A grid-parafac joint channel estimation method is proposed to solve the interference problem of super-dense networking in millimeter-wave massive multiple input multiple output systems. Firstly, the high-dimensional received signal of the large-scale antenna is mapped to the large-scale tensor space, which is decomposed into sub-tensor received signals by Grid-PARAFAC tensor decomposition, and then the tensor-received signals are decomposed in parallel to obtain the symbols, receiving antenna and the subcarriers sub-projection matrix. Finally, the channel information in the hidden high-dimensional received signal is obtained by alternate least squares accurately. The Grid-PARAFAC tensor decomposition can mine the data hiding factor under the condition of retaining the original spatial information deeply, and process it instead of the entire data tensor, which reduces the received signal dimension while retaining the spatial structure of the high-dimensional received signal. Simulation shows the proposed algorithm reduces the high-dimensional channel interference and computational complexity in ultra-dense networking and improves the performance of system .