TL;DR
This paper introduces an Adaptive Genetic Particle Swarm Optimization (AGPSO) algorithm to improve the speed and stability of integer ambiguity resolution in single-frequency GNSS positioning. By combining genetic algorithm operators with PSO and an enhanced fitness function, AGPSO demonstrates faster convergence and more reliable results compared to traditional methods, achieving centimeter-level accuracy for short baselines.
What Problem Does This Paper Address?
Accurate positioning with Global Navigation Satellite Systems (GNSS) heavily relies on carrier phase measurements. However, these measurements contain an unknown integer ambiguity that must be correctly resolved for precise positioning. Traditional methods like LAMBDA, while effective, can be computationally intensive and struggle with convergence speed and stability, especially in complex scenarios or with high-dimensional ambiguity vectors. This paper tackles the challenge of efficiently and reliably resolving these integer ambiguities, particularly for single-frequency GNSS systems, which are more susceptible to errors.
Methods
The core of this paper is the proposed Adaptive Genetic Particle Swarm Optimization (AGPSO) algorithm. Here's a breakdown of its components:
- Foundation: It starts with the standard GNSS carrier-phase double-difference model to obtain floating-point solutions and their covariance matrices.
- Decorrelation: An inverse integer Cholesky algorithm is used to reduce the correlation in the ambiguity covariance matrix, which is crucial for efficient search.
- Enhanced Fitness Function: A novel fitness function is introduced to improve convergence and search performance. This function balances the accuracy of the fit (minimizing the residual term) with a logarithmic term that smooths the objective function, helping to avoid local optima.
- Hybrid Optimization: AGPSO integrates operators from the Genetic Algorithm (GA) into the Particle Swarm Optimization (PSO) framework:
- Selection: A meritocratic approach with "half selection" identifies superior individuals based on fitness.
- Crossover: Random crossover points are used to generate new individuals, increasing the chance of escaping local optima.
- Mutation: A mutation factor is introduced to further aid in escaping local optima.
- Adaptive Weights: The algorithm employs adaptive weights for the PSO component, which gradually decrease during iteration, refining the search range as the population matures.
Datasets
The paper uses simulated data for numerical analysis and real-world data for experimental validation.
- Simulation Data: Used to compare AGPSO against Simulated Annealing (SA), Genetic Algorithm (GA), and standard Particle Swarm Optimization (PSO) for 3D and 12D integer ambiguity resolution.
- Real-World Data: Collected from two satellite receivers forming a short baseline (1.42m) and a longer baseline (263.52m) at Nanjing Forestry University. This data was used to evaluate the algorithm's performance in practical short-baseline scenarios.
Key Insights and Ideas
- The Power of Hybridization: Combining the exploration capabilities of GA (selection, crossover, mutation) with the exploitation strengths of PSO (collaboration, information sharing) creates a more robust optimization framework.
- Fitness Function Matters: A carefully designed fitness function is critical. The proposed function balances fitting accuracy with robustness against local optima, leading to better search performance.
- Escaping Local Optima: The AGPSO's mechanisms (adaptive weights, GA operators, improved fitness) are specifically designed to prevent the algorithm from getting stuck in suboptimal solutions, a common problem in optimization.
- Single-Frequency Viability: The research demonstrates that even with single-frequency GNSS data, which is more prone to errors, the AGPSO can achieve reliable integer ambiguity resolution and centimeter-level accuracy for short baselines.
Key Results
- Faster Convergence: In simulations, AGPSO converged to the optimal solution significantly faster than SA, GA, and PSO. For 3D ambiguity, AGPSO took about 5 iterations on average, compared to ~10 for PSO. For 12D ambiguity, AGPSO averaged ~25 iterations, while PSO required ~50.
- Improved Stability: AGPSO showed better stability in resolving integer ambiguities, especially in high-dimensional cases, and was less prone to falling into local optima than standard PSO.
- Practical Accuracy: In real-world experiments with a 1.42m baseline, the AGPSO algorithm achieved a baseline accuracy within 0.02m.
- Comparable Success Rate: For a longer baseline (263.52m), AGPSO achieved a success rate of 95.88%, very close to the traditional LAMBDA algorithm's 96.15%, indicating high reliability.
Limitations and Future Work
- Short Baseline Jumps: The paper notes that in the initial short-baseline (1.42m) experiment, there were still some "jumps" in the position results, possibly due to multipath effects.
- Future Directions: The authors suggest that future work could focus on reducing these jumps by improving receiver quality, increasing the number of visible satellites, and mitigating obstacle blockage. They also imply that the AGPSO algorithm's effectiveness in handling complex scenarios and its potential for further refinement warrant continued investigation.
Summary updated