Article By : Junko Yoshida
The little-noted issue of radar safety—due to its own signal interferences—is emerging as the potential Achilles heel of driver-assist and highly automated vehicles.
The little-noted issue of radar safety is emerging as a potential Achilles heel of driver-assist and highly automated vehicles: radar signals interfere with each other.
Radar has become an essential sensing modality complementing CMOS imaging cameras. Radar works in all weather conditions and enables a range of automated driving features, including automatic emergency Bbrakes (AEB). But radars can be foiled or faulty if they end up, like Ghostbusters’ particle accelerators, crossing each other’s streams.
While this is not yet a phenomenon publicly cautioned by carmakers or commonly perceived by drivers, automotive radars, operating in congested environments, will face significant interference.
Radar’s application segment ranges from adaptive cruise control and blind spot detection to forward collision warning systems and intelligent park assist. For a vehicle to get a 360-degree view, it needs both short-range and long-range radar chips. AEB typically uses radar (all-weather) and sometimes lidar and cameras to detect an imminent crash.
AEB’s rapid proliferation on the global market has become a double-edged sword for radar sensor suppliers. It is both a cause for celebration and a source of concern.
For example, China New Car Assessment Program (NCAP) already mandated AEB for all trucks coming out of the factory to the market in 2020. New cars in Japan must have the front and rear AEB functions, starting this year. In the United States, 20 carmakers have agreed to a voluntary “100-percent AEB fit rate in new cars in 2022.” Facing Euro NCAP’s 2019 requirements, 90 percent of cars sold in Europe already come with the latest crash-avoidance technology for car-to-car impacts.
NXP Semiconductors projects the automotive radar penetration rate to jump to 55 percent in 2030. In a recent interview with EE Times, Huanyu Gu, NXP Semiconductors’ senior product marketing manager responsible for ADAS and V2X, warned that radar interference is inevitable, saying that “when multiple radars transmit at the same time and in the same or overlapped frequency, and if they then share a common visible path.”
NXP’s Gu isn’t alone in worrying. Martin Duncan, general manager of ADAS and ASIC division of ST Microelectronics, also told EE Times, “The fact that we have now 25% of new vehicles with radar systems, it is already an issue. If you try to capture in real time road conditions, it is very easy to see transmissions from multiple vehicles. As we are all using the same frequency band this will potentially worsen as fitment rate increases.”
The principle of radar congestion is straightforward. The National Highway Traffic Safety Administration (NHTSA) wrote in its “Radar Congestion Study” issued in September 2018:
Radars use knowledge of radiated signals to identify echoes and estimate the range and speed of objects in the environment. These echoes are not perfect copies of the original signal, but a sum of multiple returns that constructively and destructively interfere with the signal. It is important to understand that returns from objects illuminated by radar fluctuate, especially when the relative range, aspect and other objects in the scene change. With multiple radars operating in near proximity and an environment of multiple sources of scattering, the performance of each radar degrades as the interference level rises.
That could lead to compromised safety. “Worst case scenario would be fatalities caused by radar interference. There is already a heightened use of filtering against false positives in the radar stack today, regardless of the root cause,” noted VSI Labs founder and president Phil Magney.
The industry had been warned
The more radar-equipped cars hit the road, the faster each radar must learn to deal with the presence of other radars. Radar suppliers are under pressure.
Radar interference is hardly an unexpected consequence of radar proliferation. The automotive industry had been warned. More than a decade ago, Europe put together a project called MOSARIM (MOre Safety for All by Radar Interference Mitigation) and issued a report in 2012. The project investigated “mutual vehicular radar interference and the definition and elaboration of effective countermeasures and mitigation techniques.”
More recently, NHTSA conducted a “Radar Congestion Study,” modeling and simulating radar interference with two questions in mind:
- How much power does a given radar receive from other radar transmitters?
- How does this affect the performance of a collision warning system?
The report concluded:
…Systems that operate well in environments with few other radars may suffer significant degradation of performance in radar congested environments. The results of the study show, levels of interference based on operation of current systems in congested environments will be significant. In scenarios with many vehicles operating radars in the 76-81 GHz band, the power from other radars will likely exceed the power of echoes from targets needed for specified performance, by several orders of magnitude.
Collaboration among radars?
So, the industry has known for a decade about the imminence of radar traffic jams. What actions has it taken?
With all this lead time, you might expect car OEMs and Tier 1s to develop a robust strategy to avoid interference. You might envision a radar sensor that avoids interference by adapting waveform parameters dynamically.
This is not a rocket science. The radar community has the know-how to borrow similar types of channel access rules already deployed by the telecommunication industry (as in TDMA, FDMA and CDMA). Such a “listen-before-talk” scheme should enable more structured communication among radars, said NXP’s Gu.
Unfortunately, that isn’t the interference mitigation adopted by the industry. Aside from the fact that automotive radars use the same allocated frequency spectrum (76GHz~81GHz), the radar community is under no regulations. “Radar waveform parameters are not regulated,” noted Gu.
Industry agreements, standardization and regulation have never been part of the auto industry’s DNA.
A common approach adopted today is “to limit the interference by randomizing the transmitted signals either in time or frequency,” according to Gu. Acknowledging the illogic behind this randomization, Gu said, “Today, you are doing this blindly. This is certainly fine especially if there are not that many cars on the road with radars. But if you are to improve the robustness of radar sensor to interferences, you must look for collaboration among radar sensors.”
But that would require regulation.
Nonetheless, in its own white paper on radar interference, NXP concluded:
Eventually, to support a high market penetration, some form of agreement between the manufacturers will be needed to more effectively share the sensing resources in a fair way. This last step means that all the players in the market will have to sit together to define a standardized way to access the channel while at the same time maintaining the possibility to have differentiating sensing performance.
Free for all
Radar has always been a “free for all,” observed Egil Juliussen, an independent veteran automotive industry analyst. In pursuit of innovations, radar sensor companies are typically inclined to develop new, proprietary algorithms that run on DSPs or MCUs associated with sensor chips so that their radar can improve imaging resolution and mitigate interference, he explained.
In other words, for many in the auto industry, throwing more signal processing at the radar interference problems is a more preferred approach than any industry agreements or regulations.
During our interview, NXP’s Gu posed three different approaches for radar interference mitigations: 1) avoid saturation at the front-end; 2) manage digital interference by recognizing and removing interference in the digital domain; 3) avoid interference by adapting waveform parameters dynamically.
The third approach is already deemed least likely to be adopted in the current 77GHz spectrum. Gu explained, “People think it’s too late because we already have too many radar sensors on the road, and those sensors would not collaborate.” He added that the scheme could be “applied to the 140Ghz frequency in the future, if that band is made available for radar.”
The first — more likely — approach is to devise techniques that avoid saturation of the front end. Here, at least a part of the wanted signal can be received, and the proper countermeasures taken. “You can do this by providing the radar receiver with two different gain settings,” said Gu. Alternatively, the system could include “spatial nulling,” in which the front-end uses multiple antennas to blind itself in the direction generating interference. This approach seeks to eliminate an interfering signal before it saturates the front-end, explained Gu.
Radar chip suppliers like NXP tend to focus on handling interference in the digital domain – in DSP. “Of course, its precondition is that actually the wanted signal is not being buried by the strong signals,” said Gu.
After determining that the interfering signal is relatively weak, it can be digitized together with the wanted signal, without causing the front-end to saturate.
But the name of the game is first recognizing whether the signal has been corrupted, which is easier said than done, according to NXP. The techniques to do so depend on the specific radar waveform of both the victim radar and the interference. Because today’s regulatory framework allows different ways of building radar waveforms, every radar sensor manufacturer chooses its own, rendering the process not only diverse but tricky.
The de facto standard in automotive radar is frequency-modulated continuous wave (FMCW) radar. FMCW offers very good performance that is relatively simple and elegant. It covers a large bandwidth with a low-bandwidth ADC and provides a robust estimation of target velocity, according to NXP. But it comes with some caveats.
Different manufacturers use different parameter settings of FMCW waveforms to differentiate their product proposition and cover different application requirements, such as carrier frequency, bandwidth, chirp duration, sampling time, duration of the sensing cycle and different ways of changing parameters during a sensing period.
To recap: the radar sensor first needs to recognize if there is an interferer. The detection of interference works by recognizing unique characteristics of the alien signal. Once the interference is detected, system algorithms must remove it from the received signal as completely as possible while not corrupting or removing the wanted signal.
None of this should surprise anyone in the radar community. “There are textbook signal processing algorithms out there on the market, and they are already in use by the industry,” said Gu.
Textbook algorithms, however, have limits, he noted. “They’re often limited to dealing with low correlated interferences. And they are also capable with dealing only a very limited number of interferences — one or two at a time.”
NXP’s goal is to further develop its differentiated advanced digital signal algorithms for removing interference.
ST is working on its own methodology. Duncan said, “If you are aware of what the radar chirp is meant to be you can easily filter/ignore spurious signals. It is also possible to introduce signatures between chirps.”
However, Duncan added, “If there was more standardization/sharing on what is transmitted, it would help on the countermeasures to remove unwanted signals.”
Feel radar interference?
NHTSA laid out a couple of scenarios simulating interferences expected in radar congestion.
- In the case of traffic on a two-lane highway, assuming that the radars use randomly selected carrier frequencies, NHTSA predicted that “an automotive radar would encounter power from other radars far greater than the echoes of its own transmissions needed to track other vehicles. The interference approaches four orders of magnitude, or nearly 40 dB, greater than echoes typical of a reference target, as specified for the system.”
- In radars that face rearward (as in blind-spot detection systems), “these units are vulnerable to the direct arrival of forward collision avoidance radars that utilize higher power and antenna gain.” The study said, “Our analysis shows these units could experience interfering power from a forward collision avoidance radar that is nearly five orders of magnitude, or 50 dB, greater than the reflections from their specified reference target.”
Thus far, however, the impact of radar hasn’t been felt on real-world roads.
“Radar-to-radar interference is a still unknown and as applied researchers who work with radar almost every day VSI cannot say we have ever experienced radar-to-radar interference from another vehicle while testing on public roads,” said VSI Labs’ Magney. “We can assume that we are exposed because so many vehicles on the road today have a mix of radars for short ranging to long ranging,” he added.
During the Tesla’s Q1 financial call Monday, CEO Elon Musk reiterated plans to eliminate radars from Tesla vehicles, making moot the issue of radar interference — at least for Tesla vehicles.
Other car OEMs, Tier 1s and automotive technology suppliers, however, aren’t ditching radar any time soon.
Radars in particular are critical because they’re weatherproof, stressed VSI’s Magney. “Radar is one of the most cost-effective ADAS sensors and the penetration will grow dramatically over the years ahead.”
This article was originally published on EE Times.
Former beat reporter, bureau chief, and editor in chief of EE Times, Junko Yoshida now spends a lot of her time covering the global electronics industry with a particular focus on China. Her beat has always been emerging technologies and business models that enable a new generation of consumer electronics. She is now adding the coverage of China’s semiconductor manufacturers, writing about machinations of fabs and fabless manufacturers. In addition, she covers automotive, Internet of Things, and wireless/networking for EE Times’ Designlines. She has been writing for EE Times since 1990.