Let , where has **m** components and
has **n** components.

Suppose we have **k>n** measurements , and wish to
recover .

Form matrices and by adjoining and , respectively. ** NOTE:** is not square.

The matrix can be recovered as follows. Define an
error vector with **m** components:

Adjoining the **k** error vectors, we get

The sum of the squares of the errors is

Differentiate and equate to **0**:

The term is the ** pseudoinverse**
of .

Tue Nov 7 09:13:01 EET 1995