Predictive and Adaptive Bandwidth Reservation for Hand-Offs in Cellular Networks

1. Predictive and Adaptive Bandwidth Reservation for Hand-Offs in Cellular Networks • Goal: provide a probabilistic guarantee on connection hand-off drops as mobile user moves from one cell to another • Naïve solution: for no hand-off drops, reserve bandwidth in all cells a mobile/connection might pass through • Problem: bandwidth quickly consumed and new connection blocking probability increases • Proposed solution: a cell estimates aggregate bandwidth for hand-offs from adjacent cells, to be reserved and used solely for hand-offs, not new connection requests • Predictive: estimate directions and hand-off times of ongoing connections in each cell • Adaptive: dynamically adjust amount of reserved bandwidth to account for estimation inaccuracies and varying traffic/mobility conditions

2. System Model • Cellular infrastructure with a wired backbone and base stations as access points to mobiles in their cells • A hand-off fails if new cell does not have sufficient bandwidth • Solution: reserve bandwidth in each cell for possible hand-offs from its adjacent cells • Simple admission control of new connection requests: sum of current bandwidths + new bandwidth <= cell capacity – reserved handoff bandwidth • Reserved handoff bandwidth can be static, but then can not effectively handle varying conditions • Want to update it in a predictive and adaptive way before performing the admission test • Note reserved handoff bandwidth is a target, not actual reserved bandwidth • A base station needs to communicate its hand-off load to other other base stations

3. Hand-Off Estimation • A cell’s base station maintains quadruplets: Time when mobile moved from current cell Previous cell before entering current cell Next cell to which mobile moved Residence time in current cell • Give less weight to quadruplets observed long ago • For a given previous cell, compute probability of going to some next cell given residence time in current cell

4. Bandwidth Reservation • Given current time and time elapsed in current cell, estimate probability of connection handing off to some next cell within a time (estimation) window • A cell can then estimate the bandwidth required in some next cell for its hand-offs, and inform this adjacent cell • A cell computes its total bandwidth to be reserved for hand-offs from all its adjacent cells • Large (small) estimation window may lead to over-reservation (under-reservation) • Keep track of the proportion of hand-off drops to total observed hand-offs • If it exceeds target, increase estimation window • Otherwise, decrease estimation window

5. Admission Control • AC1: simple admission control done in current cell only • Problem: cell overloaded with hand-offs from adjacent cells • Solution AC2: check available bandwidths of adjacent cells as well as current cell • Cheaper solution AC3: consider some adjacent cells only; those which “appear” to be overloaded

6. Simulations • 1-dimensional system (e.g. cars on a highway) • Voice and video connections • Poisson arrivals and exponentially-distributed cell residence times • Static reservation is not effective under varying conditions (voice ratio, mobile speed, offered load) • AC3 is effective in meeting target hand-off dropping probability • Reserved bandwidth increases with offered load, video ratio and user mobility speed increase • As hand-off drops increase, estimation window increases • AC1 gives the most hand-off drops • AC2 and AC3 perform similarly and are fair (i.e. almost same new connection blocking probability in all cells) • AC3 is a better choice since it is less complex

7. Conclusions • Meet connection-level hand-off dropping requirements by predicting hand-offs and adapting the estimation interval • Robust admission control of new connections • Higher dimensional systems, more realistic mobility patterns, use of readily available path/direction information, routing/re-routing over the wired backbone, … • Hierarchical architecture?