Radar - SW & HW Environment
1
Table of Contents
• Introduction
• Reference Architecture
• Hardware Environment
• Software Environment
• Radar Signal Chain
• From Simulation to Target
• Installing the NXP Toolchain
• Next Article in the Series
• References
2
Introduction
This article presents the NXP hardware platforms and the MathWorks and NXP software tools used to build an automotive radar application. The development workflow is anchored in the MathWorks example "Radar Signal Simulation and Processing for Automated Driving," which provides a reference architecture spanning driving-scenario simulation, radar modeling, and signal processing. The resulting signal-processing chain is then adapted and deployed onto NXP radar hardware using NXP-specific toolboxes and hardware accelerators.
This article is part of the Radar Application Development Series, which describes the complete workflow for developing, deploying, and optimizing automotive radar applications on NXP radar platforms. The purpose of this article is to introduce the overall software and hardware environment and show how the different tools, hardware components, and processing engines fit together within a radar development workflow.
The next article in this series, "Radar - Processing Chain," will explore in depth each processing block presented in the Radar Signal Processing section: Range FFT, Doppler FFT, Non-Coherent Combining, CFAR Detection, Clustering, and Direction-of-Arrival (DoA) estimation.
3
Reference Architecture
The starting point for the radar application is the MathWorks reference example, which models a complete automotive radar system end to end.
The workflow begins by defining a highway driving scenario using the Automated Driving Toolbox ( drivingScenario ), where vehicles and traffic participants are modeled. The ground-truth generated data then feeds the radar model.
Automated Driving - Bird's-Eye Plot.png
Figure 1: MathWorks example bird's-eye plot with Radar detections
A 77 GHz FMCW radar is parameterized from high-level system requirements such as:
Maximum detection range, typically 250-300 m for long-range radar
Range resolution, around 1 m
Velocity resolution
Maximum relative target velocity, up to around 230 km/h
The example then builds a transceiver model with antenna arrays, transmitter/receiver components, and signal-propagation effects, generating synthetic detections that estimate the position and velocity of surrounding vehicles.
For the NXP application, the reference architecture is divided into two domains:
Environment Simulation
Executes entirely within MATLAB, and is responsible for:
Driving scenario generation
Vehicle motion simulation
Target ground-truth generation
FMCW signal generation
Radar channel and propagation modeling
Radar Signal Processing
Contains the processing chain deployed on the S32R45 platform:
Range FFT processing
Doppler FFT processing
Non-Coherent Combining
CFAR detection
Clustering
Direction-of-Arrival (DoA) estimation
The Radar Signal Processing domain forms the basis of the embedded radar application deployed on the S32R45 Evaluation Board.
4
Hardware Environment
3.1 S32R45 Evaluation Board
The primary processing platform is the NXP S32R45 Evaluation Board, a development platform for high-performance 77 GHz radar applications such as adaptive cruise control, autonomous emergency braking, and cascaded imaging radar. It integrates several specialized processing engines optimized for radar workloads.
Processing engine
Role
4x Arm® Cortex®-A53 cores
Application-level processing, radar control, clustering, and object management
SPT Accelerator
Optimized FFTs and high-throughput radar signal-processing kernels
BBE32 DSP
Vectorized signal processing, detection algorithms, and custom radar kernels
LAX Accelerator
Matrix and linear-algebra operations for accelerated angle estimation
S32R45 Radar MPU Block Diagram.png
Figure 2: S32R45 block diagram
3.2 TEF82xx Customer Application Board
The TEF82xx is a fully integrated 76-81 GHz RFCMOS automotive radar transceiver providing the RF front end for signal generation and capture. It integrates 3 transmit channels, 4 receive channels, ADCs, a low-phase-noise VCO, and a phase rotator, and is fully compatible with the S32R45.
In the current application, the input signal is sourced from simulation rather than hardware, so the TEF82xx is not actively used. It is included as a placeholder for future hardware-in-the-loop and real-sensor integration.
5
Software Environment
The radar application combines MathWorks toolboxes for algorithm development with NXP toolboxes and tools for deployment and accelerator integration.
4.1 MathWorks Tools
Tool
Role in the workflow
Radar Toolbox
FMCW waveform generation, propagation modeling, detection, and analysis
Automated Driving Toolbox
Scenario modeling, road/vehicle simulation, and ground-truth generation
4.2 NXP Tools
Tool
Version
Role in the workflow
S32 Design Studio for S32 Platform
3.5
IDE, compiler, debugger, and deployment environment for the S32R45, including its accelerators
NXP Model-Based Design Toolbox for SPT
1.9.0
Bit-exact SPT simulator integration and rapid prototyping in MATLAB
NXP Model-Based Design Toolbox for RADAR
1.0.0
MATLAB integration for S32R45; SPT/LAX kernel execution, code generation, and PIL workflows
NXP Radar SDK (S32R45)
1.2.0
Optimized radar algorithms, accelerator libraries, SPT/LAX kernels, and embedded deployment infrastructure
Together, these tools act as the gateway between the MathWorks and NXP ecosystems, enabling algorithm development, simulation, code generation, deployment, and SIL/PIL validation within a common workflow.
6
Radar Signal Chain
Once the FMCW echoes are generated or captured, the signal-processing chain transforms the radar cube into a list of detected objects. The application currently implements the following stages.
#
Stage
What it does
Runs on
1
ADC Acquisition
Digitizes the beat signal into a radar data cube, using samples x chirps x antennas
TEF82xx ADCs → S32R45
2
Range FFT
Fast-time FFT converts beat frequency into target range
SPT accelerator
3
Doppler FFT
Slow-time FFT resolves velocity, producing the processed radar cube
SPT accelerator
4
Non-Coherent Combining
Combines the magnitude of the range/Doppler-processed radar cube across channels, producing the range-Doppler magnitude matrix
SPT accelerator
5
CFAR Detection
Applies Constant False Alarm Rate thresholding on the range-Doppler magnitude matrix to detect possible targets
BBE32 DSP
6
Clustering
Groups neighboring detections, for example using DBSCAN, into physical objects
Cortex-A53 cores
7
Angle / DoA Estimation
Estimates azimuth/elevation across the antenna array, using methods such as beamforming or MUSIC
LAX accelerator
The output of the chain is a list of detected objects with range, relative velocity, and angle of arrival.
Future Improvements
The current application focuses on signal processing and object detection. Planned enhancements include:
Multi-target tracking, including Kalman filtering and track-to-track association
Hardware-in-the-loop testing using the TEF82xx front end
7
From Simulation to Target
The MathWorks reference example executes entirely within MATLAB. During deployment, the example is partitioned into the Environment Simulation block and the Radar Signal Processing block, where the computationally intensive signal-processing functions are replaced with NXP-optimized implementations from the Radar SDK.
This delivers faster execution, reduced CPU utilization, and accelerator offloading on the S32R45.
8
Installing the NXP Toolchain
7.1 Installation Order
The development tools must be installed in the following order to ensure that all external dependencies required by the NXP MBDT for RADAR are available before it is configured:
S32 Design Studio for S32 Platform 3.5
S32R45 Radar SDK 1.2.0
NXP Model-Based Design Toolbox for SPT 1.9.0
NXP Model-Based Design Toolbox for RADAR 1.0.0
7.2 Integrating the Development Environment
After NXP MBDT for RADAR is installed, the integration of S32 Design Studio and S32R45 Radar SDK is performed using the MATLAB Live Script:
mbd_lax_dependencies_path.mlx
The script is located at the root of the NXP MBDT for RADAR installation. Running this script configures the required dependency paths and establishes the connection between MATLAB, the NXP Model-Based Design Toolbox for RADAR, S32 Design Studio, and the S32R45 Radar SDK.
Once the script is completed successfully, the environment is ready for simulation, code generation, accelerator kernel execution, and Processor-in-the-Loop (PIL) validation.
7.3 Installation Methods
The NXP toolboxes ship as MATLAB Toolbox packages (.mltbx) and can be installed in three ways:
Manual install (.mltbx) - Double-click the .mltbx file, or right-click and select Install in MATLAB. The Add-On Manager installs and registers the toolbox automatically.
Via NXP Support Package - Install NXP_Support_Package_RADAR from MATLAB Add-Ons, then follow the guided steps to download and install MBDT for RADAR and generate/activate the free license.
Via the Automotive Software Package Manager - A bundle installer that walks through toolbox installation, dependency configuration, and license activation.
9
Next Article in the Series
This article introduced the software environment, hardware environment, deployment workflow, and high-level radar signal-processing architecture.
The next article, "Radar - Processing Chain - RSDK," will provide a detailed analysis of each processing block presented in Section 5:
Range FFT
Doppler FFT
Non-Coherent Combining
CFAR Detection
Clustering
Direction-of-Arrival (DoA) Estimation
It will also explain how these algorithms are mapped onto the S32R45 processing resources and how the NXP Radar SDK accelerates the execution of each stage.
10
References
MathWorks
Radar Signal Simulation and Processing for Automated Driving
Radar Toolbox
Automated Driving Toolbox
NXP
S32R45 High-Performance Processor for Imaging Radar
S32R45 Evaluation Board
TEF82xx 77 GHz Radar Transceiver
Model-Based Design Toolbox
MBDT for RADAR - Knowledge Base
NXP Support Package for RADAR
How to install .MLTBX
查看全文
