In fact,\nFor LS problems, we usually use Givens rotations or Householder\nreflection rather than Gram-Schmidt method for VLSI implementation in\nQRD. As far as I know, the flop counts of those are almost the same.\nAnd the rotations and reflection in the usual method needs more large\nhardware as CORDIC rather than the simple multiplications and\naccumulation in Gram-Schmidt method. Of course, the normalization in\nGram-Schmidt method has problems. But this can be solved with\nworkarounds such as Look-up-tables.\n\nAnd mathematical stability of these has no problem, using modified\nversion of this or reorthogonalization technique.\n\nIn spite of these, every hardware guys ever likes Systolic arrays\nbased on rotations and reflections.\n\nWhy???????Why???????Why???????