[time-nuts] Ultra low phase noise floor measurement system for RF devices.

[Magnus Danielson]

From: SAIDJACK at aol.com
Subject: Re: [time-nuts] Ultra low phase noise floor measurement system for RF devices.
Date: Sat, 31 Mar 2007 23:33:26 EDT
Message-ID: <c73.101c34fc.33408206 at aol.com>

>  
> In a message dated 3/31/2007 17:29:21 Pacific Daylight Time,  
> bruce.griffiths at xtra.co.nz writes:
> 
> 
> Multiplying the 2 signals is essential as the non correlated  components 
> of the 2 signals will then average to zero (for infinite  measurement 
> time) whilst the correlated components of the 2 signals will  not.
> A cross spectral analysis is somewhat analogous to using an array of  
> single frequency correlators. The residual noise decreases as the  
> integration time increases. When multiple power spectra are averaged the  
> noise floor will decrease as 1/n where n is the number spectra  averaged.
> 
> Bruce
> 
> 
> Hi Bruce,
>  
> got it, thanks.
>  
> But isn't a signal source typically single-ended, and has a 50 Ohm source  
> impedance which creates noise?
>  
> Wouldn't we need a source with 1 Ohm source impedance even if the  
> measurement system is cross-correlated?
>  
> Would a differential source suffice?

In a cross-correlation system you take the (single-ended) signal of the DUT
and mix that down and measure it in two independent arms using two independent
mix down oscillators. The mix down oscillators noise as well as the noise of
the individual components in the arm does not correlate between them, but the
noise on the input port is the same and when cross-correlated between the arms,
the averaging smooths out the internal noise as only that which correlates
for both arms will remain, and that is (mainly) the input signal and its noise.

With the incremented averaging, the internal noise of the measurement arms will
reduce with 1/sqrt(N) and only the measured noise remains. Thus, this is the
way in which the noise levels of the measurement rig and its oscillator can be
overcome using cross-correlation.

The trick is to let the noise you are interested in correlate and your
measurement rig not to correlate.

In the article that was recently referred to, it was not a measurement rig for
oscillators but for transfer components, such as gain-stages, phase shifters
etc. and the approach is different then, since you do have the signal prior to
being dirtyfied. Thus, the interferometer approach makes sense. The refinement
was certainly interesting.

Cheers,
Magnus