How quiet are ADCs and DACs on micro controllers?

[Joerg]

Hello All,

There are some data sheet entries stating the SNR, for example for the
16-bit converter on the MSP430F2013. However, it isn't too clear what
kind of code it was running at that time and in my experience it is
rather difficult to get that information out of manufacturers.

Anyone with noise experience? How about the DACs? How much noise would
be riding on them if you held their data register constant and looked at
the analog out?

Other than for mundane 8-bit jobs I always used external ADC/DAC in the
past but with newer devices there are often not enough port pins to do
that. I could measure it and still hit a snag. Just want to avoid the
situation where I am told after tons of work "Yeah, but you should not
have accessed the HW multiplier thingie during integrate and hold" or
something like that.

 

[PeteS]

Hello All,

There are some data sheet entries stating the SNR, for example for the
16-bit converter on the MSP430F2013. However, it isn't too clear what
kind of code it was running at that time and in my experience it is
rather difficult to get that information out of manufacturers.

Anyone with noise experience? How about the DACs? How much noise would
be riding on them if you held their data register constant and looked at
the analog out?

Other than for mundane 8-bit jobs I always used external ADC/DAC in the
past but with newer devices there are often not enough port pins to do
that. I could measure it and still hit a snag. Just want to avoid the
situation where I am told after tons of work "Yeah, but you should not
have accessed the HW multiplier thingie during integrate and hold" or
something like that.

Well, the biggest issue is PSRR of the internal components, and whether
they are separated from the main power supplies.

Like you, I don't trust the ADC/DACs in microcontrollers for anything
except mundane, non-critical things. Therefore, I have never
characterised microcontroller ADCs/DACs as I simply use them where noise
is not a major issue.


Cheers

PeteS

 

[Kingcosmos]

Hello All,

There are some data sheet entries stating the SNR, for example for the
16-bit converter on the MSP430F2013. However, it isn't too clear what
kind of code it was running at that time and in my experience it is
rather difficult to get that information out of manufacturers.

Sometimes it takes perseverance to get the information you need. I
looked at the datasheet and there are testing conditions stated for the
SNR of the ADC. That is usually all the information that will be given
up front.
In my experience (junior engineer here), manufacturers hesitate to give
out numbers that are not explicitly in the datasheet. Through my
discussions with several manufacturers the line of reasoning is this:
Even if you were to wrench out test data from one of the senior
application engineers or even a designer, the data given is considered
to be a 'typical' and not a MIN or MAX which should be guaranteed. The
data usually comes from a test set up where a part was placed in the
configuration you asked and tested. It is not production tested nor is
it characterized.
So you are given a number and go off to purchase x thousands of
devices. A large percentage of these devices do not meet the test data
you received. WTH?! Well, the hell is that it is not a guaranteed
spec. There is not that lovely Gaussian distribution that can be found
when designed to a spec or characterized through production tests. So
manufacturers usually put the onus on you to test.
Unless someone has an idea of how much noise to expect, I would test it
in my own circuit and hopefully I will be comfortable with the results.
I would lean toward the idea that the ADC is 'isolated' from any other
internal sections of the micro-controller. Could be wrong...all
devices are different.

 

[Joerg]

Sometimes it takes perseverance to get the information you need. I
looked at the datasheet and there are testing conditions stated for the
SNR of the ADC. That is usually all the information that will be given
up front.
In my experience (junior engineer here), manufacturers hesitate to give
out numbers that are not explicitly in the datasheet. Through my
discussions with several manufacturers the line of reasoning is this:
Even if you were to wrench out test data from one of the senior
application engineers or even a designer, the data given is considered
to be a 'typical' and not a MIN or MAX which should be guaranteed. The
data usually comes from a test set up where a part was placed in the
configuration you asked and tested. It is not production tested nor is
it characterized.
So you are given a number and go off to purchase x thousands of
devices. A large percentage of these devices do not meet the test data
you received. WTH?! Well, the hell is that it is not a guaranteed
spec. There is not that lovely Gaussian distribution that can be found
when designed to a spec or characterized through production tests. So
manufacturers usually put the onus on you to test.
Unless someone has an idea of how much noise to expect, I would test it
in my own circuit and hopefully I will be comfortable with the results.
I would lean toward the idea that the ADC is 'isolated' from any other
internal sections of the micro-controller. Could be wrong...all
devices are different.

This hesitation to give out additional info can also have legal reasons.
It seems that mfgs have become much more fearful there. Recently I had a
part that misbehaved under certain conditions (sudden load changes) and
this was not mentioned in the data sheet. So I sent them the schematic
and all but I was unable to convince them to throw that onto SPICE
together with the model of their part, which I can't do because they
won't disclose that model. Maybe they were afraid to find a rat
somewhere in the innards of the chip, who knows. Instead they insisted
on scope plots which were nearly impossible to provide because the
lock-up and subsequent kablouie happened only occasionally.

I have also tried numerous times to get more info on uCs such as the MSP
series. Nada, zilch. It felt like kicking a big oak tree. Nothing beyond
what's published already. Suffice it to say that it has cost them
several design-wins because I decided to go the analog route like usual.

 

[Rene Tschaggelar]

Hello All,

There are some data sheet entries stating the SNR, for example for the
16-bit converter on the MSP430F2013. However, it isn't too clear what
kind of code it was running at that time and in my experience it is
rather difficult to get that information out of manufacturers.

Anyone with noise experience? How about the DACs? How much noise would
be riding on them if you held their data register constant and looked at
the analog out?

Other than for mundane 8-bit jobs I always used external ADC/DAC in the
past but with newer devices there are often not enough port pins to do
that. I could measure it and still hit a snag. Just want to avoid the
situation where I am told after tons of work "Yeah, but you should not
have accessed the HW multiplier thingie during integrate and hold" or
something like that.

Much is dependent on your layout. I'm not too familiar
with the mentioned family to go into details. Try
averaging to reduce the noise if cpu-time permits.

Rene

 

[Nico Coesel]

Hello All,

There are some data sheet entries stating the SNR, for example for the
16-bit converter on the MSP430F2013. However, it isn't too clear what
kind of code it was running at that time and in my experience it is
rather difficult to get that information out of manufacturers.

I doubt they can produce it. It depends too much on the usual suspects
(board layout, decoupling, running code, frequency, etc, etc). I'm
quite sure though that the specs in the datasheet are produced in a
best case scenario (which may be a simulated environment).

I think creating a test circuit and do some measurements is the only
way to know for sure.

 

[Joerg]

Hello Rene,

Much is dependent on your layout. I'm not too familiar
with the mentioned family to go into details. Try
averaging to reduce the noise if cpu-time permits.

The layout will be RF engineering style. I am more concerned about
things on the chip itself. That's behind the bond wires where there is
not much I can do other than averaging and making sure that segments of
the uC are not jerked on and off in terms of power consumption.