Multicluster, Mobile, Multimedia Radio Network

Mario Gerla and Jack Tzu-Chieh Tsai, UCLA

One-line summary: Vision paper for ARPA WAMIS, fully distributed heirarchical wireless network based on clusters, using TDMA within clusters and CDMA across them.

Overview/Main Points

Datagrams for "document" traffic, virtual circuits supported for continuous media
Passive ACK doesn�t work since SS-CDMA is used; hence need e2e as well as link-to-link flow and congestion control.
clustering provides framework for CDMA code separation, channel access, power control, etc.
TDMA used within a cluster. "Cluster head" (leader) elected by "lowest-ID" algorithm, which is more stable than "highest-connectivity".
Routing within a cluster is fully distributed; intercluster packets must go through gateways that know both CDMA codes for adjacent clusters
"Distributed gateway": a single logical gateway implemented across a pair of machines in different clusters that can hear each other. Most useful in very weak connectivity situations, unneeded complexity otherwise.
COdes can be reused in non-adjacent clusters by 4-coloring graph; since also need a code for control channel, system requires 9 orthogonal codes.
Frame is divided into "Info" and "Control" phases. During control phase, neighbors exchange a lot of data (eg power gain lists, connectivity info, code assignment, etc) using broadcast.
Size of control frame grows linearly with number of radios in network! Authors estimate 24 slots enough for a "typical" system whose properties seem to be pulled out of their butt.
Fast-reservation for virtual circuits (to handle mobility as circuits move around): shortest-path assignments using Bellman-Ford algorithm. No discussion of when to use fast reservation vs. the standard "static" scheme used in low-mobility situations.
E2E thruput is lower than link thruput by about factor of 4 (simulation); why??
Combined CDMA and TDMA yields 80 percent improvement over TDMA alone. Not surprising.
A few "blue sky" extensions to protocol proposed.

Relevance

No need for wired infrastructure; the network sets up and deploys anywhere and can handle a variety of applications.

Flaws

Vision, not implementation paper; lots of stuff glossed over or not explained.
Complexity of combined TDMA and CDMA seems formidable, yet they don�t address. How much of this have they tried to build?
Hands waved violently regarding how network can be synchronized for the TDMA part.
Scalability is claimed as a goal, but the control-phase length seems to preclude this.
Back to index