In this paper, we propose a virtual spatial modulation (VSM) scheme that performs index modulation on the virtual parallel channels resulting from the singular value decomposition of the multi-input-multi-output channels. The VSM scheme conveys information through both the indices of the virtual parallel channels and the M -ary modulated symbols. We derive a closed-form upper bound on the average bit error probability (ABEP), which considers the impact of imperfect channel estimation. Moreover, the asymptotic ABEP is also studied, which characterizes the error floor under imperfect channel estimation and the resulting diversity order as well as the coding gain under perfect channel estimation. Computer simulations verify the analysis and show that the VSM scheme can outperform the existing pre-coding aided spatial modulation schemes under the same spectral efficiency.