Industry Research

Appearance

Radar Frame Timestamping and Doppler Integration ​

Last updated: 2026-05-09

Why It Matters ​

Automotive radar measures over an integration interval. A frame is built from a sequence of chirps, and Doppler velocity comes from phase change across chirps inside that interval. A single radar message timestamp therefore needs a clear meaning: frame start, midpoint, end, first chirp, last chirp, or host receive time. Fusion also needs to know whether reported velocity is radial, ego-motion compensated, tracked-object velocity, or raw detection Doppler.

Bad radar timing creates false relative velocity, poor radar-camera association, and inconsistent dynamic-object tracking during acceleration and turns.

Deployment Contract ​

Contract itemPractical rule
Frame referenceDefine radar message timestamp as measurement midpoint unless the sensor contract requires another convention.
Integration windowPublish frame duration, chirp timing, and output cycle time where available.
Clock sourceRadar follows vehicle PTP/gPTP, hardware trigger, or a documented CAN/Ethernet timebase with bounded conversion.
Velocity semanticsDistinguish raw radial Doppler, ego-compensated velocity, and tracker-estimated object velocity.
Host timingHost receive time and CAN/Ethernet arrival time are diagnostics, not acquisition time.
ProvenanceLog sensor timestamp mode, sequence, frame counter, sync state, firmware, and radar configuration.
ReplayRaw detections/tracks plus timing and chirp/frame config must reproduce fusion alignment offline.

Radar Time Model ​

text
frame_start
  chirp 0
  chirp 1
  ...
  chirp N-1
frame_end

t_measurement ~= (frame_start + frame_end) / 2

For tracked objects, the timestamp should still identify the measurement update epoch. The tracker may predict objects to another time, but that prediction time must be explicit.

Data Semantics ​

FieldRequired interpretation
header.stampAcquisition or measurement epoch, not publish time.
frame_idRadar sensor frame with calibrated extrinsic to base_link.
Range/azimuth/elevationGeometry at the measurement epoch.
Doppler/radial velocityVelocity along the radar line of sight unless documented otherwise.
Cartesian velocityMust state whether ego compensation and tracker smoothing were applied.
RCS/SNR/noiseMeasurement quality and gating features, not timestamp substitutes.
Track ageTracker internal history; does not replace measurement time.

Failure Modes ​

Failure modeSymptomResponse
Host receive timestamp usedRadar-camera association shifts with Ethernet/CAN load.Use sensor acquisition time and log host latency separately.
Frame-start vs midpoint mismatchApparent radial velocity bias during ego acceleration.Convert timestamps to the agreed measurement epoch.
Doppler treated as Cartesian velocityObject tracker over- or under-compensates moving actors.Keep radial velocity semantics through fusion or explicitly transform with geometry.
Ego compensation undocumentedFusion subtracts ego motion twice or not at all.Version radar output mode and test with static targets during vehicle turns.
Frame counter dropsTracker sees impossible object jumps.Detect sequence gaps and avoid updating tracks with partial frames.
Radar sync lossTime drift appears as association errors rather than a timing fault.Monitor sync state and last valid time conversion.
Variable chirp/profile modeTimestamp and Doppler resolution change with mode.Log radar profile ID and integration window per frame.

Telemetry and Validation Hooks ​

  • Frame counter, sensor timestamp, converted vehicle timestamp, host receive timestamp, and latency distribution.
  • Radar profile ID, frame period, chirp count, chirp interval, Doppler resolution, and maximum unambiguous velocity.
  • Sync state, PTP/trigger lock, temperature, firmware, and dropped-frame count.
  • Static-target ego compensation residual during straight driving, braking, and constant-radius turns.
  • Radar-LiDAR-camera association residual against known reflectors, cones, and moving vehicles.
  • Replay check that raw detections and tracks align to the same measurement epoch used online.

Acceptance checks:

  1. Static reflectors report near-zero ego-compensated velocity after timestamp conversion during acceleration and turns.
  2. Dropped frames and radar profile changes are visible in diagnostics and logs.
  3. The fusion layer can tell raw detections from tracked objects and does not mix their timestamp semantics.

Sources ​

Public research notes collected from public sources.

Copyright © 2026 kvynlim