With the growing need for Internet of things (IoT) applications, wireless communications are facing the challenge of providing massive connectivity. However, as the number of users increases, the capacity of massive MIMO systems will be severely affected due to the huge channel estimation overhead. Therefore it is necessary to introduce non-orthogonal pilots to massive MIMO systems in order to ease the burden on the resources for channel estimation. In this paper, we propose a KKT-basediterativenon-orthogonalpilotdesignalgorithmwhich maximizes the mutual information between the received signal and the channel for MMSE channel estimation. For the typical antenna deployment of uniform linear array (ULA), we also provide scalable estimation schemes with low complexity based onchannelangularrepresentationasalternatives.Thesimulation results show that our schemes significantly save the estimation overhead at the cost of slight decrease in the MSE performance of the channel estimation, and the saving is even more prominent when the system accommodates more users.