Mini Project Worksheet: Dipole Antenna Array Optimization
Course: EE5606 Artificial Intelligence for Antennas in Wireless Communication
Report Submission Deadline: Week 12 (at the commencement of the lecture)
Project Background
Design a linear half-wave dipole antenna array that has a main beam direction of x as shown in Fig. 1. Traditional full-wave simulations (e.g., HFSS, CST) are computationally expensive. For simplicity, no full-wave simulations are required in this project. In this project, you will use an array factor formula to generate training data, train a fully-connected neural network (FCNN) to obtain a surrogate model, and then use an optimization algorithm to find an optimal array configuration.
Part I: Data Generation and FCNN Training
Fig. 1
Design Goal:
Generate a dataset using the array factor formula for a linear half-wave dipole antenna array, and train a FCNN to obtain a surrogate model that replaces the array factor formula.
Parameters:
1. Operating frequency: f = 2.4 GHz
2. Wavelength: λ = c/f
3. The radius of each dipole element is b = 0.5 mm
4. Number of elements: N = 32 for the entire project.
Design Variables:
1. Element spacing d (in terms of λ)
2. Progressive phase shift α (in radians) between elements
Electric Field Vector:
The electric field vector generated by the antenna array at a point in the far field.
where k = 2π/λ is the wavenumber in vacuum. For a half-wave dipole antenna, the element pattern he(θ) is given by:
where:
1. the term accounts for the spherical wave propagation, which can be regarded as a constant in this project.
2. he (θ) characterizes the radiation pattern of a single half-wave dipole element.
Analytical Formula – Array Factor:
The array factor for a linear array along the z-axis is given by:
where:
1. an = 1 (i.e., assumed uniform excitation),
2. X is the angle from the x-axis, where cos X = sinθ cosφ .
Task 1:
Train a FCNN to predict the value of |E| for a given data input of (d, α , θ, φ), with d ∈ [0.3λ,0.8λ] , α ∈ [-π, π] , θ ∈ [0°, 180°] , and φ ∈ [0°, 180°] . This FCNN (surrogate model) is used to replace the array factor formula.
Task 2:
By using the FCNN (surrogate model) and an optimization algorithm (Genetic Algorithm or PSO), find the optimized (d, α) that gives the maximum |E| at (θ = 59.8o, φ = 30.3o).
Submission Requirements
Upload your ZIP file to “Assignments” in Canvas. The ZIP file should contain:
1. Code: Provide comprehensive comments for each part (Data generation, FCNN training, Optimization algorithm).
2. Brief Report: including methodology, results (e.g., training and validation errors of FCNN, convergence curve of optimization algorithm), discussion, and conclusion (no more than 4 pages, single line spacing, New Times Romance, font 12, 1-inch margin from top, bottom, left, and right).
3. Demo video: including the processes ofFCNN training and optimization.
4. Note: In your presentation slides, use diagrams, plots, figures, etc. as far as possible. Avoid long paragraphs and keep the text as concise as possible.
5. Presentation: (10 min + 2 min Q&A)
Suggestion:
(1) Use flowcharts to explain your algorithm workflows.
(2) Include figures to visualize algorithm performance (e.g., convergence curves, prediction error). Grading Criteria
|
Category
|
Weighting
|
|
|
Code Quality
|
25%
|
Readability, modularity, and documentation
|
|
Algorithm Implementation
|
25%
|
Correctness of optimization and FCNN
|
|
Report
|
25%
|
Organization, conciseness, and analysis
|
|
Presentation
|
25%
|
Clarity and time management
|