This paper deals with the blind identification of a two channel SIMO system driven by a white input sequence. The approach we propose operates in two steps. Firstly, the variances of the additive noises are estimated by using a frequency domain errors-in-variable scheme. Secondly, the finite impulse responses (FIR) are retrievied by taking advantage of the cross-relations expressed in the frequency domain. It should be noted that the proposed approach has the advantage of addressing the so-called unbalanced case, i.e. when the noise variances are not the same. Moreover, the algorithm enables the solution to be directly obtained from the noisy data and no optimization algorithm is required.