A/D converters, spec.?

Thread Starter

Ismael O

Joined Sep 19, 2020
9
I have received an A/D converter spec that says: bandwidth > 5KHz, sampling rate 2000Samples/second, resolution = 16bits, voltage 0 to 10VDc.

Just want to get an opinion on this requirement. why the bandwidth is especified? In my opinion, For and A/D converter the bandwidth doesn't need to be especified as this value is already known that for most of A/D converters the input bandwidth is in the order of 50Mz.

I also know that the voltage sensor has a bandwidth of 10kHz while below the typical A/D input bandwidth.

Any comments?
 

Papabravo

Joined Feb 24, 2006
16,938
Your assumption about the bandwidth of A/D converters is wildly off the mark. 2000 samples per second is one sample every 500 μsec. Normally you want to have a sampling rate that is a MINIMUM of 2 times the highest frequency component in the input waveform. That limit is theoretical. In practice you want a sampling rate that is more than 2 times the highest frequency component. Since you did not provide a link to the datasheet, I cannot tell you why they quote the 5 kHz figure.
 

Deleted member 115935

Joined Dec 31, 1969
0
I have received an A/D converter spec that says: bandwidth > 5KHz, sampling rate 2000Samples/second, resolution = 16bits, voltage 0 to 10VDc.

Just want to get an opinion on this requirement. why the bandwidth is especified? In my opinion, For and A/D converter the bandwidth doesn't need to be especified as this value is already known that for most of A/D converters the input bandwidth is in the order of 50Mz.

I also know that the voltage sensor has a bandwidth of 10kHz while below the typical A/D input bandwidth.

Any comments?
The bandwidth is given because of the option to under sample.
Nyquest, is often miss quoted,

If you have a sample rate of 2 KHz,
Nyquest says that the maximum bandwidth you can un ambiguously receive is 1 KHz.

it does not say that this has to be 0 to 1 KHz,

your input signal could be say 4 KHz,
with 100 Hz bandwidth

for this the ADC is just fine,

Such techniques are normal on RF systems
we used the same back in the 80's on sonar,

This is a very old link , but a good clear foundation

https://www.analog.com/media/en/tra...actical-Analog-Design-Techniques/Section5.pdf

And this is the wiki
that has a bit more maths
https://en.wikipedia.org/wiki/Nyquist–Shannon_sampling_theorem
 

Thread Starter

Ismael O

Joined Sep 19, 2020
9
Thanks very much for your comments. great information!.

I have just double checked the requirements and it actually says bandwidth DC to 5KHz. and sampling rate of 2000 samples/cycle where the cycle is 50Hz.
Therefore, the sampling frequency will be 100KHz which will capture signals in the order of 50KHz or lower. if my understanding is correct then 5KHz signals will be measured without issues, assuming that the given bandwidth is the expected input signal frequency
Now if I do the reverse analysis, say 5KHz is the maximum frequency that need to be measured then I would need 10KHz as minimum sample rate, I choose 5 time this value then 50KHz should the sample rate of the ADC, Here is a question: why not 10 times? or 8 times? is there a formal factor (based on theory) to choose How many times above the Fs (=2 times signal freq.) should be taken?
 

Papabravo

Joined Feb 24, 2006
16,938
That's odd. I never seen ANYBODY quote samples/cycle and then give a frequency. That seems like a misleading way to do things. How about a link to the part which I suggested might be helpful in post #2. that way we can uncover any "claims" that might be inconsistent with the datasheet. there is something odd going on here, but I cannot figure out what it is.
 

MrChips

Joined Oct 2, 2009
24,401
Now if I do the reverse analysis, say 5KHz is the maximum frequency that need to be measured then I would need 10KHz as minimum sample rate, I choose 5 time this value then 50KHz should the sample rate of the ADC, Here is a question: why not 10 times? or 8 times? is there a formal factor (based on theory) to choose How many times above the Fs (=2 times signal freq.) should be taken?
If your maximum frequency of interest is 5kHz then you need to apply a low pass filter that starts to attenuate frequencies greater than 5kHz. Let us assume that your LP filter starts to roll off with -3dB at 5kHz. A 2nd-order LP filter will have a response of -9dB @ 10kHz and -14dB @ 15kHz. This means that there will still be some frequency components above 5kHz. The Nyquist theorem says that you need to sample at a minimum of 10ksps. A 7kHz component will be folded back and appear as a 3kHz signal. A 9kHz signals is folded back as 1kHz.

Instead, if you sample at 20ksps, input frequencies up to 15kHz can be digitized without adversely affecting your 0-5kHz range of interest. There is nothing gained by sampling at 8 or 10 times the maximum input frequency.
 

Deleted member 115935

Joined Dec 31, 1969
0
The filter bandwidth requirements are defined by your sampling rate ( in Hz )

so your sampling at 2 KHz,
thus the widest bandwidth signal you could ever receive is 1 Khz.
( Nyquist Shannon sampling theory )

The 2 KHz sample rate , also defies where the Nyquist zones

For 2 KHz sampling you have zones of

0 to 1 KHz - first nyquest
1 to 2 KHz - second nyquest
2 to 3 KHz - third nyquest
3 to 4 KHz -- fourth nyquest
4 to 5 KHz -- fith nyquests
5 to 6 KHz -- sixth nyquest
etc

Your signal of interest must never cross a nyquest zone, else it will alias back into the other zones and be un recoverable.

for referance , fig 5 here
https://www.analog.com/media/en/training-seminars/tutorials/MT-002.pdf

You will also note in fig 5 that the odd number nyquest zones are left to right , ie, rising frequency , whilst the even number zones are right to left, i.e. frequency is reversed. Not hard to fix in digits, but easiest to avoid if yo cna go for a odd zone.

Back to the original question:

Your devices 5 KHz bandwidth , which you ask about

This means this ADC when sampling at 2 Khz, can sample a signal in the fith Nyquist zone,

So if you had a signal of say 4.5 KHz, of 100 Hz bandwidth, then you are just fine sampling with this device

Key things to remember

a) Never cross a nyquest zone
b) you must filter "out" all the signals not in the nyquest zone you require,
c) Even number nyquest zones, the frequency is seen backwards !! ( use odd and your ok )
 
Last edited by a moderator:

Deleted member 115935

Joined Dec 31, 1969
0
here a sketch of the circuit. I think specifying the BW of the ADC is not right or is not needed.
View attachment 248461
Thank you , have just seen this


"all" you have to do is think which nyquest zone you want.

What do you want to measure,

If its the average of the 220 V DC,
then your bandwidth is for all intended purposes "zero"
Put a filter of a few Hz on the front end of the ADC,
You have 16 Bits in the ADC,

have look here

https://www.analog.com/en/analog-dialogue/articles/practical-filter-design-precision-adcs.html


You see to use all that number of bits, you need quiet a good filter

If you digitally filter off the bottom 8 bits , and only use the top 8, then the filter requirements become less onerous.


One last thing,

The ADC is only as good as the info feed it,

If you dont filter correctly , you might get16 bits, but the lower bits are alias noise, which is worse than random noise as it can not be averaged out
 
Top