Filterbank implementations of a window based Gabor transform for SAR processing

In this thesis a Window Based Gabor Transform (WBGT) is presented that is suitable for real time digital signal processing. Other Gabor Transform approaches presented in the literature such as the Orthogonal Like Discrete Gabor Transform (OLGT) are not suitable for real time processing. The reduction in analysis time using the WBGT instead of the OLGT is proportional to the ratio of the analysis window to the signal length. Using the WBGT the number of JTF coefficients required to accurately represent the original signal is far less than when using the STFT. Two filterbank implementations for the WBGT which reduce analysis times are developed, one using serial input and the second using a parallel input, each of which computes the frequency content of the signal using N parallel band pass filters. In the case of the parallel input filterbank, the analysis times are similar to those using a Short Time Fourier Transform with FFT. The parallel input filterbank can be implemented using a matrix formulation for the N band pass filters. This approach is suitable for software implementation and results in analysis times that are lower than when using an STFT or the parallel input implementation with the bank of band pass filters. The WBGT, along with the filterbank implementations result in very accurate JTF spectra. Near perfect reconstruction of the original signal from the JTF spectra was obtained for several cases for the Gabor Transform techniques presented in this thesis. Finally, the use of the parallel input filterbank in the context of SAR processing is presented and the hardware requirements are derived and shown to be achievable using commercially available products. (Abstract shortened by UMI.).