Someone's designed a WWVB signal generator. The output circuit is quite simple—some resistors (some specified as 1%) and an antenna that looks like a typical WWVB receiving antenna.
There's no source, but some inspection of the circuit plus knowledge of WWVB tends to indicate the following: At W1 you can get a ~3.5V P-P signal (if RB1 is toggling and RB0 is input) or a ~.12V P-P signal (if RB0 is toggling and RB1 is input). This is a ratio of about 30:1, or 15dB. The WWVB signal is supposed to have a 17dB reduction in carrier between mark and space.
(Alternately, since only RB0 is an actual PWM output on the PIC16LF87, maybe it's used like this: With RB1 as input, you get a ~.12V P-P signal. With RB1 as a low output, you get a ~.07V P-P signal, for a carrier ratio of only 2:1 or so. Except for RB1 not being PWM, this doesn't make as much sense to me; RB3 would seem to have been the more natural choice for a second PWM output; a write to CCPMX suffices to switch PWM between these two pins)
One is tempted to imagine an AVR (or even PC!) implementation. But a big part of this is to have correct code to generate the WWVB data to test against. Next, I turned up this WWVB subcode test generator. I assume that a guy running the website 'leapsecond.com' knows a thing or two about time signals. (however, I did spot a half bug in the program—it assumes that years that are multiples of 100 are leap years. Since WWVB doesn't include a century, it's actually impossible to tell whether year 00 is a leap year. The current code is right in the sense that the only 00 year that WWVB has been operational was 2000, and that one was a leap year.