Based on the minimum mean squared error (MMSE) trace criterion, selection of the precoding matrix indicator (PMI) from the downlink codebook of the LTE system is considered in this paper. Due to the codebook of 16 member precoding matrices and up to 1200 sub-carriers in this MIMO-OFDM system, the PMI selection needs to compute matrix inversion of up to 19,200 matrices. To satisfy this required workload, LDL H decomposition is applied at the algorithmic level, and, pipeline matrix multiplication and backward substitution modules are used at the architectural level. The VLSI implementation results of our architecture under the TSMC 90 nm CMOS technology reveals that our architecture requires 124.6K gates at operating frequency 120 MHz. Thus, our designed architecture can finish the 19,200 matrix inversions in about 1.60 ms and meets the requirement of 2 ms period of periodic channel state report.