[Gb-ccb] Caltech Backend Telecon Monday 07 July 4pm EDT

[Martin Shepherd]

On Thu, 3 Jul 2003, Richard Lacasse wrote:
> I have not see the suggestion I made last week considered.

Sorry, but I never got this email. I received a response from Roger
Norrod, and then a belatedly forwarded response from John Ford. I
haven't received any other responses to my email. Hopefully now that
there is an official mailing list, this won't happen in the future.

> I suggested using a few cables (2 or 4) that had
>     - twisted pairs,
>     - beldfoil shields over each twisted pair
>     - an overall braided shield.

Isn't this equivalent to my suggestion of using up to 4 multi-pair
Amphenol SkewClear cables? The advantage of using an off-the-shelf
cable like this is that it has documented crosstalk properties.
Amphenol claim that their use of individually shielded parallel pairs
in place of individually shielded twisted pairs, produces a cable with
better and more consistent properties, and it is hard to argue with
the claimed 70dB near-end-crosstalk.

>     - the use of spectrum control or similar filtered connectors and
>       die cast metal backshells (AMP 745171 or similar).
>
> For RFI shielding, the braided shield is connected to the connector
> backshell at both ends.  The drain wires from the twisted pairs
> shields can be connected at one end or both, depending upon what
> works best.  If they are not connected, heat shrinkable tubing
> should be used to prevent the individual shields from contacting
> each other.  We know this works well from an RFI point of view, but
> at least I have not tried it for analog data acquisition such as
> this.

Okay.

> Seems like a lot of this worry could be alleviated by using opto-isolators.

This is a suggestion that I made for the digital control signals in
the first meeting that we had at GB, but it was shot down. I would be
more than happy to consider it again, although for the analog detector
signals I would first need to convince myself that opto-isolators had
sufficient dynamic range, SNR, and linearity for our purposes.

> >   -Are there any concerns regarding my new scheme of using a low-pass
> >    filter and a 10MHz ADC, instead of an analog integrator and a 40KHz
> >    ADC?
>
> Sounds perfectly reasonable, not to mention cheaper and simpler.
> The concern mentioned about the time domain response of the filter
> may be addressed in a few ways.
>
>    - measure a similar filter.  We've got boxes full of old filters here and
> there may be quite a few laying around CIT.

Okay, although this could be difficult to do to the required accuracy.

>    - Simulate.  Simulation programs are getting very good - especially for
> passive linear devices like filters.

The problem here is that the manufacturers won't send me any details
about their filters, so I don't have anything to simulate, and the
time-domain response depends itimately on the details of the filter
(eg. a Bessel filter doesn't ring, but a Chebyschev filter with
similar characteristics does).

There is another possibility, one that eliminates the worry about the
time-domain response, provided that the filters settle within 25us. If
a step-function convolved with a smooth filtering function is a good
approximation to how phase-switch transitions appear at the output of
the anti-aliasing filter, then one can correct for the settling time,
simply by taking the difference between the sample preceding the
transition and the following sample, scaling this by a factor that
depends on the filtering function, then adding the result to the
following sample. Given that the two arms of the radiometer normally
will have very similar signals, this factor wouldn't need to be known
to great accuracy. In theory, we then wouldn't have to loose any time
to blanking around phase-switch transitions, which would be a great
advance compared to the analog integration approach.

> Also, with respect to the problem that the detector output is always
> positive, and thus requires extra range of the A/D, it is certainly
> possible to offet the output so that the mean is near-zero.

One could, but ideally one wants to locate the zero-signal level at
the zero level of the ADC, since this is where its noise and linearity
characteristics are optimized. As such, I am thinking of offsetting
the signal, but only to the extent that the detected signal level with
no sky signal, is at the optimal center of the ADC range. There will
be a small amount of noise around this level, but most of the signal
range will still only occupy the positive half of the ADC range. Note
that I would base the offset voltage on the ADC reference voltage, in
order to make the ADC insensitive to changes in the offset caused by
temperature variations etc..

> A question:
> Noise is a large concern in this design.  Is it not true though that
> the real, main concern is only about noise that will not integrate
> out?

Yes, but if one adds more noise than that implied by the
receiver-temperature, bandwidth and time, then this noise takes longer
to integrate down to the same level, so the final SNR will be
degraded.

> It seems as if a little random noise might even be a good thing.
> What if the detector output were nearly constant, to less than an
> LSB of the ADC?  Without random noise the averaged detected output
> would be less accurate than the ADC output * 2**n.  With the random
> noise, fluctuation in the signal smaller than the LSB would be
> detectable.

Indeed, and this has already been included in the determination of how
many bits of ADC resolution would be needed. As detailed in my
original CCB proposal last year, 3 bits have been assigned to sampling
the receiver noise.

Martin