Sigma Delta ADCs

[dknguyen]

Hi all. I've been looking at Sigma Delta ADCs lately and comparing them to SARs ADCs and it never occurred to me that sampling rate might be different than bandwidth. I used to think that the bandwidth of an ADC is half the sampling frequency, but apparently not? There are 200ksps ADCs out there with bandwidths of 4MHz, so I am not quite sure what this means, unless it is referring to the instananeous frequency components measured in a single sample (which doesn't seem to make sense).

I am also a bit confused about the the input bandwidth of a sigma delta ADC. The maximum bandwidth that you *should* input to a sigma delta ADC is half the output data rate, right? Not of the sample rate which is much much higher and includes oversampling?

 

[Hero999]

I've hear about this before; how the bandwidth can be higher than the sample rate but unfortunately I can't help you because I don't understand it.

 

[dknguyen]

Crazy electrical engineers...oh wait.

There is something else I am confused about. All the ADCs I have ever looked at (they just happened to be SARs ADCs) have a reference voltage that is equal to the maximum analog voltage. But when I look at the Analog Device's ADCs, their references seem to be all over the place. I thought reference voltages were always the maximum input voltage, but these ADCs seem to do things differently. They seem to use references like (Vmax-Vmin)/2 and other fractions like (4/5) x Vmax, etc. Not only that, for some reason they have ADCs that are powered by 5V, but are only able to accept 2.5V absolute maximum analog input (rather than the supply voltage 5V) which I have never seen before- and seems needlessly complicated.

Is the reference voltage used (ie. limits of the input range, midpoint of input range, some fraction of the input range) dictated by the type of ADC? Or is it completely arbitrary depending on the exact structure of the ADC?

 

[phalanx]

I've hear about this before; how the bandwidth can be higher than the sample rate but unfortunately I can't help you because I don't understand it.

Subsampling is the technique you are talking about. It only works with periodic signals. If you want to view a non-periodic waveform, it still has to have a frequency less than half your sampling rate.

Bitscope has a good writeup on subsampling that is a lot better than I can describe in the short amout of time that I have. Take a look at this link for more details:

https://www.bitscope.com/design/hardware/convertor/?p=3

 

[phalanx]

Hi all. I've been looking at Sigma Delta ADCs lately and comparing them to SARs ADCs and it never occurred to me that sampling rate might be different than bandwidth. I used to think that the bandwidth of an ADC is half the sampling frequency, but apparently not? There are 200ksps ADCs out there with bandwidths of 4MHz, so I am not quite sure what this means, unless it is referring to the instananeous frequency components measured in a single sample (which doesn't seem to make sense).

I am also a bit confused about the the input bandwidth of a sigma delta ADC. The maximum bandwidth that you *should* input to a sigma delta ADC is half the output data rate, right? Not of the sample rate which is much much higher and includes oversampling?

The 4MHz bandwidth that is listed is likely the limitation of the analog front-end found in the converter. You can feed periodic signals up to 4MHz to the converter and use subsampling to reconstruct the waveform. If you exceed 4MHz, the analog components will start to attenuate the signal you are inputing.

If you are trying to sample a non-periodic waveform, you are still limited to signals with a frequency that is half of the sampling rate. For best results and glitch capture, you want to sample as fast as is economically possible. The more samples per unit time, the higher the time resolution of your capture.

 

[hjames]

ADC's usually have a sample-and-hold circuit in front of them which actually captures the analog value. A typical design doesn't try to make this too much faster than the actual conversion time, but it is occasionally important (as phalanx mentions). In high-end apps this is occasionally used when interleaving ADC's to get higher bandwidths (i.e in the 100's of MHz - GHz).

1/2 ADC references are pretty common - I'd expect that it's easier to deal with values accurately when they aren't hitting a rail.... I don't think I've ever seen any oddball fractional ones though - do you have a part number?

 

[dknguyen]

I found out why. It's because they were bipolar ADCs isntead of unitpolar. So what happened was the reference might be +2.5V, but the input range would be +/-4x2.5 = +/-10V. IN addition, what would happen is that if the ADC had an input range if +/-2.5V, and was being operated in unipolar mode, it's would be physically limited to the range of 0-2.5V.