APPENDIX B DATA PROCESSING By George N. Landes and William R. Perret B.1 NORMAL PROCESSING Ground motion from large-yield explosions is observed as acceleration because instrument response favors this over direct observation of velocity or displacement. Required information concerning velocity and displacement must be derived thrcughiterated integration of the acceleration-time data, As observed during Operation Ivy, output of the accelerometers was recorded as a modulated carrier frequency on magnetic tape. This information was played back through a system which produced a photographic record. That record was in turn translated on Telereaders into displacements of the information trace from a reference trace as a function of time with pre- cision of +0,003 in. and +0.0001 sec. These data were recorded in digital form on IBM cards and converted according to pretest calibration data to accelerations as a function of time. B.2 TYPES OF ERROR Several types of error may be superimposed upon the data during recording and reduction. These errors are additive and, although small enough to be neglected in the raw data, may under some conditions have a serious effect on integrated results. Two ofthese errors— noise and drift—are inherent to the recording and reduction circuitry. Noise is a short-period random phenomenon. Drift is a long-period low-amplitude effect which may be linear or sinusoidal. A sinusoidal drift may be of sufficiently long period comparedto the interval over which integration is performed to appear as a constant or roughly linear or parabolic error. The third type of error — shift—is a constant which usually results from reader error in identifying the zero or balance point of the record trace but may also be a. quasi-permanent change in the balance position of the trace caused by very strong transients. ; These errors are significant only in relation to the information signal, and the signal-toerror ratio must be the criterion of importance. Integration is a cumulative process, and er- rors which are insignificant to raw data may assume dominant proportions in the integral if they persist over sufficiently long periods. Three questions arise concerning the effects of errors on integrated data: 1. How does integration affect the relation between signal amplitude and error? 2, What corrections are feasible? 3. To what extent can corrected results be depended uponto indicate true motion? 50