The document proposes a new error recovery method for signal detection in massive MIMO systems that improves upon conventional sparse error recovery. The proposed method estimates the error vector using relaxed MAP estimation, which takes advantage of both the sparsity and discreteness of the error vector. Simulation results show the proposed method using SOAV optimization outperforms the conventional method based on compressed sensing, achieving a lower bit error rate. The proposed method is also able to make use of soft decision values from tentative estimates rather than just hard decisions.