Long Distance Wireless Mesh Network Planning: Problem Formulation and Solution

Transcription

1 Long Distance Wireless Mesh Network Planning: Problem Formulation and Solution Sayandeep Sen Bhaskaran Raman Indian Institute of Technology, Kanpur

2 Outline Motivation & Background Problem statement, Uniqueness Solution Preliminaries Definitions, dependences Solution approach Overview Brief Details Evaluation Conclusions

3 to Bridge the Digital Divide Benefits: Existence of low cost equipments. Free Spectrum Example Deployments: Akshaya, Kerala, India Digital Gangetic Plains, Uttar Pradesh, India Djursland.net, Denmark Nepal Wireless

4 Digital Gangetic Plains: Testbed for last-hop access within a village Mohanpur Sept 04 Mandhana Lodhar Apr 02 Jun 02 Point-to- Point link 5.1 Km 3.5 Km 2.3 Km 0.9 Km 12 Km 12 Km MS3 Jun 02 IITK Sept 02 Safipur 17.3 Km 22.5 Km Nov 02 Bithoor 23 Km 5Km 37 Km Land-line access point (close to high-population density area) Mar 04 Rasoolabad 17.5 Km Dec 02 Sarauhan Jun 03 Banthar 22 Km Dec 03 Sawayajpur Aug 04 Dhaura 39 Km End to end distance ~80 Km Not to scale Mar 03 Rajajipuram/ Lucknow Antennae at Mandhana River Ganges

5 The Ashwini Project Byrraju foundation, West Godavari, Andhra Pradesh Envisaged to connect 34 villages Video-based health, education services

6 A WiFi Network in Djurslands, Denmark

7 Outline Motivation & Background Problem statement, Uniqueness Solution Preliminaries Definitions, dependences Solution approach Evaluation Conclusions

8 Problem Statement Motivation Problem: India has order of 600,000 villages. India has order of 600 districts. Africa has order of 100,000 sparse settlements. Long Distance WiFi network deployments envisaged No automated method exists to plan such networks. Current methods highly cost inefficient and ad-hoc.

9 Problem Statement Landline node Given: a set of village nodes, a single landline node Requirement: connect all villages to the landline in a network Primary concern: cost and bandwidth guarantees.

10 Problem Uniqueness Coverage only at village nodes (unlike cellular coverage) Line-of-Sight requirement Focus on cost optimality Cost dominated by towers (around 80% of total costs) Tower Line-of-Sight (LOS) Mast Tower/mast height (m) Cost (x1000 Rs.) Cost (U.S $)

11 Outline Motivation & Background Problem statement, Uniqueness Solution Preliminaries Definitions, dependences Solution approach Evaluation Conclusions

12 Preliminaries Assumptions: Antenna towers to be placed only at villages Tree topology 2P MAC [1] to be deployed Other MACs: Minor modifications Application requirement: Throughput requirement specified per village E.g. 384Kbps/ village (for video) Definitions... Dependences...

13 Definition: Point-to-Point Link Parabolic Grid Antenna Antenna Radiation Pattern 90 o One radio at each end 180 o 270 o 0 o Main direction Side-lobes

14 Definition: Point-to-MultiPoint Link Set Sector Antenna 90 o Antenna Radiation Pattern 180 o 0 o Single radio for both links 270 o Main direction

15 Notation: Top View Point-to-Point (P2P) links Point-to-MultiPoint (P2MP) link

16 Definitions: Transmit Power & Interference P1 Interference

17 Dependence: Throughput depends on Link/Antenna Type Point-to-Point (P2P) links Point-to-MultiPoint (P2MP) link

18 Dependence: Transmit Power (required) depends on Link (length) & Antenna Type P1 P2 P3 P2 > P1 & P3 > P1

19 Dependence: MAC (feasibility) depends on the Transmit Powers P1 Interference Signal to Interference Ratio should be above threshold Signal to Interference Ratio should be above threshold P R I R SIR reqd

20 Dependence: Tower Height(s) (required) depends on Link (length) Fresnel Zone Tower Mast

21 Dependence: Cost depends on Tower Height Tower/mast height (m) Cost (x1000 Rs.) Cost (U.S $) MASTS Cost takes quantum jumps due to change in underlying tower design Super-linear function of height Masts are extremely cheap

22 Dependences: Summary A Depends on B Throughput MAC TxPower Antenna Type Link (length) Cost Tower Height

23 Outline Motivation & Background Problem statement, Uniqueness Solution Preliminaries Definitions, dependences Solution approach Overview Evaluation Conclusions

24 Dependence Resolution Antenna assignment to be done before TxPower LP formulation of power assignment possible Throughput 2P MAC TxPower Power assignment Antenna assignment Antenna Type Link (length) Cost Tower Height

25 Solution Approach (continued) Spanning tree to be formed before assigning heights LP formulation of height assignment if possible Throughput 2P MAC TxPower Tree search Antenna Type Link (length) Cost Height assignment Tower Height

26 Solution Approach (continued) Power Assignment & Height Assignment mutually independent Throughput 2P MAC TxPower Power assignment Antenna Type Link (length) Cost Height assignment Tower Height

27 Solution Methodology Landline Landline Antennas Villages STEP 1: Tree Enumeration Villages STEP 2: Antenna + Power Assignment Towers Landline STEP 3: Tower Height Assignment Antennas Villages

28 Contributions For Tree enumeration: Branch-and-Bound based algorithm with domain knowledge based pruning strategies. For Height Assignment: Linear Programming (LP) formulation. For Antenna + Power Assignment: Heuristic Algorithm for antenna assignment. LP formulation for power assignment based on results from previous work [2].

29 Outline Motivation & Background Problem statement, Uniqueness Problem formulation Definitions, dependences Solution approach Tree Search and Height Assignment Evaluation Conclusions

30 Optimized Tree Enumeration Domain-knowledge based optimizations Eliminate long links to begin with: > 15 km (say) Trade-off between run time and search space Tree depth restriction: 2 hop topologies only Accommodates significant # practical scenarios Dynamic cost bounding: throw out costly subtrees Cyclic dependence on height assignment (to be explained).

31 Height Assignment: Problem Statement Landline Given: a topology (parentchild relationships) Villages

32 Height Assignment: Problem Statement Towers Landline Given: a topology (parentchild relationships) To determine: optimal tower/mast heights, satisfying LOS criteria Villages

33 Height Assignment: Insights Insight 1: 2-hop topology only Trade off between Runtime and Search Space In India 85% of the villages are within 20 Km radius of a fiber Point-of-Presence (PoP) [3]. One hop ~ 10km ==> 20km radius ==> 40km dia Insight 2: Tower at central location say 50m Missed Village Village Hop 1 Hop 2 Landline Common in a town with tall buildings Insight 3: link between two masts infeasible Obstructions (trees etc.)

34 Tower at Level-2 or Level-3? OR Observation-1: tower heights can be interchanged in a link, retaining the same cost Note: does not hold if terrain uncertainties are considered Observation-2: # level-3 nodes (leaves) >= # level-2 nodes Implication: towers at level-2 and masts at level-3

35 The LP Formulation Fresnel Zone M h2 Tower L d h1 Mast h2 N P D L O h1 d Q Linear equations for obstruction clearance Linear cost optimization function (forced) Use tower cost coefficient for level-2 and mast cost coefficient for level-3.

36 Dynamic Cost Bounding Decisive Leaf Non Leaf Observation-1: height of level-2 tower determined by children set Observation-2: given a link-length, can lower-bound tower height Implication: can lower-bound the cost of a sub-tree Can pre-compute lower-bounds for efficiency

37 Outline Motivation & Background Problem statement, Uniqueness Solution Preliminaries Definitions, dependences Solution approach Evaluation Conclusions

38 Evaluation: Bhimavaram topology Bhimavaram Ashwini Project: Byrraju foundation, West Godavari, Andhra Pradesh To connect 34 villages (result only for 31 nodes) Uses ONE wireless channel compared to THREE by current deployment. Careful topology planning led to 21% cost savings.

39 Evaluation: Amalapuram topology Ashwini Project: Byrraju foundation, East Godavari, Andhra Pradesh To connect 18 villages Uses ONE wireless channel.

40 Conclusions Topology construction an important problem Unique problem thus far Challenging to formulate Our contributions: Problem formulation Overall approach First-cut solution Lots of scope for further in-depth work Details in the paper

41 Bibliography [1] Design and Evaluation of a new MAC Protocol for Long-Distance Mesh Networks, B. Raman, K. Chebrolu, MOBICOM 05 [2] Long-Distance b Links: Performance Measurements and Experience, K. Chebrolu, B. Raman, S. Sen, MOBICOM 06 [3] A. Jhunjhunwala et. al, Role of Wireless technologies in connecting rural India, IJR&SP 05