[time-nuts] xtal oscillator phase noise

Bruce Griffiths bruce.griffiths at xtra.co.nz
Sun Dec 30 17:18:31 EST 2007


Grant Hodgson wrote:
> Henk
>
> Two things will dominate if you want such a low phase noise spec.:- the 
>   loaded Q of the oscillator circuit, and the flicker corner frequency 
> of the sustaining amplifier transistor.  To get a high loaded Q you need 
> a crystal with a high unloaded Q - maybe 100 000 or more - this isn't 
> difficult to achieve from a good crystal manufacturer, but you can't 
> expect any old crystal to work.  And the rest of the oscillator circuit 
> should not load the crystal too much, otherwise the loaded Q, and thus 
> phase noise, will suffer.  Good crystal manufacturers will provide the 
> necessary measurements of series resistance, motional inductance (or 
> capacitance, or unloaded Q - doesn't matter which) and static 
> capacitance.  Lesser crystal manufacturers - don't.
>
>   
A Q of 100,000 at 10MHz produces a Leeson effect phase noise corner of
100Hz
(below which the phase noise slope is 20dB/decade).
This makes it much harder to achieve the desired phase noise at 10Hz
than a crystal with a Q of 1,000,000.
> Also the flicker corner frequency of the transistor needs to be as low 
> as possible.  Generally speaking, at offsets below the flicker corner 
> frequency you will get 30dB/decade, above the flicker corner frequency 
> you should get 20dB/decade, or flat, depending on the level of the phase 
> noise floor.  If you can find a transistor with a lower corner 
> frequency, the flicker noise will be reduced.  In fact, this is one of 
> the dominant parameters when choosing a transistor as an oscillator - 
> any old transistor can be made to oscillate, but to do so with a low 
> flicker corner frequency is not so easy, and the corner frequency is 
> usually a function of bias current.
>
>   
Its not quite that simple, the flicker phase modulation depends on the
voltage and/or
current dependence of the various transistor capacitances etc and the
consequent phase
modulation produced by low frequency noise currents flowing through the
transistor
or low frequency noise voltages developed across such capacitances.
Simply choosing
a low flicker noise transistor without taking its collector base
capacitance, emitter base
capacitance, etc into account is unlikely to reduce the flicker phase noise.
Reducing the dc gain from the base to the collector and ensuring that
the emitter current
low frequency noise is low is likely to be more effective.
Increasing the collector base voltage will reduce the collector base
capacitance and its voltage dependence.
The oscillator power supply noise can also modulate the transistor
collector current and thus increase the phase noise.
> At 11MHz, most crystal oscillators use parallel resonant crystals, 
> although some are series resonant, such as the excellent Driscoll 
> oscillator which is capable of the performance you desire with a 
> suitable crystal.
>
> Then you have the added problem of the FSUP.  It's a superb instrument, 
> but it has it's limitations.  The FSUP data sheet states a phase noise 
> spec. of -130dBc at 10Hz offset for a 10MHz signal, which gives a 
> resulting sensitivity of -127dBc - 3dB worse than what you are trying to 
> achieve.  You would need option B60 (cross correlation) to significantly 
> reduce the effect of the internal source by (say) 15db or so.
>
> regards
>
> Grant
>
>   
The measured oscillator phase noise floor seems a little high for a
modern low phase noise design.
If the oscillator uses varicap diodes to adjust the oscillation
frequency, these can contribute
significantly to the phase noise especially if their tuning range is large.
Testing the oscillator without varicaps (if possible) and with lower
noise supplies may also be useful.
Failing removal of such varicaps reconfiguring the EFC so that the
varicap has a much smaller tuning range
(supplemented by a manual trimmer) may be useful in reducing the phase
noise contribution of the EFC circuit.
Use a capacitive attenuator to reduce the EFC range, not a resistive
attenuator on the EFC control voltage.

Bruce



More information about the time-nuts mailing list