"Richard Hosking" a écrit dans le message news:
...
Dear all
I want to design a DDS board which has attenuators on the output to
provide a low level output (-100dBm)
I have a venerable HP8640B which can give a calibrated output to -137dBm
To achieve this HP have gone to extraordinary lengths to shield the
oscillator attenuator/output amp circuits and any control lines - I note
there are at least two stages of bypassing/low pass filtering with an
intermediate shielded section

My question is: how do I get data and power lines into my DDS chip (in a
shielded enclosure) and prevent RF leakage out which will limit the
useful minimum level out from the DDS board?
I presume I will have to use a buffer of some sort for the data lines
and extensive bypassing on the power lines. I want reasoanbly quick
update speeds for my DDS, which is a serial port, which will mean data
rates in the MHz region.


Richard,

I don't have direct experience about this but I'm developping a VNA with
pretty tight specs so I had to think about things like this.

Here are some of my reflexions, and also some ideas provided by other people
:

Low output level is not the same thing as low leakages : it depends on where
leakages couples.
What you don't want is leakages coupled to the output port, but you may have
some leakages that will be stopped by your main enclosure (OK, not mW
though) and still maintain low level at the output port. Whether this can be
tolerated or not depends on the rest of your equipment.

Besides the DC/command lines, to achieve accurate low level output will
require good internal shielding between the DDS section and the attenuator :
1dB accuracy at -100dBm means leakages coupled to the output has to be
under -120dBm and 0.1dB lower this to -140dBm.
At this level, ground plane management can be "difficult". Depending on your
frequency range it might also help to provide separation between the DDS and
the attenuator with a small xformer.
You'll have to think to current paths for both cases, high frequency and low
frequency, if your frequency span doesn't allow an xformer, because currents
spreads differently, depending on freq.
I've used fasthenry (a freeware prog) with great success to model this kind
of things in another application.

If you really need very, very good shielding (radiated power inside the
enclosure) it might be easier/better to "double shield" the beast and to add
a lowpass section between the 2 shields.
Again, think to current paths and that common mode is your enemy ("ground"
potential that's shifted by return currents).

For the DDS command lines, optos might help, but maybe not as much as
expected if you want a high speed link.
If you really want high speed and isolation I'd go for an optic fiber. Not
very expensive and coupled with a double shielded box, you'll have
vanishingly low leakages.

Just think to neglect nothing and you're OK.


Thanks,
Fred.