WorksComputer Science
Collision Avoidance and Collision Detection Strategies to Counteract Large-Scale FMCW Automotive Radar Interference
Abstract
Automotive radars operating in dense traffic environments may experience significant mutual interference, potentially degrading sensing reliability for advanced driver assistance systems (ADAS). This issue motivates the development of techniques to mitigate interference among multiple radar-equipped vehicles sharing the same spectrum. This paper investigates interference mitigation strategies for uncoordinated radar networks through system-level simulation. Two approaches are considered: frame-by-frame collision avoidance (FbF-CA), which exploits idle intervals to measure the perceived interference and possibly adapt the operating frequency, and frame-by-frame collision detection (FbF-CD), which estimates interference during transmission based on observed chirp collisions. An event-driven radar network simulator is used to model realistic highway traffic and radar interactions. Results show that collision avoidance provides higher robustness when a larger bandwidth is available, while collision detection is more effective under bandwidth-limited conditions. The analysis highlights the trade-off between collision avoidance and collision detection mitigation strategies and provides design insights for scalable automotive radar deployments.
Introduction
Automotive radars represent a key sensing technology for advanced driver assistance systems (ADAS) and automated driving. Compared to other sensing modalities such as cameras or lidar, radar sensors provide reliable detection and tracking capabilities even under adverse weather conditions and at long distances. For these reasons, modern vehicles are increasingly equipped with multiple frequency modulated continuous wave (FMCW)-based radar sensors operating in the millimeter-wave bands in the 76–81 GHz band, with regional variations. Thanks to their robustness and relatively low cost, radar sensors are expected to play an important role in future intelligent transportation systems.
However, the growing penetration of radar-equipped vehicles also introduces new challenges. Radar signals transmitted by nearby vehicles may overlap in time and frequency, generating interference that can affect sensing reliability. In FMCW automotive radars, such interference can produce signal components that resemble legitimate reflections, potentially leading to corrupted chirps, degraded sensing performance, or even the detection of false targets. This problem is further exacerbated by the fact that automotive radars typically operate in a shared spectrum with limited coordination among devices. As radar adoption continues to increase, the probability that multiple sensors operate simultaneously within the same spectrum increases, raising concerns about the scalability of current radar spectrum allocations in dense traffic scenarios [1].
Several studies have investigated the impact of radar-to-radar interference in vehicular environments, highlighting that the currently available spectrum may become insufficient in dense traffic scenarios [2, 3, 4, 5]. While on the one hand the use of additional spectrum around 140 GHz is being investigated [6], on the other interference mitigation methods are also being considered. Interference mitigation can be achieved either by reducing it a priori, before transmission (proactive approaches) or by processing the received signal after interference occurs (reactive approaches). While reactive approaches can be used in addition to proactive ones, they are outside the scope of this work.
Focusing on proactive approaches, they can be broadly categorized into coordination-based and uncoordinated techniques. Coordination-based solutions rely on explicit cooperation among radar devices but require communication and standardization across different manufacturers, which may limit their practical deployment [7]. Uncoordinated approaches instead rely on local strategies such as randomization of transmission parameters or sensing-based adaptation of the operating frequency. Examples include frame-by-frame or chirp-by-chirp frequency hopping techniques and random-access radar transmission schemes [8, 9, 10, 11].
Very few works have included mechanisms that perform measurements to deal with interference [11, 12, 13]. All of these works consider only the sensed power and introduce actions in the time domain. In [12], [13], the authors propose to add a carrier sensing mechanism before each transmission, with random backoff in the case a signal is received. The proposal is investigated assuming no propagation delay, which means an ideal and unrealistic scenario where radars are co-located. A similar idea is explored in [11], where time is slotted and sensing is performed in the slots, possibly followed by random backoff. All results appear preliminary and are obtained with one or a few interferers, without considering system-level interactions among multiple radar-equipped vehicles operating in realistic traffic scenarios.
In this paper, we investigate mitigation strategies based on measurements for automotive radar interference through system-level simulations in realistic vehicular environments. In particular, we consider two complementary approaches: a collision avoidance (CA) mechanism that adapts the operating frequency based on interference measurements during idle periods and a collision detection (CD) mechanism that updates the operating frequency based on repeated chirp collisions observed during radar operation. An event-driven radar network simulator is used to model radar transmissions and interference interactions in dense highway scenarios with multiple vehicles.
The main contributions of this work are:
The development of a system-level simulation framework for analyzing radar interference in realistic vehicular traffic scenarios.
The definition and implementation of sensing-based mitigation strategies based on CA and CD.
A comparative evaluation of the considered strategies.
System Modeling
Modeling a highway scenario with many interfering radar-equipped vehicles requires balancing accuracy and computational complexity. In this work, we use the approach detailed in [5] and hereafter recalled.
FMCW Radar Signal Model
FMCW radars operate by transmitting a sequence of chirps, where each chirp is a signal whose frequency increases linearly over time within a given bandwidth. The transmitted chirps are reflected by surrounding objects, generating delayed echoes that are processed to estimate target range and velocity. The structure of a chirp is shown in Figure 1a. Each chirp spans a bandwidth and has duration . Chirps are transmitted periodically within a repetition interval , which includes and an idle time required to reset the oscillator. The starting instant of each chirp is assumed to vary randomly within , from chirp to chirp. A frame consists of consecutive chirps, as illustrated in Figure 1b, followed by an additional idle interval. The total frame duration, including active and inactive periods, is denoted as . The duty cycle is defined as the ratio between active transmission time and total frame duration, i.e., .
Interference Modeling and Failure Abstraction
In FMCW radars, interference arises when signals transmitted by neighboring radar sensors overlap in time and frequency with the chirps of another radar. Depending on the characteristics of the interfering signals, a distinction is commonly made between uncorrelated and correlated interference. Uncorrelated interference results from different FMCW waveforms, leading to an increase in the noise floor, whereas correlated interference occurs when interfering radars use similar chirp patterns, potentially generating ghost targets.
In this work, all radars are assumed to operate within the same frequency band, spanning from to , and to use identical waveform parameters, resulting in correlated interference. Under this assumption, interfering chirps have the same slope and may overlap, as illustrated in Figure 1. This assumption will be relaxed in future work; however, in practice, radars designed for different functions (e.g., long-range front radars and medium-range corner radars) typically use different bands.
(a)
(b)
Figure 1. Representation of (a) the chirp and (b) the frame.
Interference is modeled at the chirp level using the chirp and frame structure described in Section II-A, based on time-frequency overlap conditions. In particular, a chirp is affected only if an interfering signal falls within the analog to digital conversion (ADC) observation bandwidth during its reception interval. Following the approach in [5], a chirp is considered corrupted if such an overlap persists for at least a fraction of the chirp duration. This corresponds to severe interference, potentially leading to chirp loss within a frame. A frame is considered lost if the number of corrupted chirps is greater than or equal to . Finally, a system failure is declared when consecutive frames are lost. The symbols used in the model are summarized in Table I.
| Parameter | Symbol | Front | Corner |
| Transceiver settings | |||
| EIRP | 35 dBm | 15 dBm | |
| Rx antenna gain | 30 dBi | 23 dBi | |
| Field of view | FoV | 30° | 60° |
| Noise figure | 15 dB | 15 dB | |
| Carrier frequency | 140 GHz | 140 GHz | |
| Signal settings | |||
| Duration of a chirp | |||
| Chirp repetition time | |||
| # of chirps in a frame | 2000 | 1555 | |
| Duration of a frame | 12.8 ms | 20 ms | |
| Duty cycle | 0.5 | 0.25 | |
| Frame repetition time | 25.6 ms | 80 ms | |
| Available bandwidth | Variable | Variable | |
| Chirp bandwidth | 150 MHz | 1.5 GHz | |
| Max. beat frequency | 68.1 MHz | 97.29 MHz | |
| ADC bandwidth | 100 MHz | 100 MHz | |
| ADC sample frequency | 136.2 MHz | 194.6 MHz | |
| Radar metrics | |||
| Unambiguous range | 350 m | 100 m | |
| Range resolution | 1.00 m | 0.10 m | |
| Maximum velocity | 83.44 m/s | 41.85 m/s | |
| Velocity resolution | 0.0834 m/s | 0.0538 m/s | |
| Reflection modeling | |||
| Radar cross section | |||
| Minimum INR | 0 dB | 0 dB | |
| Maximum path length | 2694 m | 120 m | |
| Success/failure modeling | |||
| Minimum chirp overlap | 0.5 | 0.5 | |
| Corrupted chirps for frame loss | 400 (20% of ) | 311 (20% of ) | |
| Consecutive frame losses for system failure | 3 | 3 |
Table I. Symbols and settings.
Interferer Selection and Propagation Model
Given the potentially large number of radars in the scenario, considering all mutual interactions would be computationally prohibitive and often unnecessary. We therefore restrict the analysis to potential interferers, defined as those radars whose signals fall within the field of view (FOV) of a receiver with sufficient power to produce non-negligible interference.
At each time instant, potential interferers are identified from the positions of vehicles and their associated radars by considering both direct and first-order reflected propagation paths,1 and retaining only those sources for which the resulting interference to noise ratio (INR) exceeds a minimum threshold . For each interferer, only the dominant propagation path is considered, corresponding to the direct path under line of sight (LOS) conditions or the least attenuated reflected path under non line of sight (NLOS) conditions.
In the case of LOS, denoting by the transmission power, and the transmitting and receiving antenna gains, respectively, the signal wavelength, and the distance between transmitter and receiver, the received power can be calculated as
Denoting by the minimum received power required to exceed , the maximum distance is given by
where
with denoting the Boltzmann constant, the reference temperature of 290 K, the ADC bandwidth, and the receiver noise figure.
In the case of reflected interference (i.e., NLOS), the signal propagates over two paths, denoted as and , corresponding to the distances from the interferer to the reflection point and from the reflection point to the receiver, respectively. Using the same definitions as for the LOS case and introducing the radar cross-section (RCS) , the received power is given by
From (1) and (4), an equivalent distance can be derived as
which corresponds to the distance at which an interfering source in LOS would produce the same received power. A reflected interferer is therefore considered a potential interferer if and only if .
Sensing-Based Interference Mitigation Proposals
Hereafter, the frame-by-frame collision avoidance (FbF-CA) and the frame-by-frame collision detection (FbF-CD) proposals are described.
Frame-by-Frame Collision Avoidance
In the FbF-CA method, each radar starts by randomly selecting a starting frequency between and . Then, the same carrier frequency is maintained unless a certain level of interference power is measured during the idle interval between two consecutive frame transmissions.
More specifically, as shown in the flow diagram in Figure 2, the core step of this technique is the computation of the total detected interference power received from all the interfering radars during the non-transmitting period , called . Assuming that there are potential interferers, the interference power contribution from the -th potential interferer is given by
Figure 2. Frame-by-frame collision avoidance flow diagram.
where is the interference power per chirp, denotes the fraction of the interfering frame in the frequency domain that overlaps with the sensing window, and represents the number of chirps that fall within the sensing window. The total detected interference power is then obtained by summing the contribution of all interferers as
If the detected interference exceeds a predefined threshold , i.e., if , a new random starting frequency is selected between and for the transmission of the next frame.
Figure 3 portrays the proposed method from the perspective of a generic radar suffering from interference. In the figure, the inactive interval after the frame transmission is represented by a rectangle of time duration and bandwidth , also referred to as the sensing window. At the end of this interval, the given radar performs the subsequent frame transmission based on the interference detected inside the sensing window. The red and light blue rectangles in Figure 3 represent the frames transmitted by potential interferers that do and do not occupy the same time-frequency resources within the sensing window, respectively.
Figure 3. Example of interference power detection with frame-by-frame collision avoidance.
Frame-by-Frame Collision Detection
In the FbF-CD method, again each radar starts by randomly selecting a carrier frequency within and for its transmissions. Then, the starting frequency for the next frame transmission is changed if a certain interference is measured during the transmission. To identify the presence of interference, the receiver counts the number of chirps that are corrupted during the frame transmission.
The behavior of the collision detection algorithm is detailed in the flow diagram illustrated in Figure 4. The collision detection approach relies on the parameter , which represents the minimum number of corrupted chirps that triggers a change. More specifically, during the active transmission interval , the total number of collided (i.e., corrupted) chirps is computed by the receiver of the given radar. If the number of collided chirps equals or exceeds the threshold then the radar randomly changes its transmitting frequency.
Figure 4. Frame-by-frame collision detection flow diagram.
Figure 5 shows an example of the collision detection behavior obtained from the radar network simulator, with . In this example, after the first transmission, the given radar changes its frequency, because two chirp collisions are detected, i.e., the collided chirps are equal to . In contrast, during the second transmission all chirps are received without collisions and the radar preserves the same starting frequency.
Figure 5. Example of frame-by-frame collision detection mitigation method.
Performance Evaluation
This section presents the results with reference to a dense highway scenario. The main settings, discussed hereafter, are also summarized in Table I.
Simulation settings
The results are obtained using a dynamic event-driven simulator written in Matlab, with vehicle positions obtained from the traffic simulator SUMO [14] and the potential interferers obtained from the WiLabVIsim [15].
Scenario
Scenario: The scenario is a highway with 3 lanes per direction and 150 average vehicles per km, with positions and mobility reproduced by SUMO. This density represents a reference scenario with congested traffic to study a critical case from the radar interference perspective [15].
All vehicles are equipped with one long-range front radar and four mid-range corner radars, as shown in Figure 6; front and corner radars use separate bands. For the evaluation of reflected interferers, the vehicles are treated as obstacles modeled using the eight reflecting points in Figure 6, as in [16]. The RCS for reflection is assumed equal to , following the ETSI indications in [17]. The signal parameters are taken from [18] and reported in Table I. The adopted configuration targets operation around 140 GHz, as higher frequencies are being actively explored for future systems [19]. The observed trends are expected to generalize across frequency bands.
Figure 6. Modeling of the vehicle, with either one front radar or four corner radars, and with eight points of reflection for the calculation of reflected interference.
Benchmarks
We compare our proposals with the following benchmarks [5]:
Baseline: The starting frequency of each radar is randomly selected between and according to a uniform distribution. The same starting frequency is kept by the given radar and never changed.
Frame-by-frame (FbF) frequency hopping: The starting frequency of the first chirp of each new frame is randomly selected within and with uniform distribution, and then the same frequency is used for the following chirps of the same frame.
Chirp-by-chirp (CbC) frequency hopping: The starting frequency of each chirp is randomly selected within and with uniform distribution.
The baseline approach is likely representative of current market radar implementations, and the chirp-by-chirp provides a relevant benchmark while being more theoretical, since changing the frequency from chirp to chirp arbitrarily is still challenging to implement in practice.
Output Metric
The performance of the proposed techniques is evaluated in terms of the average time between two system failures, calculated as
where is the probability of having a system failure, which corresponds to the loss of consecutive frames, where each loss is caused by at least 20% of corrupted chirps. This probability is, in turn, estimated by the ratio between the number of system failures experienced by each radar and the maximum possible number of failures that would occur if all transmissions were lost.
Results
Preliminary Considerations for Collision Avoidance
Figure 7 provides preliminary results on the FbF-CAC technique, with focus on the detected interference power threshold . Front radars are here assumed, but similar conclusions were derived with corner radars.
Figure 7. Time evolution of the detected interference power and selected transmission frequency for a given radar using the FbF-CA technique; (a) with threshold dBm; (b) with threshold dBm.
Figure 7a and Figure 7b, in particular, illustrate the time evolution of the detected interference power and the corresponding channel selection for a generic radar and for two different values of . The left y-axis reports the detected interference power, while the right y-axis indicates the selected transmission frequency. The red dashed horizontal line represents the interference threshold, black circles denote the instantaneous detected interference power, and red triangles indicate the detection of a free channel. Figure 7a refers to a high threshold value, i.e., dBm. In this case, the detected interference power exceeds the threshold only occasionally and therefore the radar attempts to mitigate interference by changing its transmission frequency only when the interfering power is very high. Figure 7b shows instead the same metrics assuming a low threshold value, i.e., dBm. Under this condition, the detected interference power exceeds the threshold more frequently, leading to more frequent changes in the radar starting frequency. Notably, by comparing Figure 7a and Figure 7b, it can be observed that a lower threshold leads to more frequent frequency hopping events, which in turn increases the likelihood of identifying free channels, as indicated by the larger number of red triangles in Figure 7b. More experiments, not shown here for brevity, suggest to keep this threshold as low as possible.
Preliminary Considerations for Collision Detection
With reference to the FbF-CD technique, a critical parameter is the threshold on the number of corrupted chirps, . Also in this case, the analysis shown in Figure 8 refers only to front radars, but similar conclusions were drawn for corner radars.
Figure 8. Average time between system failure using the FbF-CD technique with different ; front radars, 150 vehicles/km.
In particular, Figure 8 illustrates the system performance in terms of average time between system failures as a function of the total system bandwidth for various values of . As can be observed, when the collision detection parameter satisfies , the performance remains nearly unchanged. Among these cases, the configuration , corresponding to the purple curve with diamond markers, exhibits a slightly better behavior. In contrast, when exceeds , the performance degrades significantly, as shown by the square-marked curve. A zoomed view further highlights that the condition achieves slightly better performance than the other cases.
Although this result is partly expected, since the optimal setting is consistent with the definition of a frame loss, an important observation is that the choice of does not appear critical, provided that it is not too conservative in triggering frequency changes.
Performance comparison
In Figure 9 and Figure 10, FbF-CA and FbF-CD are compared with the baseline, FbF, and CbC. Figure 9, in particular, shows the results in terms of average time between system failures versus the system bandwidth, assuming front radars. Looking first at the baseline, FbF, and CbC techniques (the three solid lines), an improvement of the performance can be observed as the system bandwidth increases. Among these, the CbC technique (green solid curve with diamond markers) represents the worst solution for small system bandwidth, while becoming the best performing approach when the total system bandwidth is large.
Figure 9. Average time between system failure vs. available bandwidth; front radars, 150 vehicles/km.
Figure 10. Average time between system failure vs. available bandwidth; corner radars, 150 vehicles/km.
Focusing on the FbF-CA approach, Figure 9 reports two configurations: a high detected interference threshold of dBm (yellow dashed curve with square markers) and a low threshold of dBm (grey dashed curve with triangle markers). When a high threshold is adopted, radars tend to preserve the same transmitting frequency over time, resulting in behavior similar to the baseline case where each radar maintains a fixed starting transmission frequency. Conversely, decreasing the interference threshold enables more frequency changes and this increases the probability of detecting free channels. Additional experiments, not shown here for brevity, demonstrated that using intermediate values for the threshold lead to intermediate results. Comparing the FbF-CA with the other methods, the results show that FbF-CA with a low threshold outperforms FbF, since collision avoidance triggers frequency changes only when required, leading to a more efficient use of the spectrum. CbC remains preferable only if the bandwidth is very large.
Figure 9 also reports the results obtained with FbF-CD technique, with set to 20% of the chirps in a frame. Collision detection (purple dashed curve with diamond markers) represents the best technique for low system bandwidth, as it enables frequency changes only when strictly required. However, as the system bandwidth increases, FbF-CD is outperformed by FbF-CA. This behavior can be attributed to the ability of collision avoidance to proactively detect the presence of interference and trigger a prompt frequency change. With increasing bandwidth, collision avoidance allows radars to change frequency even in the presence of low interference levels, whereas collision detection triggers frequency changes only after a frame loss has already occurred.
Figure 10 finally illustrates the same comparison for corner radars. This case implies a larger bandwidth of the chirp, thus a smaller relative available bandwidth, and a lower number of potential interferers. In this case, the FbF-CA improves the performance compared to FbF only when the available bandwidth exceeds 6 GHz, and FbF-CD outperforms FbF-CA except when 9 GHz is available. Furthermore, in this case, neither of these techniques approaches the performance achieved by the CbC technique, which, however, is difficult to implement in practice, as already stated.
Conclusion
In this work, we assessed the effectiveness of two proposed interference mitigation techniques based on measurements performed during idle intervals or during transmissions, called FbF-CA and FbF-CD, respectively. Assuming correlated interference and realistic dense highway scenarios, the results demonstrate that the two approaches significantly reduce the impact of interference compared to both a baseline solution where radars do not change the frequency configuration and a solution where the carrier frequency randomly varies at every burst of chirps (called frame). When the number of interferers is large and the available bandwidth is limited, the two solutions also proved to outperform a more ideal solution where the frequency changes at every chirp. Future work should extend the analysis to heterogeneous radar scenarios, to evaluate the robustness of the proposed techniques under radars with mixed configuration or uncorrelated interference conditions.
References
- [1]S. Alland, W. Stark, M. Ali, and M. Hegde, “Interference in automotive radar systems: Characteristics, mitigation techniques, and current and future research,” IEEE Signal Processing Magazine, 2019.DOI
- [2]“Estimating impacts of mutual interference of automotive radars,” SAE Government Industry Meeting, Jan. 2018.
- [3]F. Norouzian, A. Pirkani, E. Hoare, M. Cherniakov, and M. Gashinova, “Phenomenology of automotive radar interference,” IET Radar, Sonar & Navigation, vol. 15, no. 9, pp. 1045–1060, 2021.DOI
- [4]T. Fei, A. Becker, and M. Gardill, “Interference mitigation in automotive radar systems: A current state survey and future trends,” IEEE Transactions on Intelligent Transportation Systems, pp. 1–28, 2025.DOI
- [5]A. Bazzi, F. Miccoli, C. Coccoli, L. Facheris, and V. Martinez, “Analysis of proactive uncoordinated techniques to mitigate interference in fmcw automotive radars,” IEEE Transactions on Radar Systems, 2026.
- [6]V. Martinez, A. Filippi, and M. Vlot, “D-band vehicular radars regulatory status,” in 2024 21st European Radar Conference (EuRAD), 2024.
- [7]S. Roudiere, V. Martinez, and D. Delahaye, “A first investigation of V2X communication for radar interference mitigation,” in ITS World Congress 2021, 2021.
- [8]S. Jin and S. Roy, “Cross-layer interference modeling and performance analysis in FMCW radar multiple access network,” in IEEE VTC Fall, 2020.DOI
- [9]——, “FMCW radar network: Multiple access and interference mitigation,” IEEE Journal of Selected Topics in Signal Processing, 2021.DOI
- [10]H. K. V. B. Sukumaran, and C. Singh, “Slotted aloha for FMCW radar networks,” in National Conference on Communications (NCC), 2021.DOI
- [11]K. Haritha, V. B. Sukumaran, and C. Singh, “Slotted ALOHA and CSMA protocols for FMCW radar networks,” 2022.arxiv.org/abs/2201.09030
- [12]M. Kurosawa, T. Nozawa, M. Umehira, X. Wang, S. Takeda, and H. Kuroda, “Proposal of multiple access FMCW radar for inter-radar interference avoidance,” The Journal of Engineering, 2019.DOI
- [13]S. Ishikawa, M. Kurosawa, M. Umehira, X. Wang, S. Takeda, and H. Kuroda, “Packet-based FMCW radar using CSMA technique to avoid narrowband interference,” in 2019 International Radar Conference (RADAR), 2019, pp. 1–5.DOI
- [14]P. A. Lopez, M. Behrisch, L. Bieker-Walz, J. Erdmann, Y.-P. Flöteröd, R. Hilbrich, L. Lücken, J. Rummel, P. Wagner, and E. Wießner, “Microscopic traffic simulation using sumo,” in 21st International Conference on Intelligent Transportation Systems (ITSC), 2018.
- [15]A. Bazzi, F. Miccoli, Z. Wu, F. Cuccoli, and V. Martinez, “A tool for the statistical investigation of the automotive radar interference,” in URSI Atlantic Radio Science Conference (AT-RASC), 2024.DOI
- [16]H. Buddendieck and T. F. Eibert, “Incoherent scattering-center representations and parameterizations for automobiles [em programmer’s notebook],” IEEE Antennas and Propagation Magazine, 2012.
- [17]“System reference document (SRdoc); transmission characteristics; technical characteristics for radiodetermination equipment for ground based vehicular applications within the frequency range 77 GHz to 81 GHz,” ETSI TR 103 593 v.1.1.1, May 2020.
- [18]“Technical report on self-interference risk evaluation and sharing mechanisms for automotive radar in the 122,25 GHz – 130 GHz and 134 GHz - 148,5 GHz range,” ETSI TR 104 054, stable draft, March 2026.
- [19]European Commission, “Commission Implementing Decision (EU) 2026/883 of 21 April 2026 on the harmonisation of radio spectrum for use by radiodetermination applications in the 116-260 GHz frequency band,” Official Journal of the European Union, 2026.










