Second order blind deconvolution of single input multiple output (SIMO) FIR channels is considered herein. A major drawback of several blind diversity techniques using antenna arrays/temporal-oversampling is high sensitivity to the choice of model order. In this article, a robust method using only the second order statistics is described. It provides high estimation accuracy even for a limited sample size and an unknown model order. In contrast to other suggested approaches, that often exploit properties valid only in the large sample case, the proposed method is applicable both in the large sample and high signal-to-noise ratio (SNR) scenario; it also enjoys a simple implementation.