The MAI method developed by Jung was designed to reduce phase noise as much as possible because the MAI measurement accuracy heavily depends on the interferogram phase (Jung et al., 2009). Although it improved MAI performance by reducing interferogram phase errors and correcting phase distortions, it still had some flaws, such as filtering boundary artifacts at the boundary of incoherent areas and loss of MAI coherent phase in the presence of large and complex LOS deformation. This section describes an effective MAI processing method for overcoming such DInSAR shortcomings.
The main characteristic of this MAI processing is the construction of an MAI interferogram from residual interferograms with forward- and backward-looking components (Figure 2.5). The two residual interferograms are produced by combining a hard-filtered DInSAR interferogram with the forward- and backward-looking differential interferograms. Because it is produced by the phase difference between two interferograms with various squint angles, the residual interferogram is a type of MAI interferogram. Because different parts of radar signals are used, we can assume that the forward- and backward-looking differential interferograms are independent. However, because the radar beams partially overlap, one of the sub-aperture differential interferograms and the DInSAR interferogram are not independent and have a similar noise pattern. The DInSAR interferogram should be hard-filtered to preserve as much of the noise pattern of the sub-aperture differential interferograms as possible. Because their noise pattern differs significantly from the sub-aperture differential interferograms, the MAI interferogram may be distorted if complex filtering is not applied to generate the residual interferogram. Furthermore, the noise pattern can be canceled out. Noisy pixels in areas with high temporal decorrelation may be considered valid signals.
Figure (2. 5): Forward- and backward-looking MAI beam geometry. The aperture is divided to allow observations along the forward and reverse LOS vectors.
As a result, it is critical to hard-filter the DInSAR interferogram. Three-time adaptive filtering with different kernels can be used to implement complex filtering. Following the generation of the two residual interferograms, the second multi-looking and adaptive filtering is applied to the residual interferograms in the same manner as the InSAR processing to reduce MAI interferometric noises. The phase difference between the multi-looked and filtered forward- and backward-looking residual interferograms yields the MAI interferogram. The MAI interferogram exhibits phase distortions due to flat earth and topographic effects (Jung et al. 2009). The difference in perpendicular baseline causes the unwanted phase and squint angle between the forward- and backward-looking interferograms. A polynomial model (Jung et al. 2009) estimates the distorted MAI phase, and the final MAI interferogram is created by removing the distorted MAI phase. The corrected MAI interferogram can be subjected to yet another adaptive filter.
