Autofocus SAR
Understanding Autofocus SAR
A synthetic aperture radar creates high-resolution images by coherently combining thousands of radar echoes collected as the platform (aircraft or satellite) moves along a flight path. The SAR processor assumes it knows the exact position of the radar at every pulse. In practice, GPS/INS systems measure position to centimeter-level accuracy, but SAR at X-band (wavelength = 3 cm) requires millimeter-level path knowledge. The gap between measured and required accuracy creates residual phase errors that defocus the image.
Autofocus closes this gap by extracting the phase error directly from the radar data itself, without needing better navigation hardware. The algorithms exploit the fact that strong, point-like scatterers in the scene (building corners, metal poles, vehicles) should focus to sharp points. Any smearing of these points reveals the phase error function, which can then be estimated and removed.
φmax < π/4 radians (45°)
Equivalent Path Error at X-band (10 GHz):
ΔR = λ / (4 × 2π) × π/4 = λ/8 ≈ 3.75 mm
SAR Cross-Range Resolution:
δcr = D / 2
Where D = physical antenna length
(Independent of range, which is what makes SAR so powerful)
Autofocus Algorithm Comparison
| Algorithm | Approach | Strengths | Limitations |
|---|---|---|---|
| Phase Gradient Autofocus (PGA) | Selects prominent point targets, estimates phase gradient iteratively. | Model-free, handles arbitrary errors, proven in operational systems. | Requires isolated strong scatterers; struggles in homogeneous terrain (ocean, desert). |
| Mapdrift | Splits aperture into sub-apertures, cross-correlates sub-images to estimate drift. | Fast, simple, works on distributed scenes. | Only corrects low-order (linear/quadratic) phase errors. |
| Prominent Point Processing (PPP) | Tracks phase history of individual strong scatterers across the full aperture. | Very accurate for isolated targets. | Computationally expensive; fails in dense urban clutter. |
| Minimum Entropy | Optimizes phase correction to minimize image entropy (maximize sharpness). | Works without isolated scatterers; good for distributed scenes. | Slower convergence; risk of local minima. |
PGA has become the industry standard because it makes no assumptions about the shape of the phase error. It simply iterates until the prominent points in the image are as sharp as the diffraction limit allows. Most operational SAR processors (including those on Sentinel-1, RADARSAT, and military airborne systems) include PGA as the final processing step.
Frequently Asked Questions
Why does SAR imagery need autofocus?
SAR creates images by coherently combining radar echoes collected as the platform flies a straight path. The processor assumes a perfectly known flight path. In reality, the platform deviates due to wind gusts, vibration, or imperfect GPS/INS measurements. Even a fraction of a wavelength (millimeters at X-band) of deviation introduces phase errors that smear the image. Autofocus estimates these phase errors directly from the data and corrects them, often recovering near-diffraction-limited resolution.
What is Phase Gradient Autofocus (PGA)?
PGA is the most widely used SAR autofocus algorithm. It selects strong point-like targets in the image, centers them, windows out surrounding clutter, and estimates the phase error gradient across the aperture. The process iterates until the phase error converges to near zero. PGA is data-driven and makes no assumptions about the error model, making it robust against arbitrary phase errors from turbulence, vibration, or navigation drift.
How much phase error causes visible defocusing?
Uncompensated phase errors exceeding π/4 radians (45 degrees) across the synthetic aperture cause noticeable resolution degradation. At X-band (10 GHz, wavelength = 3 cm), π/4 radians corresponds to a path error of only 3.75 mm. For a 1-meter resolution SAR with a 500-meter synthetic aperture, the platform must maintain millimeter-level path accuracy or rely on autofocus to correct the residual errors.