[evlatests] More on 3 and 8 bit sampler comparisons

[Rick Perley]

    A further review of the 30 minute test taken Friday afternoon 
reveals some other interesting features. 

    To recap:  Michael/Ken observed 3C286 and a nearby blank field for 3 
minutes.  The two targets were observed cyclically, with 1 minute on 
3C286 and 2 minutes on the blank field.  The correlator was set up to 
observed on IF pair with the 3 bit samplers (I presume this is the AC 
pair), and the other pair with 8 bits.  The LOs were set so the 
frequency coverages were identical.  There are 8 subbands in each half, 
with 128 MHz subband width, and 2 MHz channel resolution.  All four 
correlations are present. 

    Some further observations:

    A) Phase Stability.

    Antenna 27, in the 8-bit path (BD?) is jumping phase by 50 to 100 
degrees each time we returned to the calibrator.  The jump is identical 
in the two polarizations.  The phases are stable within each 1 minute 
observation.  These jumps are not present in the 3-bit path data -- 
indicating this is an LO problem unrelated to the samplers. 
    Otherwise, the phases a varying in a manner which would be expected 
for this band at this time of year -- weather.   The in-scan variations 
are identical for the 3-bit and 8-bit paths. 

    B)  Amplitude Stability. 

    There are significant differences between the 3-bit and 8-bit paths, 
with the 3-bit being considerably worse.  Amplitude gain variations 
within the 30 minute duration are typically 5% for the 8-bit path -- but 
as high as 15% for the 3-bit path.  Variations in the 8-bit gains are 
the same for the opposite polarizations.  For the 3-bit path they are 
quite different. 
    There is in addition a very large charge in 3-bit path gain at the 
beginning of the run for most IFs.  For some, this change is enormous:  
22 RCP changes by nearly a factor of two! (That's 6 dB in power 
units!).  27 RCP changes by 2.5 dB.  These changes are not stepwise -- 
they change continuously over the initial 5 minutes, then settle down 
(more or less).   The sense of the change is that the initial power 
(both autocorrelation and X-correlation) is too high for these two IFs. 

    C) Internal Birdies.

    The 3-bit path shows strong, and frequency-variable birdies in some 
of the IFs.  These are not seen in cross-power, only in the 
autocorrrelation spectra.  These are available for antennas 15, 22, and 
27.  For some reason 28 has no autocorrelation spectra available.  For 
the three with data:
       Antenna 15 shows two birdies in both RCP and LCP, whose 
frequencies change by over 100 MHz over a ~10 minute time span, after 
which they remain at  constant frequencies.   The RCP birdies are in IF 
5 and 8.  The LCP birdies are in IFs 3 and 7. 
    Antenna 22 has a very strong birdie in IF 6, and a weak on in IF 3 
which is anti-symmetric -- its frequency decreases over the time, while 
all other birdies have an increasing frequency.  This is seen in RCP 
only.  LCP has no visible birdie. 
    Antenna 27 has birdies in both polarizations, both increasing in 
frequency with time.  The oppositely polarized birdies are at similar, 
but different frequencies. 

    These birdies are *not* seen in the 8-bit path. 

    D) External RFI.

    Both 3 and 8 bit shows the same external RFI near 6000 and 6250 
MHz.  The latter of these is very strong.  These are due to the 
microwave relay links passing over the site... 
    It would be much easier if in the future we find an empty 1 GHz 
chunk of frequency span for these tests ...

    E) Autocorrelation Spectra

    These show some peculiar problems -- they are negative on some 
subbands, at some times.  Both the 3-bit and 8-bit autocorrelation 
spectra (RR and LL) show this.  These is no relationship with external 
RFI, or other identifiable characteristics -- they just go negative. 
    As noted yesterday, the RL and LR autocorrelations show strikingly 
different delays for the odd-numbered and even-numbered subbands.  
Something is messed up ...