Es. Manolakos et Hm. Stellakis, Systematic synthesis of parallel architectures for the computation of higher order cumulants, PARALLEL C, 26(5), 2000, pp. 655-676
Fine granularity parallel architectures for the efficient estimation of hig
her order statistics (HOS) are systematically derived in this paper. A unif
ied methodology for constructing locally recursive algorithms and space-tim
e linear mapping operators that lead to highly pipelined architectures cons
isting of multiple, tightly coupled array stages is discussed first. Then a
farm of processors is synthesized that consumes second and fourth order mo
ment estimates to produce the fourth order cumulants. The unified array syn
thesis methodology allows for the characterization of all valid solutions a
nd the derivation of closed-form expressions for the permissible linear sch
eduling functions thus facilitating the search for a design instance meetin
g the architect's specified objectives. Achieving minimum latency and an op
timal spacetime matching between the farm and the moments generator archite
cture (derived in [E.S. Manolakos, H.M. Stellakis, Systematic synthesis of
parallel architectures for the real-time estimation of higher order statist
ical moments, Parallel Algorithms and Applications (to appear)]) were the t
wo main specifications driving the synthesis. A linear array solution, that
is simpler to interface with the moments generator at the expense of addin
g some control complexity is also derived. As a result, a two-stage integra
ted VLSI architecture, that may accept data samples from the host and compu
te in real-time all non-redundant moment and cumulant terms, up to the four
th order is now possible. (C) 2000 Elsevier Science B.V. All rights reserve
d.