The data table that follows captures the results of this experiment:

* An FCDEV3B modem board was powered from a bench PSU whose output voltage
  is controlled by an analog rotary knob;

* The PSU voltage knob was turned up and down to produce 0.1 V steps
  from 3.2 V to 4.2 V;

* At each PSU voltage step, VBAT MADC reading was captured with an
  mpr 30 command in fc-tmsh, and the actual voltage was measured with a
  digital multimeter.

PSU voltage	MADC reading	Voltmeter reading
-------------------------------------------------
3.2 V		473		3207 mV
3.3 V		488		3306 mV
3.4 V		502		3403 mV
3.5 V		517		3500 mV
3.6 V		532		3597 mV
3.7 V		546		3696 mV
3.8 V		562		3804 mV
3.9 V		576		3897 mV
4.0 V		592-593 [1]	4003 mV
4.1 V		606		4.08-4.09 V [2]
4.2 V		621		4.18-4.19 V [2]

[1] on the cusp of MADC step, got alternating readings

[2] My cheap multimeter's 4 V range was exceeded at this point, and I had
to switch to the next higher range in which only two digits after the
decimal point were displayed.

If we apply the firmware's default VBAT calibration values of A=7000 B=0,
we see that VBAT numbers computed from MADC readings are higher than the
true VBAT by 24 mV at the low 3.2 V end and by 45 mV at the high 4.2 V end.

Calibration math attempt from 3207 to 4003 mV, using MADC=592 at high end
=========================================================================

A=(4003-3207)/(592-473)*1024 = 6849

B parameter computed from each calibration point:

MADC reading	Voltmeter reading	Computed VBAT w/o B	B parameter
---------------------------------------------------------------------------
473		3207 mV			3163			44
488		3306 mV			3263			43
502		3403 mV			3357			46
517		3500 mV			3457			43
532		3597 mV			3558			39
546		3696 mV			3651			45
562		3804 mV			3758			46
576		3897 mV			3852			45
592		4003 mV			3959			44

B values range from 39 to 46, a very small variation - thus the behaviour of
VBAT MADC really does fit the linear model, and linear calibration really does
work well.

Calibration math attempt from 3207 to 4003 mV, using MADC=593 at high end
=========================================================================

A=(4003-3207)/(593-473)*1024 = 6792

B parameter computed from each calibration point:

MADC reading	Voltmeter reading	Computed VBAT w/o B	B parameter
---------------------------------------------------------------------------
473		3207 mV			3137			70
488		3306 mV			3236			70
502		3403 mV			3329			74
517		3500 mV			3429			71
532		3597 mV			3528			69
546		3696 mV			3621			75
562		3804 mV			3727			77
576		3897 mV			3820			77
593		4003 mV			3933			70

B values range from 69 to 77, still a very small variation - thus we are not
too terribly sensitive to errors in the A parameter resulting from MADC values
being on the cusp.