Millimeter Wave Specific Challenges mmWave Radar and Sensing Informational

What is the Doppler resolution of an FMCW radar and how does it relate to the chirp duration?

Q: What is the range-velocity ambiguity?

In FMCW radar: the beat frequency contains information about both range AND velocity. Range: causes a beat frequency proportional to range (f_beat_range = 2×BW×R/(c×T_chirp)). Velocity: causes a Doppler shift proportional to velocity (f_beat_velocity = 2v/lambda). The total beat frequency: f_beat = f_beat_range + f_beat_velocity. If only a single chirp is processed: it is impossible to separate the range and velocity contributions (ambiguity). With multiple chirps (range-Doppler processing): the range is determined from the fast-time FFT (within each chirp). The velocity is determined from the slow-time FFT (across chirps). The range and velocity are decoupled. This is why FMCW radar always processes multiple chirps per frame: a single chirp cannot unambiguously measure both range and velocity.

Q: Why does automotive radar need high velocity resolution?

Velocity resolution helps in several critical scenarios: (1) Separating vehicles traveling at similar speeds: on a highway, two cars 5 m apart traveling at 100 and 102 km/h (velocity difference = 2 km/h = 0.56 m/s). The radar must resolve this velocity difference to track both cars independently. With Δv = 0.2 m/s: the radar can distinguish 0.56 m/s difference. With Δv = 1 m/s: it cannot (they merge into one target). (2) Detecting stationary objects while moving: the ego vehicle and road clutter have the same relative velocity (zero in the vehicle frame). A stationary obstacle on the road must be detected against the road clutter. High velocity resolution helps separate the obstacle return (slightly different velocity due to different scattering behavior) from the road clutter. (3) Pedestrian classification: a walking pedestrian has a characteristic micro-Doppler signature (arm and leg swing at 0.5-2 m/s overlaid on the 1-2 m/s body velocity). High velocity resolution (< 0.3 m/s) reveals this gait signature, enabling classification.

The Doppler (velocity) resolution of an FMCW radar is determined by the total observation time per frame, not by a single chirp duration: (1) Doppler measurement principle: the FMCW radar measures the velocity of a target by tracking the phase change of the beat signal across multiple chirps. Each chirp provides one phase measurement. The phase change between consecutive chirps: Δphi = 4×pi×v×T_chirp/lambda, where v is the target velocity and T_chirp is the chirp repetition interval. (2) Velocity resolution: multiple chirps (N_chirps per frame) are processed with a Doppler FFT (across the slow-time axis). The Doppler frequency resolution: Δf_D = 1 / (N_chirps × T_chirp) = 1 / T_frame. The velocity resolution: Δv = lambda × Δf_D / 2 = lambda / (2 × N_chirps × T_chirp) = lambda / (2 × T_frame). For lambda = 3.9 mm (77 GHz), T_frame = 10 ms (N_chirps = 128 at T_chirp = 78 us): Δv = 0.0039 / (2 × 0.01) = 0.195 m/s = 0.70 km/h. This means the radar can distinguish two targets with velocity difference > 0.70 km/h. (3) Maximum unambiguous velocity: v_max = lambda / (4 × T_chirp). For T_chirp = 78 us: v_max = 0.0039 / (4 × 78e-6) = 12.5 m/s = 45 km/h. If the target velocity exceeds v_max: the velocity measurement wraps around (aliased). To increase v_max: reduce T_chirp (faster chirp repetition). But: reducing T_chirp reduces the max unambiguous range (the beat frequency bandwidth fills the ADC bandwidth). This creates a range-velocity tradeoff. (4) Resolving the tradeoff: use staggered chirp timing (different T_chirp values in alternating frames). The two frames provide different v_max values, and the correct velocity is determined by Chinese remainder theorem-like unwrapping. Modern radar ICs (TI AWR series) implement this automatically.

Category: Millimeter Wave Specific Challenges

Updated: April 2026

Product Tie-In: Radar ICs, Antennas, Signal Processors

Doppler Resolution in FMCW

In FMCW radar: range is measured within a single chirp (fast-time FFT), and velocity is measured across multiple chirps (slow-time FFT). This creates the 2D range-Doppler map that is the fundamental output of most modern radar processors.

Range-Doppler Processing

(1) Data organization: the ADC samples N_samples per chirp (fast-time), and there are N_chirps per frame. The data is organized as a 2D matrix: rows = chirps (slow-time dimension), columns = samples within each chirp (fast-time dimension). (2) Range FFT (fast-time): perform an FFT across each row (N_samples points). This resolves the targets in range. Each FFT bin corresponds to a range: R = k × c/(2×BW) × (N_samples/N_FFT). (3) Doppler FFT (slow-time): for each range bin, perform an FFT across the chirps (N_chirps points). This resolves the targets in velocity at that range. Each FFT bin corresponds to a velocity: v = m × lambda/(2×N_chirps×T_chirp). (4) The output is the range-Doppler map: a 2D image where each pixel represents the return power at a specific (range, velocity) pair. Targets appear as peaks in this map. Static clutter (ground, buildings): appears at velocity = 0 (zero Doppler). Moving targets: appear at non-zero velocities. The clutter can be removed by zeroing the v = 0 row in the range-Doppler map (MTI: moving target indication).

Design Tradeoffs

(1) Range resolution vs velocity resolution: the range resolution depends on BW (Δr = c/(2×BW)). The velocity resolution depends on T_frame (Δv = lambda/(2×T_frame)). These are independent parameters; both can be optimized simultaneously. However: the frame time (T_frame = N_chirps × T_chirp) determines the update rate. A longer frame = better velocity resolution but slower update rate. For a 30 fps radar (automotive): T_frame = 33 ms. Δv = 0.0039/(2×0.033) = 0.059 m/s = 0.21 km/h. With N_chirps = 256: T_chirp = 129 us. v_max = 0.0039/(4×129e-6) = 7.6 m/s = 27 km/h (too low for highway driving). Solution: reduce T_chirp to 30 us, increase N_chirps_per_frame = 33ms/30us = 1100 chirps. v_max = 0.0039/(4×30e-6) = 32.5 m/s = 117 km/h. Δv = 0.0039/(2×0.033) = 0.059 m/s. This works for highway driving. (2) ADC sampling rate constraint: the beat frequency for a target at range R: f_beat = 2×BW×R/(c×T_chirp). For BW = 4 GHz, R = 200 m, T_chirp = 30 us: f_beat = 2×4e9×200/(3e8×30e-6) = 178 MHz. The ADC must sample at > 356 Msps (Nyquist). This is a high sampling rate. If T_chirp is lengthened to 100 us: f_beat = 53 MHz (much more manageable). The tradeoff: longer T_chirp = lower v_max. Radar designers carefully balance T_chirp, BW, N_chirps, and ADC sampling rate to meet all specifications simultaneously.

Doppler Resolution Equations

Δv = λ/(2×N_chirps×T_chirp) = λ/(2×T_frame)
v_max = λ/(4×T_chirp)
Δφ = 4πv·T_chirp/λ per chirp
f_beat = 2×BW×R/(c×T_chirp)
Range-Doppler: 2D FFT (fast + slow time)

Common Questions

Frequently Asked Questions

Can I improve velocity resolution without a longer frame?

Not fundamentally: the velocity resolution is limited by the observation time (Δv = lambda/(2×T_frame)). No signal processing can improve this beyond the Heisenberg limit (frequency resolution × time = 1). However: (1) Non-coherent integration: accumulate range-Doppler maps across multiple frames. This improves the SNR (making weak targets detectable) but does not improve the velocity resolution. (2) Super-resolution algorithms (MUSIC, ESPRIT): can estimate the velocity of isolated point targets with precision better than Δv. But: they cannot separate two targets whose velocity difference is < Δv. (3) Longer frames with slower update rate: if the application can tolerate slower updates (e.g., 10 fps instead of 30 fps): T_frame = 100 ms, Δv = 0.02 m/s = 0.07 km/h. This is useful for: vital signs detection (heart rate), drone detection, and slow-moving targets.

What is the range-velocity ambiguity?

In FMCW radar: the beat frequency contains information about both range AND velocity. Range: causes a beat frequency proportional to range (f_beat_range = 2×BW×R/(c×T_chirp)). Velocity: causes a Doppler shift proportional to velocity (f_beat_velocity = 2v/lambda). The total beat frequency: f_beat = f_beat_range + f_beat_velocity. If only a single chirp is processed: it is impossible to separate the range and velocity contributions (ambiguity). With multiple chirps (range-Doppler processing): the range is determined from the fast-time FFT (within each chirp). The velocity is determined from the slow-time FFT (across chirps). The range and velocity are decoupled. This is why FMCW radar always processes multiple chirps per frame: a single chirp cannot unambiguously measure both range and velocity.

Why does automotive radar need high velocity resolution?

Velocity resolution helps in several critical scenarios: (1) Separating vehicles traveling at similar speeds: on a highway, two cars 5 m apart traveling at 100 and 102 km/h (velocity difference = 2 km/h = 0.56 m/s). The radar must resolve this velocity difference to track both cars independently. With Δv = 0.2 m/s: the radar can distinguish 0.56 m/s difference. With Δv = 1 m/s: it cannot (they merge into one target). (2) Detecting stationary objects while moving: the ego vehicle and road clutter have the same relative velocity (zero in the vehicle frame). A stationary obstacle on the road must be detected against the road clutter. High velocity resolution helps separate the obstacle return (slightly different velocity due to different scattering behavior) from the road clutter. (3) Pedestrian classification: a walking pedestrian has a characteristic micro-Doppler signature (arm and leg swing at 0.5-2 m/s overlaid on the 1-2 m/s body velocity). High velocity resolution (< 0.3 m/s) reveals this gait signature, enabling classification.

Keep Reading