Low Correlation Codes for Sonar Systems

Sonar is a vital technology for the detection of objects in the water. Sonarsystems have been redefined over many decades, but research is still beingconducted into optimal detection methods. Codes, and the filters thatprocess the codes, have been at the forefront of this research. An importantobjective has been the minimization of interference caused by reflections.Matched filters are commonly used in sonar systems. They are equivalent tocorrelation filters, which are bound by the Welch bound. The Welch boundgoverns the minimum peak correlation for points outside of detection.This thesis investigated matched filters and their bounds, and it wasfound that by relaxing the condition for detection, properties beyond theWelch bound could be achieved. By relaxing these conditions, the Welchbound no longer applies, and so a modified Welch bound was developedto accurately investigate the nature of these codes. In this thesis, methodsto generate codes for these new codes were investigated. Generating codesfor a matched filter is a non-convex problem, so gradient based methodswere utilised. Methods to improve correlation and power characteristicswere developed, along with methods for mapping a sequence for use witha digital transmitter having particular limitations. Mis-matched filters wereused to improve signal characteristics that may be lost due to this mapping.The performance of the generated codes was evaluated, and the rela-tionships between input parameters and output properties of the resultingsignal were observed. These performance assessments demonstrate thattradeoffs are required between various properties, and a balance is neededto obtain codes useful for sonar. The optimization was parametrized by anexample set of requirements for sonar. The signals were found to meet the given requirements, and when compared to codes typically used in sonar,the optimized signals were shown to have significantly better correlationproperties. Furthermore, compared to the general bounds for the propertiesof codes, it was found that the new codes had nearly optimal properties,and performed better than equivalent codes bounded by the Welch bound.The performance of codes were also investigated in a water tank toverify their feasibility. There were several additional considerations whichlimit codes that can be tested, and once these are taken into account thetest provided a robust method to verify the design process. Initial testsshowed results that differed from simulations, but after the inclusion ofzero padding before upscaling, the results from empirical testing agreewith simulation.Summarizing the research in this thesis, a new set of codes were devel-oped using a gradient based optimization method. The codes were mappedto a digital transmitter, and the filter adjusted using a mis-matched filter. The optimization was shown to generate near optimal codes which met allthe given sonar system requirements