This blog post details the design, construction, and successful deployment of a homemade, low-cost, polarimetric synthetic aperture radar (SAR) system mounted on a 7-inch FPV drone. The project's primary goal was to achieve high-resolution aerial imaging on a minimal budget, contrasting sharply with expensive commercial and academic systems.
The author outlines the key design choices, opting for a 6 GHz Frequency-Modulated Continuous-Wave (FMCW) radar architecture for its efficiency in short-range applications. The system is controlled by a Zynq FPGA and features custom-designed, dual-polarized stacked patch antennas with sheet metal horns, built on inexpensive FR4 PCB material. To overcome the positioning inaccuracies of standard GPS, a crucial innovation of the project is a sophisticated image processing pipeline. This pipeline uses a GPU-accelerated backprojection algorithm for rapid image formation and a novel, gradient-based minimum entropy autofocus algorithm to computationally correct for flight path errors, achieving sharp, detailed images without the need for expensive RTK-GPS.
The results demonstrate successful imaging at distances up to 1.5 km, capturing all four polarizations (HH, HV, VH, VV) to create color-composite images that differentiate terrain features. The post also showcases a "VideoSAR" capability, generated by flying an octagonal path. The entire system weighs under 1kg and was built for a fraction of the cost of comparable systems, with the author releasing the image formation software and radar schematics to the public.
Building a High-Performance SAR Drone on a Budget
Introduction: A Low-Cost Goal: The project's objective was to build a capable synthetic aperture radar drone for a personal budget, using inexpensive FPV drone parts (<€200) and aiming for a radar cost under €500, in contrast to systems costing over €15,000.
Radar design: FMCW Architecture: A 6 GHz Frequency-Modulated Continuous-Wave (FMCW) radar design was chosen over a pulse radar. FMCW offers a better signal-to-noise ratio for short-range applications and is simpler and cheaper to implement, making it ideal for a slow-moving drone platform.
RF design: Dual-Polarization: The system is designed to measure four polarizations (HH, HV, VH, VV) by using switches to control the transmit and receive antenna polarization. This allows for more detailed analysis of the imaged targets.
Link budget: Performance Estimation: Theoretical calculations predicted the radar could achieve a good signal-to-noise ratio for imaging up to 1-2 km, which was later confirmed in field tests.
FPGA: High-Speed Digital Control: A Zynq 7020 FPGA is used to manage the strict timing requirements and high data rates from the ADC (50 MHz). A custom open-source SD-card controller was implemented in the FPGA to overcome the data-writing speed limitations of the built-in processor.
PCB: Compact and Custom: A compact, 6-layer PCB (113 x 48 mm) was designed to fit the small drone frame. The author overcame a component sourcing restriction and fixed a design error using a small, custom-made interposer PCB.
Drone electronics: Autonomous Flight: The radar is mounted on a standard 7-inch FPV drone running ArduPilot firmware, which provides advanced autonomous mission planning and robust sensor fusion of GPS and IMU data.
Antennas: Novel and Inexpensive Design: Custom, dual-polarized antennas were designed using a stacked, slot-fed patch on FR4 PCB, augmented with a simple sheet metal horn. This novel design achieved wide bandwidth and a high gain of 10 dBi at a very low cost.
Mechanical design: Lightweight Integration: The entire system, including the drone, radar, and 3D-printed mounts, weighs just 752 grams without the battery, keeping it agile and efficient.
Image formation: GPU Acceleration: The computationally intensive backprojection algorithm is used for image formation. To make this practical, the author wrote a custom CUDA kernel, enabling a modern GPU to perform over 220 billion backprojections per second.
Autofocus: The Key to Sharp Images: A novel, PyTorch-based minimum entropy autofocus algorithm was developed. This software-based solution corrects for the positioning errors inherent in standard, non-RTK GPS, proving critical for achieving sharply focused images.
Measurements: Successful Field Tests: Initial tests flying a 500m straight line at 110m altitude produced blurry images using only GPS/IMU data. After applying the autofocus algorithm, the images became dramatically sharper, revealing fine details on the ground.
Full polarization measurement: Differentiating Targets: A subsequent flight captured all four polarizations. When combined into a color image (e.g., RGB = HV, HH, VV), different materials could be distinguished: forests appeared white (reflecting all polarizations), while ground and buildings showed a purple tint (reflecting HH and VV more strongly).
VideoSAR: Dynamic Imaging: By flying an octagonal path around a central point and processing the data in short, overlapping segments, a "VideoSAR" was created. This video shows how reflections from objects like bridges and power lines change based on the radar's viewing angle.
Conclusion: A Successful Project: The project successfully created a high-performance, polarimetric SAR drone for a total cost of around €800, plus development time. It is capable of imaging up to 1.5 km and demonstrates the power of advanced software to overcome hardware limitations. The author has made the image formation library and radar schematics available to the public.