Automotive Radar Signal Analysis Techniques
Automotive radar is a crucial technology for advanced driver-assistance systems (ADAS), enabling features like adaptive cruise control and collision avoidance. Understanding and analyzing radar signals is essential for designing robust and reliable radar systems. This article explores various automotive radar signal analysis techniques, including hardware modeling, signal processing, and the simulation of multipath propagation effects.
Radar Toolbox enables you to evaluate different radar architectures. You can explore different transmit and receive array configurations, waveforms, and signal processing chains. You can also evaluate your designs against different channel models to assess their robustness to different environmental conditions. This modeling helps you to identify the specific design that best fits your application requirements.
Calculating Radar Parameters from Long-Range Radar Requirements
The radar parameters are defined for the frequency-modulated continuous wave (FMCW) waveform, as described in the example Automotive Adaptive Cruise Control Using FMCW Technology. The radar operates at a center frequency of 77 GHz. This frequency is commonly used by automotive radars. For long-range operation, the radar must detect vehicles at a maximum range of 250-300 meters in front of the ego vehicle. The radar is required to resolve objects in range that are at least 1 meter apart. Because this is a forward-facing radar application, the radar also needs to handle targets with large closing speeds as high as 230 km/hr.
The radar is designed to use an FMCW waveform. These waveforms are common in automotive applications because they enable range and Doppler estimation through computationally efficient FFT operations.
When designing an automotive radar, or when the specific architecture of a radar is known, use a radar model that includes this additional information.
Modeling the Automotive Radar Transceiver
The radar uses an isotropic element to transmit and a uniform linear array (ULA) to receive the radar waveforms. Using a linear array enables the radar to estimate the azimuthal direction of the reflected energy received from the target vehicles. The long-range radar needs to detect targets across a coverage area that spans 15 degrees in front of the ego vehicle. A six-element receive array satisfies this requirement by providing a 17-degree half-power beamwidth.
Defining the Radar Signal Processing Chain
The radar collects multiple sweeps of the waveform on each of the linear phased array antenna elements. These collected sweeps form a data cube, which is defined in Radar Data Cube. These sweeps are coherently processed along the fast- and slow-time dimensions of the data cube to estimate the range and Doppler of the vehicles.
Estimate the direction-of-arrival of the received signals using a root MUSIC estimator.
Identify detections in the processed range and Doppler data by using a constant false alarm rate (CFAR) detector. CFAR detectors estimate the background noise level of the received radar data. Detections are found at locations where the signal power exceeds the estimated noise floor by a certain threshold. Low threshold values result in a higher number of reported false detections due to environmental noise. Increasing the threshold produces fewer false detections, but also reduces the probability of detection of an actual target in the scenario.
The phased.RangeEstimator and phased.DopplerEstimator objects convert the locations of the detections found in the range-Doppler data into measurements and their corresponding measurement variances. These estimators fit quadratic curves to the range-Doppler data to estimate the peak location of each detection. The resulting measurement resolutions are a fraction of the range and Doppler sampling of the data.
The root-mean-square (RMS) range resolution of the transmitted waveform is needed to compute the variance of the range measurements. The Rayleigh range resolution for the long-range radar was defined previously as 1 meter. The Rayleigh resolution is the minimum separation at which two unique targets can be resolved. This value defines the distance between range resolution cells for the radar. However, the variance of the target within a resolution cell is determined by the RMS resolution of the waveform.
To further improve the precision of the estimated vehicle locations, pass the radar detections to a tracker. Configure the tracker to use an extended Kalman filter (EKF), which converts the spherical radar measurements into the Cartesian coordinate frame of the ego vehicle. Also configure the tracker to use constant velocity dynamics for the detected vehicles.
Modeling Free Space Propagation Channel
Use the free space channel to model the propagation of the transmitted and received radar signals.
In a free space model, the radar energy propagates along a direct line-of-sight between the radar and the target vehicles, as shown in the following illustration.
Simulating the Driving Scenario
Create a highway driving scenario with three vehicles traveling in the vicinity of the ego vehicle. The vehicles are modeled as cuboids and have different velocities and positions defined in the driving scenario. The ego vehicle is moving with a velocity of 80 km/hr and the other three cars are moving at 110 km/hr, 100 km/hr, and 130 km/hr, respectively. The radar sensor is mounted on the front of the ego vehicle.
The following loop uses the drivingScenario object to advance the vehicles in the scenario. At every simulation time step, a radar data cube is assembled by collecting 192 sweeps of the radar waveform. The assembled data cube is then processed in range and Doppler. The range and Doppler processed data is then beamformed, and CFAR detection is performed on the beamformed data. Range, radial speed, and direction of arrival measurements are estimated for the CFAR detections.
The previous figure shows the radar detections and tracks for the three target vehicles at 1.1 seconds of simulation time. The plot on the upper-left side shows the chase camera view of the driving scenario from the perspective of the ego vehicle (shown in blue). For reference, the ego vehicle is traveling at 80 km/hr and the other three cars are traveling at 110 km/hr (orange car), 100 km/hr (yellow car), and 130 km/hr (purple car).
The right side of the figure shows the bird's-eye plot, which presents a top down perspective of the scenario. All of the vehicles, detections, and tracks are shown in the coordinate reference frame of the ego vehicle. The estimated signal-to-noise ratio (SNR) for each radar measurement is printed next to each detection. The vehicle location estimated by the tracker is shown in the plot using black squares with text next to them indicating the ID of each track. The velocity for each vehicle estimated by the tracker is shown as a black line pointing in the direction of the velocity of the vehicle. The length of the line corresponds to the estimated speed, with longer lines denoting vehicles with higher speeds relative to the ego vehicle. The track of the purple car (ID2) has the longest line while the track of the yellow car (ID1) has the shortest line. The tracked speeds are consistent with the modeled vehicle speeds previously listed.
The two plots on the lower-left side show the radar images generated by the signal processing. The upper plot shows how the received radar echoes from the target vehicles are distributed in range and radial speed. Here, all three vehicles are observed. The measured radial speeds correspond to the velocities estimated by the tracker, as shown in the bird's-eye plot. The lower plot shows how the received target echoes are spatially distributed in range and angle. Again, all three targets are present, and their locations match what is shown in the bird's-eye plot.
Due to its close proximity to the radar, the orange car can still be detected despite the large beamforming losses due to its position well outside of the 3 dB beamwidth of the beam. These detections have generated a track (ID3) for the orange car.
Modeling a Multipath Channel
The previous driving scenario simulation used free space propagation. This is a simple model that models only direct line-of-sight propagation between the radar and each of the targets. In reality, the radar signal propagation is much more complex, involving reflections from multiple obstacles before reaching each target and returning back to the radar. This phenomenon is known as multipath propagation. The following illustration shows one such case of multipath propagation, where the signal impinging the target is coming from two directions: line-of-sight and a single bounce from the road surface.
The overall effect of multipath propagation is that the received radar echoes can interfere constructively and destructively. This constructive and destructive interference results from path length differences between the various signal propagation paths. As the distance between the radar and the vehicles changes, these path length differences also change. When the differences between these paths result in echoes received by the radar that are almost 180 degrees out of phase, the echoes destructively combine, and the radar makes no detection for that range.
Replace the free space channel model with a two-ray channel model to demonstrate the propagation environment shown in the previous illustration.
The previous figure shows the chase plot, bird's-eye plot, and radar images at 1.1 seconds of simulation time, just as was shown for the free space channel propagation scenario. Comparing these two figures, observe that for the two-ray channel, no detection is present for the purple car at this simulation time. This detection loss is because the path length differences for this car are destructively interfering at this range, resulting in a total loss of detection.
Plot the SNR estimates generated from the CFAR processing against the range estimates of the purple car from the free space and two-ray channel simulations.
As the car approaches a range of 72 meters from the radar, a large loss in the estimated SNR from the two-ray channel is observed with respect to the free space channel. It is near this range that the multipath interference combines destructively, resulting in a loss in signal detections. However, observe that the tracker is able to coast the track during these times of signal loss and provide a predicted position and velocity for the purple car.
Radar Jamming Mitigation Techniques
As automotive radars become more widespread, the heavily occupied RF spectrum will resemble an electronic battlefield in an urban environment. An automotive radar could typically experience a denial or deceptive jamming attack. Denial jamming blinds the victim radar. This technique reduces the signal-to-noise ratio and, as a result, the probability of target detection is degraded. On the other hand, deceptive jamming makes victim radar âthinkâ there are false targets. The victim radar loses the ability to track real targets and, hence, the victimâs vehicle behavior is severely influenced.
While current jamming avoidance techniques might be adequate for today, with the proliferation of radar sensors, a resilient type of mitigation techniques will need to be used by itself or in conjunction with the avoidance approaches. The radar waveform is one of the critical system parameters that determines the sensor performance in the presence of jammers. Automotive radars in todayâs 77 GHz band mainly use FMCW type waveforms. In FMCW radar, a CW signal is linearly swept or chirped in frequency across an RF band.
An arbitrary FMCW type strong jamming signal that falls into the receiver bandwidth raises the noise floor of the victim radar. A denial attack could also purposefully be carried out by simply beaming a strong CW signal into the victim FMCW radar. If the jamming signal sweep is synchronized but delayed with the victim radar, then the impact would be deceptive false target generation at a fixed range.
More generally, deceptive jamming is based on retransmitting the victim radarâs signal with a systematic change in delay and frequency. This can be either noncoherent, in which case the jammer is called a transponder, or coherent, when itâs a repeater. A sophisticated repeater-based attack would typically require a digital RF memory (DRFM). A DRFM is capable of carrying out coordinated range delay and Doppler gate pull off attacks. So, the false target range and Doppler properties are maintained to deceive the victim radar.
Basic radar jamming mitigation techniques mostly rely on the avoidance approach:
- Spatial: Use of narrow and electronically scanned beam can reduce jamming risk.
- Spectral: Randomize FMCW chirp start and stop frequencies to reduce the probability of overlap and jamming.
The growing number of radar sensors will require more sophisticated resilience techniques to mitigate the jamming. An alternative method of avoidance could be used to repair the received waveform using signal processing algorithms. Time frequency domain techniques could be effective against the denial type of jamming attacks.
For time varying signals, a short time Fourier transform (STFT) provides more information than a regular FFT. STFT-based techniques could be used for narrow-band jamming excision. The STFT essentially moves a window through the signal and takes the FFT of the windowed region.
An amplitude-based masking could be used to filter out the jamming signal in the STFT domain. This assumes, of course, that the victim radar front end and quantization have enough dynamic range to process the stronger jammer signal and the small intended target linearly at the same time.
For deceptive jamming attacks, an STFT alone cannot authenticate whether the return signal is real or false. The elementary countermeasure to reduce the impact of deceptive jamming from repeater attacks is the use of low probability of intercept (LPI) radar waveforms. The objective of an LPI radar is to escape detection by spreading the radiated energy over a wide spectrum of frequencies, usually via quasi-random sweep, modulation, or hopping sequences. FMCW is a type of LPI waveform.
Phase coded FMCW radars could offer high jamming robustness in this use case.
Jamming in congested automobile radar sensor environments can be mitigated using advance signal processing algorithms and complex waveform generation techniques. STFT-based signal processing techniques could be used against denial type of attacks. Phase coded FMCW provides an additional layer of resistance to both noncoherent and coherent deceptive attacks by means of processing gain and interception avoidance.
| Technique | Description | Mitigation |
|---|---|---|
| Spatial | Use of narrow and electronically scanned beam | Reduces jamming risk |
| Spectral | Randomize FMCW chirp start and stop frequencies | Reduces the probability of overlap and jamming |
| Time Frequency Domain | STFT-based techniques | Effective against denial type of jamming attacks |
| Phase Coded FMCW | Provides processing gain and interception avoidance | Resistant to both noncoherent and coherent deceptive attacks |
Currently automotive radar sensors are operating in noncooperative mode without communicating with each other. Although a cooperative mode of operation would require industry-wide harmonization, the arbitration between radar sensors could help resolve the interference issue. A future radar concept including the sensor cooperation would be the fusion of communication nodes and radar sensors. Future radars with complex waveforms offer the possibility to include information in the radar signal as well. The same HW could be used for radar and communications (RadCom) simultaneously.
The trend of increasing bandwidths â today up to 4GHz â allows a better object recognition than ever before.
Analyzing Radar Signals with Oscilloscopes and Spectrum Analyzers
Oscilloscopes' wide analysis bandwidth and manifold trigger capabilities, make them a good fit to analyze radar signals for automotive applications. Multi-channel measurements on MIMO radar sensors and correlation with other signals e.g. from cameras is possible. When it comes to analyzing radar systems, a key challenge is that there is no one size fits all solution. In part, this is because both time and frequency domain information are necessary to characterize radar systems.
Pulse analysis is naturally suited to time-domain instrumentation, such as oscilloscopes. Analysis and decoding of pulse modulation is naturally suited to frequency-domain instrumentation such as spectrum analyzers. Simulation of radar systems today requires extensive digital signal processing (DSP) of received signals for pulse compression and Doppler processing.
Pulse RF carrier frequency is basic. If the available equipment does not cover the frequencies involved, then a frequency conversion device will be required in addition to the fundamental tester. Such a converter may introduce phase and flatness impairments or other distortion. Pulse bandwidth is the next consideration. Modern radars are using wider bandwidth pulses, such as faster rise times and wider modulation bandwidths.
A useful application for fast acquisition is finding baseband pulse time-domain transient errors. Below you can see that just one single pulse that has a narrower pulse width than even hundreds of thousands of correct pulses.
One of the most highly developed capabilities of the oscilloscope is triggering. For example, a high-performance oscilloscopeâs triggering system can detect transient glitches less than 200 ps wide. Advanced trigger types, such as pulse width trigger, can be used to capture and examine specific RF pulses in a series of pulses that vary in time or in amplitude. A mixed domain oscilloscope such as the Tektronix MDO4000C can add a further trigger advantage, providing the ability to simultaneously trigger and acquire signals across multiple channels.
Complementing the advanced trigger system, a fully-automated suite of pulse timing measurements available with Tektronix oscilloscopes can lead to more consistent results.
A traditional swept spectrum analyzer is a simple RF detector that effectively sweeps across a selected span of RF frequencies. This produces a display of the combined RF spectrum of all signals within the selected span of frequencies.
Occupied bandwidth is the most relevant for pulsed radar. Most radars have to meet a specified bandwidth to avoid interfering with RF systems operating on nearby frequencies. This measurement examines the RF spectrum of the signal and locates the highest amplitude value.
For more capability, however, youâll want to turn to a real-time spectrum analyzer (RSA). These instruments have an RF conversion section similar to a swept spectrum analyzer. The digitized samples are directly processed by a hardware DSP, and can be simultaneously saved in memory or on a hard disk. This hardware processor performs discrete time transforms into RF spectrum information. This can provide real-time triggering on selected frequency events, or a digital phosphor spectrum display that can discover RF transients and display same-frequency time-sharing RF signals.
Many types of techniques are being used in radar systems today with the result that a range of test equipment is required for advanced analysis and troubleshooting. To succeed in todayâs world, designers need capable tools to validate designs with advanced scanning methodologies â tools that can handle complex radar baseband, IF and RF signals as well as identify multi-system interference. In addition, analysis in both time and frequency domains is necessary to fully characterize modulated pulse radars.