This circuit is essentially the same as the original design. The 1 PPS pulse from the GPS receiver sets the latch in the 74HC4046, and it is cleared almost immediately (within one microsecond, of course) by the next 1 MHz edge from the reference divider.
As a result, the output of the latch (pin 9) has a narrow pulse on it which varies in width with the phase of the reference compared with the 1 PPS from the GPS receiver. The pulse is passed through a Schottky diode to an integrator, a small capacitor. The capacitor is slowly discharged by the 10 M resistor, but before it can, the micro reads the voltage on the capacitor in an interrupt routine, also initiated by the 1 PPS pulse.
The purpose of the Schottky diode is to prevent the capacitor from discharging back into the latch after the pulse has gone.
The Time Interval Counter circuit. (Click on the picture for a full-size diagram you can print).
This circuit is a little temperature sensitive, and best long term stability (Adev at 100 sec and 1000 sec tau) will be achieved if this circuitry is assembled quite close to the micro, and is insulated from both convective and conductive heating. The GPS receiver, the 5 V regulator, and of course the OCXO are the most significant heat sources. I placed the TIC circuit in a remote corner of the board, away from the heat sources, and later covered the TIC circuit with bubble-wrap, and this made a significant difference.
The improvement is readily illustrated in the next picture, which is an Allan Deviation plot of my prototype.
Each time I made a measurement, things improved. The blue trace is the first from the completed unit, with default parameters. While 1e-10 stability might seem wonderful, I knew (from Lars' own results) that I could do a lot better.
Allan Deviation measurements. (Click on the picture for a full-size diagram you can print).
The green trace is after I had made improvements to the time constant and damping (setting these parameters is covered well in Lars' original documents). The really interesting trace is the final (cyan coloured) one, where the stability is better everywhere, and is significantly improved at 10 out to 1000 seconds tau. This was the result of insulating the TIC circuit. It should be possible (with a lot more experimentation with layout and insulation) to achieve better then 0.01 ppb (1e-11) all the way out to 1000 seconds tau.
The TimeLab software used to generate this graph is free, and the Lars' GPSDO code sends data in a format that the program can readily use. There is no other method (including the SBSpectrum method mentioned on the Reference & Dividers page) that can 'see' stability down to better than 0.01 ppb (1e-11). SBSpectrum at 25 MHz has a resolution of about 0.4 ppb, and relies on having a significantly more stable local reference. The Time Interval method requires no fancy equipment: just the 1 PPS signal from the GPS receiver and the firmware in the LARS GPSDO.
Introduction Reference & Dividers Buffers Micro & Misc.
Copyright © M. Greenman 2020. All rights reserved. Contact the author before using any of this material.