[evlatests] first look at the newly installed three bit samplers

[Bryan Butler]

sure looks like 32 (bits 2,3,4) is the most gaussian.  but what matters 
is the noise on the sky - we've been fooled by shapes of histograms 
before...

	-bryan


Ken Sowinski wrote, On 6/22/11 17:07 PM:
> The three-bit samplers have been returned to the antennas with FPGA
> firmware to allow an arbitrary choice of three bits from the six-bit
> sampler output. All the samplers work about as well as they did before
> being removed last week.
>
> I paraphrase Mike's explanation to me of how it works. A scale factor
> is supplied which is used as a multiplier for the six bit ssample.
> After multiplication three bits are chosen from the product with due
> care given to sign and over/underflow conditions. The bits chosen
> from the product are those which result in the "traditional"bits
> 1, 2 and 3 when the scale factor is set to 64. Other powers of two
> can be used to select a diffrent set of three bits:
>
> 16 selects bits 3,4,5
> 32 selects bits 2,3,4
> 64 selects bits 1,2,3
> 128 selects bits 0,1,2
> 255 is meaningless.
>
> Bit 0 is LSB, bit 5 is MSB. Accepted values for the scale factor are
> 0 to 255. Anything over 128 is not useful. In practice there large
> differences in the state count histogram were seen when changing
> between a value of 127 and 128. The scale factor is not restricted to
> a power of two; any value between 1 and 127 may be used.
>
> To see that this really works I recorded the state count histograms
> for values of the scale factor in the set {16, 32, 64, 127, 128, 255}.
> The final value, 255, was used out of exuberance rather than thought
> and should be ignored when perusing the attached plots. At each value
> of scale factor the downconverter output attenuator was used to achieve
> a sampler RMS of about 1.65 and the state count histogram was recorded.
> The equalizers were not adjusted at all, and should not much affect
> this result. Vivek kindly assembled all the plots into one pdf file
> which is attached to this note.
>
> The best looking histograms are seen when the scale factor is between
> 32 and 64. This corresponds to using some combination of bits one
> through four. A value of 127 yields a histogram which is very centrally
> condensed; this changes dramatically for 128 to a broad, lopsided
> distribution. All the histograms are attached to this email.
>
> Next I think we should come to some consensus as to what scale factor
> we provides the best result and then make some SNR measurements.
>
> I note that this provides us with yet another gain knob which, in this
> case is equivalent to having adjustable sampler thresholds. If we choose
> to use it, this could accomodate changes in power to the sampler of at
> least 6 dB after the T304 attenuators have been set.
>
>
>
> _______________________________________________
> evlatests mailing list
> evlatests at listmgr.cv.nrao.edu
> http://listmgr.cv.nrao.edu/mailman/listinfo/evlatests