The Constant Modulus Algorithm is a common equalization method which adapts at every sample in an attempt to give a filter which produces a fixed power output. Divergence/convergence of the CMA algorithm is still a problem.
Has this ever been done in block form? For example the Least Squares equalizer uses a training sequence and computes an equalizer in block form. Is there a reason this cannot be done blind in the case of the Constant Modulus Algorithm(excluding computational requirements)? Even if a closed form expression cannot exist, you could still attempt block equalization using runs of gradient descent from a number of random starting locations or do a grid search.