Abstract: To meet the demands, low complexity and high accuracy, of multi-user detection in media based modulation (MBM), an approximate message passing (AMP) based multi-user detection algorithm was proposed. Because of the sparsity of MBM, a sparse signal reconstruction method based on compressed sensing was used for multi-user detection. The method was arranged that, when the observation matrix satisfied the condition of independent and identical distribution in the process of multi-user detection, the AMP algorithm could guarantee the performance of high precision detection and further reduce the detection complexity. At the same time, in the case of unknown noise variance, setting up the steps of estimating noise variance based on expectation maximization method, the proposed algorithm was more suitable for the actual scene. The simulation results show that the proposed multi-user detection algorithm based on AMP can provide a best performance compared with traditional multi-user detection methods and other multi-user detection methods with similar complexity.