T7 ADC errors at -2.5%FS

EDIT 2016-10-17 20:00 CET:
The error report below states an ADC dead band present for ranges 100mV and 10V only. Further tests shows that it does in fact affect all ranges.
In the test setup below, the error appears more clearly for 100mV and 10V ranges, only because the input signal is 10x smaller than full scale for these ranges.
Newly added "ADC2-5 sampling sawtooth waveform.png" shows the dead-band for all ranges 10mV thru 10V.
This time the hardware voltage divider is changed to drive the ADCs to approx +/- 10% of full scale, for all ranges. 
The setup is:
* Waveform generator: Sawtooth -1 to +1 V, connected directly to AIN5 and GND.
* Voltage divider: AIN5 - 10k - AIN4 - 1k0 - AIN3 - 100R - AIN2 - 10R - GND (I used a CB37)

----------- original report below -------------

ADC readings from my T7 have a large dead-band at around -2.5%FS, for the 100mV and 10V range only.

The attached data were obtained with the following setup: 
1) A signal generator outputs a sawtooth up-ramp with -1 to +1V range. This connects directly to both AIN4 and AIN5.
A voltage divider with 1k0 over 10R (approx 100:1) connects from there to AIN2 and AIN3, also tied together, i.e. AIN2 and AIN3 input is the same up-ramp with approx -10mV to +10mV range.

2) Analog inputs are set to single ended, ranges AIN2 10mV, AIN3 100mV, AIN4 1V, AIN5 10V, all at resolution index 8.
The result is shown in "2016-10-17_07-42-48 (up slope 30mHz AIN2-3-4-5 @ range 10m-100m-1-10 residx 8).png"
AIN4 and AIN5 are shown divided by 103 for easy comparison, across the voltage divider.
AIN3 and AIN5 have a dead-band (stuck code) around -2.5mV, i.e. about -2.5% of FS in both cases. AIN2 and AIN4 do not.
The amplitude of this error is about 300uV for AIN3 (100mV range) and 30mV for AIN5 (10V range), i.e. about 0.3%FS, while the absolute error spec for the ADCs is 0.01% for these ranges. The impact of a 300uV error in the 100mV range, used in thermocouple mode, is about 7K (this is how we first discovered the problem).

3) To determine whether the error follows even/odd ADC channels (input muxes) or ADC range, a comparison acquisition was made with identical HW setup, but reversed ADC range settings: AIN2 100mV, AIN3 10mV, AIN4 10V, AIN5 1V.
The result is shown in "2016-10-17_08-01-54 (up slope 30mHz AIN2-3-4-5 @ range 100m-10m-10-1 residx 8).txt.png"
The error followed range, not channel number.

It appears the problem is in an amplifier/attenuator used for the 100mV and 10V range only.

All data is acquired using Python LJM, in command/response mode. The .txt files contain the data. Columns are
Line, Time(s), AIN2, AIN3, AIN4, AIN5, N/A, N/A

Regards
Flemming Nyboe