Absender: "Erhard Wielandt" <320062183964-0001@T-Online.de> Empfänger: "Bob Hutt" Datum: 01. Okt 2009 14:56 Betreff: Re: GST document, 28 Sep 2009 version Dear Bob: I know that you are busy, nevertheless you may be interested in the attached three plots. They show how sensor nonlinearity is seen by CALEX and DISPCAL. The experiment was made during a lecture at Potsdam in the lecture room; the audience is responsible for occasional bursts of noise in the data. The seismometer was a Mark L4-3D passive, 1 Hz, damped electromagnetic sensor. gfzsweep.png: a sweep calibration. The signal coil was used for the stimulus as well, so the output signal is a mixture of input and output, and the resonance of the sensor at 1 Hz causes only a slight increase of the amplitude. The residual exhibits some misfit at high frequencies (5 Hz), a better fit between 2 and 1 Hz, and the second harmonic of the stimulus at lower frequencies (where the mass displacement is largest). gfzstep.png: the same with a stepped calibration current. We sent a positive, zero or negative current through the coil, as is apparent from the input trace (top). Now look at the polarity of the residual: it has opposite sign when going from -current to zero, and from zero to +current. Of course, for a linear system the residual should be the same. CALEX fits the average step and then one of the steps is smaller and the other larger. gfzsteptable.png: the same sensor on the step table. The upward steps are slower than the downward steps because the motor has to work harder. Nevertheless the final displacement must be the same. The last trace in the plot shows the deviation of the step size from the average. (The level after each step indicates the deviation; the plot clips at 1%.) The downward steps are systematically smaller than average by about 2% and the upward steps are larger by 2%. You may read these numbers from the protocol file dispotstep.out or from disptable.lst. After removing the average discrepancy between up-steps and down-steps, the generator constant scatters by only +-0.2%. So the test does in fact show sensor nonlinearity very clearly. We also had a Guralp (60 sec) broadband sensor on the same table and saw no nonlinearity. Kind regards, Erhard Anlagen: ptable.png, psweep.png, pstep.png, disptable.lst, dispotstep.out