Ultra-stable microwave sources combining high frequency stability with low phase noise are required in wireless communication, radar, metrology, quantum information processing and tests of fundamental physics. When its oscillation frequency is locked to a high-
Q sapphire resonator by Pound frequency discrimination, the cryogenic sapphire oscillator (CSO) reaches a fractional frequency instability at the 10
-16 level and delivers phase noise approaching the thermal-noise limit. In such a loop the sapphireloaded cavity (SLC) acts as the frequency discriminator, so its own frequency fluctuations ultimately limit the stability of the locked carrier. This limit is usually assessed indirectly: the sensitivity of the resonance frequency to temperature, microwave power, vibration and magnetic field is measured first, and the ambient noise is then converted into a frequency-noise contribution. Such an estimate relies on transfer relations that must remain stable and precisely known, which is difficult to guarantee in practice. In this work the SLC is instead treated as a two-port device under test, and its residual frequency noise is measured directly, so that the stability limit of the CSO can be predicted without a direct comparison between two independent ultra-stable sources.
A differential phase-noise measurement scheme is designed accordingly. A low-noise 10.8013 GHz probe is synthesized by up-converting the 10.7885 GHz output of an independent CSO with an ultralownoise 12.8 MHz radio-frequency (RF) signal, and is fed into the SLC, which is held at 5.930 K, its frequency-temperature turning point, in a pulse-tube cryocooler. The transmitted signal is mixed with the same-source 10.7885 GHz reference down to 12.8 MHz, so that the phase noise of the common local oscillator (LO) is rejected in common mode, while the frequency jitter of the resonator is transferred to the beat note with a discrimination coefficient of 1/Δ
f0.5. The whispering-gallery mode WGH
+15,0,0 at 10.8013 GHz, with a 3 dB linewidth of 30 Hz (
Q = 3.6×10
8) and an insertion loss of 16 dB, is measured in this configuration. A noise model that includes the residual LO noise, the RF beat-note noise, the additive phase noise of three cascaded low-noise amplifiers (LNAs) and the SLC frequency noise predicts a measurement floor of -106 dBc/Hz at 1 Hz offset, dominated by the 1/
f phase noise of the LNAs.
Experimentally, the phase noise of the test signal within the resonator bandwidth is rejected in common mode and a floor of -110 dBc/Hz at 1 Hz offset is obtained. Converted with Δ
f0.5 = 15 Hz, the measured frequency noise corresponds to a relative frequency stability (Allan deviation) of 2×10
-14 at averaging times of 0.01-1 s, while the 1/
f noise of the LNAs, sets a floor of 8×10
-15. Replacing the CSO with a microwave source referenced to an active hydrogen maser confirms that the common-mode rejection is bounded by the cavity linewidth, and that below 0.05 Hz the LNA noise prevails over the superior long-term stability of the maser. The differential common-mode platform itself reaches the 10
-17 level between 2 s and 30 s and retains its rejection capability after the SLC is inserted.
These results provide a direct and efficient route to the frequency-noise characterization of microwave resonators and to the prediction of the stability limit of resonator-stabilized oscillators; the scheme is also applicable to residual-noise measurements of similar resonators and of active or passive microwave devices, and, with sapphire resonators of higher
Q (up to 10
9) or LNAs of lower 1/
f noise, the measurement floor can be further reduced, making the platform attractive for dark-matter detection and precision tests of fundamental physics.