[time-nuts] DIY VNA design

[Richard (Rick) Karlquist]

On 8/21/2016 3:59 PM, Dr. David Kirkby (Kirkby Microwave Ltd) wrote:
>
> That said, I don't know why the author is using directional couplers.  A
> bridge is much wider bandwidth.  It is more lossy though.

In general, a resistive bridge will always require a
transformer/180 degree hybrid/differential amplifier
to make it work.  If you are going to go to the trouble
of making a broadband transformer or hybrid, you might
as well just build a traditional directional coupler,
because it is no more difficult.  All the resistive
bridges I have seen are followed by broadband differential
amplifiers.  The resistive bridge itself has a minimum of
something like 15 to 20 dB loss, and the differential
amplifier has a minimum NF of 7 dB or so.  This results
in a great loss of sensitivity, but you can always get
the sensitivity back by using a narrow IF bandwidth and/or
lots of averaging, or (rarely) a high drive level from
the source.

Having said that, one of the putative advantages of a resistive
bridge is accuracy.  However, with today's calibration techniques,
this is no longer all that important, so a traditional coupler
might be more practical than it used to be.  I remember attending
the retirement party of Agilent's last great designer of couplers
(pre-calibration) and let me tell you, this guy was a total guru.
He was one of greatest practitioners in this area of all time.
He freely admitted that he was now obsolete due to calibration.
Any old coupler is good enough.
>

> Anyway,  it is an interesting project,  but personally if I were going to
> go to the effort of building a 2-port VNA, I would build one with 4
> receivers.
>
> Dave
> _______________________________________________

We used to have a lot of arguments at Agilent about how many
receivers were needed.  The most I ever heard advocated was 5,
and the least was 1 or 2.  I had to intervene in some of these
arguments to bring up what I call the "back door reference"
fallacy.  If you were making a "scalar" network analyzer that
only dealt with amplitude, you could make various arguments
about why you don't need so many receivers.  In principle,
1 receiver could work.  (The achilles heel of this idea
turns out to be imperfect repeatability of switches, and
very long settling times and thermal tails in switches.
None of these calibrate out).

In any event, as soon as you start talking about vector
network analyzers, you are measuring phase.  Unlike amplitude,
phase is always a relative measurement.  That is why you
need a reference ("R" channel).  You compute A/R.  This
requires a minimum of 2 receivers, an "A" and an "R".
Concurrently, not consecutively.  Architectures that skimp
on receiver count, or ostensibly omit the reference channel,
are really a cheat.  There will be some back channel between
the instrument clock and the sampling clock in the ADC that
in essence acts as a reference channel.  If there is any
warm up drift in the phase of this channel, you will get
non-correctable errors if you try to multiplex a single
receiver.  It is also another source of crosstalk on the
PC board.

Another problem with skimping on receivers is that you
can't do full 2 port calibration, I used to
have people show me "proof of concept" why they don't need
full 2 port calibration.  They would compare a test of
some simplified architecture to some top of the line VNA
and show that the measurements were the "same".  Just like
the graphs you see comparing low cost VNA's to Agilent
VNA's (it always seems to be Agilent, not one of the other
name brands).  It would often turn out that these "benchmarks"
were not good tests of the analyzer.  Changing to more
challenging tests would reveal the true superior design.

For example, if you calibrate with a short, open, and load,
and then measure the short, it always looks perfect.  But
if you add a short length of transmission line in front of
it, the simplified architecture may not work so well any
more.  This is called a "remote short" test.

Rick