[time-nuts] Designing and building an OCXO and GPSDO

John Miles jmiles at pop.net
Thu Aug 14 22:17:27 EDT 2008

> Perhaps the local oscillator isn't the limiting factor for the low
> frequency analysers.
> The claimed noise floor is in the vicinity (within 10dB) of -120dBc/Hz
> for the analysers for which I checked the specs.
> If the 10MHz reference has a phase noise floor of around -160dBc/Hz
> this is only degraded by 50dB or so to -110dBc/Hz when multiplied by 300.
> The YIG oscillator phase noise floor may perhaps be a little better than
> this.
> However, since state of the art ADCs have a phase noise floor of around
> -150dBc/Hz one would expect a lower phase noise floor from the lower
> frequency spectrum analysers.
> Surely there's an FFT based spectrum analyser out there wit this level
> of performance.

In the 80s-era analyzers like the 8566, the sampler used in the YTO loop is
what hoses you in the end.  You can make substantial improvements in the
noise performance of both the reference and the intermediate ("M/N")
synthesizer, but you hit the sampler's noise floor at around -105 to -110

It would be interesting to learn if Agilent's high-end PSA series FFT
analyzers still use a sampler-based YTO loop, or if they are now using
traditional dividers and phase detectors.  Something tells me the service
manuals no longer contain schematics and dissertations. :)

It is true that 10 MHz @ -160 dBc/Hz * 300 = 3000 MHz @ -110 dBc/Hz, but the
devil's in the details.  The broadband floor of a 100 MHz reference is no
worse than that of a 10 MHz reference, so an obvious optimization is to lock
a 100 MHz oscillator to the 10 MHz reference and use *that* as the
reference.  This gives you more like -130 dBc/Hz in theory.  But then you
have to tune the reference, so there has to be a synthesizer to drive the
final YTO loop.  That, assuming you're not limited by sampler/PD noise, is
where the majority of the noise comes from.

Most synthesizers regardless of technology still end up with broadband
floors in the -150 dBc/Hz neighborhood, so now you're back to circa -120
dBc/Hz inband... which is what the PSA-series spec sheets show.

It's an interesting business.  I spent a lot of time redesigning my 8566B's
LO synthesizer, until I realized two things: 1) the YTO sampler noise floor
was the real problem, and I didn't feel up to redesigning *that* loop; and
2) lowering the PN floor would probably reveal spurs that the original
designers never knew/cared about, which would lead to even more

It would certainly be possible to design a cleaner analyzer LO than the PSA
has, but it would cost more, the effort would go unappreciated by most
users, and worse, Agilent would sell fewer E5500 outfits!

-- john, KE5FX

More information about the time-nuts mailing list