Reviewer 3 of CDC12 submission 903
Comments to the author
======================
The authors develop an intricate and solid analysis of
average consensus theory in non homogeneous unreliable,
directed graphs. The proofs for the successful operation of
the algorithm are clever and original . However, I have the
nagging feeling that the approach adopted is somewhat brute
force, and perhaps, a more suitable framework could yield
simpler proofs. The authors should define more clearly
terms such as broadcast mass, mass received, etc., early on
in the paper.
Reviewer 4 of CDC12 submission 903
Comments to the author
======================
This work addresses the problem of achieving
average-consensus for a multi-agent distributed system in
case of communication links that are heterogeneous and can
drop packets with generally unequal probabilities. The
underlying communication graph is a digraph. The same
algorithm has been proposed in [8] under the assumption
that the probability of a packet drop is the same for every
link but here is analyzed in the more realistic scenario
of unequal probability of a packet drop for each link. The
same probabilistic model (Bernoulli) has been used (as in
[8]) but here the probability q_{ij} differs for each link
(i,j).
The paper is quite dense theoretically but does not allow
the reader to understand the implications of having unequal
probabilities of a packet drop for every link. By
comparing the analysis with [8] it is not clear to me what
the significant adjustment of the proof methods (claimed in
the Introduction), as compared to [8], is, e.g. first and
second moment analysis seem to be the same if you replace
the scalar q with the diagonal matrices of this paper. The
reviewer believes that the authors should dilute their
analysis by pinpointing to the differences from [8] and
mainly focusing on the consequences of heterogeneous packet
dropping links (e.g. convergence rate).