Taming the Underlying Challenges of Reliable Multihop Routing in Sensor Networks By Byron E. Thornton Objective We now begin to build a Wireless Sensor Network model that better captures the operational characteristics of the real world. Introduction A More Practical Model A brief recap of last week s RF Irregularities discussion and the Radio Irregularity Model (RIM). Spherical radio energy model. Non-Isotropic Asymmetric links The model used in many WSN simulations do not resemble real-world behavior. Introduction A More Practical Model Other components may also possess operational characteristics that differ drastically from the simulation model. Our goal is to develop a more representative model. Before We Begin Other Neglected Aspects of Sensor Networks Noise, Distortion and Interference. Transmission Bandwidth and Modulation Frequency. Signal to Noise Ratio (SNR) and probability bit error P e. Digital Quantization. Coherent encoding schemes (PSK, FSK, OOK, etc.) and how they might affect WSN. Modulation Schemes (ex. PCM, PAM) What affect might it have? Introduction The Issue We now realize that the simulation model does not in all cases adequately reflect real world conditions. Suppose the assumptions we make about routing and overall route quality are also invalid. How can we develop a more accurate routing model? 1
Items Covered Link Estimator. What is it? How is it used?. How to build and maintain a good subset of reliable links for routing. Routing Protocols A look at routing as it relates to the Estimator. A Look at Making Routing Decisions We ve assumed that choosing a route was a complicated process, but with a simple formula. But as per last week s RF Irregularities we now realize that non-isotropic-ness in Wireless Sensor networks is quite common. Take a better look at making routing decisions Up until now we ve allowed route quality judgement decisions to be made by the Routing Protocol strictly at the routing layer. Developing A Quality Metric An empirical look at network neighbors. What is the best criteria for judging the quality of a neighbor connection? Creating a list of quality neighbors a priori can greatly improve the ability of the routing protocol to make good routing decisions. Line of Discussion Insertion, eviction and reinforcement. An efficient neighborhood management algorithm. Routing Protocol We examine the routing protocol with respect to how it works with the Estimator. Evaluation. A comparison of various Distance Vector Protocols with Link Estimation. Empirical Observation Defining Link Quality Through Observation We start by developing an empirical model to derive a suitable metric for link quality. Our Approach: Design a network of mote sensors, collect test data and make some observations. 2
Empirical Analysis A good link quality metric should be: Statistically Stable. Empirically accurate and verifiable. Reproducible. Estimator compatible. Empirical Analysis The Experimental Model 200 nodes linearly configured with 2-foot spacing on an outdoor tennis court. Each node transmits 200 packets at a given power level (initially 50) at a rate of 8 packets/sec. Only one node transmits while all others attempt to receive. Receiving nodes count the number of packets received from the sender. Afterwards, the power is increased and the process rerun. Empirical Model Sensor Grid with Single Sender The Results of RF Power Varying and Sampling at various distances. Each node is 2 feet apart. Each node transmits 200 packets at 8 packets/s. Only one active transmitter at a time. All the rest act as receivers. Objective is to concoct a sampling space where both mean link quality and the variance in link quality are functions of the distance from sender to receiver. Key Observations For a given power setting there is a distance in which essentially all nodes have good connectivity. Increasing the transmitter power increases the effective region. As expected, the effective region is a function of the RF transmitter power. But what about the Transitional Region? Did the size of the TR increase with increasing RF transmitter power? Stability Before our observation can be acceptable we must ensure that the link quality is stable when the nodes are immobile over time. We fix a sender and observe affects on the receivers over a long period. 3
Stability Test Experiment Transmitting node positioned 15 feet from receiver in an indoor location. Nodes transmitted at 8 packets/sec for 20 minutes. Afterwards, sender is moved from 15 ft to 8 ft and transmission continues for 4 more hours. Conclusion Distance is a good representative metric for link quality. Stable Verifiable. Reproducible. We can use it as an estimator metric. Estimator Based on our empirical model we now proceed to design an algorithm that will be used to judge good link candidates. For each link record observed Associate a quality factor (or link probability) based on the mean and variance extracted from the empirical data. Use this link probability to drive the Estimator. About EWMA Estimators A maximum likelihood estimation scheme. Employs exponential data weighting. Widely used in many different areas and applications. Examples: Round Trip Time (RTT) for TCP congestion control. A Low Pass Filter Algorithm. Assumes conditionally normal collections. About EWMA Estimators (cont) Weighting factor determines the level of agility of the estimator in following abrupt changes in the data. Problem! Too much emphasis placed on extreme returns. 4
Exponentially Weighted Moving Average (EWMA) Estimator ( ω ) ek = ek where + ωo e = new estimation, a current average at sample k k ω = weighting factor o = current observation k 1 1 k Estimator Requirements Quick reaction to large changes in link quality. Stable. Small memory footprint. Simple to compute. Estimator with other components Estimator Collecting Link Statistics Broadcast and snooping used to collect link candidates. Estimator responsible for evaluating link candidates collected. But promiscuous packet sniffing is expensive. More cost effective methods may need to be investigated. Responsible for maintaining table of viable links as determined by the Estimator. Seeks to develop and maintain a table subset of neighbor links of the highest possible quality. Rationale Nodes have limited storage, especially for overhead. So efficient management of space is an absolute necessity. Information about reliable links needs to be known beforehand to improve overall network performance. Network volatility requires constant reassessment of routing choices. 5
Additional Considerations Less used links should be less valued. Higher rating for more frequently used. Also, a higher rating should be given to proven reliable links. Functions Eviction Removes records that are obsolete or to make room for a better candidate. Insertion Adds new candidate based on Estimator evaluation. Reinforcement Increases rating for proven entries. Key Design Points Neighbor table finite and limited. Only the best quality links should be collected. Periodic review of entries need to be conducted to ensure viability. Periodic messages (i.e., beacons) should be used versus non-control type packets. Frequent insertions, updates and deletions can seriously impact performance of the overall system. Eviction and Reinforcement Strategies Round Robin with no Reinforcement. FIFO LRH -- least recently heard. CLOCK Algorithm. FREQUENCY Algorithm. Least Frequently Used. Neighbor Manager Assessment Fixed-size table as cell density increases 1st 2nd 3rd # Good neighbors > Table size Freq always maintains 50% or more good neighbors in table Routing Protocol Estimation provides quality links to Routing Protocol prior to the route acquisition phase. Quality link selection improves overall route selection. 40 Number of Potential Neighbors 6
Routing Manager Parent Selection Routing Cost Metrics Packets snooped by the Estimator for viable neighbor link candidates. Neighbor Table manager maintains the neighbor table. Periodic updates by Parent Selection. Spawns periodic parent selection Protocol Evaluation Shortest Path (SP and SP(t)) Minimum Transmission (MT) Broadcast Destination Sequenced Distance Vector (DSDV) Evaluation Graph Analysis 400 Nodes organized as 20X20 grid, 8 ft spacing. Sink placed at corner for max. network depth. Routing Protocol Evaluation Measures routing depth of nodes throughout the network. Reflects end-to-end latency and energy usage Measures the product of link quality along the path from each node in the network. Approximated end-to-end reliability of a routing path. 7