Characterizing Dynamic Effects of Oscillator Power Cycling

We create a measurement technique and easy to interpret metrics to be used in development of new oscillators specifically for applications in which the oscillator’s power is turned on and off. This is useful in predicting the performance in, for example, the frequency-difference-ofarrival (FDOA) geolocation technique, which is used to monitor and track an emitter’s location by observing its Doppler frequency shifts at a set of receivers. To conserve energy, FDOA applications compute Doppler tracks from an emitter that is powered “on” or measured for short periods (τon) after a long “off” period, called the “stride” interval (τs). For lowest size, weight, and power (SWaP) and lowest phase noise and best frequency stability, evaluations are focused on Oven-Controlled Crystal Oscillators (OCXO’s) and Temperature-Controlled Crystal Oscillators (TCXO’s). For illustration, we use τon = 3 s and τs = 60 s. This shows the need to consider the dynamic behavior during the short 3 s average frequency measurements as well as the 60 s sampling interval between measurements. Dynamic Allan Deviation (ADEV) does not accurately capture different noise types for such a short 3 s sample, so we propose using Dynamic ThêoH which characterizes the oscillator at power-on more accurately. Since RMS frequency differences vs. sampling time-intervals in multiples of 60 s cannot be used in place of the ADEV, we regard frequency differences as an uncertainty on an oscillator's predicted frequency, not on a mean frequency. This mimics ADEV and we can distinguish the dominant component of frequency prediction due to random-walk FM (RWFM) or an even more divergent noise type. This paper: (1) describes a measurement setup to obtain low-noise, fast fractional-frequency, time-series measurements, (2) motivates and illustrates Dynamic ThêoH, the hybrid of ADEV and THEO, for τon = 3 s, (3) constructs a statistic called Y(τon,τs,) which estimates a τs = 60 s frequency prediction error, and (4) transforms 3 s time-series measurements to phase noise L(f) for field applications and evaluations.