Okay, here is the abstract and summary for the provided transcript.
Abstract:
This video documents an experimental process of refining an oscillator system, focusing on minimizing phase noise and addressing stability issues. Initial attempts using a BF240-based buffer amplifier proved detrimental to noise performance. Subsequent modifications focused on optimizing the limiter stage, including impedance matching and component changes, leading to observations of phase drift potentially linked to component aging or environmental factors. Adding a second limiter and a power amplifier revealed their respective noise contributions. Various measurement techniques and instruments (TinyPFA, Linrad, PN 2060) were employed, highlighting discrepancies and measurement artifacts, particularly concerning instrument noise floor and the effects of impedance mismatches, referencing a published paper on the topic. Significant effort went into identifying and mitigating noise sources, including power supply decoupling, thermal sensitivity, mechanical vibrations impacting trimmer capacitors, and electrostatic/mechanical effects from foam insulation. Adding a second crystal filter improved close-in phase noise but introduced significant stability problems (frequency jumps, sensitivity to temperature changes, component instability). While achieving low phase noise comparable to or better than a Venzel reference oscillator in certain regions, the experimental oscillator ultimately suffered from persistent instabilities, preventing its use as a stable reference.
Oscillator Noise and Stability Experiments
0:00:04 - Initial Buffer Amplifier Attempt: Built a buffer amplifier using four parallel BF240 transistors with ferrite beads (intended to prevent oscillations) and a 50-ohm output resistor.
0:01:07 - Buffer Amplifier Issues: Found this amplifier degraded noise performance significantly, especially at higher signal levels (tested up to 14 dBm), indicating noise loading issues. The approach was scrapped.
0:03:02 - Limiter Modifications: Modified the amplitude limiter by replacing resistors with ferrite beads and adjusting a transformer for better 50-ohm impedance matching.
0:04:18 - Phase Drift Observed: Multiple measurements showed consistent phase drift over time, initially suspected to be related to glue hardening or temperature effects, affecting long-term stability.
0:06:19 - Baseline Noise Measurement: Measured the system noise floor by removing the amplifier under test, noting the need for the TinyPFA reference to stabilize.
0:08:10 - Second Limiter Added: Introduced a second identical limiting amplifier, which initially caused oscillations (requiring RC networks for suppression) and slightly degraded noise performance compared to the single limiter setup.
0:10:10 - Further Limiter Adjustment: Removed a 470-ohm resistor from the first limiter's transformer (as the second limiter now provides the load), restoring expected voltage levels and improving noise similarity between the two limiter outputs.
0:12:27 - Power Amplifier Added: Integrated a power amplifier after the limiting stage, boosting output to +19 dBm. Analysis showed it added some noise, particularly at shorter time scales (below 1 second / above 1 Hz offset).
0:16:52 - Measurement System Comparison: Compared phase noise measurements using Linrad (custom two-channel direct conversion receiver) and a PN 2060 analyzer. Good agreement was found in many regions, but discrepancies appeared at very low frequencies (<0.1 Hz) and around 10 Hz, where the PN 2060 showed higher noise. Amplitude noise was also analyzed.
0:23:41 - PN 2060 Internal Noise: Measured the internal noise floor of the PN 2060 by feeding the same signal to all inputs, finding it consistent in some regions but noting unexpected results where the oscillator measured below the instrument's apparent noise floor.
0:26:35 - Oscillator Noise Floor Issue: Observed the oscillator's wideband noise floor (-160 dBc/Hz) was much higher than expected (-185 dBc/Hz seen previously).
0:27:28 - Power Amplifier Influence Check: Disconnected the power amplifier; this didn't significantly change the close-in noise but slightly lowered the wideband noise floor. The investigation into the high noise floor continued.
0:28:38 - Venzel Oscillator Re-Check: Measured a reference Venzel oscillator again, finding its noise performance matched the current experimental oscillator, contradicting a previous, better measurement of the Venzel.
0:29:57 - Measurement Error Found (Operator Mistake): Discovered the previous high noise measurements were due to an incorrect setting on the reference oscillator's PLL time constant (knob left in the wrong position), causing correlated noise from the Rubidium source. Correcting this restored expected low noise levels.
0:32:32 - Performance After Correction: With the measurement error fixed, the experimental oscillator showed performance very close to (slightly better than) the Venzel oscillator's previous measurements, particularly below 1 Hz offset.
0:34:14 - Wideband Noise Floor Improvement: Changed transformer ratios in the amplifier, slightly worsening close-in noise (~1 dB) but significantly improving the wideband noise floor.
0:34:49 - Amplitude Noise Analysis: Compared amplitude noise measurements between Linrad and PN 2060, noting discrepancies and potential artifacts (spurs, peaks) in both systems.
0:37:03 - Bandpass Filter Test: Added a narrow crystal bandpass filter (650 Hz BW) after the oscillator. This added noise near the passband edges, as expected, but measurements revealed artifacts and inconsistencies, particularly with the PN 2060.
0:38:35 - Phase Noise with Filter: Measured phase noise with the narrow filter, showing it significantly increased noise close to the carrier but revealed the underlying measurement system noise floor further out.
0:39:51 - Phase Noise Without Filter (Baseline): Measured phase noise without the filter, establishing a baseline performance with a flat noise floor around -176 dBc/Hz between 1-10 kHz.
0:42:04 - Supply Noise Issue Identified: Suspected AC noise (20-100 kHz) on the power supply rails within the oscillator box was causing modulation sidebands.
0:42:38 - Notch Filter Analysis & Decoupling Fix: Used a notch filter setup to examine wideband sidebands. Identified strong modulation sidebands caused by AC pickup on the power supply cable. Adding a large electrolytic decoupling capacitor inside the oscillator box dramatically reduced these sidebands.
0:45:43 - Testing Different Crystal Filters: Compared the noise contribution of the narrow 650 Hz filter (noisy) with a wider 1.4 kHz filter (much lower added noise), deeming the wider one suitable for testing measurement systems.
0:47:25 - Improved AM Noise After Decoupling: Re-measured amplitude noise after adding the internal decoupling capacitor. The wideband noise floor improved significantly (around -180 dBc/Hz at 10 kHz on Linrad, slightly higher on PN 2060).
0:49:29 - Improved PM Noise After Decoupling: Re-measured phase noise. Showed a peculiar dip around 1 kHz, reaching below -180 dBc/Hz, with noise increasing at wider offsets. This was observed on both Linrad and PN 2060.
0:52:20 - Investigating Measurement Artifacts (Impedance Mismatch): Inserted a known-good filter and observed negative correlation results (blue trace in Linrad, circles in PN 2060), indicating measurement artifacts due to impedance mismatch, as described in a cited paper by Rubiola et al. Added an attenuator to improve matching, which mitigated the issue.
0:55:18 - Final Output Stage Added: Added a final "noiseless feedback" amplifier (4x BF240) to provide a stable 50-ohm output at +19.5 dBm, replacing the previous power amplifier.
0:55:56 - Investigating Attenuation Effects: Systematically added attenuation after the oscillator (0 dB to 36 dB) and observed the phase noise floor. Unlike the behavior described in the cited paper for commercial instruments (where noise floor varies significantly with attenuation due to thermal differences), this custom setup showed minimal change, likely because the attenuator and splitter were at the same temperature.
1:05:50 - Measurement Setup Details: Described the complex measurement setup involving hybrids, splitters, couplers, stepped attenuators, buffer amplifiers, and separate PLL-controlled local oscillators locked to a Rubidium reference for both Linrad and the PN 2060.
1:07:56 - Testing with External Filter for Impedance Matching: Combined the oscillator with an external crystal filter and impedance matching network (notch filter, quarter-wave cable, matchbox, attenuator) to present a stable 50-ohm impedance over a wider bandwidth (+/- 10 kHz) to the measurement system.
1:10:32 - Ultra-Low Noise Floor Measurement: Measured the phase noise with the impedance-matched filter setup over 27 hours. Achieved a noise floor below -185 dBc/Hz, demonstrating the capability of the measurement system but revealing potential artifacts (negative correlation points/circles) in the PN 2060 at these low levels, possibly due to internal crosstalk.
1:15:22 - Adding a Second Crystal Filter: Added a second crystal filter in series within the oscillator to increase the effective Q (from 1.15M to 1.83M), aiming to improve close-in phase noise.
1:17:07 - Performance with Two Filters: Initial measurements showed a potential ~3 dB improvement in close-in phase noise (e.g., -123 dBc/Hz at 1 Hz, compared to ~-121 dBc/Hz before adjustments), but stability issues became apparent.
1:18:44 - Comparison with PN 2060 (Two Filters): Measurements between Linrad and PN 2060 showed good agreement down to ~ -125 dBc/Hz @ 1 Hz and -152 dBc/Hz @ 10 Hz, though potential excess noise in the PN 2060 around 10-100 Hz seemed resolved (possibly due to better USB grounding).
1:24:53 - Stability Problems Emerge: Observed significant glitches and frequency jumps with the two-filter setup, indicating instability.
1:25:24 - Troubleshooting Instability (Soldering): Resoldered connections in the added filter, but glitches persisted, although phase noise measurements between glitches remained good (~ -126 dBc/Hz @ 1 Hz).
1:29:22 - Identifying Glitch Nature: Used waterfall plots and frequency vs. time graphs to analyze the glitches. Some appeared to be large frequency jumps (~8 mHz), while others seemed related to amplitude/phase jumps possibly originating within the measurement system (channel 1 specific).
1:35:13 - Probing with Lid Open (Infrasound Noise): Opened the oscillator box to probe components. This introduced significant noise from infrasound (~10 dB increase at 1 Hz).
1:37:08 - Trimmer Capacitor Sensitivity: Found one trimmer capacitor was extremely sensitive to mechanical touch (using a plastic straw), causing large frequency shifts. Adjusting it to use the other side of its range improved stability significantly.
1:38:58 - Persistent Problems & Thermal Sensitivity: Despite fixing the trimmer, other spontaneous glitches occurred. Experiments involving gentle cooling/heating of the box revealed extreme sensitivity to temperature changes, triggering long periods (~10 mins) of noisy operation.
1:50:18 - Foam Insulation Identified as Culprit: Found that foam plastic used for thermal isolation around the crystal oven thermostat was in contact with crystal wires and likely caused noise bursts due to movement/electrostatic effects. Replaced it with cotton.
1:50:51 - Further Glitches: Despite removing the foam, other smaller glitches persisted, some appearing related to amplitude jumps in channel 1 of the measurement system, others potentially still within the oscillator.
2:00:15 - Localized Cooling Tests: Used dust cleaner spray (cyclopentane) and small pieces of ice to apply localized cooling to different parts of the oscillator box. Identified areas near the filter amplifiers and tuning capacitors as particularly sensitive, triggering the noisy state.
2:06:54 - Heating Test: Heating a sensitive area near the first filter amplifier also triggered noisy operation, which slowly recovered. Mechanical taps caused transient noise but didn't trigger the persistent noisy state.
2:09:42 - Long Run & Final Modifications: After a 20-hour run with fewer glitches, made final modifications: added small fixed capacitors parallel to the trimmers (to reduce current through the sliding contact) and added a varactor diode for electrical frequency tuning (EFC).
2:12:53 - Final Performance & Comparison: The modified oscillator achieved good phase noise (-124 dBc/Hz @ 1 Hz, -151 dBc/Hz @ 10 Hz), significantly better than the current channel 1 reference oscillator. Allan deviation plots confirmed improved short-term stability compared to the original reference oscillators.
2:16:10 - New Oscillator as Reference (Instability): Used the new dual-crystal oscillator as the channel 2 reference. Initial results looked promising, but instabilities (amplitude jumps) were still present in the new oscillator.
2:17:32 - Power Supply Failure: Identified that the original power supply used for the Venzel oscillator had failed, explaining recent poor measurements of it. Switching power supplies restored its expected performance.
2:18:21 - Final Comparison & Conclusion: Compared the final dual-crystal oscillator (now used as channel 1 reference) against the original reference (channel 2). The dual-crystal oscillator showed better short-term stability (Allan Deviation at 10s) and phase noise at >1Hz offsets, but worse long-term stability (>50s) due to larger frequency variations and persistent amplitude instabilities. The experiment was concluded due to these instabilities, although lessons were learned for future designs.