It can be seen on the spline chart that the 1st derivative grows on intervals between X and O - which means the 2nd derivative (acceleration) is positive.

I think testing in-motion is a bad idea to start with, especially when tested object is moving back and forth. In critical situation where thrust is only observable for few hundred ns, and object is moving against it, data would be obfuscated beyond recognition. 

Why not create perfectly still floating test-bed? Change the device setup from continuous power into short, higher-powered condensed harmonic bursts. If the device will start making micro-waves, you will know something is up.

Last thing, I'm not a big fan of on-board battery, I think test-bed should only contain the RF generator and cavity, power should come via gently hanging cables coming form power source above, IMO.

Here is an analysis using Fourier transform filtering.

All time domain signals can be represented in terms of their component frequencies, in this experiment we aren't interested in the high frequencies components because the experimental perturbation isn't rapidly switched, so we can filter them out, which is what I've done.  I've also overlayed the ranges when the drive was active.

The Y axis is displacement because I subtracted the mean.

Thanx a lot. Looks good to me, the result leaves room for hope.

I´ll make another test today with longer on/off times and with the cavity turned by 180°

Simple interpolaton with 5th degree splines with knots fixed to ON (X) and OFF (O) points.

1st derivative of the interpolation (disregard the edges).

Here's some shabby analysis.  If I saw anything I would've gone deeper!

https://www.dropbox.com/s/t2syuuj8mtw2h04/emdrive.xlsx?dl=0