Implementation

For each buffer of AF_BLOCKSIZE lines and width pixels do (taking care of AF_OVERLAP)

• Take 1DFFT in azimuth direction (over the columns).

• if the Doppler centroid frequencies do not vary per column, use the same filter for all columns, else compute the correct filter foreach column and use that. (First coarse coreg, align, then evaluate FDC polynomial.)
• Take inverse 1DFFT in azimuth direction (over the columns), yielding the output.

The azimuth spectrum is also weighted for the antenna pattern,

$$frac{\sin((f_a-f_{DC})/f_{Dop})} {\pi((f_a-f_{DC})/f_{Dop})}^2$$ (O1)

Where:
$f_{Dop}=1505$ Hz, the Doppler bandwidth, see [7].

We did not de-weight (and re-weight) the spectrum for this. This might be visible in figure 15.1 as a slightly asymmetric spectrum, for master and slave. We are not convinced that this re-weighting can be performed, without changing the signal. However we believe that possible errors are small.