On matrix inversion for LTE MIMO applications using Texas Instruments floating point DSP

Multiple-Input Multiple-Output (MIMO) is one of the key technologies for the current and future broadband wireless services. Matrix inversion is the most costly computational module within the Minimum Mean-Square Error (MMSE) based MIMO receiver. For LTE Release 8, the order of the matrix to be inverted for an MMSE receiver is 2 × 2 for 2-stream MIMO. Going forward to LTE release 10, 4-stream MIMO can potentially double the throughput, but the matrix to be inverted will grow to 4 × 4, which is computationally intensive. In this paper, we will exam several software implementations of 4 × 4 matrix inversion. We will demonstrate how we can significantly reduce the cost (both in terms of cycle counts and development time) while maintaining enough output precision to meet performance requirements by using the floating-point feature of Texas Instruments' (TI) new multicore System-on-a-Chip (SoC) architecture.