We present a ''dynamical'' approach to the study of the singularity of
infinitely convolved Bernoulli measures nu(beta), for beta the golden
section. We introduce nu(beta) as the transverse measure of the maxim
um entropy measure mu on the repelling set invariant for contracting m
aps of the square, the ''fat baker's'' transformation. Our approach st
rongly relies on the Markov structure of the underlying dynamical syst
em. Indeed, if beta = golden mean, the fat baker's transformation has
a very simple Markov coding. The ''ambiguity'' (of order two) of this
coding, which appears when projecting on the line, due to passages for
the central, overlapping zone, can be expressed by means of products
of matrices (of order two). This product has a Markov distribution inh
erited by the Markov structure of the map. The dimension of the projec
ted measure is therefore associated to the growth of this product; our
dimension formula appears in a natural way as a version of the Furste
nberg-Guivarch formula. Our technique provides an explicit dimension f
ormula and, most important, provides a formalism well suited for the m
ultifractal analysis of this measure, as we will show in a forthcoming
paper.