AV Radar Signal Processing 2026 — PatSnap Eureka
Autonomous Vehicle Radar Signal Processing Technology Landscape 2026
From FMCW pipelines to end-to-end neural networks: a patent and literature analysis of 60+ records spanning 2016–2026, mapping the dominant assignees, technology clusters, and emerging IP battlegrounds in AV radar signal processing.
Four Interacting Layers of Automotive Radar Signal Processing
Automotive radar signal processing centers on extracting actionable environmental representations from raw radio-frequency returns—primarily using frequency-modulated continuous-wave (FMCW) waveforms operating in the 24 GHz and 77 GHz millimeter-wave (mmW) bands. The field decomposes into four interacting layers: front-end hardware and antenna design (MMIC, MIMO, Luneburg lens arrays); signal transformation pipelines (FFT-based range-Doppler processing, SAR imaging); learned inference engines (deep learning, end-to-end neural networks applied to radar spectra); and system-level integration (distributed computing, interference management, sensor fusion, and joint radar-communication).
A 2023 survey establishes that FMCW modulation dominates automotive radar, with the core challenge being real-time signal processing at hardware-constrained power budgets. A review of mmW radar technologies confirms that deep learning applications to radar data represent the frontier of the field, with dataset scarcity identified as the primary bottleneck. This landscape synthesizes patent and literature evidence across 60+ retrieved records spanning 2016–2026, including approximately 25 granted or pending patents. Jurisdictional coverage spans US, EP, WO, CN, IN, MX, and CA filings, with a clear acceleration in machine-learning-centric radar patents post-2022.
For regulatory context on automotive radar spectrum allocation, the ITU and ETSI provide authoritative standards on mmW band usage. The NHTSA governs ADAS safety requirements relevant to radar-dependent systems in the US market.
From Foundational Hardware to ML-Native Pipelines: 2016–2026
The patent record traces a clear arc from basic ADAS radar primitives through MIMO scaling and interference awareness to end-to-end machine learning inference operating on raw ADC samples.
Four Patent Clusters Defining the AV Radar Signal Processing Landscape
The 25 patent records decompose into four primary technology clusters, each with distinct assignees, claim strategies, and commercial trajectories.
FMCW Signal Processing & FFT-Based Range-Doppler Pipelines
The dominant classical processing paradigm applies FFT transforms to FMCW radar returns to derive range, Doppler velocity, and azimuth information. Aptiv’s MMIC compresses FFT-processed I-frames using H.264/H.265 encoding before network transmission—eliminating the need for a standalone radar processor. Motional AD’s pre-processing hardware pipeline computes range frequency and Doppler frequency spectra before ML inference. Apollo Autonomous Driving USA LLC filed an FMCW radar and lidar signal processing architecture (US, 2025). Learn more about patent landscape analytics for radar processing.
Key assignees: Aptiv, Motional AD, ApolloMachine Learning & End-to-End Neural Network Approaches
The most active recent filing cluster applies machine learning—ranging from CNNs operating on radar spectra to end-to-end models processing raw ADC samples—directly to radar signal processing tasks. Zendar Inc. is the dominant filer with a consistent IP family across WO, IN, US, and CN jurisdictions. Motional AD LLC’s 2024 WO filing processes ADC data through range-Doppler and range-azimuth-Doppler tensors for object detection and segmentation. Torc Robotics focuses on precise clock synchronization to enable accurate ML training data generation.
Key assignees: Zendar, Motional AD, Torc RoboticsDistributed Radar Computing Architectures
Several assignees protect architectures that separate radar processing across local (sensor-proximate) and centralized compute nodes. GM Cruise Holdings’ architecture performs feature vector extraction locally, then transmits compressed representations to a central processor for object identification—reducing bandwidth and latency. HL Klemove Corp. uses auto-encoder compression to transmit raw radar data to a vehicle central processor within bandwidth constraints (US, granted August 2025). This mirrors the broader automotive domain controller trend and suits zonal E/E architectures.
Key assignees: GM Cruise Holdings, HL KlemoveInterference Mitigation & Joint Radar-Communication
Mutual radar interference emerges as a critical systems problem at scale deployment. Solutions range from angle-of-arrival array processing (Baidu USA) to media access control coordination (Waymo) to learning-based interference cancellation (Aptiv). Baidu USA’s array of directional receivers determines the location of interference sources (US, 2022, active). Aptiv’s method for radar interference mitigation is active as of 2026. Vehicle-to-vehicle radar communication dual-functionality is patented by GM Cruise Holdings via heterodyne disaggregation of combined radar and message data.
Key assignees: Baidu USA, Aptiv, Waymo, GM CruisePatent Distribution by Technology Cluster and Jurisdiction
Visualising the 25-record patent dataset across technology clusters and filing jurisdictions reveals concentration patterns and geographic strategy.
Technology Cluster Distribution
ML and end-to-end neural network approaches represent the most active filing cluster in the 2022–2026 period.
Filing Jurisdiction Distribution
US is the dominant single filing jurisdiction; WO (PCT) is used for cross-border protection by Zendar, Waymo, and Motional.
Where AV Radar Signal Processing Patents Are Being Deployed
| Application Domain | Key Assignees | Representative Patent / Filing | Jurisdiction & Year | Technical Highlight |
|---|---|---|---|---|
| Autonomous Highway & Urban Driving (SAE L3–L5) | Ford, Aurora Operations, Apollo Autonomous Driving | Radar Sensor System for Vehicles | US, 2024 (active) | Complementary wide-beam and narrow-sweep rear-sector radar sensors for multi-radar vehicle coverage |
| High-Resolution Mapping & SAR Imaging | NXP USA Inc. | Systems and Methods for Automotive Synthetic Aperture Radar | EP, 2021 (active) | SAR processing without zero-padding FFT overhead; 77 GHz implementations achieve sub-degree angular resolution |
| Robotaxi & Ride-Sharing Fleet Operations | GM Cruise Holdings LLC | Radar-Lidar Extrinsic Calibration in Unstructured Environments | EP, 2023 (pending) | Multi-sensor alignment in uncontrolled environments for fleet-scale robotaxi deployments |
| ADAS Safety (Collision Avoidance, Blind-Spot) | CRadar.AI, Ford Global Technologies | Radar Target Detection and Imaging System with Ultra-Low Phase Noise Frequency Synthesizer | US, 2020 | Object detection at distance complementing camera and lidar; FMCW-based blind spot detection design |
Six Innovation Frontiers Shaping AV Radar Through 2026
The most recent filings and literature identify six converging directions that will define the next generation of autonomous vehicle radar signal processing IP.
End-to-End ML Pipelines from Raw ADC Data (2024–2026)
The most active frontier. Motional AD LLC’s 2024 WO filing processes radar ADC samples through range-Doppler and range-azimuth-Doppler tensors using E2E neural networks—bypassing classical CFAR detection entirely. Zendar’s 2026 US and CN pending filings extend this approach globally. Literature confirms that 4D imaging radar enabling high-resolution point clouds is catalyzing a deep learning era in radar perception.
Compressed Radar Data Transport via Video Codecs (2023)
Aptiv’s use of H.264/H.265 I-frame encoding on FFT-processed radar streams represents an unconventional but practical approach to bandwidth reduction in the vehicle domain network, enabling MMIC-to-domain-controller transmission without a dedicated radar processor. This signals a convergence of video compression and radar processing architectures.
Distributed Edge-Cloud Radar Processing (2023–2025)
GM Cruise Holdings and HL Klemove Corp. both protect local feature extraction followed by centralized processing. GM Cruise Holdings’ distributed radar feature-vector architecture covers both US and EP jurisdictions with active status, creating potential design constraints for competitors building similar local-centralized radar compute topologies.
V2V-Augmented Radar ML Training (2023)
Aptiv’s method uses V2V communications data as automatic label generation for radar ML models—creating a self-improving perception loop at fleet scale. This approach is protected in both US and EP jurisdictions (2023). Torc Robotics also filed on precise clock synchronization to enable accurate ML training data generation from radar sensors (US, 2025/pending).
IP Strategy Considerations for R&D and Legal Teams
Five strategic signals from the patent record with direct implications for freedom-to-operate, IP portfolio development, and competitive intelligence.
Autonomous Vehicle Radar Signal Processing — key questions answered
Frequency-modulated continuous-wave (FMCW) modulation dominates automotive radar, operating primarily in the 24 GHz and 77 GHz millimeter-wave bands. The core challenge is real-time signal processing at hardware-constrained power budgets.
Among the 25 patent records in this dataset, the top assignees by filing volume are Arizona Board of Regents (6 records), GM Cruise Holdings LLC (5 records), Zendar Inc. (4 records), Aptiv Technologies AG (4 records), and Baidu USA LLC (2 records).
End-to-end machine learning pipelines operating directly on raw ADC radar data represent the most active frontier. Motional AD LLC’s 2024 WO filing processes radar ADC samples through range-Doppler and range-azimuth-Doppler tensors using E2E neural networks, bypassing classical CFAR detection entirely.
Aptiv’s MMIC compresses FFT-processed I-frames using H.264/H.265 encoding before network transmission, eliminating the need for a standalone radar processor and reducing system complexity. This enables MMIC-to-domain-controller transmission without a dedicated radar processor.
Zendar Inc. has filed parallel pending patents for its learning-based radar processing system in WO (2025), IN (2025/pending), US (2026/pending), and CN (2026/pending), indicating a deliberate global IP consolidation strategy in advance of commercial radar ML platform licensing.
GM Cruise Holdings’ architecture performs feature vector extraction locally at the sensor, then transmits compressed representations to a central processor for object identification, reducing bandwidth and latency. HL Klemove Corp. uses auto-encoder compression to transmit raw radar data to a vehicle central processor within bandwidth constraints.
PatSnap Eureka searches patents and research literature to answer instantly.