The CM criterion [3,4] is usually implemented as CMA 2-2
where p,q=2. The steepest gradient descent algorithm
is obtained by taking the instantaneous gradient of ,
which results
in the CMA 2-2 equalizer update equation:
where is the length-Nf equalizer coefficient vector,
is the length-Nf receiver input vector (i.e. regressor),
is a (small) step-size, and
is the CM error function for CMA 2-2, where is chosen by