vector. The figure demonstrates that the (3,5) Biorthogonal wavelet basis best
describes the spatial variations. We use this wavelet basis for the remainder of
this thesis.
4.3.2 Gates on the regular grids
When gates are located on a regular grid, the two-dimensional wavelet transform
of the variations, s, can be expressed as the product of the variation vector, 1,
with the wavelet transform matrix W.
s = Wl. (4.10)
As discussed in Section 4.3.1, s is assumed sparse because of the spatial correlation
in the variations. We enforce the sparsity prior by regularizing Equation 4.9 using
the norm of s, as described in Section 2.2.2:
min ∣∣s∣∣ι + λ∣∣Al - b∣∣∣ (4.11)
or, equivalently,
min ∣∣s∣∣ι + λ∣∣AW-1s — b∣∣2, (4.12)
where λ is the regularization coefficient. Sparsity of the variations wavelet trans-
formation, s, provides a new piece of information. We call this method ^ɪ-
regularization method.
49