The project is more or less complete as it stands. The DOT050V and OH300 variants are complete and on sale now. But I have an idea for a modification of the design.

Up until now, the design has obtained its output directly from the oscillator. But there is another option. We're using Timer 1 for the PPS discipline, but Timer 0 could be used to automatically generate an output frequency. The benefit to doing so is that the timer can divide the system clock to provide user-selectable output frequency.

Unfortunately, the minimum division is 2. That means if you want 10 MHz to be one of the options, you must clock the system at 20 MHz. But Atmel controllers won't run at 20 MHz with a 3.3 volt supply. That means they must be powered at 5 volts. That furthermore means that the serial receive pin of the GPS module must be level-shifted to insure it doesn't get hit with 5 volts, and the GPS module must be fed from the 3.3v supply. In addition, a pair of DIP switches connected to two of the unused inputs on the controller will allow the user to select one of four options for output frequency. The oscillator will feed directly into the external clock input pin of the controller (with an extra loading cap), and the OC0A pin of the controller will go into the input of the fanout buffer.

There are some big questions with doing this, however. Specifically, what's going to happen to the short term stability of the output when the controller is "in the way" like this? The saving grace is that the output is going to be divided down, which should actually make it easier to achieve the stability targets. Also, since the controller's output is at least indirectly being driven by the oscillator's input, it should (in principle) be no worse.

As a bonus, the output will be a 5 volt square wave instead of a 3.3 volt one, so it'll have a bit more "oomph." Also, the feedback granularity will double, meaning that the whole system will be twice as sensitive to phase shift, which should make the discipline just that much more effective.

I'm going to try one of these as a one-off and do a bunch of tests to see if the stability changes any relative to the current models. If the result is no worse, then perhaps this is a better way.