A GPS disciplined oscillator (GPSDO) is designed to get the best of both worlds: the short-term frequency stability of a free-running, low-noise quartz oscillator and the long-term timing performance delivered by GPS. Obviously there is region between short- and long-term where the performance of a GPSDO is necessarily a compromise between the two sources.
For any given measurement averaging interval (tau) the frequency stability of a GPSDO is a combination of many internal limiting noise sources: OCXO instability, GPS receiver timing jitter, DAC and voltage reference noise, temperature variations, power supply noise and instability, buffer amplifier noise, etc. Depending on the tau, one or more of these noise sources may dominate.
Without detailed analysis, it's not always easy to determine exactly how much of noise each source contributes. But one of the easiest experiments to perform is to contrast the frequency stability of a GPSDO in two modes: free-running and locked. This reveals a great deal about the quality and implementation of the GPSDO. For each GPSDO, and in each mode, measurements were made against a very stable 10 MHz reference.
I had hoped to make detailed measurements on a dozen or more models of GPSDO that I have lying around the lab, but due to a busy schedule in 2007, I have data on only 4. Still, until I get around to measuring the rest of them, these four are quite interesting.