[time-nuts] xtal oscillator phase noise

[Bruce Griffiths]

Grant Hodgson wrote:
> Henk ten Pierick wrote:
> --snip--
>   
>> It showed to be very difficult to come lower than what I have now. If 
>> can be the crystal. How can I decide?
>>     
>
> As an absolute minimum, you need to know the crystal dynamic (or 
> motional) parameters - the crystal supplier should be able to provide 
> these.  If not, you can measure them on a network analyser whilst you 
> look for another crystal supplier.
>
>   

> Then you should be able to simulate the phase noise of the oscillator 
> using a harmonic balance or similar method as used in Microwave Office, 
> Genesys, ADS, Ansoft Desginer, QUCS etc.  Failing that, a small-signal 
> (linear) open-loop analysis would at least give an estimate of the 
> loaded Q, which can be used to predict phase noise - but ignoring 
> flicker noise.  Some SPICE-based simulators might be able to help.
>
>   
The circuit is so simple that one can easily estimate the crystal loaded
Q by hand (if one knows the crystal parameters).
Similarly the phase noise floor can easily be estimated without
requiring a simulator.
Unless the simulators include physically correct models for the flicker
phase noise generation mechanisms they will be of little help.
If they persist in using the Leeson model (which has been shown to be a
gross approximation particularly for flicker noise by Hajimiri and Lee
as well as Demir), then the results are questionable.
With the correct flicker phase noise generation mechanism model its not
too difficult to estimate the flicker phase noise if one has sufficient
data on the oscillator transistor characteristics
Since real crystals exhibit flicker noise such estimates will need to be
supplemented by actual measurements.

If the crystal is a fundamental crystal similar to those offered by
cemac in an HC49 holder and is not a strip crystal then the ESR will be
less than 35 ohms (typically 20ohms??) so with a 5mA oscillator
transistor emitter current the loaded Q will be ~25% less than the
unloaded Q with if the crystal ESR is 20 ohms.
> --snip--
>   
>> I used the BC375 for the low Rbb' and assume that the noise corner must 
>> be low as a result of that. Is this not true?
>>     
>
> There are many different types of noise - the base bulk resistance of a 
> transistor contributes to shot noise, which is close to being 'white' - 
>   
rbb adds Johnson noise not shot noise.
> i.e. equal magnitude /Hz at all frequencies.  This does not have a 
> significant effect on phase noise at offsets close to the carrier, and 
> at 30Hz offset the flicker noise dominates.
Not necessarily true, RF phase noise may have either a higher or lower
flicker noise corner frequency than the transistors low frequency
flicker noise corner.
>   Flicker noise is not 
> 'white' noise - flicker noise increases at a rate of 1/f, or 10dB/decade 
> as the offset frequency is reduced, and simply choosing a transistor 
> with low Rbb' is not sufficient - the noise mechanisms are different.
>   
Increasing the junction area of the transistor reduces the low frequency
flicker noise however this increases the junction capacitances which
increases the Rf flicker phase noise.
It has recently been shown that flicker noise has a lower limit which is
quantum mechanical in nature.
> For an 11MHz oscillator I would use 2N5179s as advocated by Rick for 
> both the sustaining amplifier and the limiter - this is a very popular 
> transistor for oscillators in this frequency range.  I'd be surprised if 
> the BC375 generated less noise than the 2N5179.  This would mean 
> changing the circuit topology to use an NPN transistor instead of the 
> BF450 which is PNP.
>
>   
Using transistors with low capacitance can be more important than using
a lower noise transistor see:

http://tf.nist.gov/timefreq/general/pdf/1134.pdf

http://tf.nist.gov/timefreq/general/pdf/1139.pdf


>   
>>> 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.
>>>       
>> I was aware that most lower frequency circuits are parallel resonant. I 
>> used series in class A because I thought is was better, it is easier to 
>> use the current though the xtal. Is there a fundamental difference 
>> between parallel ore series w.r.t performance?
>>     
>
> Not really, it's the circuit topology determines whether a parallel or 
> series resonant crystal is used.  Your circuit appears to be a variant 
> of the Driscoll oscillator, which usually uses a series resonant crystal 
> and is capable of exceptionally high performance, however there are a 
> number of differences in your circuit, which I've never seen before, 
> although I can't claim to be an expert oscillator designer.  Circuit 
> simulation is a good (no - essential) starting point, and would give you 
> a good idea of the relative merits of the features of your circuit.
>
>   
The circuit can easily accommodate a crystal specified for parallel
resonance by using an appropriate series capacitor.
If one wishes to trim the oscillation frequency of the circuit such a
series capacitor is necessary.
> --snip--
>   
>> I do normally not have access to a FSUP but borrowed the instrument for 
>> two weeks. To my luck it has the B60 option and I used this of coarse. 
>> There must be a reason for my employer to buy this fantastic tool.
>>
>> Henk
>>
>>     
>
> OK, but given that the noise level is currently way above the noise 
> floor of the FSUP, using cross-correlation doesn't add anything - it 
> just slows down the measurement.  Cross-correlation would only be of 
> benefit to reduce the noise floor of the instrument if/when the phase 
> noise of the oscillator has been reduced enough to justify it - it can 
> seriously slow down the measurement.
>
>   
Speed of measurement is usually only an issue in production testing.
> regards
>
> Grant
>   
Bruce