A low noise 14 MHz crystal oscillator with phase noise -173 dBc/Hz @10kHz
(Nov 8 2009)
 The main article is here This oscillator is an old design that I made for evaluating the WSE RXHFA unit. The text on the label is old and not quite accurate. The phase noise is -173.4 dBc/Hz and the AM noise is also -173.4 dBc/Hz for a total noise of -170.4 dBc/Hz at 10 kHz.
 This is the view from below with the lid off. Birds nest style so the oscillator is a little more microphonic than a "proper" construction.
 The block diagram is straightforward. An amplifier with feedback through a crystal filter. The transformer on the input side provides a match from 50 ohms to 4.7 ohms. The output side has a trimmer to balance the parasitic capacitance of the crystal. The input impedance of the amplifier is very low so the series resonance of the crystal is well used.
 The details of the amplifier input. The way to get very low noise is to allow the emitter to see the high impedance of the crystal at off resonance frequencies. At a frequency separation of 10 kHz the output impedance of the filter is typically 800 ohms while the crystal itself is typically 8 ohms on the resonance to give an feed impedance of about 13 ohms to the amplifier input. The bipolar transistor is four parallel MPSH10. That gives a lower input impedance than one would get with a single device for the same current. The voltage across the emitter resistor is 5 V so the current is 10mA. The gain 10 kHz away from resonance is nearly 20 dB lower than the gain on resonance so the noise floor is much lower than it would be in a conventional design where a resistor would terminate the filter. This is explained in DUBUS 2 2001 by Gerhard Hoffmann, DK4XP. Page 18 gives this formula: PNnoise-floor = -174 + NF - Pout (dB/Hz) The room temperature noise is -174 dBm/Hz. A typical NF could be 12 dB and the power delivered through the crystal could be -2 dBm. According to the formula the sideband noise far from the oscillating frequency would be -160 dBc/Hz. The phase noise would then be -163 dBc/Hz, about 10 dB worse than the noise at 10 kHz of the design presented on this. With 10 mA DC through the bipolar transistor in the peak to peak AC current could be max 20 mA for an RMS current of 7 mA. With an impedance of 8 ohms that means that the power dissipated in the crystal is 0.4 mW or less. Not having a resistive termination thus improves by about 10 dB. Another 6 dB improvement is possible mainly by not driving the transistor into saturation as is done in the design presented on this page. The +12V supply is carefully decoupled like this:
 I have investigated oscillators that use the series resonance because they can be used easily with overtone crystals in the VHF region. For fundamental frequency oscillators in the HF region there is another alternative and that is to use the parallel resonance with e.g. a Colpitts oscillator. The oscillating transistor will then see a high impedance on the resonance but low impedances at the sides. The amplifier would see the impedance at the base/gate and produce less noise at the sides where the impedance is lower. With some care Colpitts oscillators approach -170 dBc/Hz in the few experiments I have done. Sideband noise in Colpitts oscillators. To SM 5 BSZ Main Page