Contents
The main purpose of PUMA is to solve the problem
for all in the passband
where are positive polynomials of degree at most 2N and (admissible polynomials) is the set of polynomials of degree at most 2N such that the minimum phase function is admissible with
Therefore, given a load A with transmission zeros and reflection coefficient at the second port, in order to check if a polynomial is admissible, we proceed as follows
- Compute the stable function as the minimum phase factor of
- Check for the existence of a schur function such that . This is done by testing the positive definiteness of the matrix
with .
Thus we can write the previous problem as
s.t for all in the passband
This is a min-max problem that minimises the maximum of a linear criterium over the convex cone of positive polynomials P. However the matrix depends not linearly on P what makes the problem not a standard one.
Solving a min-max problem
In order to deal with the min-max problem we introduce the variable . This allows us to restate the problem as
such that:
for all real
for all in the passband
Positive polynomials
A polynomial P of degree 2N is non-negative on the real line if and only if there exist a positive semidefinite matrix such that
where is the corresponding basis vector of degree N. For instance, the monomials of degree 0 to N:
This matrix is called the Gram matrix.
From the Gram matrix it is possible to obtain the coefficients of the corresponding polynomial P with the linear operation
where Z is a matrix of size and is the column vector with the coefficients in the lower triangle of .
Positive polynomials on an interval
The polynomials of degree 2N that are positive on an interval are parametrised as
where and are polynomials positive on the real axis of degree 2N and 2N-2 respectively.
Therefore, using the previous theorem, polynomials F,G are positive if and only if there exist positive definite matrices of size and of size such that
As in the previous case, the coefficients of the polynomial Q depend linearly on the matrices
with a matrix of size
Semi-definite program
Using the previous parametrisation of positive polynomials we can use the positive definite Gram matrices to ensure that is a positive polynomial and also that is positive in the passband.
Thus we re-parametrice the problem in function of the Gram matrices
Note that by doing that it is necessary to impose that the matrix and the matrices represent the same polynomial P. We do that by imposing the linear equalities or equivalently . Using the above relations we obtain the linear system
Finally the problem remains
subject to , , , and
This is a standard non linear semi-definite program under linear equality constrains that is solved by means of the matlab toolbox NonSDP.