In this paper we use regularization, in the form of

the LASSO, as identification procedure in order to find the

parameters of the parallel Hammerstein model for behavioral

power amplifier modeling. It is shown that the LASSO chooses a

subset of the parameters of the parallel Hammerstein model in a

systematic way and thereby reduces the number of model

parameters while maintaining the performance. The values of the

parameters are also smaller than when the ordinary least-squares

algorithm is used for the parameter extraction since the LASSO

imposes a limit on the L1-norm of the parameters. Thus, the

problem with large, and sometimes very large, parameters that is

often encountered in behavioral power amplifier modeling is

avoided. Experimental results from measurements on a power

amplifier intended for the 3G WCDMA system is provided to

support the theory.