CPSC 689: Discrete Algorithms for Mobile and Wireless Systems Spring 2009 Prof. Jennifer Welch
Discrete Algs for Mobile Wireless Sys2 Lecture 20 Topic: Part III: Building and Maintaining Network Structures Topology Control Sources: Li et al. Bahramgiri et al. MIT Fall 2008 slides
Discrete Algs for Mobile Wireless Sys3 Topology Control Suppose you have a set of nodes V, where each node v has a Euclidean position and a maximum transmission power. If all nodes transmit using maximum power, they induce a set of edges E representing the communication links Starting with connected graph G = (V,E), reduce number of neighbors of each node by reducing transmission power at that node to produce a subset E' of E with G' = (V,E') still connected
Discrete Algs for Mobile Wireless Sys4 How to Optimize Power? Focus on reducing transmission power of each node minimize the maximum transmission power per node
Discrete Algs for Mobile Wireless Sys5 Cone-Based Topology Control [Li et al.] Problem studied is how to reduce power while preserving connectivity Assumes interference is implicitly handled by reducing node degree and edge set Claims not to require position of each node and not to be coupled with any radio propagation model Assumes a MAC layer that provides a reliable broadcast operation; synchronous rounds Mainly theoretical with some simulations to validate theoretical results
Discrete Algs for Mobile Wireless Sys6 Central Idea of Algorithm Parameter to algorithm Each node u starts broadcasting and collecting Acks with minimum power Node u gradually increases its transmission power stops when reaching maximum power or every cone of degree contains at least one node (neighbor)
Discrete Algs for Mobile Wireless Sys7 Sample Execution >
Discrete Algs for Mobile Wireless Sys8 CBTC( ), Code for Node u N u := Ø // neighbors discovered so far D u := Ø // directions from which Acks have arrived p u := p min while (p u < p max and there exists gap( ,D u )) p u := increase(p u ) bcast(u,p u,("Hello",p u )) wait for Acks N u := N u U {v : v just sent an Ack} D u := D u U {dir u (v) : v just sent an Ack}
Discrete Algs for Mobile Wireless Sys9 Some Definitions Let N (u) be final set of neighbors discovered by node u Let E – = {(u,v) : v is in N (u) and u is in N (v)} Let E + = {(u,v) : v is in N (u) or u is in N (v)}
Discrete Algs for Mobile Wireless Sys10 Optimizations Shrink-back: for nodes which terminate with p = p max, reduce power while keeping coverage. Asymmetric edge removal: transform E + into E –, as long as E – preserves connectivity Pairwise edge removal: remove implicit 3- cycles; use node ids to get unique edge weights and remove longest edge
Discrete Algs for Mobile Wireless Sys11 Proof of Correctness INFOCOMM paper proves that when = 2 /3, then E + is connected. PODC paper proves that when = 2 /3, then E – is connected and when = 5 /6, then E + is connected Proofs are geometrical in nature, subtle and somewhat long From MIT course notes, prove something stronger: when = 2 /3, then E – is not only connected, but contains the minimum spanning tree
Discrete Algs for Mobile Wireless Sys12 About Minimum Spanning Trees Given a set of points in the plane and all straight- line edges between them, want to find MST of the resulting graph (cost of an edge is its length). Yao's Lemma: If edge (u,v) is in MST, then v is u's closest neighbor in every /3 cone centered at u containing v.
Discrete Algs for Mobile Wireless Sys13 Using Yao's Lemma Suppose in contradiction the CBCT output graph G' is missing some edge (u,v) of the MST M of the maximum power graph G Consider the cone of angle 2 /3 centered at u with v on its bisector; each half of the cone has angle /3 By construction, CBCT guarantees that there is some neighbor w of u inside this cone in G'. Since v is not a neighbor of u, but w is, it must be that v is farther away from u than w is. Contradicts Yao's Lemma! u v w
Discrete Algs for Mobile Wireless Sys14 Extension to k-Connectivity [Bahramgiri] A connected graph is k-connected if removing any set of less than k nodes results in a graph that is still connected. How can we find minimum power needed by each node to ensure that resulting graph is k-connected? Use the cone algorithm CBTC with = 2 /3k.
Discrete Algs for Mobile Wireless Sys15 Proof of Correctness Theorem: If G is k-connected, then G , result of running CBTC( /k) with = 2 /3 on G, is also k-connected. Proof: Suppose in contradiction G /k is not k-connected: removing nodes v 1, …, v k-1 disconnects G /k.
Discrete Algs for Mobile Wireless Sys16 Proof of Correctness G remove v 1,…,v k-1 G' connected remove v 1,…,v k-1 G 1 disconnected G /k CBTC( /k) CBTC( ) G' connected by previous result = 2 /3 show G' is a subgraph of G 1, and thus G 1 is connected k-connected
Discrete Algs for Mobile Wireless Sys17 Proof of Correctness Suppose in contradiction G' is not a subgraph of G 1. So there is some edge (u,v) in G' but not in G 1. (u,v) is also not in G /k, since no edges between non- removed nodes are removed in going from G /k to G 1 So (u,v) is not needed in G /k : there are some other nodes w 1, w 2,… that are closer to u than v is, s.t. max angle b/w w i and w i+1 is at most /k
Discrete Algs for Mobile Wireless Sys18 Proof of Correctness G remove v 1,…,v k-1 G' connected remove v 1,…,v k-1 G 1 disconnected G /k CBTC( /k) CBTC( ) G' connected by previous result = 2 /3 show G' is a subgraph of G 1, and thus G 1 is connected k-connected contains (u,v) doesn't contain (u,v)
Discrete Algs for Mobile Wireless Sys19 Proof of Correctness In worst case, the k nodes removed are consecutive w-nodes. Resulting empty cone has angle at most ( /k)*k = So in G' , there is no need to use enough power to reach v; it is sufficient to reach the remaining w-nodes Thus (u,v) is not an angle of G' , contradiction.
Discrete Algs for Mobile Wireless Sys20 Proof of Correctness u u u u u
Discrete Algs for Mobile Wireless Sys21 Central Idea of Algorithm Parameter a to algorithm Each node u starts broadcasting and collecting Acks with minimum power Node u gradually increases its transmission power stops when reaching maximum power or there is no cone of degree a with no neighbors