This video presents an in-depth experimental exploration of temperature-compensated crystal oscillators (TCXOs), specifically targeting the characterization and minimization of phase noise in a 10 MHz system. The presenter investigates readily available TCXOs, noting their inherent design for long-term stability and compact size rather than ultra-low phase noise. The core experimental setup involves an array of 20 third-overtone crystals driven by custom-built amplifier stages and filter networks.
The video meticulously documents an iterative process of optimizing the oscillator's performance, including fine-tuning thermostat control, impedance matching, and power distribution. Various high-precision measurement tools, such as the Linrad spectrum analyzer and the PN 2060 phase noise analyzer, are utilized to quantify and diagnose noise characteristics. Key challenges encountered include transient "glitches" suspected to originate from micro-cracks within the quartz crystals, and unexpected noise contributions traced to systemic measurement issues (e.g., mixer problems, BNC connector integrity) or unintended feedback loops stemming from suboptimal amplifier layout. Notably, the presenter discovers that higher signal levels can induce phase modulation noise from the crystals or amplifiers, while reducing power significantly improves close-in phase noise performance. The video concludes with ongoing efforts to refine the amplifier design and layout to achieve better stability and a lower noise floor.
Characterizing and Optimizing 10 MHz Temperature Compensated Crystal Oscillators
0:00 The video introduces surplus 10 MHz Temperature Compensated Crystal Oscillators (TCXOs) which are inexpensive and produce a 2-volt square wave output.
0:28 Initial phase noise measurements indicate high sideband noise; these oscillators are primarily designed for long-term stability, not for extremely low phase noise close to the carrier.
3:10 The internal crystals are high-quality 10 MHz third-overtone crystals, with 20 of them connected in parallel for this experiment.
3:31 The optimal thermostat control voltage is determined using a network analyzer to achieve the flattest frequency response.
5:30 Adding approximately 230pF capacitance adjusts the frequency to 13 Hz above 10 MHz, which also surprisingly increases the crystal's Q factor to 1.2 million.
6:06 Power distribution to the crystals is optimized, aiming for 2mW into the crystals and 16mW into the series resistor, resulting in calculated resistor values for the crystal filter.
10:23 A high-impedance source follower (J310) and a low-noise amplifier are integrated, measuring a 2.82V signal across the crystals with a 680-ohm output impedance.
12:00 Measured thermal noise levels closely match theoretical predictions, validating the experimental setup for noise floor evaluation.
14:50 Persistent "glitches" or interference are observed in Allan variation measurements, suspected to be caused by micro-cracks in the numerous parallel quartz crystals.
15:49 Reducing the signal power by 10 dB significantly mitigates these glitches and improves the sideband noise performance.
19:47 Iterative improvements to the amplifier's components and configuration are made, showing incremental reductions in noise, though glitches may still reappear after prolonged operation.
31:11 Employing two source followers in parallel, along with provisions like ferrite beads to prevent VHF oscillations, leads to improved stability and noise performance meeting PN2060 specifications.
33:18 Systemic measurement issues are identified, including problems within the mixer and BNC connectors of the test setup, which were contributing to unexpected signal artifacts.
36:37 Adding an RC low-pass filter to the power supply eliminates spurious signals but does not significantly reduce the overall noise floor, suggesting underlying stability challenges persist.
42:06 A secondary crystal filter is introduced between the signal source and the unit under test, which adds noise at lower frequencies but not at higher ones, allowing for more isolated noise characterization of the device.
43:02 A second amplifier stage is added, featuring a step-up transformer and parallel JFETs, designed for higher power output and a stable output impedance.
44:58 Extensive noise floor characterization is performed using a 10.7 MHz crystal filter and a noise figure meter, revealing small discrepancies (e.g., a 4 dB error) between different measurement techniques, highlighting calibration challenges.
59:36 Further investigations uncover and mitigate high-frequency self-oscillations (at 80 MHz and several hundred MHz) in the amplifier stages by adding specific capacitive and resistive components, leading to a higher output power of 92 mW.
1:01:07 Unintended feedback due to poor circuit layout is identified as a significant issue, coupling the output of the second amplifier stage back to the input crystals, which negatively impacts Q factor and frequency stability.
1:08:10 A key finding: reducing the signal level by 15 dB significantly improves the close-in phase noise, indicating that at full power, the noise is dominated by phase modulation induced by the crystals or amplifier non-linearities.
1:09:47 Noise added at a 1 Hz offset is found to be independent of the power level, suggesting it might stem from a varying reactive component seen by the crystals or the crystals themselves.
1:11:50 A new, simplified amplifier design (step-up transformer + source follower) is tested; while it produces 20 mW, it is highly prone to self-oscillations, requiring extensive stabilization.
1:14:54 The video concludes, acknowledging its length, with plans for a future video detailing a revised layout and design to address the remaining high-frequency oscillation and noise issues.
1:14:59 Visual evidence of instabilities and oscillations is provided through slow-motion capture, demonstrating the complex behavior of the system.
1:15:28 The thermal noise of the crystal is observed directly, showing a higher level at resonance due to higher impedance, and highlighting the need for low-noise amplification to effectively analyze sideband noise.