Range Filtering

This operator filters the spectras, of stack of SLC images, in the range direction. This is an optional step in interferometric processing chain. The filtering in range direction of master and slave image increases Signal-to-Noise Ration (SNR) in the interferogram. This noise reduction results from filtering out non overlapping parts of the spectrum. This spectral non overlap in range between master and slave is caused by a slightly different viewing angle of both sensors. The longer the perpendicular baseline, the smaller the overlapping part. Eventually a baseline of about 1100 m results in no overlap at all (that is also critical baseline for ERS). Assuming no local terrain slope, a reduction of typically 10-20% in the number of residues can be achieved.

The range filtering should be performed after coregistration (after slave images are resampled/warped to the master grid), because the fringe frequency is estimated from the interferogram (that is temporary computed). It is performed simultaneously for the master and slave image, unless there are multiple slave images in the stack. If later is the case, as with Azimuth Filtering Operator, only slave images will be filtered while the master image will be left in its original state.

Implementation Details

Currently, only a so-called "adaptive" filtering is implemented, while method based on orbital data and terrain slope will be implemented in coming releases.

Adaptive range filtering algorithm builds on the local fringe frequency estimated from the locally computed interferogram. After the warping/resampling of the slave on the master grid the local fringe frequency is estimated using peak analysis of the power of the spectrum of the complex interferogram. The warping/resampling is required since the local fringe frequency is estimated from the interferogram. The fringe frequency is directly related to the spectral shift in range direction. (Note: that this shift is not an actual shift, but it is an indication that the different frequencies are mapped on places with this shift.

Input parameters

FFT Window Length: Length of the estimation window, a peak is estimated for parts of this length.
Hamming Alpha: Weight for hamming filter. (Note that, if alpha is set to 1, the weighting window function will be of rectangular type).
Walking Mean Window: Number of lines over which the (walking) mean will be computed. This parameter reduces noise for the peak estimation. The parameter has to be an odd number. Logically, the walking mean can be compared with the principles of periodogram estimation.
SNR Threshold: In peak estimation, weight values to bias higher frequencies. The reasoning for this parameter is that the low frequencies are (for small Oversample factors) aliased after interferogram generation. The de-weighting is done by a dividing by a triangle function (convolution of 2 rect window functions, the shape of the range spectrum). Effect of this parameter may be negligible to overall results.
Oversampling factor: Oversample master and slave(s) with this factor before computing the complex interferogram for the peak estimation. This factor has to be a power of 2. 2 is default, and with this factor the filter is able to estimate the peak for frequency shifts larger than half the bandwidth. A factor of 4 for example might give a better estimate, since the interval between shifts that can be estimated is in that case halfed (fixed FFT Window Length).

Source bands:

Source Bands are set of coregistered bands of the complex product.

Output bands:

Output Bands are set of bands with spectras filtered in range direction.