You are on page 1of 22

Joint Flow Routing and Relay Node Assignment in Cooperative Multihop Networks

Sushant Sharma et al. IEEE Journal on Selected Areas in Communications, vol. 30, no.2, Feb. 2012
Speaker: Pham Tran Anh Quang
1

Presentation History
Y.Li et al., Enhancing Real-Time Delivery in WSNs with Two-Hop Information, in IEEE TII, Vol. 5, No.2, May 2009 Shuo Guo et al., Opportunistic Flooding in Low-Duty-Cycle Wireless Sensor Networks with Unreliable Links, in Mobicom 09 Sinem Coleri Ergen and Pravin Varaiya, TDMA scheduling algorithms for wireless sensor networks, Wireless Network, Springer Z. Liang et al., Delay Performance Analysis for Supporting Real-Time Traffic in a Cognitive Radio Sensor Network, IEEE TWC, Vol. 10, No. 1, Jan. 2011 Vehbi Cargi Gungor et al., A Real-time and Reliable Transport Protocol for WSANs, IEEE ToN, Vol.16, No.2, April 2008 P.T.A. Quang and Dong-Sung Kim, Enhancing Real-time Delivery of Gradient Routing for Industrial Wireless Sensor Networks, IEEE TII, Vol. 8, No.2, May 2012 Emanuele Toscano et al., Multichannel Superframe Scheduling for IEEE 802.15.4 Industrial Wireless Sensor Networks, IEEE TII (early access)
2

Outline
Motivations and Goals Models Constraints and problem formulation Algorithms:
Overview Feasible solution construction (FSC)

Simulation results Conclusions Future works


3

Motivations and goals


Motivations:
Achieve better throughput using cooperative scheme Joint problem: (1)relay node assignment and (2) multihop flow High computational complexity (MILP with large solution space) long calculation time

Goals:
Combine Branch and Bound (BB) and Gomory-cutting planes(CP) speed up computation

Limitations:
Complexity: exponential !!!
4

Models (Overview)

s d: direct transmission s r d: cooperative communications


(1) CC with amplify and forward (PHY layer) (2) CC with decode and forward (MAC layer) (3) direct transmission

Models (Channel capacity)


(1) CC with amplify and forward

(2) CC with decode-and-forward

(3) Direct transmission


Above equations can be achieved by using Shannons theorem and Information theory
6

Models (Node)
2 types of relay nodes: Cooperative relay (CR) and Multi-hop relay (MR) Full-duplex transmission (1 transmitter + 1 receiver) a relay node will be CR or MR single in-stream and single out-stream S != CR D != CR S !=MR D !=MR Ns=Nd Nr+Ns+Nd=N S: source, D: dest., CR=coop. relay, MR= multi-hop relay Ns: # souce nodes, Nd:#dest. nodes, Nr: #relay nodes
7

Mathematic constraints
(1) Role of relay nodes:
w is CR on hop (u,v) - Link u,v is active
1 CR 1 hop

Single in-stream MR

CR

MR In-stream = out-stream One CR can be assigned for only one-hop


8

Mathematic constraints
(2) Flow routing
Source always transmits to another node A node (except dest.) can receive (=1) or not (=0) Destination always receives from another node Destination can forward to another node

Flow balance at the intermediate node

Mathematic constraints
(3) Rate constraints
Direct transmission

Amplify and forward (AF) transmission

All streams go through hop (u,v) are either Direct transmission or CC transmission

10

Problem formulation
For a given session (si, di), e2e flow-rate:

Optimize the minimum flow rate

MILP problem:

11

Algorithms: Overview
MILP can be solved by using Branch and Bound (BB) or Gomory Cutting Plane (CP)
BB: partition relaxed problem into 2 subproblems. Then solving sub-problems until satisfying integer requirement. CP: add linear constraint to reduce feasible solution region until satisfying integer requirement Upper bound is improved (moce accurate)
12

Algorithms: Overview
Proposed algorithm: Combined BB and CP and FSC (finding lower bound) Proposed selection conditions to reduce calculation time Obtain (1-e)-optimal solution acceptable solution belongs to [U, (1-e)U] (1-e)U > L (with U: upper bound and L: lower bound of optimal solution)
13

Algorithms: Overview
L=max{li} U=max{ui} li is the lower bound of problem i (determined by FCS) ui is the upper bound of problem i A cutting planes can improve lower bound and upper bound by restricting solution region Cutting planes are added until the lower and upper bounds are not improved or the gap is smaller than e

Check r=(U-L)/U If r>e continue Else Finish


Assume that: r>e and (u2 >u1 or (1-e)u1 <= L) Sub-problem 1 will be removed

14

Algorithms: FSC
Feasible solution construction (FSC) is a local search algorithm Upper bound (determining by relaxed MILP):
Optimal value but not meet integer conditions

Lower bound (FSC):


Based on solution of upper bound, satisfy integer requirement but not optimal value

15

Algorithms: FSC
Phase 1: Path determination
Finding path based on throughput (widest pipe)
If (route si encounters relay nodes) finish and move to next hop; (0) else (route si encounters source (sj) or destination (dk) node) { If (encounter sj) sj L (1) else { if (sk in L;) L=L \{sk}; (2)

1-0-0-2

1-0-1-0-4

else if (dk is di)


{ if (L is empty) finish and move on to the next remaining nodes; (3) else { move on to the sm in L ; L = L \{sm}; (4) } } else { dk must not be included in the path; (5) } } }

1-0-0-5

16

Question: Find an example go to condition (3)

Algorithms: FSC
Phase 2: CR assignment: (for free CR)
Capacity-flow-ratio (CFR): hops capacity to the number of overlapping sessions CR assignments start with the minimum CFR hop and so on.

Phase 3: Flow recalculation After phase 2, every integer variables are determined MILP became LP solve the problem to find throughput for each flow lower bound of branching process
17

Simulation results
Bandwidth = 22MHz / channel Transmission power = 1W Path loss exponential = 4 (Multi-path models) Noise variance 10^-10 W e=0.1 40 nodes (Ns=Nd=8, Nr=24)

18

Simulation results
Routing map with CC Routing map without CC

19

Conclusions
The most important contributes are:
Combined BB and Gomory- CP to reduce computational time Proposed FSC to find down lower bound
Weakness: - Cannot reduce computational complexity (exponential complexity!!!) - Not prove that proposed scheme can reduce computing time !!! (even that it is the most important contribution) - Lack of simulations - Theoretical solution not stick on any standard - Ideal assumption (how can they handle interference???)

20

Future works
Combine that solution to n-hop region routing (n-hop region is much simpler than whole network feasible solution for WSN) what else??? I have not found out yet ~~
2 R 1 R s
21

2 R R Its much simpler than the problem of this paper 2 1

Thank you for your listening


22

You might also like