The ultra-dense radio access network with distributed MU-MIMO is considered as a promising approach to improve the coverage and the link capacity in 5G advanced systems. However, large-scale MU-MIMO requires prohibitively high computational complexity. Our previous work has proved that grouping neighborhood users/antennas into a number of clusters and performing cluster-wise MU-MIMO in parallel can alleviate the computational complexity problem, but the link capacity improvement is limited by the severe interference. In this paper, we propose a 2-step graph coloring algorithm that can eliminate both the inter-cell interferences and the inter-cluster interferences. The first step is to apply the graph coloring algorithm on the cell edge in order to reduce the inter-cell interferences. Once the color of the cell edge has been decided, the second step is to utilize the conditional graph coloring to the clusters within each BS cell. As a preliminary research, we focus on the second step and propose a Restricted Color Number (RCN) algorithm to mitigate the inter-cluster interferences. The computer simulation results show that our RCN algorithm can improve the link capacity compared with no coloring case.