Balancing ensemble averaging and signal stationarity when processing acoustic data recorded during a rocket launch

There are many important details to consider when collecting acoustical data during a rocket launch including system noise floor, dynamic range and transducer sensitivity. In this talk, the choice of signal processing parameters will also be shown to be important. Since the pressures are created by turbulence, ensemble averaging is key to reducing random variation. As such, the use of a Hanning window with 50% overlap would be customary for performing this type of processing. However, as the rocket lifts off, the source location moves, and the data can only be considered stationary for short blocks of time. Therefore, the ability to average is limited by non-stationarity. To balance these competing phenomena, an adjustable windowing function such as the Tukey window may be more appropriate than Hanning. Noise data collected at two different launches will be used to compare of the Hanning window with 50% overlap to that using of the Tukey window with a taper of α = 0.25, which filters 12.5% of the signal at each extremity, with 87.5% overlap. The use of both window types will be compared when estimating auto spectral densities and integrated overall levels. The effect of blocksize will also be determined.