Radar Target and Clutter Simulation for an ACC Sensor

The problem

An ACC radar must see the lead vehicle reliably while ignoring guardrails, bridges, road surface returns and vehicles in adjacent lanes. Testing this on real roads is costly and unsafe at the design stage, so engineers use simulated scenarios and sensor models. Understanding what each level of radar model captures, and what it misses, is essential for using simulation correctly.

Objectives

• Build ACC-relevant scenarios in the Driving Scenario Designer.
• Configure a statistical radar detection model to an ACC specification.
• Simulate clutter and static roadside objects.
• Compare detection rate, false alarms and accuracy across scenarios.
• Compare statistical and waveform-level radar models on one scenario.

What you need

Key concepts

Statistical sensor model

Generates detections from object truth using configured detection probability, false alarm rate and measurement noise, without simulating waveforms. Fast and suited to tracking and control studies.

Waveform-level model

Simulates transmitted signals, propagation, reflections and signal processing. Slower, but it shows effects such as sidelobes and resolution limits.

Radar cross section

A measure of how much radar energy a target reflects back. Trucks are large reflectors, pedestrians and two-wheelers small ones.

Clutter

Unwanted returns from the road, guardrails, signs and bridges that can produce false detections or hide real targets.

Step-by-step guide

1. 1

   Define the radar specification

   Write an ACC radar specification: field of view about ±10 degrees, range up to 150 to 200 m, range accuracy and resolution, azimuth accuracy, range-rate accuracy, update rate around 20 Hz and a target detection probability at a reference range for a car-sized target. Write the scenarios the radar must handle, including a lead car, cut-in, stationary vehicle ahead, guardrails and a motorcycle in the adjacent lane.

   Checkpoint: A specification table and a scenario list with what the radar must and must not report.

2. 2

   Build scenarios in the Driving Scenario Designer

   Create a straight and a curved highway with three lanes, guardrails and a few signs. Add the ego vehicle and actors with trajectories for each scenario: steady following, cut-in from the adjacent lane, approach to a stopped vehicle and a motorcycle overtaking. Export each scenario to a MATLAB function so it can be run from a script with fixed seeds.

   Checkpoint: Exported scenario functions that play back correctly in a bird's-eye view.

3. 3

   Configure the statistical radar

   Create a radarDataGenerator mounted at the front bumper and set field of view, range limits, resolution, accuracy, detection probability at a reference range and RCS, and false alarm rate from the specification. Enable range-rate measurement. Set actor RCS values: larger for trucks, smaller for motorcycles and pedestrians. Plot the detections with the actors in a birdsEyePlot to check placement and coverage.

   Checkpoint: A bird's-eye animation with radar coverage and detections appearing on the correct actors.

4. 4

   Add roadside objects and clutter

   Represent guardrails and signs as static actors with appropriate RCS and enable false alarms in the radar model. Run the scenarios and look at detections from static objects, especially on the curved road where guardrails enter the field of view ahead. Note which detections an ACC function would have to reject, and add the ego speed so stationary objects show a range rate equal to minus ego speed.

   Checkpoint: Logs showing detections from guardrails and signs alongside vehicle detections.

5. 5

   Run Monte Carlo simulations

   Run each scenario 50 to 100 times with different random seeds. For each run, match detections to the closest truth object within a gate and log matched detections, misses and false alarms with range, azimuth and range-rate errors. Store everything in a table so statistics can be computed by scenario, object type and range band. Choose the gate size from the configured measurement accuracy, and record how many detections fall inside the gates of two truth objects at once, which signals ambiguity. Save the seeds so any run can be reproduced.

   Checkpoint: A results table with counts and errors from all Monte Carlo runs.

6. 6

   Analyse detection performance

   Plot detection probability against range for cars, trucks and motorcycles, and compare with the specification. Compute false alarms per scan and the share of detections from static objects. Show range and azimuth error distributions. Check whether the cut-in vehicle is detected early enough, measuring the time from it entering the ego lane to its first detection. Compare each curve with the specification line and mark where the radar falls short, as these gaps define what a tracker or a second sensor must cover. Summarise the results by scenario.

   Checkpoint: Detection probability curves, false alarm rates and error histograms by object type.

7. 7

   Separate moving and stationary targets

   Implement a simple classification of detections as moving or stationary using range rate compensated for ego speed and azimuth. Count how many guardrail detections remain as moving after compensation and how many stopped vehicles are wrongly treated as stationary clutter. This shows the classic difficulty of radar ACC with stationary objects, which matters for approaching a stopped vehicle. Report results for the curved-road scenario separately, because azimuth errors make stationary guardrails look like they move there.

   Checkpoint: A confusion table of moving against stationary classification for vehicles and roadside objects.

8. 8

   Compare with a waveform-level model

   For the two-vehicle following scenario, simulate an FMCW waveform with Phased Array System Toolbox objects, propagate to the actors, form a range–Doppler map and detect with CFAR. Compare detections, accuracy and runtime with the statistical model. Note effects only the waveform model shows, such as two close targets merging, and discuss when each model is appropriate. Use the same target positions and speeds in both models at each time step, so that any difference comes from the model level and not from the scenario itself.

   Checkpoint: A comparison of detections and runtime for the two model levels on one scenario.

9. 9

   Report

   Present the specification, scenarios, radar configuration, Monte Carlo results, stationary-target analysis and model comparison. Explain how the radar output would feed a tracker and ACC controller. List the limitations of both models, such as no multipath in the statistical model and simplified target scattering, and state what real-data validation would be needed before relying on the simulation. Attach the specification table again at the end with each requirement marked as met, partly met or not met by the simulated radar. Explain the reason for each shortfall.

   Checkpoint: A report with scenario files, scripts and result tables.

Testing & validation

• Detection probability against range by object type, compared with the specification.
• False alarms per scan and share of detections from static objects.
• Measurement error distributions in range, azimuth and range rate.
• Statistical against waveform-level model comparison on one scenario.

Expected results

A set of ACC radar scenarios with configured radar models, Monte Carlo detection statistics, an analysis of stationary-object handling and a clear view of what each radar model level can show.

Take it further

• Feed detections into a multi-object tracker and measure track quality.
• Add a camera model and fuse it for lead-vehicle selection.
• Export scenarios in ASAM OpenSCENARIO for use in other tools.
• Build the same scenarios in RoadRunner with realistic road geometry.

Viva and review questions

Learn it at DIYguru

These DIYguru courses and programmes teach what this project needs.