Back to index
Architecture and Methodology for Mobile-Executed Handoff in Cellular
Anthony Acampora and Mahmoud Naghshineh
One-line summary: A virtual connection tree, overlaid on
the graph of connected ATM switches ("basestations"), allows low-latency
handoff without involving the Network Call Controller in every handoff and
requires minimal extra hardware at the root node of the tree. Two QoS metrics,
basestation overloading probability and expected duration of overload, are
computed using binomial-distribution basestation assignment as well as a
dynamic model that models call connect/disconnect and handoffs.
- Idea: set up static rooted tree of ATM switches. Each path from root
to a leaf is assigned a virtual conn. number (VCN); root is connected to
- At call setup, call is assigned a pair of VCN's (one in each direction)
corresponding to the leaf (BS) at which the call was attached.
- Standard ATM switch routing (lookup table) translates between VCN's
and port numbers in the switch.
- Root determines (from VCN) which BS is serving the mobile, so it can
assign the appropriate VCN to route returning packets back down the tree.
(So it's really a pair of overlapping half-duplex trees.)
- To handoff: mobile moves to a new BS and starts transmitting packets
w/the new VCN of that path. Root recognizes this as a handoff event, and
now assigns appropriate new VCN to returning packets.
- Special hardware (VCN monitor/translator) required at root to do this,
which has a "last-known-location" register and location comparator
to detect handoffs.
- To do mobile-to-mobile calls: roots of connection trees can be connected
together. Network call processor must become involved for this one. Expect
this to be rare since trees cover large geographic area.
- Base station overloading: it can happen. Statistical multiplexing suggests
that this should be rare. Authors derive the probability of overloading
and expected overload period as a function of number of users in total tree
and number of users each BS can handle, using binomial distribution to assign
users to BS's. (Assumption: cellular link is bottleneck; wired infrastructure
has bandwidth to spare.)
- Efficiency (utilization of switch resources): for large M (no. of calls
supported by a single BS), traffic tends toward being evenly balanced, as
you would expect. For a given overload probability, you trade off utilization
efficiency vs. size of the tree (area covered).
- Authors also (in appendix) derive a dynamic model that accounts for
call arrivals, completions, and handoffs. Under heavy traffic, its result
converges to that of the simple model.
- Authors show how their scheme can be implemented in ATM cells using
the modified-polling MAC scheme (each BS periodically polls its mobiles
and passes a token among them for transmission; mobile requests to join
BS's polling sequence during a specific segment of the frame). Authors
show how the info required by their system fits into the signalling field
of the frame.
- Future work: nonuniform prob. of assignment of calls to BS's; choosing
handoff cell site based on chan quality; call dropping in case of severe
In the common case, handoff doesn't have to be complicated and can be based
mostly on existing (fast) ATM switch routing algorithm.
- What cellular wireless systems currently use ATM?
- Trees are a non-optimal routing topology, though authors do derive the
- Tries to address everything from mathematical characterization of routing
tree behavior to frame format for the implementation; maybe should have
been separate papers.
Back to index