symmetricom_-_allan_variance_-_challenges_and_opportunities.pdf
The attached white paper from Symmetricom may be of interest.
Among other things it explains why a common technique is largely a waste of time for tau greater than the inverse PLL bandwidth.
It also indicates that using a bandpass filter centered on the beat frequency really should be used in DMTD systems.
However the associated filter phase shift tempco will require that the filter components have low tempcos and will require elaborate temperature control if analog filters are used.
The best compromise would be to use a high end sound card and use digital filters.
One can then interpolate to accurately time stamp the zero crossings of the bandpass filter signals.
Although this method doesn't make optimum use of the signal samples it will implement a more ideal DMTD albeit with a noise level of around 1E-13/Tau.
However, one is still left with the problem of calibrating the sampling clock.
Bruce