Performance analysis of a blind HOS separation criterion

We analyze the performance of a blind multiuser detection criterion. Blind multiuser detection belongs to the general family of blind source separation problems and therefore the same method of analysis can be applied to a wide range of problems. We use Taylor series expansion to obtain error expressions for finite observation length. The results can be used to evaluate the separation criterion errors due to the different parameters of the problem including number of sources, channel gain, data length, etc. We verify the results by Monte-Carlo simulations and show remarkable agreement between theory and experiments.