Title: An Application-Specific Protocol Architecture for Wireless Microsensor Networks By: W. Heinzelman, A. Chandrakasan
1An Application-Specific Protocol Architecture for
Wireless Microsensor NetworksBy W. Heinzelman,
A. Chandrakasan H. Balakrishnan
- A review prepared for CEG 790
- By Patrick Flaherty
2Presentation Outline
- What is a Wireless Sensor Network?
- Why are Protocols for Self-Organizing an issue?
- The proposed Low-energy adaptive clustering
hierarchy (LEACH) protocol - Concept
- Algorithms
- Analysis and simulation of LEACH
- Conclusions
3What Is A Wireless Sensor Network?
Network Layer Tasks (partial)
Application Source code
Transport Packets, congestion control
Network Routing
Data-Link MAC, error correction
Physical Wireless
4Wireless Sensor Networks
- 10s to 1,000s of wireless sensors placed into
an environment - May be put into structures where truckloads of
cabling would be required to connect to the data
collection point (e.g., Golden Gate Bridge) - May be placed in a natural environment to monitor
wildlife (e.g., study relationship between
weather conditions and animal behavior) - May be used in hostile environments to detect
movement of opponents
5Wireless Sensors
6Energy Consumption is a Function of Device
Activity
- Telos is a recently released microsensor platform
design
Source Telos Fourth Generation WSN
PlatformPresented at TinyOS Technology
Exchange, Feb. 11, 2005
7Wireless Transmission Issues
- Line-of-sight (LOS) transmission attenuation
- Power falls off as d 2
- Multiple paths lead to reflection and scattering
- Power falls off as d 4 beyond a certain distance
- Receiver may fail to discriminate valid signals
due to - Interference from other Tx
- Noise (internal to the Rx)
8Characteristics That Affect Wireless Sensor
Network Design
Characteristic Implication
Severe power constraint Network protocols must take power into consideration
Number of nodes could be in the thousands, scattered in any order Network topology and routing not pre-engineered protocols must handle establishing the network
Radio consumes large amounts of power when on Keep radio off whenever feasible
Transmit power loss rises as d2 in close, as d4 further out Minimize number (and duty cycle) of distant nodes transmitting to base station
9Authors Design Assumptions
- All nodes can transmit with enough power to reach
the base station if needed - Individual nodes can adjust the amount of
transmit power - Each node has sufficient computational power to
support different MAC protocols, perform signal
processing, etc. - Nodes always have data to send
- Nodes that are sufficiently close have correlated
data
10Consider 3 Different Network Protocols To Clarify
Concepts
- Simple network -- every node talks directly to
the base station - Minimum Transfer Energy (MTE) nodes minimize
transmit distance energy loss - Static Cluster group nodes spatially, aggregate
data, and assign one node to handle
communications with the base station
11Network Scenario -- Simple
- All nodes communicate directly with the base
station always on
- Problems
- Who talks next?
- Out of power fast!
- Especially distant nodes
Recall Signal strength is inversely
proportional to the square of the distance
(best case)
Base Station
12Network Scenario MTE
- Each node discovers the best hop-by-hop path to
the base station during an initialization phase
Problem close-in nodes overused
- Data transmitted every tdelay seconds
- Collision avoidance via CSMA protocol
- As nodes run out of energy, routes are recomputed
to maintain connection to base station - Problems
Base Station
Problem multiple hops --gt latency
13Network Scenario -- Static Clustering
- Organize nodes into clusters
Base Station
14Summary of Previous Protocols
Issue Simple MTE Static Cluster
Rate of energy Consumption Very Rapid Rapid Slower (but cluster head failure is an issue)
Control of Access Not addressed CSMA TDMA
Impact from Loss of 1 Node One node lost One node lost (after rerouting) Entire cluster lost (if a cluster head)
Latency One hop Multiple hops Two hops
Less Energy Consumption
15Low-Energy Adaptive Clustering Hierarchy (LEACH)
Protocol
- Structure of rounds occurring over time
- Nodes organize themselves into local clusters
- One node acts as the cluster head
- Member nodes transmit data to the cluster head
during the timeslot allocated by a TDMA algorithm - Cluster head aggregates the data from the member
nodes (e.g., computes mean value) - Cluster head transmits aggregated data to base
station - Repeat until time to begin a new round
16Cluster Behavior During a Round
Round
- Each member node (in turn) transmits their data
to the cluster head during the assigned timeslot
- Cluster head processes the data
- Cluster head transmits to base station
Base Station
17Clusters Reform Periodically
- Each round consists of a setup period and some
number of frames
Round 1
- Each round establishes a new structure of clusters
- Each cluster has a new cluster head
Round 2
- Repeat rounds until the network fails (due to
energy depletion)
18Now For Some Details
- Cluster head selection algorithms
- Cluster formation algorithm
- Steady-state phase
- Alternate scheme LAECH-C
19Cluster Head Selection Algorithms
- Need distributed Algorithms
- Desired results
- Specified number of cluster heads formed for each
round - Cluster head duties rotated among nodes so as to
evenly draw power from the nodes over time (no
overly-utilized nodes) - Case 1
- Nodes begin with equal energy
- All nodes transmit data during each frame
- Case 2
- Nodes begin with unequal energy, and/or
- Nodes transmit upon event
20Cluster Head Selection Case 1
- Each node elects itself to be a cluster head with
probability Pi(t) such that for N total network
nodes
Where k cluster heads
- To ensure that each node becomes a cluster head
only once in each of N/k rounds, assign Ci(t) 0
if the node has already been a cluster head in
the current round and Ci(t) 1 otherwise.
- Each individual node chooses to become a cluster
head in round r with probability
21Cluster Head Selection Case 1 (continued)
- Value of N - k(r mod N/k) represents the number
of unpicked nodes - Use of r mod N/k ensures restarting after all
nodes have been picked
Example
End of Round (r) Un-picked Nodes Pi(t) of remaining nodes
0 9 .22
1 7 .29
2 5 .40
3 3 .67
4 1 1.00
5 8 .25
N 9 k 2
22Cluster Head Selection Case 2
- Nodes with more energy should have a higher
probability of being chosen than nodes with less
energy
- Thus, the probability that a given node will be
chosen is determined by that nodes share of the
total remaining energy
- Where Ei(t) is the energy of the ith node and
23Cluster Head Selection Case 2 (continued)
- Notice that this algorithm requires each node to
know (or have an estimate for) the value of
Etotal(t) - To know the exact value would take time and
consume energy - As an estimate we could compute the average
energy of each node in a given cluster and
multiply by N - Nodes report current energy to cluster head
- Cluster head computes estimated Etotal(t) and
returns the value to all nodes in the cluster
24Distributed Cluster Formation
- Cluster heads broadcasts advertisement message
(ADV) using CSMA MAC protocol - Non-cluster head nodes measure received strength
of ADV and select strongest sender as their
cluster head - Nodes notify cluster head of their selection with
a Join-REQ message - Cluster head creates TDMA schedule for nodes in
its cluster
25Steady State Phase
- Recall Rounds are divided into frames
- Each node sends data once per frame
- TDMA requires accurate synchronization
- Possible method ? base station sends
synchronization signals - Energy saved at non-cluster head nodes since
- Power is reduced to only that required to reach
local cluster head - Radio turned off except for short period provided
by TDMA - Cluster head steady state tasks
- Listen to non-cluster head nodes
- Aggregate the data
- Transmit the data to the base station
26Steady State Phase (continued)
- Transmissions must succeed even though other
nodes and cluster heads are broadcasting - LEACH uses Direct-Sequence Spread Spectrum (DSSS)
- Each cluster uses a unique spreading code
- Chosen from a pre-defined list
- With enough spreading, potentially interfering
signals can be filtered out during de-correlation - Easier to implement than dynamically assigned
frequency bands - Difficulty is need for tight timing
synchronization - How does DSSS work? (Not addressed in this paper)
27DSSS Key Ideas
- A wireless transmission technology that enables
multiple users to share the same bandwidth
- Spreading Data signal is multiplied by a unique,
high rate code which spreads the bandwidth
before sending (1 data bit now represented by
many bits)
- The resulting Spread Spectrum is less
susceptible to interference
- Receiver must have the same code to recover the
original data
28LEACH-C -- A Variation of LEACH
- Idea using a central control algorithm may
produce better clusterings - Each node sends location information and energy
level to the base station - Base station
- Eliminates low energy nodes from consideration
- Finds k optimal clusters (since this is NP-hard,
uses the Simulated annealing algorithm ) - Goal is to minimize the total sum of squared
distances between non-cluster heads and the
nearest cluster head
29Performance Analysis
- Simulate the performance of four protocols
(Static Clustering, MTE, LEACH, LEACH-C) - How?
- Set up the simulation
- Find the optimal number of clusters
- Compare the protocols energy consumption
performance - Conclusions
30Simulation of Protocol Performance
- Analytical model of even moderately-sized
realistic networks is difficult - Authors used the network simulator ns
- Comparison of performance over four metrics
- System lifetime
- Energy dissipation
- Amount of data transferred
- Latency
31Experiment - Setup
Base Station
- 100 nodes randomly distributed over a 100 X 100
grid (0, 0) to (100, 100) - Base station placed outside the grid (50, 175)
- Channel bandwidth 1 Mb/s
- Packets have 25 byte header and 500 byte data
size - Power loss determined by distance d
Free space model
Multi-path model
32Experiment - Setup (continued)
- Radio energy dissipation model
- lEelec energy consumed by the electronics to
process l bits - l efs energy consumed by the amplifier to
transmit l bits over distance d where d lt do - l emp energy consumed by the amplifier to
transmit l bits over distance d where d gt do
- Then total energy consumed by the transmitter
- Total energy consumed by the receiver
33Experiment - Setup (continued)
- Energy parameters used
- Eelec 50 nJ/bit
- efs 10 pJ/bit/m2
- emp 0.0013pJ/bit/m4
- Energy for Data Aggregation EDA 5
nJ/bit/signal - Question How many clusters should be used?
34How Will The Number Of Clusters Affect Results?
- Case 1 Baseline (BL)
- S non-cluster head energy ENCHBL
- S Cluster head energy ECHBL
Case 1
- Case 2 Fewer Clusters (FC)
- ENCHFC gt ENCHBL
- ECHFC lt ECHBL
- Does Case 2 use less energy than case 1?
- Case 3 More Clusters (MC)
- ENCHMC lt ENCHBL
- ECHMC gt ECHBL
- Does Case 3 use less energy than case 1?
Is there an optimal number of clusters?
35Optimal Number Of Clusters
- With a given spatial distribution of nodes and
known energy consumption parameters, we can
compute an optimal number of cluster heads (k)
- Step 1 Develop expressions for node energy use
- Cluster heads (always on)
- (assumes dtoBS gt do)
- Non-cluster heads
- (assumes dtoCH lt do)
- Step 2 Develop an expression for the expected
squared distance from the nodes in a cluster to
the cluster head
36Optimum Number Of Clusters (continued)
- Step 2 (continued) Assumptions
- In an M x M grid, each cluster occupies an area
of M2/k - Clusters have a node distribution of p(x, y)
- The cluster head is at the center of mass of the
cluster
- Then the expected d2 from the nodes to the
cluster head is
(in Cartesian coordinates)
(in polar coordinates)
- Further assume the area is a circle ? radius R
(M/(pk)1/2) - And p(r, q) constant for r and q, then
37Optimum Number Of Clusters (continued)
- Step 2 (continued) Assumptions
- Node density is uniform across all clusters ? p
(1/( M2/k)) - Then simplifies to
- Step 3
- Combine energy and distance expression for
non-cluster heads
- Then for the entire cluster
- (During a single frame)
38Optimum Number Of Clusters (continued)
- Step 3 (continued)
- Total energy for a frame
- Step 4 Set derivative of Etotal with respect to
k to zero
Simulation results agree with analytical
prediction
- Results for this case (100 nodes, etc.)
Analytical method predicts 1 lt kopt lt 6
39Comparison of Algorithms
- Each node was given 2 Joules of energy (def J
Ws) - This is equivalent to a 5 volt device _at_ 20 mA for
20 s - Parameters tracked during simulations
- Rate at which data packets were transferred to
the BS - Energy required to get the data to the BS
- What is not in the simulation
- No static energy loss (e.g., RTC energy use)
- Energy for CSMA is ignored (? CSMA energy use in
MTE is understated) - Energy expended during cluster organization (not
mentioned in the paper)
40Simulation Results Data Received
40 more data for the same energy as LEACH
- LEACH-C and LEACH deliver far more data than MTE
and Static Clustering (SC) and they are far more
energy efficient (as measured by signals per
Joule) - SC fails when all cluster heads die, even with
most energy still unused - MTE slow to deliver data due to multi-hops
- LEACH-C is the best performer due to optimal
cluster design
41Simulation Results Nodes Alive
LEACH-C delivers more data due to higher data
rate/J
- LEACH-C and LEACH maintain full network
availability far longer than MTE and SC - MTE lasts the longest, but at the price of very
limited effective data delivery due to - Lack of data aggregation
- Energy wasted in CSMA collisions
- LEACH-C is again the best performer
42Conclusions
- Wireless Sensor Networks which meet the original
assumptions will benefit from - Rotating the cluster head position among all
nodes - Adapting cluster organization to new cluster
heads - Aggregating data
- Disadvantages
- LEACH LEACH-C are very dependent on nodes
having correlated data - Both require tight time synchronization
- LEACH-C requires location information
43Future Work
- If nodes send data on condition
- They can be on standby for longer periods than
TDMA permits - Efficient bandwidth use will require a different
communication protocol - If nodes are beyond max possible communication
range - Multi-hop protocols may be required
- Super cluster heads may prove a better solution
- If the original cluster is kept and the nodes
within the existing clusters just rotate the
cluster head job - No setup overhead is used after round one
- Downside ? nodes may expend more energy
communicating since current cluster head may be
far away