A limited feedback based learning algorithm of interference channel is proposed to tackle the issue of interference channel learning in cognitive Multi Input Multi Output(MIMO) system. The null space of interference channel instead of the whole channel matrix is fed back from the primary receiver to secondary transmitter. The time correlation of quantization result of interference channel null space is analyzed based on the derivation of null space codeword selection criteria, and the null space codebook of current frame is constructed by perturbing the quantization result of previous frame based on a rotation codebook. Furthermore，to avoid the propagation of quantization error，the update of rotation parameter is derived. Theoretical analysis and simulation results indicate that proposed algorithm tracks the null space of slow time-varying MIMO interference channel better than the Grassmannian Subspace Packing codebook.