16-bit a/d

[jones2]

Posted on August 25, 2004 at 11:07

16-bit a/d

[jones2]

Posted on May 17, 2011 at 12:02

Is there a way to combine, two 8-bit analog inputs into one 16-bits.

I dont mind the delay between the two samples taken.

[danielh1]

Posted on May 17, 2011 at 12:02

I hoped than somebody else would answer... I am sorry to say that even with ten 8-bits A/D converter you will not get a 9-bits converter. By comparison, if you can measure a distance to a precision of 1mm with a ruler, ten identical rulers won't give you 0.1mm...

So, look for a 16 bits A/D converter if you are sure you need it. And keep smiling: every problem comes to a solution. Daniel

[jdaniel]

Posted on May 17, 2011 at 12:02

Asterix,

I'm not entirely sure that's true. I actually think your analogy might be flawed. Assuming you had access to a rule with a 0.1mm precision SOMEWHERE, you could carefully stagger the starting point of your 1mm rulers in 0.1mm increments and with their combined readings get the precision you want. In the A/D example, he could possibly achieve his result if a huge swath of uncertain values in the middle of the range were acceptable (which is unlikely). That is, imagine you have two (stand-alone, not the ones in the uPSD) 8-bit A/D converters that can be scaled externally. If you have a 0-5V range on your input signal, one could be scaled so it's full-scale range is from 0-2.5V and the other from 2.5V-5.0V and their combined reading could give you ''more LSB's essentially.'' The problem is that resistor tolerances and a host of other factors will make the few LSB's surrounding that crossover reading exceedingly ugly. Regards, pHaze426

[danielh1]

Posted on May 17, 2011 at 12:02

Hi Phase426

First, for the ruler example: if you can evaluate 0.1mm on your ruler, a second ruler won't give you more than that. Back to the 8-bits A/D converter: it gives you an 8 bits digital value, period, no staggering possibility. It is true that cascading two converters and re-scaling the second one, would give you an additional bit, if it worked... But, as you stated it yourself, the whole problem lies in re-scaling, and not only because of the resistor divider limited precision. By reducing the reference voltage, you increase the converter's noise, so that the gain (if any) in ENOB (effective number of bits) is certainly less than one. By tripling the price (two A/D converters + a high precision divider) you get much less than simply using a 10-bits A/D converter for a few additional cents. Another approach could be averaging the A/D's output. This could work if the noise was truely white, which is usually not the case. After all, sigma-delta converters are actually 1-bit converters with digitally filtered white noise. The bottom line is: if you need a 16-bits converter, just buy one, or build one from scratch if you like hard work. By the way, this discussion is completely out of scope, but why not have some fun... Daniel

[jdaniel]

Posted on May 17, 2011 at 12:02

Asterix,

My comparison to your ruler example suggested that if *I* had a single ruler that could measure 0.1mm and used it to arrange 10 rulers that could only measure 1mm, I could then sell that complicated sub-standard arrangement of rulers to someone and it could be said to be able to measure with greater precision in its composite. Why would someone want to do that? I guess if you had a huge box of rulers laying around. Along a similar vein, I agree that scaling a few different A/D's is not nearly as good as getting a better precision A/D to begin with. The same scenario as above applies, however. Why would you want to do this? If this person has a box of 8-bit A/D's laying around and wants to unload them in some design, then he could probably buy himself a bit more precision if he was willing to make some (very serious) concessions. As far as noise gain, the specs on most modern A/D's are exceedingly good even at relatively low price-points. If someone is willing to make the concessions about the uncertainty around the crossover region, I sincerely doubt if a doubling of the noise gain is going to be what bothers them. And I guess we are out of topic here, but hey... there's A/D's in the uPSD series. It's relevant enough for me. Best Regards, pHaze426