clipped.

https://mediatech.edu/32-bit-audio-end-audio-clipping/

Possibly in a DAW with floating point coding, but has nothing to do with converters. A signal actually clipped during conversion stays clipped.