dual receiver setup for interferometry measurements

I have a question concerning connecting two DVB-T dongles on the same clock source for interferometric (or passive radar) measurements, as described at http://kaira.sgo.fi/2013/09/16-dual-channel-coherent-digital.html I have assembled the same system with one dongle used as oscillator on a 28.8 MHz resonator and the second one as a slave to this clock. All works fine, solved the issue when the oscillator would not start, now I have a reliable source of measurements. Initial tests (these are R820T-based dongles) exhibits significant random phase drift which I attributed to heating of the chips (they get above 50 degC when running continuously), so after gluing two heat sink with heat-conducting epoxy, I more or less managed to get a stable phase measurement when recording a same oscillator (200 MHz) with the two dongles and displaying the phase as angle(conjugate(signal1)*signal2). The question is as follows: at http://jmfriedt.sequanux.org/ph_tout.pdf I have shown one graph, quite representative of all my experiments, displaying the evolution of the phase difference between both dongles connected to the same 200 MHz oscillator. I *always* start with a quite stable phase difference (red curve -- inset in a zoom of this particular measurement) after plugging in my USB hub fitted with the two dongles and starting gnuradio-companion for recording the dongle I/Q stream (notice the abscissa sampling rate of 10 Hz => the full graph is about 1-hour long). After about an hour, I stop recording the red curve, and all I do is switch off gnuradio-companion and start it back => green curve with a quickly falling phase. Switch off again, disconnect-reconnect USB hub, restart an acquisition => blue curve. Same procedure => magenta curve. Can anyone hint at an explanation as to why I always start with a reasonably stable phase difference (yet not constant -- is the phase fluctuation indeed due to heating of the fractional PLL in each RF frontend, drifting below the feedback loop time constant ?), but more worrisome why I always get this huge drift after launching a new acquisition ? The fact that I always get the same slope hints at a sofware/hardware communication issue, but how it is possible since both dongles are clocked by the same source and receive the same commands from the software ? Thanks, JM -- JM Friedt, FEMTO-ST Time & Frequency/SENSeOR, 32 av. observatoire, 25044 Besancon, France