Testing mmWave Radars in the Loop

University of Texas researchers develop a real-time HIL testbed that beats the incumbents on cost, time and flexibility.

October 1, 2017

Preeti Kumari, Alaa Zaki Abdulghafoor, Robert W. Heath Jr. and Marcus Monroe

Fig. 1 - The mmWave testbed for ADAS prototypes consists of an automotive HIL, a radar target emulator, an ECU and a Delphi sensor mounted on a rotator and a tripod.

Millimeter-wave radars are a key component of advanced driver-assistance systems (ADAS). Engineers know, however, that setting up real-time radar experiments with multiple vehicles is challenging in an indoor laboratory environment. On-road drive testing is thus the typical approach when engineers need to evaluate ADAS systems.

With the goal of significantly improving on this process, our team at the University of Texas at Austin developed a real-time ADAS testbed, its main feature being an automotive hardware-in-the-loop (HIL) simulator combined with a mmWave radar target emulator (RTE) hardware. We believe our testbed is better than conventional HIL testbeds that either incorporate unrealistic model-based radar sensor and target simulation or perform prohibitively expensive and time-consuming testing with real vehicles.

The testbed also follows a software-defined modular approach that enables flexibility for different configuration options required to characterize the system.

System configuration

To demonstrate the application of our real-time ADAS testbed, we develop algorithms and perform measurements for determining latency between the components of the ADAS testbed. The latency measurements will help engineers decide whether enhancements, such as changing the type of automotive communication bus, are necessary to speed up a vehicle’s reaction time.

Our closed-loop ADAS testbed (Fig. 1) has four main components: the automotive HIL, the RTE, the electronic control unit (ECU), and the mounted radar sensor. The HIL consists of a real-time processor, I/O interfaces, and an operator interface that communicates with the real-time processor.

The RTE consists of a National Instruments PXIe-5646R vector signal transceiver and a 76-81 GHz radio-frequency transceiver. The vector signal transceiver is comprised of analog front-ends (up to 6.5 GHz intermediate-frequency (IF)), an analog-to-digital converter, a digital-to-analog converter, a baseband FPGA and digital input/output (I/O) interfaces. The third component is a real ECU used in vehicles. The last component in the testbed is a Delphi radar sensor mounted on a rotator and tripod for angle of arrival emulation and alignment purposes.

The components in the closed-loop ADAS test platform perform the following actions and interactions within one simulation step. The automotive HIL is responsible for sending a CAN message to command the RTE for target simulation. After receiving the CAN message from the automotive HIL and a transmitted signal from the radar sensor, the vector signal transceiver in the RTE generates and sends a radar target pulse to the mmWave transceiver through SMA connections.

The simulated target signal is then up-converted in the mmWave transceiver, transmitted over the air, and detected by the Delphi radar sensor. Once the target is detected, the radar will send a CAN message to communicate the target presence and information to the ECU to trigger an action for avoiding collision with the target.

The automotive HIL uses its I/O (analog, digital, and bus) interfaces to generate stimulus signals, acquire data, and provide the sensor/ actuator interactions between the ECU and the virtual environment being simulated.

Fig. 2 - Radar target simulation algorithm that will add velocity, delay, RCS, and path loss effects to the radar transmitted signal.

The vector signal transceiver in the RTE simulates a target by modifying the signal received from the Delphi radar based on the target parameters (e.g., location) sent by the automotive HIL, as shown in Fig. 2. In this algorithm, the target velocity is simulated by mixing the Doppler-shift frequency with the transmitted signal, and the target range is simulated by adding integer delays and fractional delay filters (FDF) to the signal. The signal is then amplified/attenuated to reflect the RCS and path loss associated with the target.

Additionally, the target direction can be simulated by mechanically changing the rotation of the Delphi sensor. The orientation of the Delphi sensor is also controlled by a LabVIEW module which sends control signals to the motor of the rotator.

Measuring latency

Fig. 3 - Sequence of events (not to scale) during the latency measurement of the communication between the HIL and the RTE components.

We developed an algorithm in LabVIEW to measure the latency in the communication between the automotive HIL and the RTE components of our mmWave testbed. Initially, a pulse is generated every one second in the HIL to trigger a clock in the RTE FPGA, as shown in Fig. 3. After the pulse is sent, the HIL sends the CAN message to command the target simulation.

Then, the RTE detects the CAN message and generate a target radar pulse. Once the radar pulse is generated, the RTE FPGA clock is latched. Thus, the latency is measured from the time the clock started to the time the clock stopped.

Fig. 4 - Delphi radar sensor output for the target simulated at 150 m.

For simulation purposes, we chose a target with 150 m (492 ft) range and -5 m/s velocity. The result of the testbed was a successful target detection by the radar, as shown by the thirty-eighth target block in Fig. 4. The estimated range and velocity parameters of the thirty-eighth detected target is close to the actual simulated parameters.

Fig. 5 - The characterization of Delphi radar sensor. The top figures indicate the beam width in the horizontal and vertical directions and the bottom figure indicates the power of the transmitted signal.

The Delphi radar sensor was characterized using our mmWave testbed, including power and beam width measurements, as shown in Fig. 5. This is realized by using the same RTE hardware but with a different software module. The radar transmitted signal contains two chirps of different power levels. The higher power chirp will be used for long-range targets and the lower power chirp will be used for short-range targets.

Additionally, the beam width of the Delphi radar sensor is wide in the azimuth direction but is quite narrow in the elevation direction. The wide-azimuth beam width will allow Delphi radar to detect multiple targets in the horizontal direction and the narrow-elevation beam width will enable it to minimize the clutter in the vertical direction.

Fig. 6 - Latency measurement between the HIL and the RTE communication.

We also measured the latency in the communication of the CAN messages from the automotive HIL to the RTE. The measurement was essential to test that the system was successfully transferring CAN messages. The measurement results indicated that the time-delay of communication was around 0.2 ms as shown in Fig. 6. The mean of the communication delay based on several measurements is found to be 0.2057 ms with a minimum delay of 0.1913 ms and a maximum delay of 0.3898 ms.

Fig. 7 - Latency measurement histogram of the communication between the HIL and the RTE.

Fig. 7 shows a histogram of the delay measurements. The histogram shows that most of the readings are around the 0.2 ms delay range indicating that the result is consistent. The latency in the communication between the HIL and the RTE of our set-up is well within the simulation step-size of 500 s to 2 ms.

Our closed-loop ADAS testbed provides a real-time, reliable, modular and efficient HIL testing for automotive radar applications. Using our testbed, we characterized Delphi radar sensor and performed latency measurements of the communication between the automotive HIL and the RTE. Our latency measurement set-up can be further extended to determine the time delay of one ADAS simulation step and to estimate the communication latency between the radar sensor and the ECU.

The knowledge of this delay would allow engineers to look for ways to minimize the overall simulation rate and to speed up the ECU reaction time for avoiding collisions once a target is detected.

This research was supported by National Instruments and is a part of the UT SAVES initiative. For further information, contact the authors at the Department of Electrical and Computer Engineering, The University of Texas at Austin: This email address is being protected from spambots. You need JavaScript enabled to view it.; This email address is being protected from spambots. You need JavaScript enabled to view it.; This email address is being protected from spambots. You need JavaScript enabled to view it., This email address is being protected from spambots. You need JavaScript enabled to view it..