Double-Loop Receiver-Initiated MAC for Cooperative Data Dissemination via Roadside WLANs

Transcription

1 Double-Loop Receiver-Initiated MAC for Cooperative Data Dissemination via Roadside WLANs Presented by: Hao Liang Broadband Communications Research (BBCR) Lab

2 Outline Introduction and Related Work System Model The Double-Loop Receiver-Initiated MAC Scheme Performance Analysis (Brief) Numerical Results Summary and Discussions 2

3 Introduction and Related Work 3

4 Data Dissemination Service Main characteristics: 1) Data traffic is generated by a content server in the Internet and destined to a group of nomadic nodes roaming in the network region; 2) Can tolerate some delay Example 1: Traffic information downloading Example 2: Entertainment content distribution Example 3: Commercial advertising (e.g., a flyer from Food Basics) Question: How to deliver the service? 4

5 Available Network Infrastructures Wireline networks e.g., home Wireless local area networks (WLANs) e.g., home/work Cellular network e.g., anywhere with $ cost Any others (@ anywhere with less/no $ cost)? 5

6 Roadside WLAN (RS-WLAN) Idea: Use what we have rather than deploy new infrastructures (e.g., RSUs)! FON (a company operating five million hotspots in the world) London, England (picture from Google Maps) Hotspots are privately owned, but mutually shared (i.e., free) among Foneros 6

7 Research Issues How to improve the data delivery rate to nomadic nodes walking/ driving/passing(by other means) through the coverage area of an RS-WLAN? How to design an efficient medium access control (MAC) scheme? Related Work Drive-thru Internet Direct transmission from the AP (like RSU) Cooperation Exploits spatial/temporal diversity among local users Delay tolerant cooperation (DTCoop) Exploits the delay tolerance Transmitter-initiated cooperative MAC 1) Achieves channel diversity; 2) Overhead of RTS/CTS message transmission Receiver-initiated MAC 1) Reduces MAC overhead by suppressing RTS message transmission; 2) Without channel diversity 7

8 System Model 8

9 Cooperative Data Dissemination (1/2) Resource management within each RS-WLAN Bandwidth: a) A superframe structure; b) Each dedicated phase is assigned to one nomadic node, while different dedicated phases are shared among nomadic nodes (because of the delay tolerance) Buffer space: a) Storage local nodes; b) Non-storage local nodes Energy: Participation in cooperation (e.g., laptops) 9

10 Cooperative Data Dissemination (2/2) Delay tolerant cooperation (DTCoop) Step 1) Packet pre-downloading Step 2) Packet scheduling Benefits 1) Relieves the limitation on wireline bandwidth 2) Spatial diversity: Enlarges the coverage area of an RS-WLAN 3) Temporal diversity: Improves the packet transmission rate Limited Bandwidth Focus of this talk: How to design an efficient MAC scheme to realize DTCoop? 10

11 Assumptions (1/2) Packet pre-downloading is completed before the arrival of a nomadic node (e.g., on-demand pre-downloading based on cellular networks or stochastic pre-downloading based on historic mobility information) Pedestrian nomadic node Large wireline bandwidth Single-hop (all connected) RS-WLAN There are M possible wireless transmission rates. The signal-tonoise ratio (SNR) threshold of the mth rate is 11

12 Assumptions (2/2) The traffic of the data dissemination service is partitioned into packets with equal size The packet transmission rate from local node i to the visiting nomadic node (including the transmission of the ACK message and packet relaying overhead) Storage local node Time of first-hop packet transmission (information available to all local nodes) T P (m) is the packet transmission time at the mth transmission rate Non-storage local node T SIFS, T HR, and T ACK are the time durations of short interframe space (SIFS), helping request message, and ACK message, respectively 12

13 The Double-Loop Receiver-Initiated MAC (DRMAC) Scheme 13

14 Concept and Overview (1/2) Time Correlation of a Wireless Channel Example: A Rayleigh fading 5 GHz frequency 1.2 m/s mobility speed => maximum Doppler frequency = 20 Hz Channel coherence time Packet transmission time Channel coherence time = 1 / maximum Doppler frequency 14

15 Concept and Overview (2/2) The outer-loop MAC is performed at a low frequency to determine the contention group (CG) membership for spatial diversity Only the CG members can participate in channel contention The inner-loop MAC is performed at a high frequency to select the transmitter node for temporal diversity The ACK message is used as a receiver-initiated contention invitation to reduce MAC overhead 15

16 The Outer-Loop MAC Each local node i estimates its instantaneous packet transmission rate ( ) to the nomadic node based on the signal-to-noise ratio (SNR) of each ACK message it received Based on the historic estimations, the average packet transmission rate ( ) is calculated by each local node and reported to the AP A group of N C local nodes (including the AP) with the highest average transmission rates are selected as the CG members The AP broadcasts a CG notification message to notify all local nodes about the CG membership and a descending order in terms of the average transmission rate 16

17 The Inner-Loop MAC (1/5) Contention by Invitation (i.e., Not Invited => Not Contend) Each ACK message from the nomadic node carries the rate index of the previous transmission (m Tx ) The rate of the previous transmission (R Tx ) can be calculated based on the receiver address field of the ACK message (corresponding to the ID of the previous transmitter node, i Tx ) and the value of m Tx If a CG member has a higher rate than (R Tx ), it is invited and will join the channel contention 17

18 The Inner-Loop MAC (2/5) Contention by Invitation (Cont d) The invited CG member sends a short burst after the previous ACK message, as a contention request A deterministic backoff begins with the kth potential (invited) CG member backoffs for (k - 1) slots before its transmission kth potential CG member wins the contention 18

19 The Inner-Loop MAC (3/5) Contention by Invitation (Cont d) The group of all potential invited CG members Previous transmitter node Rate of previous transmission Group of CG members The potential invited CG members are sorted according to a descending order of the one-step higher transmission rate 19

20 The Inner-Loop MAC (4/5) CG attachment At the beginning of a new dedicated phase, the value of m Tx indicated in the previous ACK message becomes out-of-date Each dedicated phase begins with a dedicated phase assignment message broadcasted by the CG member with the highest average transmission rate to the nomadic node 20

21 The Inner-Loop MAC (5/5) CG attachment (Cont d) Deterministic backoff and rate notification The nth CG member will start transmission immediately if no other transmissions are detected for a duration of T SIFS and (N C - n) slots Nomadic Node SIFS RR SIFS Beginning of Deterministic Backoff DA: dedicated phase assignment HR: helping request RN: rate notification RR: receiving request ACK (N C -1) slots 1st CG Member n th CG Member DA Beginning of Dedicated Phase RN SIFS Deferred Backoff T O L S (n-2) slots RN SIFS T O L S Deferred Transmission (N C -n) slots HR Data SIFS Traffic Source of the n th CG Member T O L S nth CG member wins the rate notification T O L S T O L S T O L S T O L S SIFS Data SIFS 21

22 Performance Analysis (Brief) 22

23 Key: Finite-State Markov Chain based Channel Model Time duration One-step higher transition probability One-step lower transition probability Probability of transmission rate m Maximum Doppler frequency Average SNR Probability of no rate change 23

24 How the Channel Model is Used? Duration of contention-by-invitation Use the one-step transition probabilities, given the current transmission rate of CG members Duration of CG attachment Iterative calculation based on the success/failure of transmitting each dedicated phase assignment message (like coin flipping, but with one-step memory) 24

25 Numerical Results 25

26 System Parameters Parameter Speed of Nomadic Node Value 1.5 m/s Pathloss Exponent 3 Radius of RS-WLAN Neighbour Discovery Range Rayleigh Fading Model Length of Packet 25 m 50 m Jake s Simulator 1000 bytes Circular topology: Local nodes are evenly distributed on two circles with radius r/2 and 3r/4, respectively. The two circles include the same number of local nodes, and half of the local nodes can provide packet storage capabilities 26

27 Performance Evaluation of the DRMAC Scheme The effect of N C on the proposed MAC scheme, without MAC overhead The effect of N C on the proposed MAC scheme, including MAC overhead 27

28 Performance Comparison among Different MAC Schemes Receiver-initiated MAC: CTS/data/ACK. Modified by inviting the storage local node with the highest average transmission rate to the nomadic node Transmitter-initiated cooperative MAC: RTS/CTS/data(direct transmission or relay transmission)/ack IEEE MAC: RTS/CTS/data/ACK Direct transmission 28

29 Performance Comparison among Different MAC Schemes (Cont d) The effect of movement speed on different MAC schemes 12 local nodes, (π/8, 0) The effect of movement speed on different MAC schemes 20 local nodes, (π/16, 3π/4) 29

30 Summary and Discussions 30

31 Main Contributions Achieve the cooperative diversity gain based on a receiverinitiated MAC Scheme Utilize the time correlation of a wireless channel to reduce the signalling overhead A novel analytical model is established using a finite-state Markov chain based channel model to characterize the time correlation 31

32 Discussions Extension to Vehicular Nomadic Nodes Issues on wireless channel: 1) High mobility => Difficult to obtain accurate channel state information; 2) Predictable mobility => Predictable channel? Issues on packet pre-downloading: 1) Less predictable mobility than train or bus; 2) More predictable mobility than pedestrian 32

33 Thank you! 33