Introduction Wireless Ad-Hoc Network Set of transceivers communicating by radio
Introduction Wireless Ad-Hoc Network Each transceiver has a transmission power which results in a transmission range
Introduction Wireless Ad-Hoc Network Transceiver receives transmission from only if
Introduction Wireless Ad-Hoc Network As a result a directed communication graph is induced
Model & Problems Definition A set of transceivers
Model & Problems Definition A set of transceivers is the power assignment
Model & Problems Definition A set of transceivers is the power assignment
Model & Problems Definitions A set of transceivers is the power assignment is the communication graph
Model & Problems Definitions A set of transceivers is the power assignment is the communication graph is the cost of the assignment
Outline Connectivity problems Bounded hop broadcast Spanners Interference-free broadcast
Connectivity Definitions A graph is k-vertex-connected if for any two nodes there exist k-vertex-disjoint paths connecting to 2-vertex-connected
Connectivity Definitions For graph , a subset is a connected backbone if restricted to is strongly connected and for each there exists so that Connected backbone
Connectivity Problem 1 (k-vertex-connectivity) Input: A set of transceivers, and a parameter Output: A power assignment with minimal possible cost , where is k-vertex connected
Connectivity Problem 1 (k-vertex-connectivity) Input: A set of transceivers, and a parameter Output: A power assignment with minimal possible cost , where is k-vertex connected -approximation algorithm
Connectivity Problem 2 (connected backbone) Input: A set of transceivers Output: A subset of and a power assignment with minimal possible cost , where (restricted to ) is strongly connected, and for each , there exists , such that
Connectivity Problem 2 (connected backbone) Input: A set of transceivers Output: A subset of and a power assignment with minimal possible cost , where (restricted to ) is strongly connected, and for each , there exists , such that Constant-factor approximation algorithm in time
Fault-Tolerant Power Assignment Definitions For each , let be a set of closest nodes to
Fault-Tolerant Power Assignment Definitions For each , let be a set of closest nodes to
Fault-Tolerant Power Assignment Definitions For each , let be a set of closest nodes to Let
Fault-Tolerant Power Assignment The algorithm Assign each the range (denote ) Compute an of
Fault-Tolerant Power Assignment The algorithm Assign each the range (denote ) Compute an of
Fault-Tolerant Power Assignment The algorithm For each edge of increase the range of the nodes in such that each node can reach all nodes in , and vice versa (denote )
Fault-Tolerant Power Assignment The algorithm For each edge of increase the range of the nodes in such that each node can reach all nodes in , and vice versa (denote )
Fault-Tolerant Power Assignment Proof sketch Let In each is assigned at most Case 1:
Fault-Tolerant Power Assignment Proof sketch Let In each is assigned at most Case 1:
Fault-Tolerant Power Assignment Proof sketch Let In each is assigned at most Case 2:
Fault-Tolerant Power Assignment Proof sketch Let In each is assigned at most Case 2:
Fault-Tolerant Power Assignment Proof sketch Let In each is assigned at most Easy to see
Fault-Tolerant Power Assignment Proof sketch Let In each is assigned at most Easy to see Kirousis et al. proved
Fault-Tolerant Power Assignment Proof sketch Let In each is assigned at most Easy to see Kirousis et al. proved As a result and since degree of MST is constant
Connected Backbone Power Assignment Definitions Given the of , for any node , let be the size of the longest edge adjacent to
Connected Backbone Power Assignment Definitions Given the of , for any node , let be the size of the longest edge adjacent to
Connected Backbone Power Assignment The algorithm Compute an of
Connected Backbone Power Assignment The algorithm Compute an of
Connected Backbone Power Assignment The algorithm Compute an of Let be the set of all internal nodes of
Connected Backbone Power Assignment The algorithm Compute an of Let be the set of all internal nodes of Assign each with (denote )
Connected Backbone Power Assignment The algorithm Compute an of Let be the set of all internal nodes of Assign each with (denote )
Connected Backbone Power Assignment The algorithm Compute an of Let be the set of all internal nodes of Assign each with (denote )
Connected Backbone Power Assignment Proof sketch Construct a power assignment for which it holds and , as a result obtaining is derived from
Connected Backbone Power Assignment Proof sketch Let be the connected backbone in For each node let be the transmission range of in
Connected Backbone Power Assignment Proof sketch For each node let be all the nodes within distance from
Connected Backbone Power Assignment Proof sketch For each node let be all the nodes within distance from
Connected Backbone Power Assignment Proof sketch For each node let be all the nodes within distance from For each node compute of
Connected Backbone Power Assignment Proof sketch For each node let be all the nodes within distance from For each node compute of
Connected Backbone Power Assignment Proof sketch In : Each node is assigned
Connected Backbone Power Assignment Proof sketch In : Each node is assigned
Connected Backbone Power Assignment Proof sketch In : Each node is assigned Each node is assigned
Connected Backbone Power Assignment Proof sketch Carmi et al. showed that
Connected Backbone Power Assignment Proof sketch Carmi et al. showed that
Connected Backbone Power Assignment Proof sketch Carmi et al. showed that
Connected Backbone Power Assignment Proof sketch Carmi et al. showed that + + +
Connected Backbone Power Assignment Proof sketch Carmi et al. showed that Using this and is at least longest edge in we obtain
Connected Backbone Power Assignment is at least longest edge in and Thus (summing over all v),
Connected Backbone Power Assignment Proof sketch Kirousis et al. proved that given an assigning each node with yields a 2-factor approximation for strong-connectivity (denote )
Connected Backbone Power Assignment Proof sketch Kirousis et al. proved that given an assigning each node with yields a 2-factor approximation for strong-connectivity (denote ) Using this fact and that B gives us strong connectivity, we obtain
Connected Backbone Power Assignment Proof sketch Therefore,
Broadcast A graph is a broadcast graph rooted at if there is a path from to any
Broadcast A graph is a broadcast graph rooted at if there is a path from to any
Broadcast A graph is a h-bounded-hop broadcast graph rooted at if there is a path from to any and the number of hops is limited by 4-bounded-hop broadcast
Broadcast A graph is a k-h-broadcast graph if it remains h-bounded-hop broadcast graph even with the removal of up to nodes 2-4-bounded-hop broadcast
Broadcast A graph is a k-h-broadcast graph if it remains h-bounded-hop broadcast graph even with the removal of up to nodes 2-vertex disjoint paths under 4 hops
Broadcast A graph is a k-h-broadcast graph if it remains h-bounded-hop broadcast graph even with the removal of up to nodes 2-vertex disjoint paths under 4 hops
Problem 3 (k-h-bounded broadcast) Input: A set of transceivers in , root node and parameters Output: A power assignment so that is k-h-broadcast and is minimized
Planar Case The Algorithm Take a power assignment so that is 1-h-bounded hop graph
Planar Case The Algorithm Take a power assignment so that is 1-h-bounded hop graph Let be a directed spanning tree of Max distance – h hops
Planar Case The Algorithm Take a power assignment so that is 1-h-bounded hop graph Let be a directed spanning tree of Max distance – h hops
Planar Case The Algorithm Add edges from to its grandchildren
Planar Case The Algorithm Add edges from to its grandchildren Remove edges from the children of
Planar Case The Algorithm Add edges from to its grandchildren Remove edges from the children of Denote the resulting tree Max distance – h-1 hops
Planar Case The Algorithm No power is assigned yet! We have a skeleton with a bounded cost
Planar Case The Algorithm Assign
Planar Case The Algorithm Assign to reach k closest neighbors.
Planar Case The Algorithm For each directed edge in increase the range of all nodes in to reach all nodes in
Planar Case The Algorithm For each directed edge in increase the range of all nodes in to reach all nodes in
Planar Case The Algorithm For each directed edge in increase the range of all nodes in to reach all nodes in
Planar Case The Algorithm For each directed edge in increase the range of all nodes in to reach all nodes in
Planar Case The Algorithm For each directed edge in increase the range of all nodes in to reach all nodes in
Planar Case The Algorithm For each directed edge in increase the range of all nodes in to reach all nodes in
Planar Case The Algorithm For each directed edge in increase the range of all nodes in to reach all nodes in
Planar Case The Algorithm For each directed edge in increase the range of all nodes in to reach all nodes in
Planar Case The Algorithm For each directed edge in increase the range of all nodes in to reach all nodes in
Planar Case The Algorithm For each directed edge in increase the range of all nodes in to reach all nodes in
Planar Case The Algorithm Denote the resulting power assignment
Planar Case The Algorithm Denote the resulting power assignment Along each path in there are vertex-disjoint paths in of at most hops
Analysis For a single edge in the power increase of is bounded by:
Analysis For a single edge in the power increase of is bounded by:
Analysis For a single edge in the power increase of is bounded by: Power assignment in
Planar Case Analysis For a single edge in the power increase of is bounded by: Node can be in many -s
Planar Case Analysis For a single edge in the power increase of is bounded by: Node can be in many -s, with many edges
Planar Case Analysis For a single edge in the power increase of is bounded by: Node can be in many -s, with many edges But eventually only one ‘dominates’ the bound
Planar Case Analysis A node can be dominated only by the outgoing edges of in
Planar Case Analysis A node can be dominated only by the outgoing edges of in A single edge can dominate at most nodes (those in )
Analysis A node can be dominated only by the outgoing edges of in A single edge can dominate at most nodes (those in ) Recall,
Analysis A node can be dominated only by the outgoing edges of in A single edge can dominate at most nodes (those in ) As a result,
Analysis
Analysis Due to
Analysis PTAS due to Funke and Laue [24]
Analysis Let be the optimal power assignment for the k-h-broadcast problem From ,
Analysis Let be the optimal power assignment for the k-h-broadcast problem From , We need to bound
Analysis Let be a power assignment so that each node has at least neighbors Clearly,
Analysis - Hamiltonian cycle based power assignment for the k-(n-1)-broadcast problem, so that
Analysis - Hamiltonian cycle based power assignment for the k-(n-1)-broadcast problem, so that In each node has at least neighbors
Analysis – Hamiltonian cycle based power assignment for the k-(n-1)-broadcast problem, so that In each node has at least neighbors From ,
k-(n-1)-broadcast The Algorithm
k-(n-1)-broadcast The Algorithm Compute an MST of
k-(n-1)-broadcast The Algorithm Compute an MST of Construct a Hamiltonian cycle with cost
k-(n-1)-broadcast The Algorithm Compute an MST of Construct a Hamiltonian cycle with cost Assign each node to reach nodes in both directions of the cycle Example: k=4
k-(n-1)-broadcast The Algorithm Compute an MST of Construct a Hamiltonian cycle with cost Assign each node to reach nodes in both directions of the cycle As a result,
k-(n-1)-broadcast Hamiltonian Cycle Stage Compute an MST of
k-(n-1)-broadcast Hamiltonian Cycle Stage Compute an MST of Apply MST-Augmentation (Calinescu and Wan)
k-(n-1)-broadcast Hamiltonian Cycle Stage Compute an MST of Apply MST-Augmentation (Calinescu and Wan)
k-(n-1)-broadcast Hamiltonian Cycle Stage Compute an MST of Apply MST-Augmentation (Calinescu and Wan)
k-(n-1)-broadcast Hamiltonian Cycle Stage Compute an MST of Apply MST-Augmentation (Calinescu and Wan)
k-(n-1)-broadcast Hamiltonian Cycle Stage Compute an MST of Apply MST-Augmentation (Calinescu and Wan)
k-(n-1)-broadcast Hamiltonian Cycle Stage Compute an MST of Apply MST-Augmentation (Calinescu and Wan)
k-(n-1)-broadcast Hamiltonian Cycle Stage Compute an MST of Apply MST-Augmentation (Calinescu and Wan) 2-strongly connected undirected graph
k-(n-1)-broadcast Hamiltonian Cycle Stage Compute an MST of Apply MST-Augmentation (Calinescu and Wan) Apply TSP-Approx (Bender and Checkuri) Square of every biconnected graph is Hamiltonian (Fleischner)
k-(n-1)-broadcast Hamiltonian Cycle Stage Compute an MST of Apply MST-Augmentation (Calinescu and Wan) Apply TSP-Approx (Bender and Checkuri) As a result, The cost of the Hamiltonian cycle
Back to k-h-broadcast Analysis - A simple approximation due to: For any it holds:
Back to k-h-broadcast Analysis - Take as before
Back to k-h-broadcast Analysis - Take as before The most distant node at most hops away
Back to k-h-broadcast Analysis - Take as before The most distant node at most hops away Assign the root to reach all!
Spanners What is a spanner? A spanning subgraph that approximates some measure of the original graph
Spanners What is a spanner? A spanning subgraph that approximates some measure of the original graph E.g., Euclidean distance
Spanners What is a spanner? A spanning subgraph that approximates some measure of the original graph E.g., Euclidean distance
Spanners What is a spanner? A spanning subgraph that approximates some measure of the original graph E.g., Euclidean distance Shortest path is at most times longer than in
Spanners What is a spanner? A spanning subgraph that approximates some measure of the original graph E.g., Euclidean distance Shortest path is at most times longer than in stretch factor
Spanners We propose two spanner optimization measures Distance – reducing transmission latency Energy – increasing network lifetime
Spanner optimization measures The original graph Let be the wireless nodes in the plane Let be a weighted complete graph Weight function: The Euclidean distance
Spanner optimization measures The original graph Let be the wireless nodes in the plane Let be a weighted complete graph Weight function: The Euclidean distance Proportional to the energy required to transmit from to
Spanner optimization measures The original graph Let be the wireless nodes in the plane Let be a weighted complete graph Weight function:
Spanner optimization measures The spanner Let p be a power assignment
Spanner optimization measures The spanner Let p be a power assignment is an induced directed graph, where
Spanner optimization measures The spanner Let p be a power assignment is an induced directed graph, where The cost:
Spanner optimization measures Energy measure (stretch factor) The energy of some path is its weight
Spanner optimization measures Energy measure (stretch factor) The energy of some path is its weight The minimum energy from to in
Spanner optimization measures Energy measure (stretch factor) The energy of some path is its weight The minimum energy from to in The minimum energy from to in
Spanner optimization measures Energy measure (stretch factor) The energy stretch factor of
Spanner optimization measures Energy measure (stretch factor) The energy stretch factor of We aim to minimize both and
Spanner optimization measures Energy measure (stretch factor) The energy stretch factor of We aim to minimize both and Clear benefits Prolonged network lifetime Low cost Low interference…
Spanner optimization measures Distance measure (stretch factor) The distance of some path
Spanner optimization measures Distance measure (stretch factor) The distance of some path The minimum distance from to in
Spanner optimization measures Distance measure (stretch factor) The distance stretch factor of
Spanner optimization measures Distance measure (stretch factor) The distance stretch factor of We aim to minimize both and
Spanner optimization measures Distance measure (stretch factor) The distance stretch factor of We aim to minimize both and Clear benefits Low delay in message delivery Low cost
Main results Preliminaries We consider a random, independent, and uniform node distribution in a unit square The probability of our results converges to 1 as the number of nodes, n, increases
Main results Preliminaries We consider a random, independent, and uniform node distribution in a unit square Spanners make sense only if the induced graph is strongly connected
Main results Preliminaries We consider a random, independent, and uniform node distribution in a unit square Spanners make sense only if the induced graph is strongly connected Otherwise, the stretch factor is infinity Path does not exist
Main results Preliminaries We consider a random, independent, and uniform node distribution in a unit square Spanners make sense only if the induced graph is strongly connected The cost of any spanner is at least the minimum cost of strong connectivity
Main results Preliminaries We consider a random, independent, and uniform node distribution in a unit square Spanners make sense only if the induced graph is strongly connected The cost of any spanner is at least the minimum cost of strong connectivity (denote this cost )
Main results Energy spanner Develop power assignment so that where , ,
Main results Distance spanner Develop a power assignment so that
Lower bound on the cost of any spanner Technical details Some bounds… Using [Zhang and Hou ‘05] Lower bound on the cost of any spanner
Minimum spanning tree of G Technical details Some bounds… Using [Zhang and Hou ‘05] From [Kirousis et al. ‘00] Minimum spanning tree of G The weight of the tree
Technical details Some bounds… Using [Zhang and Hou ‘05] From [Kirousis et al. ‘00] Using [Berend et al. ‘08] & [Penrose ‘97] Maximum length edge of MST
Energy spanner [power assignment] Technical details Energy spanner [power assignment]
Energy spanner [power assignment] Technical details Energy spanner [power assignment] Find the minimum spanning tree (MST)
Energy spanner [power assignment] Technical details Energy spanner [power assignment] Find the minimum spanning tree (MST) Lemma: We can find nodes so that any node is within hops from some node in U
Energy spanner [power assignment] Technical details Energy spanner [power assignment] Find the minimum spanning tree (MST) Lemma: We can find nodes so that any node is within hops from some node in U Take diameter
Energy spanner [power assignment] Technical details Energy spanner [power assignment] Find the minimum spanning tree (MST) Lemma: We can find nodes so that any node is within hops from some node in U Take diameter Add the -th node to U
Energy spanner [power assignment] Technical details Energy spanner [power assignment] Find the minimum spanning tree (MST) Lemma: We can find nodes so that any node is within hops from some node in U Take diameter Add the -th node to U Remove first nodes from the diameter
Energy spanner [power assignment] Technical details Energy spanner [power assignment] Find the minimum spanning tree (MST) Lemma: We can find nodes so that any node is within hops from some node in U Take diameter Add the -th node to U Remove first nodes from the diameter
Energy spanner [power assignment] Technical details Energy spanner [power assignment] Find the minimum spanning tree (MST) Lemma: We can find nodes so that any node is within hops from some node in U Take diameter Add the -th node to U Remove first nodes from the diameter
Energy spanner [power assignment] Technical details Energy spanner [power assignment] Find the minimum spanning tree (MST) Lemma: We can find nodes so that any node is within hops from some node in U Take diameter Add the -th node to U Remove first nodes from the diameter
Energy spanner [power assignment] Technical details Energy spanner [power assignment] Find the minimum spanning tree (MST) Lemma: We can find nodes so that any node is within hops from some node in U Take diameter Add the -th node to U Remove first nodes from the diameter
Energy spanner [power assignment] Technical details Energy spanner [power assignment] Find the minimum spanning tree (MST) Lemma: We can find nodes so that any node is within hops from some node in U
Energy spanner [power assignment] Technical details Energy spanner [power assignment] Find the minimum spanning tree (MST) Lemma: We can find nodes so that any node is within hops from some node in U Let be a LAST rooted at LAST [Khuller et al. ’93] is a spanning tree T of G, rooted at some so that and
Energy spanner [power assignment] Technical details Energy spanner [power assignment] Define the power assignment p so that
Energy spanner [power assignment] Technical details Energy spanner [power assignment] Define the power assignment p so that Let
Energy spanner [power assignment] Technical details Energy spanner [power assignment] Define the power assignment p so that Let Finally, For technical reasons
Energy spanner [cost analysis] Technical details Energy spanner [cost analysis]
Energy spanner [stretch analysis] Technical details Energy spanner [stretch analysis] If , there is a path P in G, so that and
Energy spanner [stretch analysis] Technical details Energy spanner [stretch analysis] If , there is a path P in G, so that and Therefore, since for every u, path P also exists in
Energy spanner [stretch analysis] Technical details Energy spanner [stretch analysis] Otherwise,
Energy spanner [stretch analysis] Technical details Energy spanner [stretch analysis] For any two nodes, s and t, the path in first arrives at some LAST origin by using the MST edges (denote P’)
Energy spanner [stretch analysis] Technical details Energy spanner [stretch analysis] For any two nodes, s and t, the path in first arrives at some LAST origin by using the MST edges (denote P’)
Energy spanner [stretch analysis] Technical details Energy spanner [stretch analysis] For any two nodes, s and t, the path in first arrives at some LAST origin by using the MST edges (denote P’) second travels through the edges of from to t (denote P’’)
Energy spanner [stretch analysis] Technical details Energy spanner [stretch analysis] For any two nodes, s and t, the path in first arrives at some LAST origin by using the MST edges (denote P’) second travels through the edges of from to t (denote P’’)
Energy spanner [stretch analysis] Technical details Energy spanner [stretch analysis] Otherwise, We bound the weight of P’ and P’’ Lemma Maximum edge of MST
Energy spanner [stretch analysis] Technical details Energy spanner [stretch analysis] Otherwise, We bound the weight of P’ and P’’ A possible path goes through s
Energy spanner [stretch analysis] Technical details Energy spanner [stretch analysis] Otherwise, We bound the weight of P’ and P’’ Eventually,
Distance spanner [power assignment] Technical details Distance spanner [power assignment] The general idea is that for uniformly distributed nodes, we can always find “good” relays between any pair of nodes
Distance spanner [power assignment] Technical details Distance spanner [power assignment] To find these relays, for any pair of nodes, s and t, we start a recursive process
Distance spanner [power assignment] Technical details Distance spanner [power assignment] To find these relays, for any pair of nodes, s and t, we start a recursive process At step i, we place adjacent disks along the edge The diameter of a disk at step i is
Distance spanner [power assignment] Technical details Distance spanner [power assignment] To find these relays, for any pair of nodes, s and t, we start a recursive process At step i, we place adjacent disks along the edge The diameter of a disk at step i is
Distance spanner [power assignment] Technical details Distance spanner [power assignment] To find these relays, for any pair of nodes, s and t, we start a recursive process At step i, we place adjacent disks along the edge The diameter of a disk at step i is
Distance spanner [power assignment] Technical details Distance spanner [power assignment] To find these relays, for any pair of nodes, s and t, we start a recursive process At step i, we place adjacent disks along the edge The diameter of a disk at step i is The process ends when one of the disks has no relay nodes
Distance spanner [power assignment] Technical details Distance spanner [power assignment] To find these relays, for any pair of nodes, s and t, we start a recursive process At step i, we place adjacent disks along the edge Finally, we use relay nodes to obtain a path We use an arbitrary node in each disk at the last non-empty step
Distance spanner [power assignment] Technical details Distance spanner [power assignment] The power assignment p is obtained by ensuring that all paths are in
Distance spanner [power assignment] Technical details Distance spanner [power assignment] The power assignment p is obtained by ensuring that all paths are in Let be the constructed path from s to t
Distance spanner [power assignment] Technical details Distance spanner [power assignment] The power assignment p is obtained by ensuring that all paths are in Let be the constructed path from s to t And be all the edges from u in all the paths
Distance spanner [power assignment] Technical details Distance spanner [power assignment] The power assignment p is obtained by ensuring that all paths are in Let be the constructed path from s to t And be all the edges from u in all the paths Finally,
Distance spanner [analysis] Technical details Distance spanner [analysis] Lemma: Let D be the maximum radius disk which can be placed inside the unit square, so there are no nodes in D Let r be the radius of D
Distance spanner [analysis] Technical details Distance spanner [analysis] Lemma: Let D be the maximum radius disk which can be placed inside the unit square, so there are no nodes in D Then, Let r be the radius of D
Distance spanner [analysis] Technical details Distance spanner [analysis] From Lemma,
Distance spanner [analysis] Technical details Distance spanner [analysis] From Lemma, Clearly,
Extended wireless network model Power assignment Nodes have no fixed power supply Each node has an initial battery charge b(v) The lifetime of node v is The network lifetime is
Wireless network model Power assignment Interference is a direct consequence of a power assignment p ?
Wireless network model Power assignment Interference is a direct consequence of a power assignment p Several interference models exist Number of nodes affected by transmission Number of edges affected by transmission
Wireless network model Power assignment Interference is a direct consequence of a power assignment p Several interference models exist Number of nodes affected by transmission Number of edges affected by transmission We combine several common models by defining the interference to be
Main results Contribution We develop power assignment: can be computed in time where n is the number of nodes and
Technical details The construction The power assignment is computed by dividing the unit square into k grid cells
Technical details The construction The power assignment is computed by dividing the unit square into k grid cells Then we compute a k shortest path trees rooted at an arbitrary node in each cell
Technical details The construction The power assignment is computed by dividing the unit square into k grid cells Then we compute a k shortest path trees rooted at an arbitrary node in each cell The power assignment of nodes is increased to assure all these k trees are included
Technical details The construction The power assignment is computed by dividing the unit square into k grid cells Then we compute a k shortest path trees rooted at an arbitrary node in each cell The power assignment of nodes is increased to assure all these k trees are included The power assignment of nodes is increased again to be at least