Presentation is loading. Please wait.

Presentation is loading. Please wait.

Dynamic Sensor Networks Project Review of UCLA’s Activities Mani Srivastava UCLA.

Similar presentations


Presentation on theme: "Dynamic Sensor Networks Project Review of UCLA’s Activities Mani Srivastava UCLA."— Presentation transcript:

1 Dynamic Sensor Networks Project Review of UCLA’s Activities Mani Srivastava UCLA

2 This Review Two Separate Projects at UCLA DSN (Subcontract from USC/ISI)  Sole PI: Mani Srivastava  Focus: networking Low-power/low-latency link, MAC, and routing GPS-less discovery & distribution of location Capability and attribute based addressing and connectivity Sensor network simulation and emulation Protocols for GPS-synchronized communication subsystem Sensorware (Subcontract from Rockwell Science Center)  Two PIs: Mani Srivastava, Miodrag Potkonjak  Focus: distributed middleware services Network coverage service for sensor networks Sensor control scripts: light-weight, mobile, platform independent, secure Spatial addressing and communications, timing synchronization Implementation on Rockwell’s nodes

3 This Review: Selected Recent Activities I. Update on Sensorsim II. GPS-less ad hoc localization III. Low-latency packet forwarding IV. Dynamic assignment of MAC addresses V. Low-power multihop routing

4 I. SensorSim Update Simulation framework for modeling sensor networks built on top of ns-2  sensing channel and sensor models  scenario generation tool (SensorViz)  light weight protocol stacks  hybrid simulation  battery/power model (further model development under PAC/C) Alpha release at http://nesl.ee.ucla.edu/sensorsim/http://nesl.ee.ucla.edu/sensorsim/ Selected features being migrated to official ns-2 through Deborah’s group In use by external groups such as U. Maryland

5 SensorSim Architecture Target Node Sensor Layer Physical Layer Sensor Stack Sensor Channel Target Application Wireless Channel User Application User Node Network Layer Physical Layer Network Stack MAC Layer Sensor Layer Physical Layer Sensor Stack3 Sensor Layer Physical Layer Sensor Stack2 Functional Model Sensor Node SensorWare Power Model Battery Model Radio CPU ADC (Sensor) Wireless Channel Sensor Channel1 Network Layer MAC Layer Physical Layer Network Stack Sensor Layer Physical Layer Sensor Stack1 Sensor App Sensor Channel2 Sensor Channel3 Sensor Node Wireless Channel Sensor Channel User Node Target Node

6 Scenario Generation & Visualization SensorViz Features: Diverse scenario generation  Node deployment patterns  Target trajectories  Sensor characteristics  Node attributes Can be slaved to a running simulation (SensorSim) Monitor real sensor nodes Planned: XML output Read in SITEX format scenarios

7 II. Dynamic Location Discovery Discovery of absolute and relative location important  location attribute based naming and addressing of nodes  geographical routing  tracking of moving phenomena (targets) GPS not enough  not work everywhere due to requirement of LOS to satellites (trees, indoors)  not on all nodes (costly, large, power-hungry) No infrastructure in sensor networks precludes solutions based on trilateration with special high power beacons  also, susceptible to failure Problem: given a network of sensor nodes where a few nodes know their location (e.g. through GPS) how do we calculate the location of the other nodes? Known Location Unknown Location

8 Ad-Hoc Localization System(AHLoS) Every node contributes to process Small fraction of initial beacons Distributed  Robust  Energy Efficient Inter-node ranging uses  RSSI  Ultrasound Integrated with routing messages  Location discovery almost free! Adapts to channel conditions via a joint estimation of location & channel parameters Iterative Multilateration Collaborative Multilateration

9 Centralized vs. Distributed Localization Distributed Pros More robust to node failure Less traffic => less power Better handling of local environment variations  Speed of ultrasound  Radio path loss Rapid updates upon topology changes No time synch. required Centralized Cons A route to a central point Time synchronization is required High latencies for location updates Central node requires preplanning More traffic => higher power consumption

10 Basic Multilateration Repeat until δ becomes 0 a 1 2 3 Linearize using Taylor Expansion Residual of measured and estimated distance Linear form MMSE Solution

11 Iterative Multilateration Basic multilateration can be applied iteratively across the network Step 1 Step 2 Step 3

12 Node vs. Initial Beacon Densities % Initial Beacons Total Nodes % Resolved Nodes Uniformly distributed deployment in a field 100x100. Node range = 10.

13 Challenges Iterative multilateration may stall if  the network is very sparse  the percentage of beacons is very low  terrain obstacles If the network is large, error will accumulate from iterative multilateration

14 Collaborative Multilateration ab 1 2 3 4 Uses location information over multiple hops Linearize residuals over 2 types of edges: Both equations have the form Follow the same solution procedure as basic multilateration

15 Collaborative Multilateration (contd.) Execute Update Until

16 Collaborative Sub-trees Necessary conditions: Each unknown node must have at least 3 participating neighbors A participating node is either a beacon node or an unknown node connected to 3 participating nodes 18 equations 16 unknowns Over-determined!

17 Distributed Ad-Hoc Operation Location estimation takes place at the scope of a neighborhood Collaborative sub-trees can zoom in and out to  Form a well-determined system  Avoid degenerate cases  Avoid obstacles  Reduce Error Propagation Error can be further reduced if computation takes place at a central point.

18 Platform Characterization Ultrasound TDoARSSI in football field

19 Iterative Multilateration Accuracy 50 Nodes 10% beacons 20mm white gaussian ranging error

20 Implementation Status Initial prototype competed: Medusa Design of Medusa II (using non-SensIT resources)  Longer range ultrasound (15-20m)  Radio Power Control & RSSI circuitry  More computation (Atmel THUMB) Goal: Hybrid Radio-acoustical localization  use radio for long-range when ultrasound is unable to find a neighbor  Medusa used standalone or as a location coprocessor to sensor nodes Atmel AVR RFM Radio Ultrasound Receiver Ultrasound Transmitter INT

21 III. Low Latency Packet Forwarding Problem: node often simply relays packets in multihop network  NS-2 simulation: 1000x1000 terrain, 30 nodes, DSR, CBR traffic from random SRC and DEST Traditional approach: packets sent from radio to main CPU  long latency (serial bus), power hungry (main CPU woken up) Action% of received packets ACCEPT34.300 FORWARD65.567 DROP0.133 Communication Subsystem Radio Modem GPS Micro Controller Rest of the Node CPU Sensor Multihop Packet Traditional Approach

22 Our Packet Forwarding Architecture Our approach: Embedded Packet Processor in the Radio  exploit programmable microcontrollers in the radios to handle common cases of packet routing  can also do operations such as combining of packets with redundant information Packets are redirected as low in the protocol stack as possible  reduced latency (and, incidentally, also reduced power…) Key challenge: how to do it so that every new routing protocol will not require a new radio firmware? Communication Subsystem Radio Modem GPS Micro Controller Rest of the Node CPU Sensor Multihop Packet

23 Application-defined Routing Framework for Radio Firmware Packet-classifier and packet-modifier driven by application defined matching rules and actions  Matching rules: and/or expressions using =,, range operators on arbitrary packet fields (offset, length)  Actions: accept, forward, drop, field increment/decrement etc. Rules and actions operate on arbitrary packet fields (any layer)  fields specified as (offset, length) For complex cases packet sent to the main processor  only simple, common cases handled at the radio Expressiveness: implemented the following as test cases  Node ID-based addressing and routing (DSR-like)  Geographical point-cast (send to a circular area specified as destination) Communication Subsystem Radio Modem GPS Micro Controller Packet Classifier Packet Modifier Application-Defined Matching Rules & Actions

24 Proof-of-concept Implementation Rockwell nodes with a prototype radio Prototype radio because Rockwell’s radio firmware is not open RFM radio with FPSLIC (microcontroller with FPGA) Mixed software/FPGA implementation FPGA used to accelerate packet matching/modification

25 Performance Analysis Difference in packet DELAY between the traditional approach and our approach: Serial port delay is the dominant factor Packet Distribution Delay Overhead for ACCEPT Delay Overhead for FWD Serial port delay Measurements Given the measurements the difference in delay is: When Pr FW > 3% Pr AC the traditional approach delay is more than our approach For our simulation traffic data D diff = 44ms ParameterValue (ms) D MCUAC 4.182 D MCUFW 4.894 D SR 36.532 D CPUAC 0.111 D CPUFW 0.125

26 IV. Dynamic MAC Address Allocation Wireless spectrum is broadcast medium MAC addresses are required In wireless sensor networks, data size is small Unique MAC address would present too much overhead Employ spatial address reuse (similar to reuse in cellular systems) Two aspects  Dynamic assignment algorithm  Address representation 4 0 1 1 2 2 3 5

27 Distributed Assignment Algorithm 0 0 1 2 0 1 3 2 0 1 0 0 1 2 4 1 3 1. Network is operational (nodes have valid address) 2. Listen to periodic broadcasts of neighboring nodes 3. In case of conflict, notify node (this node resends a broadcast) 4. Choose non-conflicting address and broadcast address in a periodic cycle. At this point the new node has joined the network. Additive convergence: network remains operational during address selection Mapping: unique ID to spatially reusable address Algorithm also valid when unidirectional links

28 Encoded Address Representation Address range 0-1112-1718-1920-2223… Codewor d size (bits) 45678… Encoded (bits/address)1.7 Fixed size (bits/address)2 Address Frequency D av = 0.01 nodes/m 2 0 1 2 3 0 10 110 111 0 1 2 3 Freq. of occurrence 0.5 0.3 0.1 Size of the address field? Non-uniform address frequency  Huffman encoding  Robust: can represent any address Practical address selection  All addresses with same codeword size are equivalent  Choose random address in that range to reduce conflict messages

29 Network Density Parameter 1. Taking only bulk nodes eliminates edge effects  Virtually extends network size to infinity (so independent of L)  Suggests that only close proximity is critical 2. Characterization of node density  Connectivity is key  Average degree Address Frequency All nodes Bulk nodes = 10 D av = 0.01 nodes/m 2  N = 125  N = 250  N = 1000 2R2R Bulk nodes L

30 Non-uniform Network Density X(m) Nodes/m 2 Y (m) X(m) Average address size (bits) Y (m)

31 Effect of Packet Losses ( = 10) Address Frequency P drop Convergence time (s)

32 Scalability Address assignment  Distributed algorithm with periodic localized communication Address representation  Encoded addresses depend only on distribution Scales perfectly (neglecting edge effects) Off-lineCentralizedDistributed ++-+ Unique Fixed reusable Encoded reusable --  + Representation Assignment Average address size (bits) =10 =5 =15 =20 Unique address Number of nodes

33 Simulation Results Our schemes Scheme Address selection type Av. size (bits) Address size scalability Globally uniqueManufacturing128+ Network wide uniqueDeployment14  Centr. / Distr.4.7  Encoded dynamicDistributed4.4+ Fixed size dynamic

34 Implementation Issues Functionality  Dynamic address assignment  Address resolution (mapping) Address Resolution & Assignment Protocol (ARAP) Unique receiver ID is mapped into MAC address without being included into the packet The own MAC address is modified by the ARAP Link Layer MAC PHY APP/NETW ARAP (~ARP) R x -ID R x -Addr Own Address APP Data Dest R x -Addr T x -Addr

35 Dynamic Address Allocation: Summary Spatial reuse of address Dynamic assignment algorithm  Localized: scalability  Additive convergence: robustness Encoded address representation  Independent of network size: scalability  Variable length addresses: robustness

36 V. Low-power Multihop Routing ATHENA: Adaptive Transmission-power Heuristic and Energy- optimizing ad-hoc Network routing Algorithm  adapts transmission power to find power-optimal multi-hop paths.  uses alternate routes to maximizes lifetime of the network. Recent work from Maryland  offers the same benefit: combine alternate routes with tx power control  but is not easy to implement (cost of algorithm vs. convergence) E(x) =energy to send a packet over distance x E(a) + E(b) < E(c) ab c Principle in adapting tx power

37 Constant Tx Power Case Constant power case (for comparison) On-demand algorithm  using request and reply messages  resemble DSR, AODV: source path carried to avoid loops  siblings not visited

38 Adaptable Tx Case Increased # of requests and replies  if destination reached, algorithm not over  siblings have to be asked Three main rules to prevent explosion of requests  one of them produces suboptimal routes, but simulations show the cost savings are worth it

39 Example Constant tx power case (level 8): 6 req, 8 replies 2.48*10 -4 Joules/packet Adaptablet tx power case: 8 req, 30 replies 6.638*10 -5 Joules/packet 10 tx levels [10m - 250m] packet = 125bytes signal attenuation ~ 1/d 3

40 25-node Network requests replies

41 25-node Network requests replies

42 Average Gains signal attenuation: ~ 1/d 3 ~ 1/d 4 25-node network 50-node network signal attenuation: ~ 1/d 3 ~ 1/d 4

43 Using Alternate Paths Self_Energy, Next-Hop_Energy should affect path cost. Each time the node’s energy changes 10%:  notify neighbors  recalculate best paths Heuristic used: Remaining_Energy self -x1 Power_Cost next_hop + Remaining_Energy next_hop -x1 Power_Cost destination Simulations show best x1 = 2 30% more packets routed than vanilla ATHENA 96% of the packets routed in the optimal case.

44 Recent Accomplishment Summary I. Sensorsim II. GPS-less ad hoc localization III. Low-latency packet forwarding IV. Dynamic assignment of MAC addresses V. Low-power multihop routing


Download ppt "Dynamic Sensor Networks Project Review of UCLA’s Activities Mani Srivastava UCLA."

Similar presentations


Ads by Google