PaperlayerView exact revision on arXiv

Source-linked AI summary

RSSI-Based Distributed Self-Localization for Wireless Sensor Networks used in Precision Agriculture

Pooyan Abouzar, David G. Michelson, Maziyar Hamdi

arXiv:1509.02400v1cs.DC

TL;DR

Large agricultural WSNs need localization that accommodates sparse landmarks and realistic deployment constraints. The paper proposes distributed RSSI-based Bayesian location updates, with the total messages needed for convergence growing slower than O(n).

  • Problem

    The paper targets localization in large agricultural WSNs where connectivity to landmarks and a small number of landmarks are relevant constraints.

    Source: §V · p. 20

  • Method

    The algorithm uses RSSI-based Bayesian information aggregation to update node locations recursively.

    Source: §I · p. 3

  • Results

    The total number of messages needed for the algorithm to converge grows slower than O(n).

    Source: B. Results · p. 19

  • Takeaways & Limitations

    The authors identify compatibility with realistic WSN deployment scenarios as the algorithm’s main strength.

    Source: §V · p. 20

Abstract

from arXiv · show

Node localization algorithms that can be easily integrated into deployed wireless sensor networks (WSNs) and which run seamlessly with proprietary lower layer communication protocols running on off-the-shelf modules can help operators of large farms and orchards avoid the difficulty, cost and/or time involved with manual or satellite-based node localization techniques. Even though the state-of-the-art node localization algorithms can achieve low error rates using distributed techniques such as belief propagation (BP), they are not well suited to WSNs deployed for precision agriculture applications with large number of nodes, few number of landmarks and lack real time update capability. The algorithm proposed here is designed for applications such as pest control and irrigation in large farms and orchards where greater power efficiency and scalability are required but location accuracy requirements are less demanding. Our algorithm uses received signal strength indicator (RSSI) values to estimate the distribution of distance between nodes then updates the location probability mass function (pmf) of nodes in a distributed manner. At every time step, the most recently communicated path loss samples and location prior pmf received from neighbouring nodes is sufficient for nodes with unknown location to update their location pmf. This renders the algorithm recursive, hence results in lower computational complexity at each time step. We propose a particular realization of the method in which only one node multicasts at each time step and neighbouring nodes update their location pmf conditioned on all communicated samples over previous time steps. This is highly compatible with realistic WSN deployments, e.g., ZigBee which are based upon the ad hoc on-demand distance vector (AODV) where nodes flood route request (RREQ) and route reply (RREP) packets in the network.

I. INTRODUCTION

The paper targets low-cost, scalable localization for agricultural WSNs and proposes a distributed RSSI-based Bayesian approach for static networks. Its recursive updates and compatibility with route discovery address deployment constraints that burden graph-based localization methods.

I. INTRODUCTION

  • Agricultural deployments need cost-effective, scalable, fast localization robust to seasonal changes, while their accuracy requirements are comparatively loose.Sensor locations matter for feature maps and resource management, but equipping every sensor with GPS is costly.Source: §I · p. 2
  • BP-based methods can fail to converge on loopy graphs and require global distance information, while spanning-tree formation can add at least O(n) multihop messages.These constraints make the methods less suited to large agricultural networks with many connected unknown nodes and a need for real-time updates.Source: §I · p. 3
  • The proposed RSSI-based method updates locations through single-hop communication without requiring spanning-tree construction or a designated starting node or landmark.The authors position it for DIY mesh networks using ZigBee or other proprietary protocols over IEEE802.15.4.Source: §I · p. 4
  • A recursive Bayesian update combines path-loss likelihood with the generating node’s latest location estimate, and constraints from dependent paths multiply.This avoids spanning-tree construction and intermediate-node tracking while using constraints from paths traversed by flooded RREQ packets.Source: §I · p. 4
  • Measurements in high-density apple orchards support a log-normal path-loss model for different transmitter and receiver antenna heights.The measurements were collected across three campaigns over two consecutive summer seasons.Source: §I · p. 4

II. THE LOCALIZATION PROBLEM AND PATH LOSS LIKELIHOOD FUNCTION

For precision-agriculture tasks where approximate sensor locations suffice, the paper formulates localization over a discrete field grid. Grid resolution adjusts precision, but gains flatten beyond a threshold.

II. THE LOCALIZATION PROBLEM AND PATH LOSS LIKELIHOOD FUNCTION

  • For applications such as pest control, approximate sensor locations can suffice to trigger relevant actuators.Source: §II · p. 5
  • The field is divided into square cells, and each unknown node is assigned the centroid of one cell.The number of grid cells controls adjustable localization precision.Source: §II · p. 5
  • Increasing grid resolution stops improving precision once it exceeds a threshold.Source: §II · p. 5

A. Problem Formulation

The formulation estimates which grid cell contains each node from inter-node communications. It represents path-loss observations accumulated over time and uses them to estimate node locations.

A. Problem Formulation

  • The field is modeled as equal-area cells, with each cell a possible location for an unknown node.Source: A. Problem Formulation · p. 5
  • The objective is to use inter-node communications to identify the grid cell containing each node.Source: A. Problem Formulation · p. 5
  • For each node, the formulation tracks path-loss samples collected from connected neighbours across time steps.The samples may be absent for a neighbour at a given time step.Source: A. Problem Formulation · p. 5
  • The location estimate at a later time is based on previously aggregated observations and a probability distribution over possible locations.Source: A. Problem Formulation · pp. 5–6

B. A Representative Path Loss Model for Orchard Environments

The paper develops an orchard path-loss model from measurements across canopy modes, directions, and locations. It uses a log-distance model to represent path loss in vegetated environments.

B. A Representative Path Loss Model for Orchard Environments

  • Measurements covered standard- and high-density apple orchards, five directions relative to tree rows, and transmitter heights below and above the canopy.The receiver was at 2.5 m; the transmitter was at 2.5 m or 4 m, with samples collected over roughly 0–100 m.Source: B. A Representative Path Loss Model… · pp. 6–7
  • The path-loss model uses a log-distance form with a zero-mean normal random variable and parameters for reference-distance loss and path-loss exponent.Source: B. A Representative Path Loss Model… · p. 6
  • The measurement setup reflects deployments with above-canopy gateways and sensors or actuators inside the canopy; GPS-equipped gateways serve as landmarks.Source: B. A Representative Path Loss Model… · p. 6
  • The measurements came from three campaigns spanning two summer seasons, and the paper reports model-fit statistics and parameter confidence intervals.The samples were collected at nine orchard locations; Figure 3 contrasts the two transmitter-height modes.Source: B. A Representative Path Loss Model… · pp. 6–7

C. Path Loss Likelihood Function

The paper derives a path-loss likelihood relating observations to candidate node locations. It then uses Bayesian information aggregation to update location probabilities recursively, including a realization aligned with AODV route discovery.

C. Path Loss Likelihood Function

  • The likelihood function connects path-loss values to inter-node distances for use in the localization algorithm.The model derives the conditional likelihood from candidate node locations and their corresponding distances.Source: C. Path Loss Likelihood Function · pp. 7, 9
  • The likelihood is approximated by sampling the normal distribution amplitude at 1 dB steps and normalizing the values to sum to one.Source: C. Path Loss Likelihood Function · p. 9

III. LOCALIZATION ALGORITHM FOR PRECISION AGRICULTURE APPLICATIONS

  • Bayesian information aggregation updates the location probability mass function when new RSSI evidence arrives, using a recursive expression.Source: §III · p. 9
  • A special realization, simulated in the paper, is designed to match AODV route discovery, with one node multicasting at each time step.The general formulation also allows arbitrary packets exchanged between one or more node pairs per step.Source: §III · p. 9

A. General Case

The general-case method recursively updates a node’s location pmf using recent path-loss samples and neighboring nodes’ priors. Its assumptions support local updates, while its communication design limits per-node overhead.

A. General Case

  • Measurements found shadowing correlation below 0.1 between links in the vegetated environment studied, with links around 50 m for pest-management applications.The authors report that their measurements verify the assumption in (8).Source: A. General Case · p. 10
  • Each update uses neighboring-node priors and recent path-loss samples to recursively update a node’s location posterior.The required channel information and priors are combined with new observations; the update relies on stated conditional-independence assumptions.Source: A. General Case · pp. 9–11
  • Per node, computational complexity is O(n) and communication complexity is O(1); nodes communicate only with single-hop neighbors.The authors state that computational complexity matches BP-based techniques, while communication overhead is significantly lower.Source: A. General Case · p. 11

B. Localization Algorithm Compatible with Wireless Sensor Networks

The specific realization has one node multicast at each time step, using route-discovery traffic to update neighboring nodes. Its payload constraints motivate compression, and its supported application scope is pest management and mating disruption.

B. Localization Algorithm Compatible with Wireless Sensor Networks

  • At each time step, only one node multicasts, and connected nodes update their location posteriors from observed path loss.This realization is presented as compatible with AODV-based ZigBee route discovery and its RREQ/RREP traffic.Source: B. Localization Algorithm Compatible… · p. 11
  • The 102-byte maximum payload imposed by PHY and MAC layers limits exchanged pmf-message resolution, potentially preventing the localization algorithm from achieving desired accuracy in large orchards.Exchanging multiple packets causes excessive power consumption in addition to delay.Source: B. Localization Algorithm Compatible… · pp. 11–12
  • DCT and 6-bit quantization achieve compression up to 8/1, corresponding to coverage of a 100 ha orchard at 7 nodes/ha for pest management.The passage identifies the application as pest management (mating disruption).Source: B. Localization Algorithm Compatible… · p. 12
  • The algorithm is mostly suited to pest-management and mating-disruption applications, where error from simplifying assumptions or a mistuned path-loss model is tolerated.The authors note that seasonal environmental variation makes global awareness of the path-loss model unrealistic.Source: B. Localization Algorithm Compatible… · p. 12

IV. PERFORMANCE EVALUATION OF THE LOCALIZATION ALGORITHM

The evaluation uses simulation scenarios to assess localization error under different deployments and communication settings. Its methodology focuses on landmark placement, orchard assumptions, and connectivity-related measures.

IV. PERFORMANCE EVALUATION OF THE LOCALIZATION ALGORITHM

  • The simulations include random and grid deployments on a square field and examine how average landmark and unknown-node degrees relate to accuracy.Node degree counts connected landmarks or unknown nodes, respectively.Source: §IV · p. 12
  • The study evaluates localization accuracy with 2DRMS: 2DRMS = r means 95% confidence that a location estimate falls within a circle of radius r around the actual node’s location.Location estimation is random due to the random nature of path-loss samples.Source: §IV · p. 13
  • The study does not optimize landmark locations; it explains the rationale for its adopted arrangements instead.The evaluation proceeds through simulation setup and numerical examples.Source: §IV · p. 13

A. Methodology

  • The authors place landmarks at borders or corners because these arrangements provide more information about unknown-node locations.A higher landmark degree does not necessarily improve accuracy: closely spaced distances with noisy path-loss estimates can make location estimates less accurate.Source: A. Methodology · pp. 14–15
  • The simulations model random node deployments in 6 ha and 20 ha orchards at densities of 3 nodes/ha and 7 nodes/ha; WSN simulations also use random and deterministic grid deployment on a square field.Source: §IV · p. 12, A. Methodology · p. 16
  • Nodes are considered connected when the probability of RSSI falling below receiver sensitivity is below 1%, or connectivity probability exceeds 99%.Maximum transmission distance is calculated using the measurement-based path-loss model and the assumed Synapse RF200 module.Source: A. Methodology · p. 17

B. Results

The simulations report convergence after a few multicasts and show that accuracy depends on node degree, landmark arrangement, and transmit power. The authors report acceptable pest-management accuracy in a 20 ha orchard with eight landmarks under the stated radio-module setup.

B. Results

  • With average landmark degree around 1.5, an average unknown-node degree of 8 yields the desired 2DRMS of approximately 20 m.The reported surface plot shows error dropping as average unknown-node degree increases.Source: B. Results · p. 18
  • The algorithm converges after a few multicasts, and its total message count grows slower than O(n).The authors contrast this with BP spanning-tree variants, which require at least O(n) messages to form the tree before iterative messaging; they describe their method as faster and less communication-energy-intensive at the expense of accuracy.Source: B. Results · p. 19
  • Six- and eight-landmark scenarios generally meet pest-management accuracy requirements, while soil-moisture sensing requires more landmarks or higher maximum transmit power.The authors also report that finer pmf resolution does not affect accuracy when cell dimensions already support the application’s inter-node distance.Source: B. Results · p. 19
  • At low transmit power, node density has greater influence on localization error; at higher power, error drops to a minimum once unknown-node degree exceeds the required threshold.For a fixed landmark degree, increasing transmit power raises average unknown-node degree.Source: B. Results · p. 20
  • Above 90% acceptable probability for pest-management requirements is reported in a 20 ha orchard with 8 landmarks using Synapse RF200 modules.The authors state that larger orchards need more landmarks to meet the average landmark-degree requirement.Source: B. Results · p. 20

V. CONCLUSION

Large agricultural WSNs pose localization challenges that existing approaches are described as ill-suited to address. The conclusion presents an RSSI-based distributed method intended to scale while fitting realistic deployments.

V. CONCLUSION

  • Large farms and orchards have limited landmark connectivity and loopy graphs, while existing methods are described as overly complex and unable to provide real-time updates.Connectivity is limited by excessive path loss and large fields; existing approaches require completing inter-node distance estimates before updating locations.Source: §V · p. 20
  • The algorithm addresses these constraints using local neighbor-distance estimates and few landmarks relative to the total number of nodes.Source: §V · p. 20
  • The field grid can be made coarser or finer according to the application, available processing power, and desired localization accuracy.The conclusion also identifies the transceiver-module microcontroller as a consideration.Source: §V · p. 20
  • A Bayesian information-aggregation model divides computation across nodes and time steps, and location decisions can be made after stopping at any time step.The conclusion describes communication and computational complexity with respect to the number of nodes as scalable.Source: §V · p. 20
  • The method is designed for realistic WSN deployment with low added communication overhead, using multicast route discovery such as AODV's RREQ/RREP behavior.The conclusion describes integration as convenient and inexpensive.Source: §V · p. 20

Algorithm 1 Localization Algorithm For Agricultural Environments

Algorithm 1 initializes landmark and unknown-node locations, then uses landmark advertisements and multicast routing for distributed updates. It returns landmark replies and makes location decisions after M time steps.

Algorithm 1 Localization Algorithm For Agricultural Environments

Loading 1509.02400v1…