We demonstrate what we anticipated theoretically that we can overcome
the minimum phase assumption in blind deconvolution. Our process is non-linear,
but (we claim) not extremely so. To be successful it does require a
non-Gaussian distribution of impulses. Likewise, the iteration has a
few adjustable parameters which makes its use a little more difficult,
but we do not anticipate serious difficulties in practice.
One interesting phenomenon about the bidirectional deconvolution (Figure 1(e)1(f) and figure
2) is that it was
able to compress a mixed-phase wavelet to a spike but
without obtaining the correct causal and anti-causal parts.
We do not yet understand this. In addition,
it is more costly because it requires multiple iterations.