Automotive and Industrial RF Automotive Radar Informational

How do I design the signal processing chain for an automotive FMCW radar?

The signal processing chain for an automotive FMCW radar converts the raw ADC samples from the radar receiver into a list of detected objects with range, velocity, and angle attributes through a multi-stage pipeline. The processing stages are: (1) Beat signal digitization, where the mixer output (beat frequency proportional to target range) is sampled by the ADC at 10-50 MSa/s. (2) Range FFT (first FFT), applied to each chirp to convert the time-domain beat signal into a range spectrum where each frequency bin corresponds to a specific target distance. (3) Doppler FFT (second FFT), applied across chirps at each range bin to extract velocity information from the phase change between consecutive chirps. The output is a 2D range-Doppler map. (4) MIMO phase compensation to separate signals from different TX antennas (for TDM-MIMO or BPM-MIMO). (5) CFAR detection (Constant False Alarm Rate), which adaptively sets a detection threshold based on the local noise floor to identify target peaks while maintaining a specified false alarm rate. (6) Angle estimation using the phase differences across RX channels to determine azimuth and elevation angles for each detected target (using FFT, Capon, or MUSIC algorithms). (7) Clustering and tracking using extended Kalman filter or unscented Kalman filter to group detections into objects and maintain target tracks over time.

Category: Automotive and Industrial RF

Updated: April 2026

Product Tie-In: Radar ICs, PCB Materials, Antennas

Automotive FMCW Radar Signal Processing Pipeline

The signal processing chain transforms millions of raw ADC samples per second into a compact list of tracked objects that the ADAS controller can use for decision-making. The processing must complete within the frame time (typically 30-50 ms) for real-time operation.

Technical Considerations

Each chirp produces N_s ADC samples. An N_s-point FFT (typically 256-2048 points with zero-padding) converts the beat signal to a range spectrum. The range associated with each bin k is R_k = k x c/(2 x B_chirp) x (f_s/N_s), where f_s is the ADC sample rate. Windowing (Hanning, Blackman) is applied before the FFT to reduce spectral leakage that would smear strong targets into adjacent range bins.

Performance verification: confirm specifications against the application requirements before finalizing the design
Environmental factors: temperature range, humidity, and vibration affect long-term reliability and parameter drift
Cost vs. performance: evaluate whether the application demands premium components or standard commercial grades

Performance Analysis

At each range bin, N_c chirps (typically 64-512) are collected across the frame. An N_c-point FFT across chirps extracts the Doppler velocity for each range bin, creating the range-Doppler map. Velocity resolution is lambda/(2 x N_c x T_c), where T_c is the chirp repetition interval. Maximum unambiguous velocity is lambda/(4 x T_c) for a simple chirp sequence, or reduced by N_TX for TDM-MIMO.

Common Questions

Frequently Asked Questions

How much processing power does automotive radar signal processing require?

A basic single-chip radar (4 RX, 256 range bins, 128 Doppler bins) requires approximately 1-5 GOPS for the FFT processing chain. A 4-chip cascade imaging radar (16 RX, 512 range bins, 256 Doppler bins, 2D angle estimation with MUSIC) requires 50-500 GOPS. This is typically handled by dedicated radar DSPs (C674x on TI AWR), FPGAs, or automotive-grade application processors.

What frame rate does automotive radar achieve?

Modern automotive radar operates at 15-30 frames per second for long-range modes and up to 50-100 frames per second for short-range/parking modes. Each frame consists of a chirp sequence (64-512 chirps at 20-100 us per chirp), producing a complete range-Doppler-angle measurement. Higher frame rates provide better temporal resolution for tracking fast-moving objects.

How does the radar track objects over time?

After CFAR detection and angle estimation, a tracking algorithm (typically extended Kalman filter or unscented Kalman filter) associates new detections with existing tracks using gating and data association (global nearest neighbor or joint probabilistic data association). The tracker maintains estimated position, velocity, and acceleration of each object, smoothing measurement noise and filling in missed detections.

Keep Reading