[time-nuts] xtal oscillator phase noise
bruce.griffiths at xtra.co.nz
Sat Jan 5 18:06:33 EST 2008
Henk ten Pierick wrote:
>> Meanwhile look at:
> I did and it was very helpful
>> This crystal oscillator circuit is a variant of the low noise
>> proposed by Wenzel for use with fundamental crystals. It uses a common
>> base buffer and demonstrates several low noise biasing techniques that
>> can be used.
> The schematic is easy to understand, but it is not clear to me what
> defines the xtal current. The loading of the oscillator due to the
> base resistor of 10k is surprising. I expected a controlled loop for
> the xtal current in a well designed oscillator.
> How high or low is the loaded Q. How is noise matching for the
> oscillator defined?
Crystal current is defined by the dc current in Q104.
To adjust the crystal current adjust R116.
To see the effect you need to either build or simulate the circuit
(difficult without the right software).
The 10K is connected in parallel (as far as the RF is concerned) with
C107 and C108. (C107 has a reactance of around )
Its actually difficult to make a choke that has an equivalent shunt R as
large as 10K.
If you are worried about it R114 it can be increased to 100K.
The oscillator transistor turns on for a small part of the RF cycle.
Optimising the duty cycle by adjusting the ratio of C107 to C108 can
improve the phase noise.
Noise matching has little relevance for flicker phase noise.
Loaded Q is quite high.
>> The 2nd and third references above the oscillator schematic explain
>> mechanisms for generating AM and PM noise in a BJT RF amplifier. The
>> derivation is quite mathematical (statistics and calculus) but the
>> conclusions are relatively simple.
> Nice reading with clear conclusions. A lot of the cited measures and
> comments from you and other time nuts are now better understood.
>> To drive an HC04 the common base buffer can use a load consisting
>> of an
>> inductor shunted by a resistor to develop the drive. The shunt
>> reduces the dc gain (from base to collector) of the buffer stage and
>> hence the low frequency noise voltage developed across the collector
>> base capacitance. Such noise voltages modulate the output capacitance
>> and hence the phase shift of the buffer, increasing the buffer phase
>> noise. Nonlinearities such as hfe variation with current tend to
>> increase the buffer output AM noise not the PM noise.
> It is funny that I always tried to avoid coils because of sensitivity
> for magnetic fields and now I learn that I have to use coils for the
> lowest phase noise. Why is there not an inductor used i.s.o a
> resistor of 10k in the base of Q104? Avoids the loading with the
> consequence that xtail current control has to be done in an other way.
Actually real inductors have finite equivalent shunt R, in practice its
difficult to get much above 10K without special techniques and
relatively large physical size.
Increasing R114 to 100K is probably a good idea in that it reduces the
series equivalent from 2.5ohms (a bit high)) to about 0.25 ohms.
This will not increase the flicker phase noise.
Increasing C107 and C108 will also reduce the effective ESR due to R114.
> I assume that the loaded Q of the oscillator is important for a low
> phase noise. The circuit suggests different. Am I missing something?
Circuit Q is higher than you think.
If you only want a CMOS output a lower phase noise oscillator is
possible using 1 diode and 1 transistor less than in your circuit.
I'll add it to the collection of additional crystal oscillator circuits at:
Yes, biasing of the CMOS gate appears weird but it works well.
More information about the time-nuts