Millimeter Wave Specific Challenges mmWave Radar and Sensing Informational

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).
Category: Millimeter Wave Specific Challenges
Updated: April 2026
Product Tie-In: Radar ICs, Antennas, Signal Processors

Radar Clutter Rejection

Effective clutter rejection is essential for reliable radar operation. In automotive and industrial radar: clutter can be orders of magnitude stronger than the target return, and false detections from clutter are a safety concern.

Clutter Sources at mmWave

(1) Ground clutter: the road surface reflects the radar signal. The return is distributed across all ranges where the beam intersects the ground. At 77 GHz: the ground backscatter depends on the surface type: smooth asphalt: low backscatter (σ° = -20 to -15 dBsm/m²). Rough concrete: moderate (σ° = -15 to -10 dBsm/m²). Wet road: variable (standing water can create strong specular reflections). (2) Building and infrastructure clutter: walls, guardrails, signs, and poles return strong echoes. These are point or extended reflectors with high RCS (1-100 m²). They appear as strong, stationary targets in the range-Doppler map. (3) Rain clutter: rain drops scatter the radar signal. The backscatter is distributed in range and has a velocity spread (rain drops fall at 2-9 m/s depending on drop size). Rain clutter appears as a diffuse return at downward-looking velocities. At 77 GHz with fine range resolution (3.75 cm): the rain clutter per resolution cell is low (the clutter volume is small). Rain is generally not a major clutter source for automotive radar with > 1 GHz bandwidth. (4) Multipath: reflections from the ground or structures create ghost targets. A signal reflects off the road, hits a target, reflects off the road again, and returns to the radar. The ghost appears at a longer range (the reflection path is longer). Multipath ghosts can be identified by: their range (2× the single-reflection range), their Doppler (same velocity as the real target), and their amplitude (weaker, since the ground reflection adds loss).

Advanced Clutter Mitigation

(1) Doppler filtering: design the chirp timing so that specific clutter velocities are placed in the blind zones of the Doppler spectrum. For automotive radar on a moving vehicle: the ground clutter has the Doppler of the vehicle velocity (the road surface moves toward the radar at the vehicle speed). At 100 km/h (27.8 m/s) at 77 GHz: f_D_ground = 2×27.8/0.0039 = 14.3 kHz. The radar signal processor applies a bandpass filter around the zero-Doppler bin (removing the clutter at ±v_vehicle). Moving targets have velocities different from v_vehicle and pass through the filter. (2) Space-time adaptive processing (STAP): combines spatial filtering (beamforming) with temporal filtering (Doppler) to reject clutter. The optimal STAP filter maximizes the target SNR in the presence of both ground clutter and interference. This is the gold standard for airborne radar (where the ground clutter spectrum is spread by the aircraft motion). For automotive: simplified STAP or Doppler-only processing is typically sufficient. (3) Machine learning classification: train a neural network on the range-Doppler-angle features to distinguish between targets (cars, pedestrians, cyclists) and clutter (guardrails, signs, multipath ghosts). The ML classifier uses: the target RCS pattern, velocity profile, range trajectory over time, and angular extent. State-of-the-art automotive radar processors include ML classification as a standard feature.

Clutter Rejection Methods
MTI: zero Doppler bin = 0 → reject clutter
CFAR threshold: T = α × P_noise_estimated
CA-CFAR: average of training cells
f_D_ground = 2v_vehicle/λ
Doppler filter: pass v ≠ v_vehicle
Common Questions

Frequently Asked Questions

How does CFAR handle a target near strong clutter?

If a target is next to a strong clutter source (e.g., a pedestrian near a guardrail): the CA-CFAR training cells include the strong clutter. The estimated noise power is high (dominated by the clutter). The threshold is set high. The pedestrian, being weaker than the clutter, may be below the threshold and missed. This is the "masking" problem. Solution: OS-CFAR: uses the k-th largest training cell value (not the average). If the clutter is in only a few cells: the OS-CFAR ignores the outliers and sets a lower threshold. Alternative: use a smaller training window (fewer cells) to reduce the contamination from nearby clutter. Or: use range-Doppler processing first (the clutter and pedestrian are at different velocities) and then apply CFAR in the velocity-filtered domain.

Can I reject ground clutter from a stationary radar?

Yes. For a stationary radar (traffic monitoring, security, industrial): ground clutter is stationary (zero Doppler). Moving targets (people, vehicles) have non-zero Doppler. Simple MTI removes the ground completely. Challenge: very slowly moving targets (walking at < 0.5 m/s, or a drone hovering at near-zero velocity): these may fall within the clutter rejection notch around zero Doppler. Increase the velocity resolution (longer frame time) to narrow the rejection notch. Or: use background subtraction (subtract the average range profile from each measurement; any change is a target). Background subtraction detects even zero-velocity targets (as long as they were not part of the original background).

What about clutter rejection in imaging radar?

Imaging radar (4D radar with elevation resolution): clutter rejection benefits from the elevation dimension. Ground clutter comes from below the sensor. Targets (vehicles, pedestrians) are above the ground. By beamforming in elevation and selecting the above-ground elevation bins: ground clutter is spatially rejected. This is more robust than Doppler-based MTI (which requires the target to be moving). Static objects above the ground (parked cars, traffic signs) are preserved after elevation filtering, while ground clutter is suppressed. The elevation processing typically provides 10-20 dB of clutter rejection beyond what Doppler processing alone achieves.

Need expert RF components?

Request a Quote

RF Essentials supplies precision components for noise-critical, high-linearity, and impedance-matched systems.

Get in Touch