Hi,

I remember during my first interview (in which I got rejected ) I was asked a question of how to realise a 16-bit ADC using two 8-bit ADC and Mux(es)? One is free to use discretes and opamp.

Does anybody know the answer?

 

I know how to do it if you will allow the use of a D/A converter instead of muxes.

 

I think it is a trick question, and there is no way to do it. That interviewer was pulling your leg.

 

A recent Electronics Design has an article that suggests it's possible to get 10 bits resolution out of a single 8 bit ADC, but you cannot string two together for improved resolution.

 

Hi,

I remember during my first interview (in which I got rejected ) I was asked a question of how to realise a 16-bit ADC using two 8-bit ADC and Mux(es)? One is free to use discretes and opamp.

Does anybody know the answer?

Why didn't you ask the interviewer for the answer? Since you didn't get
the job at least you would have learned the answer ;-)

I have never heard of using two 8 bit ADCs to make a 16 bit ADC. I have
heard of using an ADC and a DAC to do a null-and-difference measurement
(which increases the resolution of the ADC).

You measure the input voltage (Vin) with the ADC. You then set the DAC to
the value measured on the ADC (Vdac). You change the gain of the ADC input and
then measure difference between Vin and Vdac.

You need to be careful with the DAC and ADC specs. An DAC with 8 bit resolution
does not usually have 8 bit accuracy. I would be surprised if connecting two
8 bit devices together would get you 16 bits.

(* jcl *)

 

Why didn't you ask the interviewer for the answer? Since you didn't get
the job at least you would have learned the answer ;-)

I have never heard of using two 8 bit ADCs to make a 16 bit ADC. I have
heard of using an ADC and a DAC to do a null-and-difference measurement
(which increases the resolution of the ADC).

You measure the input voltage (Vin) with the ADC. You then set the DAC to
the value measured on the ADC (Vdac). You change the gain of the ADC input and
then measure difference between Vin and Vdac.

You need to be careful with the DAC and ADC specs. An DAC with 8 bit resolution
does not usually have 8 bit accuracy. I would be surprised if connecting two
8 bit devices together would get you 16 bits.

(* jcl *)

That is the method I know also. The first 8 bit A/D has to have 16 bit accuracy if you want true 16 bit accuracy on the output.

 

I've interviewed many people over the years. He was most likely not looking for the *correct* answer as much as to see how you logic through it.

Without thinking about it too much, basically you need to create a situation where each count of the first one covers a range. Then you gain it up and offset it to be zero based for the second to read. The gain and offset depend on the result of the first one. So the 16 bit answer is (RESULT OF 1st)*2^8 + RESULT OF 2nd

 

I don't think you could actually get such a scheme to work with just A/D converters and Multiplexers. Do you have any idea how difficult it is to get 16 bit accuracy on an actual A/D converter?

What do you suppose happens to your system noise when you put in a gain stage of 48 dB?

 

Papabravo, you are completely right that it would be very difficult to get it to work. But, there is nothing in the architecture to make it theoretically impossible.

The point was not to build it, or make it robust, or make it manufacturable. It was to show that it is possible.

 

He didn't say anything about accuracy, only resolution. You would need to split the signal into two halves, then apply them to the ADCs. If you can change references, you can use 1/2ref for the first ADC, then subtract the signal from the first reference. You'd then need to detect when the signal is negative to trigger the multiplexers to go to ground for the MSB.

Steve

note: clearly this is an impractical way to make a 16-bit ADC, since it would work like crap, take up a lot of space, and cost a lot more than any 16-bit adc

 

Steve,

That would give 256+256 total counts range, which is a 9bit ADC, not 16-bit.

 

He didn't say anything about accuracy, only resolution. You would need to split the signal into two halves, then apply them to the ADCs. If you can change references, you can use 1/2ref for the first ADC, then subtract the signal from the first reference. You'd then need to detect when the signal is negative to trigger the multiplexers to go to ground for the MSB.

Steve

note: clearly this is an impractical way to make a 16-bit ADC, since it would work like crap, take up a lot of space, and cost a lot more than any 16-bit adc

Unless I misunderstand what you mean, this will only give you 512 separate levels, equivalent to a 9-bit A/D.