AD7606C System Gain Calibration

Hi Rehfi,

The gain register does not compensate for LSBs, but for kΩ of series resistor in front of AD7606 inputs. If you have a 10kΩ in series with the analog inputs (e.g. as part of an RC filter), writing 0x0A into the gain register will remove any gain error introduced by that resistor. Please find attached a couple of resources that may help understand this feature.

What are the benefits of Software mode compared to Hardware mode or AD7606 - Documents - Precision ADCs - EngineerZone (analog.com)

The On-Chip Calibration Benefits of New Simultaneous SAR Analog-to-Digital Converters | Analog Devices

Regards,

Lluis.

Lluis,

that article was very helpful, thanks a lot. I didn't find it when I searched - seems that I was unlucky with my choice of search terms. Can I assume that this is the "A.N. 2011" mentioned in the FAQ?

I understand the intention of the gain calibration, and why it is in kΩ. But still, an equivalent in LSB (even when its only a guesstimate) would be nice, an if its just for an estimate if the calibration is good enough for my application.

But let's check the article:
- Figure 8 has its labels wrong (AD7606B should be dark blue, AD7606C light blue), otherwise I can reproduce this graph.
- Figure 14 is as similar graph also in the AD7606B datasheet, but it introduces a factor of 0,5 for the "Error (% of FSR)". Error (LSB) is 2^16*Error(% of FSR)/100, which seems correct to me
- Figure 15 is as similar graph also in the AD7606C datasheet, but I have no idea how this was calculated. Even Error (LSB) / Error (% of FSR) relationship is broken. And, even better, the graph fits in no way the value given in Table 2 (which seems to be about right, given the 0,5 factor)

Where does the factor of 0,5 for the "Error (% of FSR)" come from?
What's going on with Figure 15?

Thanks a lot,
rehfi

Hi rehfi,

The A.N. 2011 is the pdf posted in first link: (+) What are the benefits of Software mode compared to Hardware mode or AD7606 - Documents - Precision ADCs - EngineerZone (analog.com), but it was never released to the webpage. will follow up.

-Fig 8. You are right, the larger the R_in, the lower impact an Rfilter has on the gain error.

-Fig. 14. You're right. Depends on whether you express it as a function or FS or FSR you divide by 2^25 or 2^16.

-Fig 15. You are refering to the left y-axis on tha a) figure?  wondering if I mixed up AD7606C-18 and AD7606-16 plots? i.e. divided by 2^17 instead of 2^18. I could look find the code I used to plot this data, thanks for the spot.

Regards,

Lluis.

Hi Lluis,

Lluis said:

-Fig 15. You are refering to the left y-axis on tha a) figure?  wondering if I mixed up AD7606C-18 and AD7606-16 plots? i.e. divided by 2^17 instead of 2^18. I could look find the code I used to plot this data, thanks for the spot.

Seems like you used a Rin of 5,2 MΩ and 17 bit. Maybe intermingled AD7606B & AD7606C-18?

Regards,

rehfi

Looks like it, yeah. 

Regards,

Lluis.

Hi Lluis,

With the article and the corrections given I can now calculate want i wanted to know. Thanks a lot.

I would love to see the datasheet corrected, but knowing how it is... 

As a side note: I like the idea behind the System Calibration Features, and letting the user correct for the external resistor while compensating the internal input impedance is quite nifty. One or two more bits would have been nice, and basing the calibration on a even 1000 R resistor, but there may be trade-offs involved, so this is what was possible.

Thanks again, also for your quick responses!

Regards,

rehfi

Hi rehfi,

Thanks for the feedback and for reporting those errors, I will try to get those corrections in, but unfortunately the process takes some time. 

What do you mean by " basing the calibration on a even 1000 R resistor". This is the minimum change on resistance that you can compensate for, if you look at the saw-shaped plots, the worst-case error is still quite good. Would you have expected finer resolution on the ch_gain register?

Thanks again!

Lluis.

Hi Lluis,

what I meant was that it would be nice when the steps I can compensate for would be 1000Ω, and not 1024Ω. I guess I have an idea where the "1024" come from, but 1k steps is what I can usually buy and what will be commonly used.

For the fun of it I looked up if I can buy 1024Ω resistors at Digikey, and it turns out that I actually can (Y11211K02400T), but in a clunky package and it costs a fortune.

But that's mostly nitpicking. And in the end 1000Ω vs 1024Ω doesn't really matter much.

I would not expect a finer resolution on the ch_gain register (although it has free bits!), but I would not say "no" if I could compensate the gain error even further down to 8 or 4 LSB PFS and NFS Error. But, as said, there may be trade-off involved when implementing this and just may not be so easy. It is how it is.

Regards,

rehfi

Hi Lluis,

what I meant was that it would be nice when the steps I can compensate for would be 1000Ω, and not 1024Ω. I guess I have an idea where the "1024" come from, but 1k steps is what I can usually buy and what will be commonly used.

For the fun of it I looked up if I can buy 1024Ω resistors at Digikey, and it turns out that I actually can (Y11211K02400T), but in a clunky package and it costs a fortune.

But that's mostly nitpicking. And in the end 1000Ω vs 1024Ω doesn't really matter much.

I would not expect a finer resolution on the ch_gain register (although it has free bits!), but I would not say "no" if I could compensate the gain error even further down to 8 or 4 LSB PFS and NFS Error. But, as said, there may be trade-off involved when implementing this and just may not be so easy. It is how it is.

Regards,

rehfi