Range, Velocity and Angle Measurement

Introduction

Radar converts measurements of propagation delay, frequency or phase change, and spatial wavefront variation into target range, radial velocity, and direction. These quantities form the measurement vector used by surveillance, navigation, tracking, imaging, automotive, weather, and defence systems.

Measurement is not merely peak detection. A practical estimator must address finite bandwidth, finite observation time, noise, ambiguity, synchronization errors, array calibration, clutter, multipath, and interference. Resolution must also be distinguished from accuracy: resolution concerns separating nearby targets, whereas accuracy concerns the error in estimating a target parameter.

This chapter develops conventional and statistically efficient estimators, including matched filtering, Doppler processing, monopulse and array processing, maximum-likelihood estimation, Fisher-information bounds, and super-resolution methods.

Learning Objectives

After studying this chapter, the reader should be able to:

• derive range, Doppler, and angle measurement equations;
• design matched-filter and coherent processing algorithms;
• distinguish resolution, accuracy, precision, and ambiguity;
• evaluate the PRF trade-off between unambiguous range and velocity;
• formulate maximum-likelihood estimators for delay, Doppler, and direction;
• derive and interpret relevant Cramér–Rao lower bounds;
• apply MUSIC, ESPRIT, and related super-resolution techniques;
• quantify synchronization, calibration, and multipath errors; and
• construct measurement uncertainty budgets for practical radar systems.

1. Radar Measurement Model

For a monostatic radar observing a point target, a useful complex-baseband model is

\[

y_m(t)=\alpha a_m(\theta,\phi)s(t-\tau)e^{j2\pi f_Dt}+w_m(t),

\]

where \(s(t)\) is the transmitted waveform, \(\alpha\) is the complex target amplitude, \(\tau\) is round-trip delay, \(f_D\) is Doppler frequency, \(a_m\) is the response of array element \(m\), and \(w_m(t)\) is noise and unmodelled interference. The target range and radial velocity are

\[

R=\frac{c\tau}{2},\qquad v_r=\frac{\lambda f_D}{2},

\]

where \(c\) is propagation speed and \(\lambda\) is wavelength. Positive and negative Doppler conventions depend on the receiver definition; the magnitude relationship remains the same.

The factors of two arise because the wave travels to the target and back. Bistatic and passive radars require bistatic-range and bistatic-Doppler geometry rather than these monostatic equations.

2. Range Measurement

2.1 Time-of-flight estimation

A pulsed radar estimates delay by correlating received data with a waveform replica. For additive white Gaussian noise, the matched-filter output is

\[

z(\tau')=\int y(t)s^*(t-\tau')\,dt.

\]

The peak location estimates delay:

\[

\hat\tau=\arg\max_{\tau'}|z(\tau')|,

\qquad \hat R=\frac{c\hat\tau}{2}.

\]

For sampled data, correlation is implemented by finite impulse response filtering or FFT-based convolution. Sub-sample accuracy can be obtained through interpolation, fractional-delay filtering, or direct continuous-parameter optimization. Interpolation reduces quantization error but cannot restore information absent from the signal bandwidth.

2.2 Range resolution

For a waveform with effective bandwidth \(B\), the nominal range resolution is

\[

\Delta R\approx\frac{c}{2B}.

\]

This expression characterizes the main-lobe scale of the compressed waveform; exact separability depends on waveform spectrum, windowing, relative target amplitudes, and the chosen resolution criterion.

A \(100\text{ MHz}\) waveform therefore has

\[

\Delta R=\frac{3\times10^8}{2(100\times10^6)}=1.5\text{ m}.

\]

A long coded pulse may simultaneously provide high transmitted energy and fine resolution. Its matched-filter sidelobes can mask weak targets near strong ones. Amplitude tapering lowers sidelobes but broadens the main lobe and usually reduces output SNR.

2.3 Range ambiguity

For pulse repetition interval \(T_r\) and pulse repetition frequency \(f_r=1/T_r\), a return may be associated only with its delay modulo \(T_r\). The nominal unambiguous range is

\[

R_u=\frac{cT_r}{2}=\frac{c}{2f_r}.

\]

Thus, a radar with \(f_r=5\text{ kHz}\) has \(R_u=30\text{ km}\). Targets beyond this interval appear at a folded range. Pulse duration, receiver recovery, eclipsing, and processing gates can make the usable interval smaller than the nominal value.

Ambiguity can be mitigated by:

• low or staggered PRFs;
• multiple-PRF residue association;
• coded pulse-to-pulse transmission;
• track-based ambiguity resolution; or
• waveform diversity and joint range–Doppler processing.

Multiple-PRF unfolding searches for a physical range whose predicted residues match all measured ambiguous ranges. Chinese-remainder reasoning applies to exactly commensurate measurements, but practical systems should use likelihood-based association because noise and missed detections disturb exact congruences.

2.4 FMCW range estimation

For a linear frequency-modulated continuous-wave radar with chirp slope \(S=B/T_c\), a stationary target produces an approximate beat frequency

\[

f_b=S\tau=\frac{2SR}{c}.

\]

Hence

\[

\hat R=\frac{c\hat f_b}{2S}.

\]

When the target moves, range and Doppler are coupled. Under the usual narrowband, low-acceleration approximation, triangular up- and down-chirps give

\[

f_{b,u}=S\tau+f_D,\qquad f_{b,d}=S\tau-f_D,

\]

subject to the receiver's sign convention. Therefore,

\[

\tau=\frac{f_{b,u}+f_{b,d}}{2S},

\qquad f_D=\frac{f_{b,u}-f_{b,d}}{2}.

\]

High-speed or long-range conditions may require correction for target motion during a chirp and nonlinear chirp calibration.

3. Radial Velocity Measurement

3.1 Doppler principle

For monostatic narrowband radar,

\[

f_D=\frac{2v_r}{\lambda}.

\]

Velocity is commonly estimated from pulse-to-pulse phase progression. For pulses separated by \(T_r\), the ideal phase increment is

\[

\Delta\varphi=2\pi f_DT_r.

\]

A simple estimator is

\[

\hat f_D=\frac{1}{2\pi T_r}\arg\left(\sum_{n=0}^{N-2}x_{n+1}x_n^*\right),

\]

where \(x_n\) is the complex sample from a selected range cell. This pulse-pair estimator is efficient for a dominant spectral component but may be biased by clutter or multiple targets.

An \(N\)-point slow-time DFT produces Doppler bins separated by

\[

\Delta f_D=\frac{f_r}{N}=\frac{1}{NT_r},

\qquad

\Delta v=\frac{\lambda}{2NT_r}.

\]

Windowing suppresses Doppler sidelobes at the cost of broader peaks and reduced coherent gain.

3.2 Doppler ambiguity and the PRF trade-off

Sampling once per pulse makes Doppler periodic with period \(f_r\). If the unambiguous Doppler interval is chosen as \([-f_r/2,f_r/2)\), then

\[

|v_r|<v_{u}=\frac{\lambda f_r}{4}.

\]

Combining this with unambiguous range gives

\[

R_uv_u=\frac{c\lambda}{8}.

\]

Consequently, increasing PRF enlarges the unambiguous velocity interval but reduces unambiguous range. A single uniform PRF cannot independently maximize both.

For \(f_r=5\text{ kHz}\) and \(\lambda=0.03\text{ m}\),

\[

R_u=30\text{ km},\qquad v_u=37.5\text{ m s}^{-1}.

\]

Staggered PRFs allow range and Doppler residues to be jointly unfolded. Medium-PRF airborne radars often require multidimensional ambiguity resolution involving range, Doppler, angle, amplitude, and track continuity.

A delay-line MTI canceller also has blind speeds when the pulse-to-pulse phase change is an integer multiple of \(2\pi\):

\[

v_b=n\frac{\lambda f_r}{2},\qquad n=0,1,2,\ldots

\]

These blind speeds are distinct from the limits of the chosen unambiguous velocity interval.

4. Angle Measurement

4.1 Beam scanning and monopulse

A scanning radar may estimate direction from the antenna pointing angle at maximum response. Its accuracy depends on beamwidth, SNR, scan sampling, and target fluctuations.

Amplitude-comparison monopulse forms sum and difference channels. Near boresight, the normalized error signal is approximately linear:

\[

e(\theta)=\Re\left\{\frac{\Delta(\theta)}{\Sigma(\theta)}\right\}\approx K_m(\theta-\theta_0),

\]

where \(K_m\) is the calibrated monopulse slope. The direction estimate is \(\hat\theta=\theta_0+e/K_m\). Monopulse obtains an angular error estimate from one received pulse and is therefore less sensitive to pulse-to-pulse amplitude fluctuation than sequential lobing.

4.2 Interferometry and array steering

For two sensors separated by baseline \(d\), the phase difference for a plane wave at angle \(\theta\), measured from broadside, is

\[

\Delta\phi=\frac{2\pi d}{\lambda}\sin\theta\pmod{2\pi}.

\]

If \(d\leq\lambda/2\), the visible region is generally free from grating-lobe ambiguity. Longer baselines improve phase sensitivity but require ambiguity resolution.

For an \(M\)-element uniform linear array (ULA),

\[

\mathbf a(\theta)=

[1,e^{jkd\sin\theta},\ldots,e^{jk(M-1)d\sin\theta}]^T,

\qquad k=\frac{2\pi}{\lambda}.

\]

A conventional beamformer evaluates

\[

P_B(\theta)=|\mathbf a^H(\theta)\mathbf y|^2

equation

\]

or its normalized version. The estimate is the direction that maximizes the spatial response.

4.3 Angle resolution

For a uniformly weighted, large ULA, the approximate half-power beamwidth in \(u=\sin\theta\) is

\[

\Delta u_{3\text{dB}}\approx0.886\frac{\lambda}{Md}.

\]

Around angle \(\theta\), this corresponds approximately to

\[

\Delta\theta_{3\text{dB}}\approx

0.886\frac{\lambda}{Md\cos\theta}

\]

in radians. This approximation assumes a sufficiently large array and operation away from endfire. The first-null scale is approximately \(\lambda/(Md\cos\theta)\).

Resolution should not be equated with single-target accuracy. At high SNR, the centroid or phase of one isolated target can be estimated much more accurately than a beamwidth. Conversely, two targets separated by less than a beamwidth may remain inseparable using conventional beamforming.

5. Maximum-Likelihood Estimation

Let sampled observations satisfy

\[

\mathbf y=\alpha\mathbf s(\boldsymbol\eta)+\mathbf w,

\qquad \mathbf w\sim\mathcal{CN}(\mathbf0,\sigma^2\mathbf I),

\]

where \(\boldsymbol\eta\) may contain delay, Doppler, and angle. The negative log-likelihood, apart from constants, is

\[

L=\frac{1}{\sigma^2}\|\mathbf y-\alpha\mathbf s(\boldsymbol\eta)\|^2.

\]

For fixed \(\boldsymbol\eta\), differentiating with respect to the unknown complex amplitude gives

\[

\hat\alpha(\boldsymbol\eta)=

\frac{\mathbf s^H(\boldsymbol\eta)\mathbf y}

{\mathbf s^H(\boldsymbol\eta)\mathbf s(\boldsymbol\eta)}.

\]

Substitution yields the concentrated maximum-likelihood estimator

\[

\hat{\boldsymbol\eta}_{ML}=

\arg\max_{\boldsymbol\eta}

\frac{|\mathbf s^H(\boldsymbol\eta)\mathbf y|^2}

{\mathbf s^H(\boldsymbol\eta)\mathbf s(\boldsymbol\eta)}.

\]

For delay and Doppler, this is a search over the matched-filter ambiguity surface. For direction, it becomes a normalized matched spatial filter. A practical algorithm is:

1. Compute a coarse range–Doppler–angle grid using FFTs and beamforming.

2. Retain statistically significant local maxima.

3. Refine each maximum using interpolation, Newton iteration, or gradient optimization.

4. Estimate complex amplitudes by least squares.

5. Re-estimate jointly if neighbouring targets interact.

Multi-target ML estimation minimizes

\[

\left\|\mathbf y-\sum_{q=1}^{K}\alpha_q\mathbf s(\boldsymbol\eta_q)\right\|^2.

\]

It can outperform independent peak picking, but it is non-convex, computationally demanding, and sensitive to model-order errors and initialization.

6. Fisher Information and Accuracy Bounds

6.1 General result with unknown amplitude

For the preceding complex Gaussian model, with unknown complex amplitude, the Fisher information for real parameters \(\eta_i\) is

\[

J_{ij}=\frac{2|\alpha|^2}{\sigma^2}

\Re\left\{

\frac{\partial\mathbf s^H}{\partial\eta_i}

\mathbf P_\perp

\frac{\partial\mathbf s}{\partial\eta_j}

\right\},

\]

where

\[

\mathbf P_\perp=\mathbf I-

\frac{\mathbf s\mathbf s^H}{\mathbf s^H\mathbf s}.

\]

The projection removes information indistinguishable from a change in unknown complex amplitude. Under regularity conditions, every unbiased estimator satisfies

\[

\operatorname{cov}(\hat{\boldsymbol\eta})\succeq\mathbf J^{-1}.

\]

The bound is local and model-dependent. It may not predict threshold failures, ambiguity selection errors, or bias at low SNR.

6.2 Range CRLB

Let \(\rho=|\alpha|^2E_s/N_0\) be the received energy SNR and let the waveform's mean-square RMS bandwidth be

\[

\beta^2=\frac{\int(f-\bar f)^2|S(f)|^2df}{\int|S(f)|^2df}.

\]

For a single delay in white Gaussian noise with unknown complex gain,

\[

\operatorname{var}(\hat\tau)\geq

\frac{1}{8\pi^2\beta^2\rho},

\]

and therefore

\[

\operatorname{var}(\hat R)\geq

\frac{c^2}{32\pi^2\beta^2\rho}.

\]

Thus, bandwidth controls both resolution and attainable delay accuracy, while SNR improves accuracy but does not fundamentally narrow the deterministic matched-filter main lobe.

6.3 Doppler CRLB

For energy-weighted RMS time duration

\[

T_{\rm rms}^2=

\frac{\int(t-\bar t)^2|s(t)|^2dt}{\int|s(t)|^2dt},

\]

the corresponding single-component bound is

\[

\operatorname{var}(\hat f_D)\geq

\frac{1}{8\pi^2\rho T_{\rm rms}^2}.

\]

For \(N\) equally spaced pulse samples with spacing \(T_r\), unknown complex amplitude, and per-pulse SNR \(\rho_p=|\alpha|^2/\sigma^2\), direct evaluation gives

\[

\operatorname{var}(\hat f_D)\geq

\frac{3}{2\pi^2\rho_pT_r^2N(N^2-1)}.

\]

Long coherent observation intervals therefore provide substantial velocity accuracy, provided target acceleration, phase noise, and oscillator instability do not destroy coherence.

6.4 ULA angle CRLB

For one narrowband source observed by an \(M\)-element ULA, define per-element SNR \(\rho=|\alpha|^2/\sigma^2\). Since

\[

\sum_{m=0}^{M-1}\left(m-\frac{M-1}{2}\right)^2

=\frac{M(M^2-1)}{12},

\]

the unknown-amplitude angle bound for one snapshot is

\[

\operatorname{var}(\hat\theta)\geq

\frac{6}{\rho k^2d^2\cos^2\theta\,M(M^2-1)}.

\]

For \(L\) independent, identically informative snapshots, the right-hand side is divided by \(L\). The result assumes an accurately calibrated ULA, spatially white Gaussian noise, one far-field narrowband source, and no angle-dependent unknown element errors.

For \(M=8\), \(d=\lambda/2\), \(\theta=20^\circ\), and \(\rho=10\),

\[

\sigma_\theta\geq

\sqrt{\frac{6}{10\pi^2\cos^2(20^\circ)(8)(63)}}

\approx0.0117\text{ rad}\approx0.67^\circ.

\]

The degradation near endfire follows from the \(\cos^2\theta\) term. Larger aperture, more elements, higher SNR, and additional snapshots improve the ideal bound.

7. Super-Resolution Estimation

7.1 MUSIC

For \(K<M\) narrowband sources,

\[

\mathbf R=\mathbf A\mathbf R_s\mathbf A^H+\sigma^2\mathbf I.

\]

An eigendecomposition separates signal and noise subspaces. If \(\mathbf E_n\) contains the noise eigenvectors, MUSIC evaluates

\[

P_{MU}(\theta)=

\frac{1}{\mathbf a^H(\theta)\mathbf E_n\mathbf E_n^H\mathbf a(\theta)}.

\]

True steering vectors are ideally orthogonal to the noise subspace and produce spectral peaks. The algorithm requires an estimated model order, an accurate steering model, sufficient snapshots, and adequate SNR. Fully coherent multipath can make the source covariance rank deficient; spatial smoothing may restore rank for suitable arrays but reduces effective aperture.

7.2 ESPRIT

ESPRIT uses two shift-invariant subarrays. If \(\mathbf E_1\) and \(\mathbf E_2\) are corresponding rows of the estimated signal subspace, it solves

\[

\mathbf E_2\approx\mathbf E_1\mathbf\Psi.

\]

The eigenvalues \(\mu_q\) of \(\mathbf\Psi\) satisfy

\[

\mu_q=e^{jkd\sin\theta_q},

\qquad

\hat\theta_q=\sin^{-1}\left(\frac{\arg\mu_q}{kd}\right),

\]

provided spatial aliasing is absent or independently resolved. ESPRIT avoids an angular grid, but it requires array shift invariance.

7.3 Sparse and continuous methods

Range, Doppler, and angle estimation can be posed as sparse reconstruction:

\[

\hat{\mathbf x}=

\arg\min_{\mathbf x}

\|\mathbf y-\mathbf D\mathbf x\|_2^2+\lambda_s\|\mathbf x\|_1.

\]

The dictionary \(\mathbf D\) contains candidate parameter responses. Fine grids increase computation and coherence, while off-grid targets cause basis mismatch. Atomic-norm, matrix-pencil, annihilating-filter, and gridless sparse methods model continuous parameters and avoid fixed-grid quantization.

Super-resolution is not unlimited. Performance deteriorates for closely spaced targets, low SNR, few snapshots, coherent multipath, calibration errors, and incorrect model order. Claimed resolution should therefore be accompanied by detection probability, false-alarm rate, bias, and robustness analysis.

8. Error Sources and Measurement Uncertainty

8.1 Synchronization and oscillator errors

A timing offset \(\delta t\) produces a monostatic range bias

\[

\delta R=\frac{c\delta t}{2}.

\]

A \(10\text{ ns}\) offset gives \(1.5\text{ m}\) range bias. Timing jitter adds random range variance \((c/2)^2\sigma_t^2\).

A receiver frequency-reference error \(\delta f\), if interpreted as target Doppler, causes velocity bias

\[

\delta v=\frac{\lambda\delta f}{2}.

\]

Oscillator phase noise spreads Doppler spectra and limits coherent integration. Distributed, bistatic, and networked radars additionally require clock alignment, frequency coherence, and knowledge of transmitter and receiver positions.

8.2 Array calibration

Element position errors, gain and phase mismatch, mutual coupling, radome distortion, and temperature drift perturb the steering vector. A phase error may be interpreted as an angle shift. Calibration may use internal reference networks, known external sources, satellite signals, or joint estimation of calibration and target parameters.

8.3 Multipath and propagation

Multipath creates delayed or angle-shifted replicas. These may generate ghost targets, biased centroids, fading, or correlated sources that invalidate standard subspace assumptions. Low-elevation radar is particularly affected by direct and reflected-path interference. Mitigation includes elevation diversity, polarization, terrain models, spatial smoothing, multi-frequency observations, and explicit multipath estimation.

Atmospheric refractivity bends propagation paths and alters effective range and elevation. Weather radars must also account for volume scattering and velocity spread, while automotive radars face strong reflections from roads, barriers, and vehicles.

8.4 Uncertainty propagation

Let measured parameters \(\mathbf q=[R,v_r,\theta,\phi]^T\) have covariance \(\mathbf C_q\), and let Cartesian state be \(\mathbf x=g(\mathbf q)\). First-order propagation gives

\[

\mathbf C_x\approx\mathbf G\mathbf C_q\mathbf G^T,

\qquad

\mathbf G=\left.\frac{\partial g}{\partial\mathbf q}\right|_{\hat{\mathbf q}}.

\]

Systematic biases should be estimated separately rather than hidden within random covariance. A defensible uncertainty budget identifies thermal noise, quantization, synchronization, calibration, propagation, algorithmic bias, and ambiguity-selection error. Monte Carlo simulation or nonlinear transforms are preferable when errors are large, non-Gaussian, or strongly nonlinear.

9. Illustrative Case: Roadside Radar Design

Background and problem

Consider an illustrative \(10\text{ GHz}\) roadside surveillance radar required to measure vehicles out to \(55\text{ km}\), radial speeds up to \(60\text{ m s}^{-1}\), and closely spaced lane directions. The waveform bandwidth is \(100\text{ MHz}\), and the radar uses a 12-element half-wavelength ULA.

Analysis

The wavelength is

\[

\lambda=\frac{3\times10^8}{10^{10}}=0.03\text{ m},

\]

and range resolution is \(1.5\text{ m}\). A single \(5\text{ kHz}\) PRF gives only \(30\text{ km}\) unambiguous range and \(37.5\text{ m s}^{-1}\) unambiguous speed, so both requirements exceed the single-PRF limits.

For a target at \(55\text{ km}\), the round-trip delay is

\[

\tau=\frac{2R}{c}=366.67\ \mu\text{s}.

\]

With PRFs of \(4.8\) and \(5.0\text{ kHz}\), the corresponding ambiguous delays are approximately \(158.33\ \mu\text{s}\) and \(166.67\ \mu\text{s}\), giving ambiguous ranges of \(23.75\) and \(25.0\text{ km}\). Candidate-range association recovers \(55\text{ km}\).

At \(60\text{ m s}^{-1}\), Doppler is

\[

f_D=\frac{2(60)}{0.03}=4\text{ kHz}.

\]

It appears as \(-0.8\text{ kHz}\) under a \(4.8\text{ kHz}\) PRF and \(-1.0\text{ kHz}\) under a \(5.0\text{ kHz}\) PRF when centered Nyquist intervals are used. Joint residue matching recovers \(4\text{ kHz}\).

For \(M=12\), \(d=\lambda/2\), \(\theta=20^\circ\), and per-element SNR of 10, the one-snapshot angle CRLB is approximately

\[

\sigma_\theta\geq0.0063\text{ rad}\approx0.36^\circ.

\]

This is much smaller than the approximate conventional half-power beamwidth of about \(9^\circ\), illustrating the difference between isolated-target accuracy and two-target resolution.

Approach and outcome

The design uses separate coherent bursts at staggered PRFs, likelihood-based ambiguity association, calibrated digital beamforming, and MUSIC only when model-order and snapshot-quality tests are satisfied. Track continuity rejects implausible range–velocity ambiguity combinations. Timing and phase-reference health are monitored continuously.

The radar meets the nominal coverage requirement without claiming that staggered PRFs eliminate all ambiguity. Dense scenes can still yield incorrect residue associations, while multipath and array errors may dominate theoretical CRLB values.

Learning points

• Range and velocity coverage must be designed jointly with PRF.
• Beamwidth, estimator accuracy, and multi-target resolution are different quantities.
• Super-resolution should supplement, not replace, calibration and ambiguity management.
• Operational uncertainty is generally larger than an ideal thermal-noise bound.

10. Implementation, Governance and Best Practice

Efficient systems use FFT-based range and Doppler processing, GPU or FPGA acceleration, adaptive detection, and track-before-detect methods. Machine learning may classify interference, select waveforms, or assist multipath rejection, but training data must represent weather, traffic, hardware drift, and rare failure conditions. Physics-based estimators remain important for interpretability and extrapolation.

Radar operation must comply with spectrum allocations, emission masks, electromagnetic exposure limits, and interference coordination requirements. Surveillance radar also raises privacy and governance concerns. Stored trajectory data should have defined access controls, retention limits, cybersecurity protection, and documented purposes. Safety-critical systems should report confidence and integrity measures rather than outputting unqualified point estimates.

Recommended engineering practices include traceable calibration, explicit sign conventions, separate bias and variance reporting, ambiguity stress testing, waveform validation over temperature, uncertainty-aware tracking, and field trials against independent reference instruments.

Chapter Summary

Range is estimated from round-trip delay, radial velocity from Doppler, and direction from spatial phase or amplitude variation. Bandwidth governs nominal range resolution, coherent observation time governs Doppler resolution, and aperture governs conventional angle resolution. PRF creates a fundamental trade-off between unambiguous range and unambiguous velocity, often requiring staggered PRFs and multidimensional association.

Maximum-likelihood estimation provides a unified matched-search framework. Fisher information quantifies ideal local accuracy and shows how bandwidth, dwell time, aperture, SNR, and snapshot count affect estimation variance. MUSIC, ESPRIT, and sparse methods can exceed conventional beamforming resolution under suitable modelling conditions but are sensitive to coherence, calibration, model order, and low SNR. Reliable radar measurement therefore combines statistical estimation with synchronization, calibration, ambiguity resolution, multipath mitigation, and defensible uncertainty reporting.

Key Terms

Ambiguity function: Matched-filter response as a function of delay and Doppler mismatch.

Angle accuracy: Closeness of an estimated direction to the true direction.

Cramér–Rao lower bound: Lower covariance bound for unbiased estimators under a specified statistical model.

Doppler ambiguity: Indistinguishability of Doppler frequencies separated by integer multiples of PRF.

Doppler resolution: Ability to separate targets with nearby radial velocities.

ESPRIT: Gridless subspace estimator based on array shift invariance.

Fisher information: Measure of parameter sensitivity contained in a probability model.

Matched filter: Filter maximizing output SNR for a known signal in white Gaussian noise.

Maximum-likelihood estimation: Parameter selection that maximizes the probability of observed data.

Monopulse: Simultaneous sum-and-difference angle estimation technique.

MUSIC: Super-resolution method exploiting signal and noise subspace orthogonality.

Pulse repetition frequency: Number of transmitted radar pulses per second.

Range ambiguity: Folding of delays exceeding one pulse repetition interval.

Range resolution: Ability to separate targets at nearby ranges.

RMS bandwidth: Energy-weighted spectral spread governing delay information.

Spatial aliasing: Ambiguous direction response caused by excessive sensor spacing.

Steering vector: Array response to a wave from a specified direction.

Super-resolution: Parametric estimation capable of separating components more closely than conventional main-lobe criteria.

Synchronization error: Measurement error caused by timing, frequency, or phase-reference mismatch.

Uncertainty budget: Quantified accounting of random and systematic measurement-error sources.

Review Questions

Short Answer Questions

1. Why does monostatic range contain a factor of one-half?

2. Distinguish range resolution from range accuracy.

3. What causes Doppler ambiguity in a pulsed radar?

4. Why does angle accuracy degrade near the endfire direction of a ULA?

5. State two conditions required for reliable MUSIC estimation.

Long Answer Questions

1. Derive the unambiguous range and velocity limits of a uniformly sampled pulsed radar and discuss their PRF trade-off.

2. Develop the concentrated maximum-likelihood estimator for delay, Doppler, and direction when target amplitude is unknown.

3. Derive the single-source angle CRLB for a uniform linear array and interpret its dependence on array size and SNR.

4. Compare conventional beamforming, monopulse, MUSIC, and ESPRIT in terms of assumptions, resolution, and robustness.

5. Explain how synchronization errors, array calibration, multipath, and atmospheric propagation affect radar measurement uncertainty.

Analytical / Application Questions

1. A radar operating at \(6\text{ GHz}\) has a PRF of \(8\text{ kHz}\) and bandwidth of \(50\text{ MHz}\). Calculate nominal range resolution, unambiguous range, and maximum unambiguous radial-speed magnitude.

2. Design a two-PRF strategy for a radar whose required range and velocity exceed the limits of either individual PRF. Explain the ambiguity-association algorithm and its failure modes.

3. For a calibrated half-wavelength ULA, evaluate how doubling the number of elements changes conventional beamwidth and the ideal high-SNR angle CRLB.

Multiple Choice Questions with Answers

Question 1

A monostatic radar measures a round-trip delay of 2 µs. Assuming c = 3 × 10^8 m/s, what is the target range?

A. 150 m

B. 300 m

C. 600 m

D. 900 m

Correct Answer: B. 300 m

Question 2

What is the nominal range resolution of a radar waveform having 100 MHz effective bandwidth?

A. 0.75 m

B. 1.0 m

C. 1.5 m

D. 3.0 m

Correct Answer: C. 1.5 m

Question 3

A monostatic pulsed radar operates at wavelength 0.03 m with a PRF of 10 kHz. Using a centered Doppler interval, what is the maximum unambiguous radial-speed magnitude?

A. 37.5 m/s

B. 75 m/s

C. 150 m/s

D. 300 m/s

Correct Answer: B. 75 m/s

Question 4

What is the nominal unambiguous range for a PRF of 5 kHz?

A. 15 km

B. 30 km

C. 60 km

D. 150 km

Correct Answer: B. 30 km

Question 5

For a calibrated ULA with fixed spacing and per-element SNR, the single-snapshot angle CRLB is inversely proportional to which array-size factor?

A. M

B. M squared

C. M(M squared minus 1)

D. The square root of M

Correct Answer: C. M(M squared minus 1)

Question 6

For white Gaussian noise and unknown complex target amplitude, concentrated maximum-likelihood estimation maximizes which quantity?

A. The unnormalized signal energy only

B. The normalized squared correlation between observation and signal model

C. The noise variance alone

D. The derivative of target amplitude

Correct Answer: B. The normalized squared correlation between observation and signal model

Question 7

Which technique is commonly used to restore rank before applying MUSIC to coherent sources on a suitable array?

A. Spatial smoothing

B. Range gating

C. Pulse blanking

D. Envelope detection

Correct Answer: A. Spatial smoothing

Question 8

What monostatic range bias is produced by a 10 ns timing offset?

A. 0.75 m

B. 1.5 m

C. 3.0 m

D. 15 m

Correct Answer: B. 1.5 m

Question 9

Why are both up- and down-chirps useful in triangular FMCW radar?

A. They eliminate thermal noise completely

B. They allow range and Doppler contributions to be separated

C. They double the propagation speed

D. They remove the need for frequency calibration

Correct Answer: B. They allow range and Doppler contributions to be separated

Question 10

Under first-order uncertainty propagation, which expression maps measurement covariance Cq through x = g(q)?

A. Cx = G Cq G transpose

B. Cx = G plus Cq

C. Cx = Cq divided by G

D. Cx = G transpose Cq G transpose

Correct Answer: A. Cx = G Cq G transpose