Digital Signal Processing to Enhance Oceanographic Observations

Abstract

Quantization noise, the difference between a continuous physical signal and its discrete integer approximation, is an unavoidable consequence of data sampling. The problem is particularly acute for oceanographic data because these signals are usually red while the quantization noise is white, and this spectral mismatch limits our ability to detect short-term (high-frequency) fluctuations. A method of preemphasis and deconvolution is presented that reduces quantization noise and increases the resolution of short-term fluctuations by a factor of several hundred without any reduction in the full-scale range of the measurements. Examples are presented of a 12-bit thermometer with a range of ?5° to 35°C and a resolution of 60 µ°C, and a 14-bit pressure gauge with a range of 600 db and a resolution of 1 ? 10?4 db. The preemphasis consists of summing a signal and its scaled time derivative before sampling. The enhanced version of the signal is recovered by convolving the preemphasized signal with a discrete single-pole low-pass filter with a time constant determined by the scale factor applied to the derivative. Alternatively, the signal and its derivative can be sampled separately and then combined in the discrete domain before deconvolution.