Small transforms with $N < 70$ can often be implemented quite efficiently in AI Engines using a “vector x matrix””” approach. Instead of using the “divide-and-conquer” approach of FFT, compute the transform explicitly using the DFT. This section provides such an example.