This video documents the third and final phase of an experiment to create an ultra-low phase noise 10 MHz oscillator by using twenty crystals in parallel. The video begins by evaluating the 20-crystal array as a selective amplifier, investigating and refining measurement techniques to accurately determine its noise floor, which is found to be an exceptionally low -188 to -190 dBc/Hz.
The amplifier is then converted into an oscillator by adding a feedback loop. Initial results show excellent performance at 10 Hz offset but poor performance close to the carrier. The cause is identified as flicker noise in the source follower, which is significantly mitigated by implementing a bootstrap capacitor. This improvement pushes the oscillator's performance beyond the capabilities of the existing measurement setup, prompting an extensive debugging and troubleshooting process of the reference oscillators and phase-locked loops. After addressing these system limitations, the final multi-crystal oscillator demonstrates record-breaking performance, achieving -126 dBc/Hz at 1 Hz offset and -156 dBc/Hz at 10 Hz offset, surpassing benchmarks found in scientific literature for crystal-based oscillators.
Building and Measuring a Record-Breaking Low-Noise Multi-Crystal Oscillator
0:01 Introduction and Amplifier Testing: The video continues from "nerds108," starting with the 20-crystal parallel array configured as a selective amplifier. A new low-impedance output amplifier is added to the circuit.
4:59 Measurement Challenges and Techniques: The initial noise floor measurement is worse than expected. The creator discovers his previous, more optimistic measurement was flawed due to using a 90-degree phase shift cable, which introduces a time delay that corrupts high-offset frequency measurements. This leads to a deep dive into alternative, more accurate methods for measuring the noise floor.
15:28 The Source of Low Noise: By analyzing the circuit with the input signal off, it's shown that the high-Q crystal filter itself significantly reduces thermal noise away from its center frequency (by over 17 dB), which is a key reason for the extremely low wideband noise floor.
23:23 Conversion to an Oscillator: A feedback loop is added to turn the amplifier into an oscillator. The first "ugly" attempt shows promise but has poor close-in performance.
24:45 Identifying a Critical Flaw: After building a proper version in a shielded box, the oscillator exhibits excellent noise at 10 Hz (-155 dBc/Hz) but very poor noise closer to the carrier. The creator hypothesizes this is due to flicker noise in the source follower transistor modulating the phase.
37:18 The Breakthrough "Bootstrap" Fix: To solve the flicker noise issue, a bootstrap capacitor is added to the source follower. This dramatically improves close-in performance, achieving a potentially record-breaking -125 dBc/Hz at 1 Hz offset.
40:06 Hitting the Limits of the Test Gear: The improved oscillator is now so quiet that the noise from the measurement system's reference oscillators is dominant, making further analysis impossible. This begins a long and detailed troubleshooting process to improve the stability of the test setup itself.
53:04 Advanced Troubleshooting with tinyPFA: To isolate problems, the creator uses a tinyPFA (Phase/Frequency Analyzer) to measure the phase stability of the amplifier chain alone, discovering an instability where the phase jumps between two states, particularly at higher power levels.
1:13:04 Final Assembly and Measurement: After fixing the amplifier instabilities and partially improving the reference system, the multi-crystal oscillator is re-tested. A 16-hour measurement run is performed to get reliable data.
1:17:20 Record-Breaking Performance Achieved: The final measurements confirm the oscillator's exceptional performance: a flat noise floor of -188 dBc/Hz, -125.9 dBc/Hz @ 1 Hz offset, and -156.1 dBc/Hz @ 10 Hz offset.
1:26:00 Comparison to Scientific Literature: The creator compares his results to published papers on ultra-low noise oscillators. His design is shown to be significantly better than other state-of-the-art crystal oscillators, outperforming them by 3 dB at 1 Hz and 8 dB at 10 Hz. He concludes that while optical methods can achieve even lower noise, his design represents a new benchmark for what is possible with a crystal-based architecture.