I haven't look at Stephane Redon's work so I cant comment on that.
But I have worked on the Impulse scheme by Brian Mirtich. If you haven't
seen it already here's the link
It describes a scheme for estimating the Time of Impact (TOI) by using
the following information:
-objects' dynamic information (linear and angular velocity)
-shortest distance between the 2 objects (using CD algoritm called Lin-Canny)
-bounding radius of the 2 objects
In practice, when use in games, this scheme suffer a few problems.
In particular, when 2 objects are in near static contact, you get a series of
very small time-steps, causing your simulator to slow down alot. And precision
problem could also cause your objects to have penetrated when at the TOI.
So I would recommend that when using Impulse, have a minimum time-step
plus a back-stepping scheme as well.