I finished reading section 8 of the Technical Report, which is the material germane to this thread. It is a good clear explanation of the problem/approaches and I found it very useful.
However, there is something missing from it and parenthetically from all of the material I have read so far. Specially, how do you use the Allan variance (or its square root, the Allan deviation) once you have computed it. I imagine someone setting up a system might want to know if a particular oscillator is suitable for use in that system. For example, a hobbyiest may wish to measure signal characteristics of some radio transmission system he is building. He wants to use a 10 MHz reference clock passed through a distribution amplifier to synchronize the instruments he is using to test his system.
For the sake of convenience, let's refer to the paper by Rutman & Walls (linked by GerryBags) instead of the NBS Monograph, since it contains everything we need and everybody has access to it.
There are (generally) two ways to view the stability of an oscillator -- in the frequency domain and in the time domain. In the frequency domain, one is concerned with the spectral densities of phase and frequency fluctuations and the resulting spectral density of the oscillator itself. In the time domain, one is concerned with variances, i.e. standard deviations. The utility of the Allan Variance is that it provides a link between these two domains. (See Table 1 of the cited paper.) In brief, the Allan Variance allows one to determine the dominant noise process of an oscillator on any given time scale. For example, if the (root) Allan Variance of an oscillator has 1/root(Tau) dependence over some time scale, it can be concluded that the dominant noise source on that time scale is white phase noise. If the (root) Alllan Variance of this same oscillator flattens out at longer times, it can be concluded that the dominant noise source is flicker frequency at longer times.
I understand from the reading that I have done that computing the traditional variance of fractional frequency data doesn't work because it doesn't converge as the sample size increases. That is one of the reasons Allan created his variance measure. But, with the traditional variance (actually its square root, the standard deviation), if you assume a gaussian distribution of the fractional frequency process, 99.7% of the values lie within a 3 sigma band around the mean. So, if the designer had a traditional variance to work with, he could look at the range of frequencies within the 3 sigma band and decide whether that sort of jitter was acceptible for testing his system.
You cannot make the assumption that the distribution is Gaussian and, therefore, you can't conclude much from the standard deviation of the frequency of an oscillator. In general, a traditional variance provides very little information about the underlying behavior of an oscillator. As an example, consider two 10 MHz oscillators monitored over some long time period. Oscillator 1 might have very low "jitter" but it drifts linearly about 10 Hz during the monitoring period. Oscillator 2 might not have any linear drift, but it has about 10 Hz of random jitter. Despite having very different characteristics, these oscillators will have nearly the same traditional variance. Their Allan Variances, however, will be very different.
So far, I have found nothing like this for the Allan variance. How do you use it in a practical situation?
I'm guessing a lot of people on this forum are familiar with GPS-disciplined oscillators, where a local oscillator is locked to a signal extracted from GPS satellites. The big advantage of these devices is the ability to impose the long term stability of the GPS (Cesium) clocks onto a less expensive, local clock. But, how does one marry the local clock to the GPS signal? In other words, how quickly or how tightly does one lock the local oscillator onto the GPS signal? The answer to that question depends on the characteristics of both the local oscillator and the GPS signal, and the Allan Variance gives you those characteristics. For instance, the Allan Variance of a simple TXCO might indicate it should be locked onto the GPS signal with a short time constant, whereas the Allan Variance of a Rb oscillator might indicate that it should be locked onto the GPS signal with a relatively long time constant.