In the ALMA LO, for reasons of simplicity and reliability it
has been specified that mechanical tuners will not be employed.  This
is a departure from designs traditionally used in total power radio
telescopes.  

As a paradigm, consider the NRAO12m.  Since the demise of
Klystrons, the ultimate LO source was a Gunn diode, probably made by
John Carlstrom.  This device could be tuned, rejecting noise in the LO
signal to a very high degree.  At higher frequencies, the Gunn was
located in series with multipliers, also tunable, sometimes vexingly so.
John Payne estimates:
" It is instructive to calculate both the intensity 
and stability of any spurious signals that may accompany the LO.  A few 
assumptions are needed. Make the following: 
  the minimum bandwidth is 100 KHz: 
  the position switching time is 30 secs: 
  the LO power required is 0.5 micro Watts: 
  the bandwidth of the LO signal is 1 Hz and 
  any spurious signals in the bandpass are stable to better than one part in 
       one thousand over one minute( the usual position switching cycle).
  This latter point establishes the residual amplitude of any spurious 
signals in the  I.F. passband to be around one Kelvin. Of course a 
crucial point here is the bandwidth of the spurious signal. If the same 
as the LO its intensity will be lessened by a factor of 10E5 by the 
resolution of the spectrometer.  The strength of the LO signal in Kelvin is 
around 10E16 so any spurious signal for this very simple scenario should be 
around 160dB down on the LO signal." 

Now consider ALMA.  There is no mechanical tuning of the YIG oscillator,
so noise in the LO signal is not rejected but passes to the amplifiers and
multiplier(s).  Thacker notes that the YIG itself is very clean but
operates at intrinsically lower frequencies than does a Gunn.  Owing to this,
more amplification and multiplication must be included to reach the very
high ALMA frequencies.  Within the YIG, all those 
heavy nuclei precessing in lockstep make a very good resonator which has
the advantage of being electrically tunable but still a tuned resonator which
rejects noise just as the mechanical resonator in the Gunn.
The multiplier is not tunable either, and passes noise not only in the desired
harmonic but perhaps in others as well,  From spectrum analyzer observations,
it is known that this noise is present and it can be roughly characterized
in bandwidth and strength but not as well in temporal stability.  If this noise
is injected early in the signal chain, it is subject to further instability
in any amplification as even a perfectly stable signal will have superposed
upon it gain fluctuations from the amplifying device.  The same gain
fluctuations are affecting the system noise and the astronomical
signal, so the spur does not necessarily make things any worse, as Larry
notes.  

How are we to know whether the ALMA total power system will be capable of
the science tasks assigned to it?  It is planned to test this system as
completely as possible using an ALMA Front End in conjunction with the Green
Bank Spectrometer.  This test will take place in late May or early June and
involve many of the elements of a complete ALMA system.  Skip Thacker
will carry out this test, accompanied by someone from the science group (AW).
This test will not include a laser synthesizer; the 100 GHz signal will
come from a harmonic multiplier as traditionally provided as a clean test
is needed.  Skip plans to set the system up and integrate for several hours
to reach levels of several mK with spectral resolution provided by the
spectrometer in its widest spectral mode, a few hundred kHz.  The system 
will be switched, so a non-stable signal will show up as a 
spectral feature within the bandpass.  The result of the test will be a
demonstration that the ALMA LO can provide stable clean signals down to
a level which will be determined.  

If the system provides a clean bandpass down to the several mK level,
it is still possible that the output of the multiplier produces power
at multiples of the desired frequency.  This should be characterized also,
by a future test.  A further worry is that the LO from the WVR, at 183 GHz,
will also provide an unwanted harmonic at 366... GHz which may affect some
receivers.  This will also be characterized when the system becomes
available.