I wonder If it is possile
to impose a monotone condition on the overall function
(in order not to result in oscillations, due to
scattering data points).
Offhand, I'm not aware of how to directly impose a monotonicity
condition using B-splines. Of course, you could always choose a basis
with fewer degrees of freedom than your number of sample points and
then perform a least squares fit. You may still get some oscillations
for oscillatory data, but they should be much slower ones. Tune as
appropriate for your application.
- Rhys