Amplifier driving SAR ADC 1

[sanlog]

Does anyone know of a "true" 16 bit SAR ADC that has an on-chip amplifier? It seems like all of the converters require an external amplifier. Given that the dataconverter speed is known, load is known, input voltage level is known, power supply votlage is known, and the noise known...what's left? And the data sheets always show an optimum amplifier part number.

Driving these huge switched cap loads would sure be easier if they simply gave you a pin where you hang a cap!

 

[itsmoked]

Part of the problem is that 16bits is dang tough!! Very, very, very, hard to achieve. An A/D is primarily digital and a 'good' amp is analog. Mixing those two processes on one die means some 'compromises'. 16bits allow no compromises! Also the digital noise will tend to pollute the small analog signals arriving so the A/D would probably have too much noise to even achieve 16b. I don't think you'll find what you're looking for. You might, be I doubt it.

Keith Cress
Flamin Systems, Inc.-

http://www.flaminsystems.com

 

[sanlog]

Too bad about the part... What got me to thinking about this is that many companies sell these 16 bit converters on microcontrollers. I'm thinking about the Silicon Labs parts (I don't remember the number). They have DACs, temperature sensors, multiple ADCs, amplifiers etc....AND a big honking microprocessor clocking at multiple megahertz. It just seems that if they can do it with all that junk on board, why can't I simply buy an amp/ADC part?

Not to beat a dead horse...but, It seems like it would make more sense to have amplifiers and ADCs on the same part, and the microprocessor separately [heck of alot more digital on the uP than in the ADC!!...I think!!]. You can then pick the best analog solution and the best digital solution separately...just venting my frustration I guess...driving a 100pF switched input capacitance with a 200ns settling time is a tough problem to solve.

 

[itsmoked]

Whoa! Whoa! You have been 'sucker played' by the turkey chip makers marketing! You would be very hard pressed to get even 12bits out of any processor based A2D regardless of their lying claims! When they say 16b or often even 12bits all they are saying is, "when you take a reading from the A2D unit the answer will be contained in a 16bit or 12 bit number".
The [useful] result will certainly not be 16b or 12b. It is a total scam. Look at the noise levels and you will see that the lower bits are drowning in noise. Sometimes using massive oversampling you can claw back a bit or two but then it really isn't a 'that many' bit A2D - you are just stretching a lower bit A2D with sampling.

Keith Cress
Flamin Systems, Inc.-

http://www.flaminsystems.com

 

[sanlog]

Sounds like you've been here before. Your link says you do analog circuit development. How much experience do you have in this area (i.e. low noise, 16 bit data acquisition, 1 MHz conversion rate)? Are you able to do work in this area under contract?

 

[Skogsgurra]

Sanlog,

Are you in a hurry? Do you need the speed of an SAR AD? There are many other techniques that do a decent job. Like the sigma/delta, which is a onebitter used with lots and lots of oversampling and filtering. But does a good job if you are in the (lower) sub MHz range. Those used to have poor DC performance but are good even in that respect these days.

Gunnar Englund

http://www.gke.org

 

[sanlog]

Yes, I need the speed of the SAR. Multiplex about 16 signals. A delta-sigma can never work. I think the best part gives a conversion rate of about 1kbps.

 

[itsmoked]

Well a system that does MSPS is more than just a fast A2D. There is, of course, a truckload a data spewing out of the conversion system that must be funneled somewhere and messed with. I do a lot of A2Ds usually not real fast and rarely in the 16bit arena. A great deal of any success resides in the minutia of the board layout. 16 bits takes attention to every detail and a making certain concessions. Would certainly be a challenge.

Can't solicit in these forums so I shan't be able to respond to your question in here. Sorry.

Keith Cress
Flamin Systems, Inc.-

http://www.flaminsystems.com

 

[MacGyverS2000]

At my "current" job (I write "current" as I'm currently looking for another one... anyone need a DSP/embedded software engineer? [] ), we have 16-bit A/D and D/A converters on the board. One of my first tasks was to create a self-test mode for our ATPs (automated test procedure) that looped back the two units and checked for errors.

After a few simple tests, I tried to convince the higher ups the first four bits could be absolutely disregarded, and possibly the next two. Even the datasheet said the converter's lower 3 bits were in the noise floor! The looks I received were less than friendly, and essentially I was told to "make it work".

Here we are a year later, and the hardware guys are asking me to "fix" my tests... "They never worked in the past, almost every board has failed your test, so we ignored all but the most glaring errors." Well, they're starting to find out their 16-bit units are only good for a whopping 8 bits. Yes, that's right, most microcontrollers with built in units would do a better job.

I guess that broken test of mine was actually doing its job after all. Point is, getting 16 bits isn't nearly as easy as some of the engineers would like you to think, and certainly not the marketing guys.


Dan - Owner

http://www.Hi-TecDesigns.com

 

[sanlog]

You said : <<Yes, that's right, most microcontrollers with built in units would do a better job.>>

I'm confused. Are you saying that you think the microcontrollers with 16bit ADCs DO WORK???

 

[itsmoked]

Easy.. Not even in the same paradigm as the word "easy". Cracks me up. If you go to the A2D web sites and look at the high bit count A2D's, if they are honest, like National Semi,(not like Linear Tech), somewhere they will show ENOB or Effective Number of Bits. Their 16b 2.5MSPS A2D is good for about 12-13ENOB.

Keith Cress
Flamin Systems, Inc.-

http://www.flaminsystems.com

 

[Warpspeed]

I can confirm the difficulty of all this. While the manufacturers spec sheets look fine and dandy, doing it on a real circuit board with real signals sure ain't as easy as it looks in the book. Usually the lower three bits just end up as noise.

Sometimes, the only way to get back anything like the full clean sixteen bits is by averaging several readings in software. Obviously that kills speed stone dead, but it is somewhat effective at recovering the resolution.

Another approach is to use several a/d converters instead of multiplexing multiple incoming analog signals. Not vey elegant either, but it cuts down on the requirement for conversion speed, and lower cost parts could be used.

I am glad I am now retired, and can spend more time in the garden away from all this stuff, hehehe.

 

[MacGyverS2000]

sanlog, my "better job" comment was in regards to a microcontroller's A/D doing a better job than we were with a separate 16-bitter, considering we only had 8 bits of useful resolution. To repeat what others have said, though, you will not get 16 bits from a microcontroller (unless you're willing to oversample 1,024 times). You can get it from a stand-alone chip, but you have to play a lot of games (and win) to do it). I don't envy the guys who try to work with 24-bit ADCs and are expected to get something useful out of them.

Dan - Owner

http://www.Hi-TecDesigns.com

 

[sanlog]

Okay...I get the message loud and clear. Thanks for the advice from everyone. I've got to put together a budget outline by Friday and obviously price is what wins...unfortunately. I may be forced into a microcontroller solution. These standalone 16 bit devices run over $20 EACH!!

...a few of you guys have said I should oversample to extend the resolution. I studied that this weekend. It is my understanding that oversampling will only increase the resolution if the noise is white (i.e. truly random). It seems that the noise on the microcontroller will not be random but rather deterministic (i.e. at least a clock and maybe looping instructions). (1) Do you guys agree that the microcontroller noise is deterministic? (2) If so, how does this improve the resolution on the microcontroller? It seems like I would get a new rms error everytime the deterministic part changed (i.e. Loop on Code segment A, then Loop on Code segment B).

Maybe I'm missing something here??

 

[MacGyverS2000]

If the noise is deterministic (and you can determine its exact nature... there's the rub), you can arguably remove it from the samples. For all intents and purposes, though, I would say you could consider the noise "white" in this case (however, I highly suggest running some tests).

That said, I think you may be chasing your tail. If you really need 16 bits, you're not going to get it out of a microcontroller-based ADC... you need to go with a separate chip. If you don't need 16-bits, the stand-alone ADC will be cheaper, and you may save yourself some misery by using a separate chip. Remember, bit for bit, you're going to get better performance out of a stand-alone chip compared to a micro with an embedded ADC.

From the sound of it, you need to either relax your range requirements or expect to get rid of noise by oversampling in software and/or hardware.


Dan - Owner

http://www.Hi-TecDesigns.com

 

[sanlog]

Dan + all - At what resolution would you recommend an on-chip ADC? These 12 bit solutions from Microchip? Like you said, maybe I argue to go down in precision to save money. I don't think I can go below 12 though.

 

[MacGyverS2000]

I've successfully used 8 bits without issue, 10 in some cases with a bit of oversampling. I question the accuracy of 12 bits, and at that level it will vary significantly with your layout (possibly from chip to chip).

Your increased SNR (in dB) becomes 10logN, where N is the number of times you oversample. In particular, for every extra bit of accuracy you desire, you need to oversample by 4 times (1 bit is roughly 6dB). Assuming you can get 10 bits of accuracy from a 12-bit micro's A/D, you'll need to oversample 4,096 times to get your extra accuracy.

You can try to build a Ferrari for the price of a Yugo, but in the end all you'll have is vehicle that performs like a Yugo. You need to either convince the powers that be to up your budget or expect a Yugo in the end.


Dan - Owner

http://www.Hi-TecDesigns.com

 

[itsmoked]

Yep, right with Dan on all this!

8 bits: No problem. $0.80

10 bits: No problem if you watch things like layout, and filtering. $1.80

11 bits: Now you must pay close attention to all details. $3

12 bits: The point 'difficultly' and the real possibility of failure rears it's head. Requires multilayer boards, filtering , off CPU A2Ds, sample and holds. $9

13-14 bits: (See 12 bits) Plus shielding, excellent parts, and layout. $17

15-16 bits: (See 13-14 bits) Plus, nearly impossible, entire design must be hand crafted to serve the front-end converter, large concessions to the front-end and layout, expensive parts, thermal management, thermocouple junction effects, shielding from local radio emissions, solder types, lead chemistry, highest grade references, luck. $65+

No 12 bits out of a CPU, sorry. (Not allowed!)[flush]

Keith Cress
Flamin Systems, Inc.-

http://www.flaminsystems.com

 

[sanlog]

I think I've got to go back and reevaluate the requirements. Thanks for all the insight, guys (and gals if any).

 

[Skogsgurra]

Smoked,

That's a very clear picture of the problem. Only, I guessed $$$ to rise even more sharply at the end.

Gunnar Englund

http://www.gke.org