NoiseReject
Real-world radar doesn’t only return the targets you care about. Signals bounce off walls, guardrails and nearby vehicles and arrive back as ghosts — phantom detections at positions where nothing actually exists. NoiseReject is Provizio’s signal-processing layer for suppressing these multipath ghosts inside the radar’s own DSP chain, so the point cloud reaching perception is cleaner than the one the antenna sees.
Applications
Multipath is worst where the world is reflective.
On-road & urban
Walls, tunnels, barriers and multi-storey car parks throw strong specular reflections, and signals bounce between the ego vehicle and surrounding traffic. NoiseReject keeps ADAS and autonomous-driving perception working from genuine detections.
Off-road & industrial
Mining, construction, agriculture and ports are full of large metal machinery, trailers, implements and shipping containers — exactly the big flat reflectors that throw ghost targets. NoiseReject clears them so industrial autonomy and AGVs plan from a clean point cloud.
Smart cities & infrastructure
Roadside and infrastructure radar sits among buildings, gantries and signage that reflect signals back as phantom vehicles and pedestrians. NoiseReject keeps intersection monitoring and traffic-counting data trustworthy.
How it works
Telling ghosts from real targets, in the signal chain.
STEP 01
Read the geometry
A ghost from a single bounce off its target appears at the same bearing but further away; a ghost reflected off another surface appears at an angle that can’t be physically real. NoiseReject uses these geometric signatures to flag candidate ghosts.
STEP 02
Track it over time
Real targets move predictably and hold together from frame to frame. Ghosts flicker, jump and fail to track consistently. Watching how each return behaves across frames confirms which detections are genuine.
STEP 03
Reject it early
Confirmed ghosts are removed early in the DSP chain — before detection, classification or tracking ever see them. The cleaner the input, the better every downstream stage performs.
The two kinds of ghost
Type-1 and type-2 multipath.
Multipath ghosts come in two forms, depending on where the signal made its last bounce. Each has its own signature — and NoiseReject targets both.
TYPE 1
Bounced off its own target
The signal’s last reflection happens at the real object — for example bouncing between the ego vehicle and the target, or underneath a vehicle’s body. Because the signal travels further than the direct path, the ghost lands behind the real target.
Signature: same bearing, greater range
TYPE 2
Bounced off another surface
The last reflection happens at a different surface — a wall, barrier or large reflector — before the signal reaches the radar. The ghost is placed at the angle of that surface rather than the true target, so it appears where nothing exists.
Signature: wrong angle, harder to predict
Integration
Part of the on-chip backend
NoiseReject runs on the radar SoC as part of Provizio’s signal-processing backend. It’s a layer in the DSP chain, not a separate microservice. Downstream perception code sees the same point-cloud format, just with the ghosts removed.
- Runs on the radar SoC
- A layer in the DSP chain, not a separate service
- Same point-cloud format downstream
- Software-defined, OTA-updatable
- Antenna-agnostic — works on the point-cloud output
Questions, answered
NoiseReject FAQ
Multipath ghosts — phantom detections created when the radar signal bounces off a surface before returning. A single bounce off the target object produces a type-1 ghost at the same bearing but a longer range; a bounce off another surface produces a type-2 ghost at the wrong angle entirely. NoiseReject identifies and suppresses both.
It combines signatures rather than relying on a single test. Ghosts have characteristic geometry — they sit at the same bearing as a real target but further away, or at angles that aren’t physically plausible. They also behave differently over time: real targets track smoothly frame to frame, while ghosts flicker and fail to track. NoiseReject uses both cues to flag and reject them.
On the radar SoC, as part of Provizio’s signal-processing backend. It is a layer in the DSP chain, not a separate microservice — downstream perception code sees the same point-cloud format, just cleaner.
No — those are handled elsewhere in the stack. Static ground clutter is the domain of Provizio’s Ground Detection processing, and mutual-radar interference is addressed separately. NoiseReject is focused specifically on multipath ghosts, so each layer does one job well.
NoiseReject is software-defined and antenna-agnostic — it operates on the radar’s point-cloud output, not on any specific hardware. Provizio deploys it as part of its own SoC backend, and the approach is portable to any radar producing standard point-cloud output.
Stop chasing ghosts in your point cloud.
If your radar deployment is fighting phantom detections from reflective infrastructure, dense traffic or large nearby objects, NoiseReject suppresses them in the signal chain — before they ever reach perception.