bjt oscillator
which part number would you recommend for the lowest phase noise crystal oscillator design? The transistor should have very low 1/f noise (datasheets don't provide this data). Some designs prefer NEC CEL's NE688xx family for the low 1/f.
The oscillator will use either 3rd or 5th overtone at approx. 100-200MHz. Would 5th overtone crystal produce better or worse phase noise performance - why?
The Infineon's app note suggests that the best oscillator transistor is a device with the lowest possible fT. A commonly used criteria is: fT ≤ 2 x fosz. What is the reason that the lower fT would give better phase noise?
Rohde suggested in one of his books an old transistor BFT66 to be used in his circuit achieving -168dBc/Hz at some kHz offset.
Thanks for any constructive ideas and suggestions,
rfmw
I know that's the transistors Rohde always uses for all of his oscillator designs. I have been looking into this problem recently, and tried to find the right parameters of the spice model describing flicker noise. Problem is that most models didn't supply any parameters for flicker noise.
The rule of thumb is:
* Si transistors are best
* SiGe is intermediary but you can go up to higher frequencies with this
* FET/HEMT is worst due to many surface defects causing large flicker noise
If the resonator would be lossless, the 5th overtone would have a better Q-factor than the 3rd overtone (phase transition much faster). In real life, loss mechanisms become worse at higher freqs, so typically the Q-factor at 5th overtone would be lower.
Thing is that the higher the fT, the more harmonics you generate. All harmonics can mix flicker noise up and mix back to the ground tone. This causes higher phase noise. Therefore, try to calm down your device as much as possible.
AT41533
Mandi
Actually J-FET?s have the lowest flicker noise 1/f, followed by BJT?s. CMOS is the worst.
Low ft transistors have higher parasitic capacitance to the ground, resulting in a decrease of the flicker noise. RFIC designers chose bigger transistors (biased at low current) for this reason.
HP (Agilent) recommend this method way before Infineon.
In regard to the BJT vs. JFET I found this:
"
The JFET is appealing because of the inherently low white noise. The noise temperature can be as low as a fraction of a degree Kelvin. Unfortunately, the low noise of the JFET derives from low input current, hence a high input resistance (some MOhm) is necessary. The JFET noise voltage is hardly lower than 5 nV/√Hz, some five to six times higher than the thermal noise of a 50 Ohm resistor (√4kTR = 0.89 nV/√Hz). The JFET is therefore discarded in favor of the BJT.
"
BJTs achieve lower noise density than 5 nV/√Hz.
EDIT: above discussion in the quote relates to the JFET as an amplifier of DC-low frequencies.
Above paragraph specify the white noise and not the flicker noise, which is a different story.
rfmw
could you please give me a reference or upload the Infineon application note you mention?
Mazz
http://www.infineon.com/dgdl/AN023.p...14271546af06cf
Merci
Using the BFR93 and obviously quite good crystal I get currently -160dBc/Hz @ 10kHz offset at 109MHz (probably 3rd overtone). Nice for the first try, but I'll test the super low 1/f transistors from CEL also.
rfmw that's interesting. With wich topology do you reach such a low wideband noise floor? In one design I came across one day using BFR93 it was only -150dBc/Hz, and that is typical.
Another topic I'd like to hear from you guys is FM noise from the crystal. I was trying to find out why we could not reach our spec @ 10/100 Hz offset, and I read that the phase noise is predominantly caused by acoustic FM noise in the crystal itself! Can anyone shed a light on this? If you look on the website of famous crystal oscillator suppliers like Vectron, they are able to reach much better specs @ 10/100 Hz than what I measured from a self-made design. Is this because they sort the crystals and keep the best ones for their own oscillators?
Measuring phase noise that low and close to the carrier is a non-trivial task. It is quite possible that the oscillator is better than you measured.
One trick used in the past is to use a second crystal in a BPF after the oscillator to slice off some of the phase noise.
@radiohead
see the attached circuit. The PN was measured using the e5052b with 100x correlation.
@flatulent: I heard that before, but I wonder how you deal with tolerances then? Or do you make the filter auto-tunable for maximum output power?
@rfmw: Thanks, I'll try that when I have the opportunity. The circuit I saw was a Butler common base 5th overtone oscillator.
Who has a good book/reference comparing different topologies and actually explaining WHY the phase noise is better instead of stating that is is that way?
The filter is factory tuned.
I've collected quite a few books on crystal oscillators, but more or less only Rohde gives practical designs with actual measurements.
For low noise crystal oscillators , depending on the frequency, a FET or a BIP transistor with the right size, fT and proper bias condition should be selected. More information about flicker noise feeback will be found in
Ulrich L. Rohde, Ajay K. Poddar, Georg B?ck "The Design of Modern Microwave Oscillators for Wireless Applications ", John Wiley & Sons, New York, NY,
Ulrich
I am very glad to see you Ulrich on our forum. Definitely this is very beneficial for all of us.
I see now that you changed your callsign from KA2WEU to N1UL. This one sound well in CW contests :D
Dear Ulrich,
welcome to the forum and thanks for the help. I think its needless to say that many of us appreciate your excellent work and that you'd visit this forum more often in the future.
br,
rfmw