Planning and Optimization of Broadband Power Line Communications Access Networks: Analysis, Modeling and Solution

Transcription

1 Technische Universität Dresden Chair for Telecommunications 1 ITG-Fachgruppe Workshop Planning and Optimization of Broadband Power Line Communications Access Networks: Analysis, Modeling and Solution Dr.-Ing. Abdelfatteh Haidine Prof. Dr.-Ing. Ralf Lehnert haidine@ifn.et.tu-dresden.de Munich,

2 Content 2 Introduction/Motivation B-PCL and Planning Process Generalized Base Station Placement Problem Channel Allocation problem Numerical Experiments Conclusions

3 Introduction 3 WAP ISP Internet & PSTN TS BS R R Low-voltage network 220V-110V HVN MVN Education building High building R: Repeater BS: Base station M: PLC modem TS: Transformer station MVN: Medium-voltage network WAP: WAN Access Point HVN: High-voltage network B-PLC uses low-voltage grid to build access network No new wiring: Huge saving in investment costs Power signal 60Hz OFDM subcarrier 100% penetration (in rural and urban areas) Possible very high bit rates (today ~200Mbps) 3MHz B-PLC signal Power line spectrum (MHz) PLC channel with capacity K Ch 30MHz Goal: Analysis of B-PLC AN planning process, i.e.: Description Modeling Solution

4 B-PLC Access Networks Planning 4 Goal: build a PLC site on a low-voltage network BS1 WAP1 BS2 WAP2 BS3 Transformer station Open cut-off point Repeaters WAN House Street Cabinet Connection Access are placed where necessary for other Points are placed users connection Base Channels Users Potential Station are are places connected assigned placed for Base WAN to in to the some to the WAN Stations Access Base potential Access Points Stations (BSs) places Points

5 Sub-tasks of Planning Process 5 Base stations placement Users allocation to BSs Repeaters placement Connecting BSs to WAPs Generalized Base Station Placement (GBSP) Allocation of a channel sub-set to each BS/cell PLC Channel Allocation Problem (P-CAP) Strongly similarity to wireless networks: GBSP and P-CAP probably NP-hard like in wireless

6 Sub-problems Interdependence 6 GBSP and P-CAP are interdependent sub-problems General approaches solving of such inter-sub-problems: Input data Input data Sub-prob. 1 optimize Sub-prob. 1 Sub-prob. 2 Sub-prob. k optimize Solution 1 Cost C 1 Cost C Global Sub-prob. 2 optimize Solution 1 Solution 2 Solution k Solution 2 Cost C 2 Cost C 1 Cost C 2 Cost C k Sub-prob. k Solution k optimize Cost C k a) Sequential approach: less complexity, more probable sub-optimum b) Integrated approach: high complexity, more probable global optimum

7 7 Placement of Base Stations in Broadband Power Line Communications Access Networks by Means of Multi-Criteria Optimization

8 GBSP Description 8 Given: Tasks: - LVN topology - Cost of network elements - Traffic demand - Medium access strategy - Available WAPs - Realize all GBSP sub-tasks Objectives: - Minimize network costs - Minimize delay Constraints: - In-sight constraint - System (BS/TDR) coverage (L max ) - WAP capacities

9 Main GBSP Constraints 9 In-sight and ˆ1= 1 y j u 1 B k No in-sight and yˆ1 m =0 B m B j Cable between BS j and user u 1 of length l u 1 B j u 4 R 1 (j) u 4 Cable between BS j and user u 4 of length l In-sight constraint between B-PLC subscriber and his base station B-PLC systems (BS/TDR) coverage limit constraint

10 BS/TDR Locations: European Examples 10 Assume all potential locations have enough free space Source: DREWAG, Dresden, Germany

11 BS/TDR Locations: Some Examples 11 Repeater Repeater Source: Linz Strom (AUT), OPERA Project, D14 Source: OPERA Project, D14 Source: DREWAG, Dresden

12 Network Costs 12 Network costs: Summing costs of all implemented Network Elements (NE) -10C < Temperature < 70C Weak ventilation Humidity < 95% Very dusty Risk of flooding, etc. C NE = C Hardware + C Installation C NE = C Hardware = Constant Source: DREWAG, Dresden, Germany

13 Delay: The Types 13 Downlink: BS-to-end-user (D B2e ); Uplink: end-user-to-bs (D e2b ) Supposing Time Division MUX Repeaters (TDR) BS 1 st Time Slot (TS) 2 nd TS 3 rd TS u#1 u#2 u#3 R1 R2 R3 R4 Packet Propagation u#4 4 th TS Time In downlink: packet broadcasting In uplink: need for medium access control mechanism assuming a Round-Robin Polling protocol

14 a) BS is the server (1 server) Delay: The Model b) Data flow from each user is a Poisson process c) Service time is different from user to an other B-PLC AN as a M/G/1-queueing system From backbone network B-PLC access network 14 To PLC user E { } { } { 2 λ E X } D = E X + 2 (1 ρ) E{ X } = 1/ µ E{ X 2 ρ = λ / µ } = i p( X i ) X i Average service time 2 Average service rate Average arrival rate Service time: In downlink: just the transmission time (function of repeater number) In uplink: waiting time for polling + transmission time

15 Solving Multi-objective Optimization 15 By conversion into Single-Objective Optimization (SOO): scaling: constraining: minimize f Problem =α + β minimize f f 2 Problem F = (2) Threshold 1β1 f1 α 2 2 f2 scaling factors f 1 C omparison By Pareto approach: minimize f = [ f, f Problem 1 2 ] GBSP as MOP: Evolutionary search-based metaheuristics widely used

16 Multi-Objective Optimization (MOO) 16 MOO optimize the different objective quasi-independently MOO algorithm finds an approximation set A (set of trade-off solutions) f 2 f 2 (S 2 ) f 2 (S 1 ) S 1 S 2 e.g.: minimize f = [f 1,f 2 ] f 2 (S 3 ) S 3 S 4 S 5 A={ S 1, S 3, S 4, S 5 } f 1 (S 1 ) f 1 (S 3 ) f 1 (S 2 ) f 1

17 Motivation for MOO Application 17 Comparison of MOO with SOO evolutionary algorithm Source: [Huan01]: Automatic cell planning for mobile network design: optimization models and algorithms, PhD Thesis, Fakultät für Elektrotechnik, Universität Karlsruhe

18 Another Motivation for MOO Application 18 Eckart Zitzler, Lothar Thiele: Comparison of Multiobjective Evolutionary Algorithms: Empirical Results, Research Report, Swiss Federal Institute of Technology, 2000) minimize f Test = [f 1,f 2 ]

19 How to compare SOO and MOO SOO supplies 1 solution, MOO an approximation set Using front reduction for the comparison f 2 f = 1, Thr f ( SOO) 1 19 f 2,min f 1,Thr Multi-Objective Evolutionary Algorithm (MOEA) variants: - Non-dominated Sorting Genetic Algorithm (NSGA); - its 2 nd variant (NSGA-II); - Strength Pareto Evolutionary Algorithm (SPEA) f I

20 Numerical Experiments: Parameters 20 4 test scenarios small network 7 loc.; 45 users large network 14 loc.; 934 users system#1 (30Mbps) system#2 (15Mbps) Characteristics of PLC systems Algorithm: 100 generation, 100 individuals, 100 Individuals, crossover proba. p C =1, mutation proba. p m =0.01, 30 runs

21 SOO-MOO: Quantitative Comparison 21 Large network Small network

22 SOO-MOO: Qualitative Comparison 22 Solution 1: (f costs =850cu, f e2bs =27.44ms) 4 5 BS5 Solution 2: (f costs =1100cu, f e2bs =15.6ms) BS4 4 5 BS5 SPEA Small net. System#2 1 2 BS3 3 3 BS6 6 7 BS2 1 2 BS S 1 =[850, 27.44] Small_Network_system#2 SPEA Solution 3: (f costs =1400cu, f e2bs =7.83ms) 25 BS4 4 5 BS5 Uplink delay (ms) S 2 =[1100, 15.6] BS1 BS BS3 3 3 BS6 6 BS S 3 =[1400, 7.83] Costs (cu)

23 SOO-MOO: Qualitative Comparison 23

24 24 Modeling of Channel Allocation in Broadband Powerline Communications Access Networks as a Multi-Criteria Optimization Problem

25 Spectrum utilization and Cells Structure 25 Spectrum utilization OFDM subcarrier PLC channel with capacity K Ch 3 30 Power line spectrum (MHz) Network structure Backbone access point B j u 1 (j) u 2 (j) B l R r (j) u 3 (j) B k BS4 BS5 BS6 B m R 1 (j) R 1 (m) u 4 (j) R 2 (j) BS3 BS7 u 5 (j) Street cabinet with BS Repeater User PLC cell on an European low-voltage grid BS2 PLC site BS1

26 Wireless CAP Models 26 CAP is a standard problem in wireless communications networks Different models have been used: a) Maximum service (Maniezzo and Carbonaro (2000)) b) Minimum blocking (Koster (1999)) c) Minimum span (MS-FAP) (Aardal et al. (2001)) d) Minimum interferences (MI-FAP) (Schulz (2003)) P-CAP model as Multi-objective Optimization Problem (MOP): Optimize f CAP = [maximize Resource Reuse, minimize Interference]

27 P-CAP Description 27 Given: - PLC site structure (GBSP output) - Interference distance matrix (M ID ) - Set of available PLC channel (F S ) - Locally blocked channel (F (j) block) - Users traffic demand Tasks: - Allocate to cell j channel sub-set (F (j) alloc) Objectives: - Maximize resource reuse - Minimize interferences

28 Resource Reuse 28 Maximum resource reuse (i.e. maximum service) objective: ( j) 1 F = alloc maximize f RR ( j) B F Such that : a F j, f ( j) alloc = S j B j F S S a { } 0,1 ; : = 0; j, f demand j B j B S, f S j B F S S, f f F F S ( j) block B S F (j) demand F (j) block a j,f : Set of installed BS by GBSP : Channel demand inside cell j : Sub-set of channel blocked inside cell j : Decision variable (1 if channel f allocated to cell j; 0 otherwise)

29 B-PLC Site Interferences: Types 29 Interferences {source; victim; coupling path} Placed BS Street Cabinet House connection Transformer station BS4 BS5 BS6 BS3 BS7 BS2 Common segment conducted interferences BS1 Wiring too close radiated interferences 2 types of interferences in B-PLC site: a) in-line interferences b) in-space interference

30 Interferences Optimization Modeling 30 In mobile networks called co-site interference modeled as: a) Hard constraint (Eisenblätter (2001), Beckmann (2001)); b) Minimize interference occurrence Constraint Satisfaction Problem (Maniezzo et al. (2000)) P-CAP: δ jj : minimal channel distance to avoid interference between cell j and j (Interf ) Iin line : intensity of in-line interference

31 Solving P-CAP: I. Problem instance 31 Low-voltage grid: 14 possible BS locations ; 934 users 4 P-CAP problem scenarios Small site 9 BSs Large site 12 BSs system#1 (30Mbps) system#2 (15Mbps) P-CAP as MOP: Goal: find adequate crossover and mutation schemes for solving P-CAP by multi-objective optimization

32 4 Different Crossover Schemes 32 1-Point Simple Crossover (1-PSC) 1-Point Multiple Crossover (1-PMC) 2-Point Simple Crossover (2-PSC) 2-Point Multiple Crossover (2-PMC)

33 Evaluation of Crossover Schemes 33

34 Conclusions (GBSP) 34 Planning of B-PLC access network GBSP + P-CAP GBSP is multi-objective optimization: min. costs vs min. delay P-CAP as MOP: max. resource reuse vs min. interferences Can be solved either by SOO or MOO Evolutionary algorithms well known/performing search algorithms (in both SOO and MOO) For a given delay threshold, SOO supplies often minimum costs However, MOO supplies a diversity of optimum solutions, allowing an easier decision making in the practice

35 35 THANK YOU FOR YOUR ATTENTION!

36 36 Open issues: Simplification of GBSP formulation GBSP in form (min. costs, min. delay, max. reliability); Consideration of medium access protocol from standard Use of tailored heuristics may be more efficient