It looks like the injection actually does hit the 400 count limit (plot 1). It saturates right at the end when the injection chirps up to high frequency. There's some kind of ringing as well (plot 2). From the spectrogram (plot 3) and the zoom (plot 4) this looks like a feature at just above 300 Hz. I thought it might be a notch for the PCal line, but that's 331.9 Hz. So someone will have to check the inverse actuation filter and see what's happening at that frequency.

It's possible to see the overflow from the first injection in the ETMY L3 MASTER channel (plot 5). It happens at -131072 counts, and the injection is trying to push it past -200000. The blind injection caused an overflow as well, but since this channel is only recorded at 2048 Hz, it looks like it falls short of overflow (plot 6). There's a faster readback whose name escapes me at the moment.

Unless the blind injection is made a factor of about 10 smaller, or rolled off at high frequency, it will be trivial to detect it by looking at the drive to the ETM.

FYI, the injected waveform was fiducial waveform from ER7:

https://alog.ligo-la.caltech.edu/aLOG/index.php?callRep=16125

It's a 1.4-1.4 BNS at 45 Mpc, optimal orientation.

There are a couple of things to watch out for when performing CBC hardware injections, based on iLIGO experience:

• In iLIGO, the actuation function had a notch at the violin-mode frequencies. If you just invert A(f) without smoothing this out, the notch will get inverted and you'll give the the mirrors a hard kick as the inspiral sweeps through the notch frequency. (My guess is that Adam and Jeff have already dealt with this, though.)
•  If the end of the inspiral signal is not smoothly rolled off and the signal drops from some loudish amplitude to zero then you put a step function in. This caused the iLIGO test mass to swing at its pendulum frequency, putting a huge ringdown in the data (called the "whooper" in iLIGO). In iLIGO, the CBC group generated waveforms in counts, by applying 1/A(f) ourselves using the adiabatic inverse calibration, so we smoothed 1/A(f) ourselves before applying it. In aLIGO, 1/A(f) is: (i) more complicated, and (ii) applied in the front end by IIR filters. It is possible that the sharp turnoff of the CBC injection at the end is causing the filters that apply 1/A(f) to ring and that's what's giving us garbage at the end of the signal.

For the ER7 injection we used an SEOBNRv2 waveform that has a ringdown at the end, hoping that this turn off would not trigger an impulse. However, for BNS masses, the turn off and ringdown is pretty sharp. I've asked Chris check that there are no "whooper" effects with the SEOBNRv2 waveform, but we haven't had chance to do this yet. For a SpinTaylorT4 waveform (the other waveform CBC wants to inject), there will definitely be a step, so this needs to be checked and rolled off carefully.

One other comment on the test: what scaling in awgstream did you use? That waveform looks monstously loud (eyeball SNR > 20). That's much louder than would be useful for a blind injections, but good for helping us find whooper effects.

Just for completeness, because I didn't see it posted, here's an Omega scan of the injections in h(t). The first is the non-blind injection, the second is the blind injection. I think the glitch ten seconds after the blind injection is unrelated. I thought it might be a filter turning off or being reset, but it's not on a GPS second (it's at 1124603210.28). It does cause an overflow of the ETMY ESD DAC.

I verified that the blind injection was correctly recorded in the raw frame file.