Abstract—The celebrated Active Set method proposed by
Goldfarb for solving convex quadratic problems (QP) was implemented
on-chip in fixed-point arithmetic. This implementation
was tested on random QPs and using the standard benchmark
system of oscillating masses for which I designed a feedback
model predictive controller. The solution times were compared
with the ones of the Dual Gradient Projection (DGP) algorithm,
also implemented in fixed-point. The solution times of the Active
Set algorithm were found to be lower, although no theoretical bounds on convergence are known.