I've built a GPS Disciplined Oscillator based on the design by G3RUH (with a Rockwell Jupiter TU30-D140-xxx GPS board) and a Morion Inc. MV89a 10MHz Double Oven Controlled Xtal Oscillator (DOCXO). The MV89a was found on ebay for just € 3, unfortunately the price has increased ever since and can now be found for around € 20! Here's a local copy of MV89a datasheet.
By itself the DOCXO already has a stability in the 10e-9 range, and it has a tuning range of +/-0,25ppm (+/-2,5Hz at 10MHz) with external control voltage range (Uin) 0…+5 V
Locking this DOCXO to a 10kHz clock, derived from a GPS receiver, can give a nice long-term accuracy improvement.
I've designed a PCB* (dimensions 70mm(w) x 60mm(h), based on the circuit described in "A Simple GPS Disciplined Oscillator", published by James G3RUH (see links page). The component placement for the pcb is here and also the schematic diagram upon request.
Also a supporting PCB (65mm x 95mm) to neatly mount the MV89a DOCXO was quickly designed*.
Update january 2013: New version of the MV89a support PCB v2. It has provisions for on-board fine-tune (multiturn) potmeter / voltage divider from Vref to Vin (only needed for fine-tuning when the DOCXO is used stand-alone i.e. without GPS) and a SMA-socket for 10MHz output.
By request, here is the component layout for the MV89a support PCB v2 .
The PCB I've designed is missing three traces. This can be solved with short wire-bridges (all from the "1CLR" and "2CLR" pins of the 74HC390 divider IC's to GND). It's also best to ground the inputs of the unused half of U4 74HC390 to prevent oscillation.
I've built my version on single-sided pc board but I had some problems with spurious (1MHz, 100kHz and 10kHz) leakage from the dividers on the 10MHz output signal, even after applying a 10MHz filter and adding a lot of decoupling caps on the trace-side.
It's best to use 2-sided copper clad board, leaving one side as ground plane like in proper HF-designs. Wire-thru-holes should be placed connect the ground-traces on the other size.
Loop filter: The component values of the loop filter R1 R2 C1 C2 need to be adjusted / optimized for the OCXO used.
For my MV89a: C1=4u7 X7R C2=30uF (3x10uF parallel) X7R, R1=22k and R2=12k but these values are still not optimal.
Edit 2013/06/03: Improved loop filter values for the MV89a DOCXO: R1=47k C1=4u7 tantalum. R2 and C2 not used. Note: C1 now mounted close to the MV89a Vin pin on the support board, this greatly improved suppression of the 10kHz sidebands (from the phase comparator) on the 10MHz signal.