FMCW Radar: Principles & Applications

• FMCW radar is a coherent sensing system that transmits continuously modulated chirp signals to achieve fine range, Doppler, and angle estimation.
• Advanced processing techniques like CFAR, VMD, and DFrFT enhance interference mitigation and enable high-resolution imaging and robust vital sign monitoring.
• MIMO array configurations and realistic simulations facilitate precise 2D/3D mapping and joint radar–communication waveform coexistence for diverse applications.

Frequency Modulated Continuous Wave (FMCW) radar is a coherent radar architecture that enables simultaneous range, Doppler, and, with antenna arrays, angle estimation, achieving fine range resolution and efficient hardware implementation through the use of linear frequency-modulated (LFM) chirps. FMCW radar has become the dominant modality for automotive, industrial, biomedical, and imaging radar applications, largely due to its capability to operate continuously, its intrinsic time–bandwidth compression, and its compatibility with low-power and low-cost RF front-ends. Recent advances encompass mutual interference mitigation, leakage cancellation, joint radar–communication coexistence, high-resolution 3D imaging, signal decomposition for clutter/interference suppression, and robust vital sign monitoring.

1. Principles of FMCW Radar Signal Generation and Processing

The core of FMCW radar is the transmission of a continuous wave with a linearly ramped frequency—a chirp—across a bandwidth $B$ within a period $T$. The baseband chirp is typically represented as

$$s_{\rm tx}(t) = \exp\left(j2\pi f_0 t + \frac{1}{2}j2\pi K t^2\right), \quad K = \frac{B}{T},$$

where $f_0$ is the start (carrier) frequency. A target at range $R$ produces a delayed echo, $s_{\rm rx}(t) = A s_{\rm tx}(t-\tau)$ with $\tau = 2R/c$, superimposed with noise and interference.

The received signal is mixed with a copy of the transmit chirp (dechirped), resulting in a low-frequency beat signal whose frequency $f_b$ is proportional to the time delay:

$f_b = K \tau = \frac{2B R}{c T},$

so that range $R$ is recovered by $R = (c T f_b) / (2B)$ (Moon et al., 2020, Gao et al., 2020). Range resolution is $\Delta R = c/(2B)$, established by the total sweep bandwidth, and can reach the centimeter or millimeter regime for $B > 1$ GHz (Hamidi et al., 2020, Jha et al., 7 Sep 2025).

Velocity (Doppler) is typically estimated across multiple chirps via the phase evolution over repeated sweeps. For $M$ chirps, Doppler resolution is $\Delta v = \lambda/(2 MT)$, with $\lambda = c/f_c$. In MIMO configurations, joint range, Doppler, and angle estimation is enabled via virtual array synthesis, exploiting the spatial diversity in transmit and receive antenna geometries (Kim et al., 2018, Hamidi et al., 2020).

2. Advanced Interference and Leakage Mitigation Techniques

The proliferation of FMCW radars in dense environments—particularly automotive—has intensified the mutual interference problem, manifesting as spurious chirp-like signals with time–frequency characteristics distinct from genuine target echoes. Numerous signal processing techniques have been developed for robust suppression:

• Random Frequency Hopping (BlueFMCW): By subdividing each chirp into $N$ sub-chirps and pseudorandomizing their frequency offsets using LFSR or PRNG-generated permutations, BlueFMCW diffuses the energy of interfering chirps across $m$ beat bins, yielding $10 \log_{10} m$ dB suppression of interference. Correct phase alignment via inverse permutation reconstructs continuous-phase beat signals, preserving range and velocity resolution (Moon et al., 2020).
• CFAR-based Masking in Time–Frequency Domain: 1-D CFAR detectors along STFT frequency bins identify interference ridges (slanted lines) in the t–f plane. Detected interference regions are dilated and subjected to amplitude-correction or zeroing, enabling preservation of target amplitude/phase and real-time computational efficiency (wang, 2021).
• Variational Mode Decomposition (VAFER): The beat signal is adaptively decomposed into narrowband modes via VMD; FSST is then applied for precise time–frequency localization. Energy–entropy thresholding retains only tone-like modes corresponding to valid targets, yielding SINR improvements exceeding 14 dB in simulation and nearly 10 dB in experimental automotive radar data (Gaur et al., 2022).
• Fractional Fourier Transform-based Suppression (DFrFT): Interference chirps are compressed into localized pulses in the fractional domain through DFrFT of tunable angle, detected via a CFAR energy-ratio test and zeroed. The EMDFrFT implementation substantially reduces computational complexity below naive multi-angle searches, supporting multi-interference scenarios (Oswald et al., 4 Apr 2025).
• RF-Domain Leakage Cancellation: For transmitter–receiver leakage, a replica IQ-mixer and Wilkinson combiner are inserted at the RF front end, digitally generating a matched leakage tone to subtract from the RX path. A TX IQ-mixer enables precise beat-frequency tuning, ensuring that leakage aligns with a single FFT bin for accurate estimation and >20 dB suppression (Chen et al., 2024).

3. Joint Radar and Communication Waveform Coexistence

Emerging paradigms in automotive radar demand cooperative coexistence of sensing and communication within shared spectrum. The non-orthogonal superposition of FMCW chirps and OFDM data streams enables dual-functionality:

• The radar channel impulse response, inferred from FMCW echo processing (dechirp and LS fitting), is used for real-time OFDM channel equalization, obviating the need for pilot symbols and incurring only a marginal ($0.6$ dB) SNR penalty at 1% BER relative to perfect CSI (Sahin et al., 2020).
• Range and velocity resolution ($\Delta R = c/(2B)$, $\Delta v = \lambda/(2T)$) are unaffected, and multi-target detection is possible. The FMCW pulse at the start of each frame is leveraged for noise/ambiguity-free channel estimation before the communication signal commences.

4. MIMO Array Architectures: Virtual Aperture Expansion and Joint Estimation

Modern FMCW radar systems exploit spatial diversity using MIMO front-ends and virtual array synthesis to achieve high angular resolution and robust multi-target discrimination:

• Time, Frequency, and Code Division Multiplexing: TDM, FDM, and CDM allow virtual enlargement of the array by separating simultaneous or sequential transmitter emissions. CDM, employing orthogonal Walsh-Hadamard codes, enables all transmitters to operate concurrently over the full bandwidth, maximizing SNR and range resolution with minimal cross-channel leakage (Hamidi et al., 2020).
• High-Resolution 2D and 3D Imaging: FMCW-MIMO radars, particularly when integrated into SAR scan geometries, afford three-dimensional reflectivity mapping with millimeter accuracy. Processing includes range FFT, 2D spatial FFT, Stolt interpolation, and multi-dimensional IFFT reconstruction. At $B=3.5$ GHz and $L_y=340$ mm, axial and cross-range resolutions approach 4 cm and 3–5 mm, respectively (Jha et al., 7 Sep 2025).
• Bias-Free Joint Range–Angle Estimation: Conventional 2D-FFT approaches incur deterministic coupling biases in range–angle matrices. Maximum-likelihood iterative algorithms, as derived in (Kim et al., 2018), rectify these errors and attain CRB for both range and angle, supporting SLAM applications with sub-centimeter precision over multi-target scenarios.

5. Biomedical Applications: Vital Sign Monitoring via mm-Wave FMCW Radar

Millimeter-wave FMCW radar systems employing MIMO-Virtual arrays have demonstrated robust, non-contact monitoring of physiological parameters such as respiration and heart rate:

• Signal Model: Sub-centimeter range bins are extracted via range–FFT; phase unwrapping of the slow-time IQ sequence isolates chest-wall motion, with $\phi(t) = (4\pi / \lambda) d(t)$ for displacement $d(t)$. Bandpass filtering isolates RR (0.05–0.7 Hz) and HR (0.6–4 Hz) bands (Sadeghi et al., 28 Aug 2025, Sadeghi et al., 2024).
• Advanced Estimation Techniques: Parametric (Prony) and subspace (MUSIC) algorithms, operating on phase-differenced, clutter-suppressed signals, achieve robust separation from strong respiration harmonics and impulsive noise. Empirically, heart rate MAE is 0.81 BPM (Prony), 1.8 BPM (MUSIC); respiration MAE is $\leq 1$ RPM (Prony) over diverse scenarios including post-exercise and clinical extremes (Sadeghi et al., 28 Aug 2025, Sadeghi et al., 2024).
• Dataset and Validation: Publicly available datasets at 77–81 GHz, comprising raw IQ radar data and matched ground truth (Polar H10, manual breath counts), support the development and benchmarking of advanced vital sign extraction algorithms in multi-scenario conditions (Sadeghi et al., 2024).

6. Simulation, Channel Modeling, and Environmental Effects

Highly flexible FMCW radar simulators leveraging 3D ray-tracing environments (e.g., Blender) allow precise modeling of radar channel propagation, signal strength, and multipath effects:

• Full Beat Signal Generation: Ray tracing provides path lengths, received power, and angles for arbitrary scenes. TDM-MIMO signal models integrate these parameters into realistic radar data cubes, with fast-time samples, slow-time chirp blocks, and virtual array indices (Liu et al., 2023).
• Validation: Comparisons with analytic ground truth confirm typical mean velocity errors near 1.3 m/s and sub-bin range errors for static reflectors. Multipath and attenuation are captured via the emission pass in Blender, with adjustable material properties and dynamic scenarios (e.g., moving pedestrians) (Liu et al., 2023).

7. Practical Implementation, Hardware Considerations, and Performance Metrics

• ADC, Matched Filter, and Stretch Architectures: Stretch (dechirp) architectures require only low-rate ADCs (sampling up to $f_{b,\max}$); combining digital up-sampling and re-chirping with matched filtering attains optimal SNR without increasing hardware cost (Movafagh et al., 2021).
• CFAR Thresholding and Clutter Models: Weibull and other non-Gaussian clutter models necessitate adaptive CFAR thresholding, both for detection and interference mitigation. Algorithms are robust against dense clutter in terrain (e.g., low-grazing-angle micro-UAV detection at 24 GHz with sub-10 cm/s Doppler resolution) (Ezuma et al., 2019).
• Computational Complexity: Advanced interference suppression methods, such as EMDFrFT, operate at $O(N^2 \log M)$, enabling practical deployment in radar DSPs/FPGA hardware (Oswald et al., 4 Apr 2025). CFAR methods and VMD+FSST pipelines are similarly parallelizable and suited for real-time automotive applications.
• Quantitative Performance: State-of-the-art interference mitigation techniques deliver SINR gains of 14–23 dB (simulation and experiment), with minimal impact on target amplitude and phase. Range and angle localization errors under joint ML estimation are sub-millimeter and sub-degree, supporting autonomous vehicle SLAM (Kim et al., 2018, Gaur et al., 2022, wang, 2021, Oswald et al., 4 Apr 2025).

---

FMCW radar continues to enable extreme sensing accuracies and environmental robustness through innovations in front-end architecture, multiplexed array processing, advanced signal decomposition, and interference mitigation. Its scalability to massive sensor deployments and integration with cooperative communication links positions FMCW as a central pillar of next-generation automotive, biomedical, and industrial radar systems.