[evlatests] Calibrating 3-bit data

[Vivek Dhawan]

Fans of 'evlatests' know that Teledyne 3-bit digitizers show ~10%
lower SNR in visibility data, compared to ideal 3-bit devices. In
addition, the Psum and Pdif [*] are expected to deviate from their
analog equivalents. The 8-bit data have been shown to calibrate to
a few %, compared to the classic method with 3C286 etc. The goal
is to calibrate at least as well with 3-bit Ps and Pd.

[*] Total power & switched power calibration signals, corresponding
to 2*Tsys and the noise diode Tcal, synchronously detected per sub-
band on the WIDAR station boards.

I compared 3-bit vs. 8-bit cal data, to see if the deviations are
predictable; or at least stable enough to derive a correction to
the former.

A 'sampler non-ideality factor' or SNIF is examined (see below).

The SNIF curve for a sampler is found to be stable and repeatable
over Psum excursions of ~3dB, as long as attenuator, equalizer, and
requantizer settings are held fixed. A straight line, or at most a
parabola, corrects the 3-bit pdif to its 8-bit 'good twin', measured
at the same sky frequency. However, most units only vaguely resemble
the analytic prediction [from a preview of a Rick Perley EVLA memo.]

Moreover, when the levels are re-jiggered, changes occur in some
units e.g. when changing bands from K to Q, or just repeating an obs
script with reset gains. [See plots]. Most, but not all units, retain
their slope; some change their intercept, for reasons that are still
opaque. So no joy yet. I hope the plots stir some ideas.

Vivek

attached plots:

X-axis: P_sum, 3bit, with gain correction undone (i.e. what came
out the digitizer into the FIR filter before gain adjustment).

Y-axis: (pdif3/psum3) / (pdif8/psum8). This is (Tcal/2*Tsys) for a
3-bit subband, normalized by the same quantity for its 8-bit twin
at the same sky frequency. The setup uses 8-bit BD overlapping A1C1
and A2C2.  black, red, green, blue are A1/B, A2/B, C1/D, C2/D. The
coloured curves would be 1.0 and flat for an analog system. They are
very nearly so for 8-bit vs. 8-bit (not shown)

Light brown curve: expected from ideal 8-level digitizer with equal
spacing and no bias, compared to a true analog system. [Rick memo]

Light vertical line: Psum value (~15) servoed by the gain adjustment,
before Psum is changed by the antenna tipping.




-------------- next part --------------
A non-text attachment was scrubbed...
Name: kt.pdf
Type: application/pdf
Size: 153586 bytes
Desc: not available
URL: <http://listmgr.nrao.edu/pipermail/evlatests/attachments/20120129/513f0101/attachment.pdf>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: qt.pdf
Type: application/pdf
Size: 150376 bytes
Desc: not available
URL: <http://listmgr.nrao.edu/pipermail/evlatests/attachments/20120129/513f0101/attachment-0001.pdf>