Abstract:Localization based on received signal strength (RSS) has drawn great interest in the wireless sensor network (WSN). In this paper, we investigate the RSS-based multi-sources localization problem with unknown transmitted power under shadow fading. The log-normal shadowing effect is approximated through Fenton-Wilkinson (F-W) method and maximum likelihood estimation is adopted to optimize the RSS-based multiple sources localization problem. Moreover, we exploit a sparse recovery and weighted average of candidates (SR-WAC) based method to set up an initiation, which can efficiently approach a superior local optimal solution. It is shown from the simulation results that the proposed method has a much higher localization accuracy and outperforms the other
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the multi - source localization problem based on Received Signal Strength (RSS) in an environment with log - normal shadow fading, especially when the transmit power is unknown. Specifically, the paper focuses on how to estimate the positions of multiple target sources through RSS measurements in a Wireless Sensor Network (WSN), while taking into account the influence of shadow fading. This is a multiplicative noise caused by obstacles in the environment and has a significant impact on RSS - based localization. The paper proposes a method based on Maximum Likelihood Estimation (MLE) and combines Sparse Recovery and Candidate Weighted Averaging (SR - WAC) techniques to improve localization accuracy and algorithm performance.
### Main Contributions
1. **Proposing a probability - based maximum likelihood estimation algorithm**: Different from the previous Minimum Mean Square Error (MMSE) method, this paper proposes a probability - based algorithm based on maximum likelihood estimation. It transforms it into an optimization problem and uses the Gradient Projection (GP) method to solve the non - convex problem, thereby obtaining the positions of the sources.
2. **Proposing a new method for determining the initial point**: Since the optimization problem is nonlinear and non - convex and has multiple local optimal solutions, this paper proposes a new method based on Sparse Recovery and Candidate Weighted Averaging (SR - WAC) to approach a better local optimal solution, solving the problem that the gradient descent method is sensitive to the initial value.
### Technical Details
- **System Model**: In the log - normal shadow fading environment, the RSS measurement model is defined, taking into account the signal superposition of multiple co - channel target sources.
- **Maximum Likelihood Estimation**: Using the Fenton - Wilkinson (F - W) method to approximate the Sum of Log - normal (SLN) distribution, a probability - based maximum likelihood estimation method is derived.
- **Sparse Recovery and Candidate Weighted Averaging**: Through the Compressive Sensing theory, sparse recovery is first carried out, and then the k - means clustering and weighted average rules are used to determine the initial point.
### Simulation Results
- **Performance Evaluation**: Through numerical simulation, the performance of the proposed method is compared with that of existing methods (such as MMSE) under different shadow fading intensities. The results show that the proposed method is significantly superior to other methods in terms of localization accuracy, especially in the case of strong shadow fading.
### Formula Presentation
- **RSS Measurement Model**:
\[
r_m=\sum_{k = 1}^{K}c_0P_kd_{mk}^{-\alpha}10^{\frac{n_{mk}}{10}}, \quad m = 1,2,\ldots,M
\]
where \(P_k\) is the transmit power of the \(k\)-th target source, \(d_{mk}\) is the Euclidean distance from the \(m\)-th sensor node to the \(k\)-th target source, and \(n_{mk}\) is a zero - mean Gaussian random variable representing log - normal shadow fading.
- **Objective Function of Maximum Likelihood Estimation**:
\[
\min_{u_k,v_k,P_k}\sum_{m = 1}^{M}\left(\ln(\sigma_m^2)+\frac{(\ln(\hat{r}_m)-\mu_m)^2}{\sigma_m^2}\right)
\]
- **Optimization Problem of Sparse Recovery**:
\[
\min_s\frac{1}{2}\|\hat{r}-\Phi s\|^2_2+\lambda\|s\|_1
\]
where \(\Phi\) is the energy propagation matrix and \(s\) is the sparse representation vector.
These technical details and formula presentations show the innovation and contribution of the paper in solving the multi - source localization problem.