Lattice filter implementation with ETSI math operations on the TMS320C62xx

The specific lattice filter implementations considered here are the short-term analysis and synthesis filter realizations in the GSM Full Rate (06.10) specifications. With the ETSI math operations like multiply with round (mult-r) to be performed, and with TMS320C62xx's instruction set not consisting of any mult-r or even round operation, the C-compiler generated code fails to achieve the best performance. With techniques like unrolling the loop, intelligent scheduling of the math operations in the pipeline etc., and hand coding in assembly, we show how to achieve a performance better than that of a processor with operations like mult-r in its instruction set.