How does clutter rejection work in a millimeter wave radar system?

Clutter in a mmWave radar system is the unwanted return from stationary objects (ground, buildings, vegetation) and environmental effects (rain, multipath). Clutter rejection methods: (1) Moving Target Indication (MTI): exploits the fact that stationary clutter has zero Doppler velocity, while moving targets have non-zero velocity. Implementation: in the range-Doppler map, the zero-velocity column (Doppler bin = 0) contains all stationary clutter. Moving targets appear in non-zero Doppler bins. Simple MTI: remove (zero out) the Doppler bin = 0 column (and its neighbors ±1-2 bins to remove clutter at near-zero velocities). All remaining signal is from moving targets. Limitation: targets with zero radial velocity (moving perpendicular to the radar beam) are also removed. (2) CFAR (Constant False Alarm Rate) detection: an adaptive threshold algorithm that adjusts the detection threshold based on the local noise + clutter level. For each cell in the range-Doppler map: estimate the clutter + noise power from the surrounding cells (training cells). Set the detection threshold = (estimated noise power) × alpha, where alpha is chosen for the desired false alarm probability (P_fa). If the cell under test exceeds the threshold: declare a detection. The CFAR automatically adapts to varying clutter levels across the range-Doppler map (strong clutter regions have higher thresholds). Common CFAR variants: Cell-Averaging CFAR (CA-CFAR): averages the power of the training cells. Simple, effective for homogeneous clutter. Ordered-Statistics CFAR (OS-CFAR): uses the k-th largest value of the training cells. More robust to non-homogeneous clutter (e.g., clutter edge transitions). (3) Static clutter removal: accumulate a background model (the average return from stationary objects over multiple frames). Subtract the background from each new frame. Only the deviation from the background (new or moving objects) remains. This removes all static clutter perfectly (including clutter at non-zero Doppler bins caused by multipath). Limitation: the background model must be updated when the radar environment changes (e.g., the radar is mounted on a moving vehicle).