Information Technology Reference
In-Depth Information
constructed from a subset of the clients (called parent nodes) [10]. The
topology assumes that all the parent nodes are fully connected, and that
each remaining node (called child node) is connected to only one parent
node. When a call is set up, the client that initiated the call collects delay
and loss information among the clients. Due to the prohibitive nature
of enumerating all topologies, we use a greedy algorithm to find a good
topology. The heuristic first determines the client pair with ME2ED (called
bottleneck pair) in a fully connected topology. It then finds a single-parent
topology that minimizes ME2ED. If the improvement in ME2ED is small
(say less than 50 ms), then it uses the best single-parent topology as the
overlay network; otherwise, it adds a second parent node. It iteratively
increases the number of parents until either the difference between the
ME2EDs of the current topology and the fully connected topology is small
or the bottleneck pair in the current topology is directly connected. The
process can be repeated whenever there is a significant change in the net-
work conditions.
We are in the process of developing admissions-control algorithms for
deciding whether a new client can be added without adversely affecting the
conversational quality of all the listeners, and the dynamic dissemination
of network information for operating the control algorithms. We plan to ini-
tially use UDP as the transport protocol but will consider RTP and other
protocols at a later stage.
2.6 Conclusions
In this chapter, we have presented some solutions on the study of real-time,
two-party, and multi-party VoIP systems that can achieve high perceptual
conversational quality. Our solutions focus on the fundamental understand-
ing of conversational quality and its trade-offs among the design of speech
codecs and strategies for network control, POS, and loss concealments.
The degradation in the perceptual quality of an interactive conversa-
tion over a network connection is caused by a combination of the decrease
in speech quality when packets are lost or delayed, and the asymmetry in
silence periods when the conversation switches from one speaker to another.
Its study is largely unexplored because there is no single objective metric for
assessing the quality of a VoIP conversation whose results match well with
subjective results. Indiscriminate subjective testing is not feasible because it
is prohibitively expensive to carry out many such tests under various conver-
sational and network conditions. Moreover, there is no systematic method to
generalize the subjective test results to unseen conditions. To this end, there
are three issues studied in this chapter that address the limitations of exist-
ing work and that improve the perceptual quality of VoIP systems.
