Potter Square Root Filter in RLS for Parameter estimation

I have to know how can I implement potters square root filter in following Markov parameter estimation problem.

   yk = [u01 u11 u12 u13 u14][h0 h1 h2 h3 h4]'u denotes jth component of k times integral ou u(t)

    yk = (phi)k*(th)k
To estimate (th)k, kamlan gain RLS method I have used.

       ek = yk - (phi)k*(th)k
       (th)k = (th)k-1 + qk*ek
       qk = Pk-1*(phi)k/[1+(phi)k'*Pk-1*(phi)]
       Pk = Pk-1 - qk*(phi)k'*Pk-1
Now to avoid problem of ill-conditioning of covariance matrix, How can I apply Potter's Square Root Filter???