Subspace-based robust blind channel identification algorithm[1] using sign covariance matrix has recently been proposed to mitigate the adverse effect of impulsive noise. Since batch eigen-decomposition, which is computationally very expensive, needs to be implemented to estimate subspace of the sign covariance matrix for channel vector estimation, a subspace-tracking-based channel identification scheme is naturally more preferable to reduce complexity. Unfortunately, conventional RLS-based subspace tracking algorithms[4], which has much lower computational complexity though, are widely known to be sensitive to impulse noise in nature. In order to overcome this dilemma, we herein propose a robust channel identification scheme based on robust statistics subspace tracking algorithm. This scheme is shown to be able to further improves the robustness of channel identification in impulsive ambient noise in comparison with it's sign covariance matrix based counterpart, while it's computational complexity is substantially lower than that of latter. Moreover, even if the estimation of sign covariance matrix can be carried out recursively in a symbol-by-symbol fashion and then adapt to the variance of channel coefficients, our simulation results show that the robust approach we propose in this paper has comparable tracking capability in term of tracking speed, but smaller steady state error resulting from it's higher robustness, for both sudden change and slow time-varying channel with impulsive noise.