TL;DR
This paper introduces DEUDC, a novel method for estimating distances in wireless sensor networks (WSNs) by leveraging interval data clustering to handle the inherent uncertainties in Received Signal Strength Indicator (RSSI) readings. By modeling RSSI as interval data and employing both hard and soft clustering techniques, DEUDC achieves improved accuracy and efficiency compared to existing methods, particularly in environments with high signal variability.
What Problem Does This Paper Address?
Accurate distance estimation is crucial for many Wireless Sensor Network (WSN) applications, including localization, tracking, and object identification. The most common approach for distance estimation in WSNs relies on the Received Signal Strength Indicator (RSSI), which is assumed to have a linear relationship with the logarithm of the communication distance. However, this assumption breaks down in real-world scenarios due to various environmental factors like obstacles, multipath propagation, and interference, which introduce significant uncertainties and noise into RSSI measurements. These uncertainties can lead to substantial errors (up to 50% or worse) in distance estimations if not properly addressed. Existing methods like Kalman filters and particle filters require accurate system models and can be computationally expensive, making them less suitable for resource-constrained WSNs. This paper aims to tackle the challenge of RSSI uncertainty to achieve more reliable and efficient distance estimation in WSNs.
Methods
The proposed method, Distance Estimation using Uncertain Data Clustering (DEUDC), consists of two main phases:
- Off-line Environment Measurement:
- RSSI Sample Measurement: RSSI values are collected at various known communication distances within different representative environments (corridor, hall, open air). For each distance, approximately 150 RSSI samples are taken to capture the signal's behavior.
- Statistical Computation and Cluster Center Representation: The collected RSSI data for each distance is statistically analyzed to compute the mean ($\mu$) and standard deviation ($\sigma$). This statistical information is then used to define a "cluster center" as an interval data range: $[\mu - k \times \sigma, \mu + k \times \sigma]$, where $k$ is a coverage factor. These cluster centers represent the distribution characteristics of RSSI values for a given distance.
- On-line Distance Estimation:
- Distance Calculation between RSSI Value and Cluster Center: For a new, unknown RSSI value ($Y$), the distance between $Y$ and each cluster center ($c_j$) is calculated. This distance is represented as an interval data, $D_j(c_j, Y) = [\lambda \times d_{j,min} + (1-\lambda) \times d_{j,max}]$, where $\lambda$ is a correlation factor that balances the minimum and maximum distance extremes.
- Uncertain Data Clustering: * Hard Clustering (DEUDHC): The RSSI value $Y$ is assigned to the cluster center with the minimum calculated distance. The communication distance associated with that closest cluster center is then taken as the estimated distance.
- Soft Clustering (DEUDSC): Fuzzy C-Means (FCM) is used to determine the membership degrees of $Y$ to the top three closest cluster centers. The final distance estimate is a weighted sum of the distances to these three centers, weighted by their respective membership degrees. This approach accounts for the possibility of an RSSI value belonging to multiple clusters to some extent.
- Efficiency Improvement (Micro-Cluster Based Clustering): To further enhance efficiency, the UK-means algorithm is applied to group the initial cluster centers into "macro-clusters." This two-level clustering (macro-cluster then micro-cluster) reduces the search space for the closest cluster center.
Datasets
The paper uses real RSSI measurements collected in three distinct wireless environments: a corridor, a hall, and an open-air environment. These measurements form the basis for both training (creating the RSSI-D mapping models) and evaluation. No pre-existing public datasets are explicitly mentioned or used.
Key Insights and Ideas
- Interval Data for RSSI Uncertainty: Representing RSSI values as interval data $[\mu - k \times \sigma, \mu + k \times \sigma]$ is a novel way to explicitly model and handle the inherent uncertainty and variability in RSSI measurements.
- Distribution-Based Mapping: Instead of relying on sample-based mapping (like in RADAR or ARIADNE), DEUDC uses distribution-based mapping derived from statistical properties of RSSI, which is more robust to noise and uncertainty.
- Hard vs. Soft Clustering for Different Uncertainty Levels: The paper proposes both hard and soft clustering approaches. Hard clustering is simpler but can be less accurate in highly uncertain environments, while soft clustering (FCM) offers better accuracy by considering partial memberships, especially in environments with high RSSI variability (e.g., corridors, halls).
- Micro-Clustering for Efficiency: The introduction of a hierarchical clustering approach (macro-clusters and micro-clusters) significantly improves the computational efficiency of the distance estimation process without a substantial loss in accuracy.
Key Results
- Accuracy Improvement: DEUDC (both DEUDHC and DEUDSC) significantly improves RSSI-D estimation accuracy compared to methods like LSLF, SRLF, BP, and LS-SVM, achieving accuracy improvements of 11.31% to 72.15% on average across the tested environments.
- Environmental Impact: The accuracy of RSSI-D estimation is heavily influenced by the environment. Complex environments with high uncertainty (corridor, hall) show greater improvement with DEUDC, while open-air environments with low uncertainty see less dramatic but still positive gains.
- Efficiency Gains: DEUDC demonstrates superior estimation efficiency. DEUDHC improves efficiency by 98.59% to 99.97% compared to SRLF, BP, and LS-SVM. DEUDSC offers improvements of 85.80% to 99.87%. The modeling and estimation processes in DEUDC are simpler and faster.
- Parameter Sensitivity: The correlation factor ($\lambda$) has a minimal impact on estimation accuracy, allowing for flexible selection. The coverage factor ($k$) is more critical and needs to be tuned based on the level of RSSI uncertainty in the environment, with smaller values generally preferred in less uncertain environments.
Limitations and Future Work
- Off-line Measurement Requirement: The current DEUDC method requires an off-line measurement phase to build the RSSI-D mapping models. This can be a bottleneck if wireless measurements are not feasible or if the environment changes frequently.
- Adaptive Estimation: The authors suggest exploring adaptive WSN RSSI-D estimation methods for future work. These methods would aim to update model parameters iteratively in real-time, potentially using maximum likelihood or least-square techniques.
- Computational Costs for Adaptive Methods: Acknowledged challenges for adaptive methods include higher computation costs and energy consumption, which need to be addressed.
- Distributed Implementation: While the current work uses a central processing strategy, the authors note that the distance estimation method can be implemented in a distributed manner within WSNs, with each unknown node performing its own estimation.
Summary updated