To solve the user selection and channel estimation problem in multi-user MISO system, a new data transmission frame structure combined with the decentralized user self-selecting strategy in TDD mode is designed. Then, the base station receiving uplink random pilot sequence signatured with the user identity is built as a block sparse linear model based on the natural signal sparsity from users′ low active degree and the channel impulse response sparsity in delay-spread domain. In addtion, to resolve such an objective optimization problem, an efficient block-sparse signal recovery algorithm is proposed based on l2/l1 reconstruction model. In the novel algorithm, the objective function is transformed through variable splitting and four variables are alternately updated in the framework of alternating direction method (ADM) until the prespecified convergence criterion is satisfied. During the alternate updating procedure, Aiming at unobtainable closed form solution of the signal variable item, the block coordinate descent (BCD) method is utilized to acquire an iterative solution. Simulation results demonstrate that the proposed method can achieve higher computational efficiency and the better estimation accuracy compared with two state-of-art fast algorithms, such as block orthogonal matching pursuit (Block OMP) and block compressive sampling matching pursuit (Block CoSaMP).