Unfortunately the original message has been filtered out by the large
attachment filter.
So sorry for the bad quality of the reszized picture.

----- Original Message -----

Hello Mickle,

ADC #13 is just my "serial number" of ADCs
Up to now all (#1-#14) are based on the 24 bit sigma delta LTC2400.

On the photo the left pcb has:
lower left photocoupler to isolate from PC (RS232 either direct or via USB
serial cable).
upper left PIC12F675 for reading ADC  + storage of all calibration constants
like
nominal reference voltage, 3rd order temperature correction,
2nd order linearity correction, serial number etc.
upper right: AD586LQ. Note that only Pin 4 is soldered directly to the pcb.
Other pins with Vero wire to remove PCB stress (humidity) from the
reference.
The orange part at the AGND pin is the temperature NTC.
lower right: LTC2400.

Foreground: voltage stabilisation with low noise LT1763 to 14V (minimum
needed for AD586)

right PCB:
Capacitive 2:1 voltage divider with LTC1043 + LT1050 buffer.
The noise from ADC input is isolated from LT1050 buffer
by a  R/C low pass. Otherwise the switching from ADC input will
increase noise level well above the datasheet spec.

When measuring all (ADCs + other references) is mounted on a large metal
ground plane.
Clothes over the ADCs + connections keep air currents away.

Since the noise level of LTC2400 is rather high, I have to use large
integration times
(averaging several hundred measurements over 1 or 5 minutes) to reduce
the noise from a single measurement of 10uVpp (0.3ppm RMS in the datasheet)
to below 1uVpp.

For further experiments I will probably have to change to a other ADC with
less noise
like LTC2440 or AD7190.

But the LTC2400 has some advantages that I'm missing on other ADCs.
first is the +/-12.5% (or at least some millivolts when regarding linearity)
overrange without clipping of the ADC-value.
This makes offset + full scale calibration more easy than on other devices
which are clipping the reading at zero + full scale.
The next is the very low gain (0.02ppm/K) + offset drift (0.01ppm/K)
compared to other sigma delta converters.
And finally the 4 sub-bits below the 24 bits (giving a readout of totally 28
bits) would help
on a low noise ADC like AD7190 to get a more gaussian distribution of the
values.

With best regards

Andreas

----- Original Message -----
From: "Mickle" timka2k@yandex.ru
To: "Discussion of precise voltage measurement" volt-nuts@febo.com
Sent: Tuesday, January 29, 2013 8:31 AM
Subject: Re: [volt-nuts] Some questions to zeners (1N823-1N829)

Thanks for the well-explained photo Andreas, I guess you etched your own PCBs?

By the way, the data sheet for the LTC2440 shows that it sends a 32 bit word, including 4 LSBs beyond the 24 bit result. I found that while those bits are logically present, only two of them are "real" data bits, the last two are fixed. b1 is always low and b0 is always high. I contacted Linear Tech support and they confirmed it:

"I ran your schematics and questions through our applications engineer and he mentioned what you are seeing is normal. The two LSBs are far below the noise floor and have no real influence but I agree this should be clarified in the datasheet." -Uma D, Product Marketing Engineer, Linear Tech  Nov.5 2012

If enclosed PNG image makes it, the scope screenshot shows an example. Data is valid on the rising edge of the clock. Light blue traces on the DATA channel (2) show previous data words (scope display-persist mode).

Both the LTC2440 and AD7190 can run at higher output rates, with higher noise. So, for example if you run at 10 Hz output rate (or even 50 Hz or 60 Hz, depending on your local AC frequency) and then do your own averaging, I think you can get useful data beyond 24 bits if your Vin / Vref is clean enough.

I previously sent Andreas a screen capture of the AD7190 eval board software which included a lopsided histogram, but that was in fact a software problem on the PC host (LT was sloppy in making their LabView app).  For histograms with only a few bins, the labview auto-scale function sometimes crops the right-hand edge off the histogram. You see this clearly if you watch it acquiring data live. If you export the raw AD7190 data and look at it in an external graphing program, you can confirm the histogram is symmetric as expected, for any large enough dataset.

Hello John,

Yes this PCB is etched by myself.
Good to know that the LTC2440 does behave somewhat different in transferring
the data than my LTC2400.
I have looked to the data and on the LTC2400 all 4 sub LSBs are toggling.

To the 50 or 60 Hz output rate:
The output rate is not necessarily equal to integration time.
When I measure the input of a LT2400 with a oscilloscope,
I see the swiching noise from sigma delta converter
only the 2nd half 80 ms out of  160 ms output rate (configured for 50 Hz).
The first half of the measurement time the input is passive.

I guess that the LTC2400 does a kind of self-calibration every measurement
cycle.
Otherwise he would not have the offset and gain drift specs.

Good also to know that the AD7190 has gaussian noise.
So I will have to look up the two devices for a further ADC.

With best regards

Andreas

----- Original Message -----
From: "beale" beale@bealecorner.com
To: "Discussion of precise voltage measurement" volt-nuts@febo.com
Sent: Wednesday, January 30, 2013 11:46 PM
Subject: Re: [volt-nuts] bits beyond 24 from the ADC

The data sheets for the LTC2400 and LTC2440 are very specific that
offset and full-scale calibrations are performed on every conversion
cycle.

On Sat, 2 Feb 2013 14:58:19 +0100, "Andreas Jahn"
Andreas_-_Jahn@t-online.de wrote: