The TDM mixer accepts I/Os in an interleaved fashion, passing one sample from each channel in turn and repeating this pattern for subsequent samples. Applying a mixer function directly to this TDM stream requires generating a different frequency for each sample, assuming each channel uses its own carrier. Vectorization across eight samples, using cint16 data types, means the AI Engine must generate eight unique frequency samples per cycle. However, the AI Engine’s non-linear sincos() generator belongs to the scalar processor and produces only a single sample per cycle. Vectorization then requires another approach, such as parallel lookup tables or more expensive Taylor series expansions.