Multi-View, Multi-Frequency Sensing Geometry

The Multi-View Geometry of RF Imaging

The research problem that motivates this book involves multiple Tx-Rx terminals observing a scene from different angles and at multiple frequencies. This section develops the geometry of this multi-view, multi-frequency sensing configuration:

  • Monostatic, bistatic, and multi-static geometries.
  • Bistatic range and isorange ellipses.
  • k-space coverage analysis (extending Chapter 8).
  • How the combination of view angles and frequencies fills the spatial frequency plane and determines image resolution.

The key insight: more diverse measurement geometries yield a richer sensing matrix A\mathbf{A} and ultimately better images.

,

Definition:

Monostatic, Bistatic, and Multi-Static Configurations

  • Monostatic: Transmitter and receiver are co-located (si=rj\mathbf{s}_{i} = \mathbf{r}_{j}). The two-way propagation path has length 2βˆ₯pβˆ’siβˆ₯2\|\mathbf{p} - \mathbf{s}_{i}\|.

  • Bistatic: A single Tx at si\mathbf{s}_{i} and a single Rx at rjβ‰ si\mathbf{r}_{j} \neq \mathbf{s}_{i}. The two-way path length is: Rb=βˆ₯pβˆ’siβˆ₯+βˆ₯pβˆ’rjβˆ₯.R_b = \|\mathbf{p} - \mathbf{s}_{i}\| + \|\mathbf{p} - \mathbf{r}_{j}\|. The bistatic angle is: Ξ²=∠(pβˆ’si,β€…β€Špβˆ’rj).\beta = \angle(\mathbf{p} - \mathbf{s}_{i},\; \mathbf{p} - \mathbf{r}_{j}).

  • Multi-static: Multiple Tx and Rx at different locations, creating NtNrN_tN_r bistatic pairs. This is the general MIMO radar geometry.

The monostatic case is a special case of bistatic with Ξ²=0\beta = 0 and Rb=2RR_b = 2R. Most of our analysis uses the general bistatic form and specialises when needed.

Definition:

Bistatic Range and Isorange Contours

The bistatic range for a Tx at si\mathbf{s}_{i} and Rx at rj\mathbf{r}_{j} is the total propagation path length:

Rb(p)=βˆ₯pβˆ’siβˆ₯+βˆ₯pβˆ’rjβˆ₯.R_b(\mathbf{p}) = \|\mathbf{p} - \mathbf{s}_{i}\| + \|\mathbf{p} - \mathbf{r}_{j}\|.

The locus of points with constant bistatic range Rb=constR_b = \text{const} is an ellipse (in 2D) or ellipsoid (in 3D) with foci at si\mathbf{s}_{i} and rj\mathbf{r}_{j}.

The bistatic range resolution depends on the signal bandwidth WW and the bistatic geometry:

Ξ”Rb=c2Wcos⁑(Ξ²/2)\Delta R_b = \frac{c}{2W \cos(\beta/2)}

where Ξ²\beta is the bistatic angle and cc is the speed of light. Note that as Ξ²β†’Ο€\beta \to \pi (forward scatter), the range resolution degrades.

In the monostatic limit (Ξ²=0\beta = 0), this reduces to the familiar Ξ”R=c/(2W)\Delta R = c / (2W) from Chapter 7. The isorange contours become circles centred on the radar.

Bistatic Isorange Ellipses

Visualises isorange contours (ellipses) for a bistatic Tx-Rx pair. The foci are the Tx (red) and Rx (blue) positions. Each ellipse represents a constant total path length RbR_b.

Notice how the range resolution cell (distance between adjacent ellipses) depends on both the bandwidth and the bistatic angle: cells are narrowest along the baseline and widest perpendicular to it.

Move the Tx/Rx positions to see how the geometry affects the range resolution and the coverage of the scene.

Parameters
-5
0
5
0
1
10

Theorem: k-Space Coverage of MIMO Sensing

For a MIMO system with Tx at si\mathbf{s}_{i}, Rx at rj\mathbf{r}_{j}, and carrier frequency ff, the spatial frequency (k-space) sample corresponding to a target at position p\mathbf{p} is:

ΞΊij(f)=ΞΊ(f)(e^(si,p)+e^(rj,p))\boldsymbol{\kappa}_{ij}(f) = \kappa(f) \Bigl( \hat{\mathbf{e}}(\mathbf{s}_{i}, \mathbf{p}) + \hat{\mathbf{e}}(\mathbf{r}_{j}, \mathbf{p}) \Bigr)

where ΞΊ(f)=2Ο€f/c\kappa(f) = 2\pi f / c is the wavenumber and e^(a,b)=(bβˆ’a)/βˆ₯bβˆ’aβˆ₯\hat{\mathbf{e}}(\mathbf{a}, \mathbf{b}) = (\mathbf{b} - \mathbf{a}) / \|\mathbf{b} - \mathbf{a}\| is the unit vector from a\mathbf{a} to b\mathbf{b}.

The total k-space coverage of the MIMO system is:

K={ΞΊij(fk):i∈[Nt],β€…β€Šj∈[Nr],β€…β€Šk∈[Nf]}.\mathcal{K} = \bigl\{\boldsymbol{\kappa}_{ij}(f_k) : i \in [N_t],\; j \in [N_r],\; k \in [N_f]\bigr\}.

The extent of K\mathcal{K} determines the image resolution; the density determines the sidelobe level and conditioning of A\mathbf{A}.

Each Tx-Rx-frequency triple contributes one sample in k-space. Different frequencies scale the k-space vector radially (changing the wavenumber magnitude). Different Tx-Rx pairs rotate it (changing the bistatic angle). Together, they tile the k-space plane --- and the better the tiling, the better the image.

Proof
Derivation from the Born model

Under the Born approximation (Chapter 6), the scattered field measured at frequency ff for Tx at si\mathbf{s}_{i} and Rx at rj\mathbf{r}_{j} is a Fourier integral of the reflectivity:

yij(f)∝∫Ωc(p) eβˆ’jΞΊij(f)β‹…p dp.y_{ij}(f) \propto \int_{\Omega} c(\mathbf{p})\, e^{-j\boldsymbol{\kappa}_{ij}(f) \cdot \mathbf{p}} \, d\mathbf{p}.

Each measurement yij(f)y_{ij}(f) thus samples the Fourier transform of cc at spatial frequency ΞΊij(f)\boldsymbol{\kappa}_{ij}(f).

Decomposition into Tx and Rx wavenumbers

Writing ΞΊij(f)=ΞΊtx,i(f)+ΞΊrx,j(f)\boldsymbol{\kappa}_{ij}(f) = \boldsymbol{\kappa}_{\mathrm{tx},i}(f) + \boldsymbol{\kappa}_{\mathrm{rx},j}(f) where each term depends on only one antenna position. The combined wavenumber ΞΊs,r\kappa_{\mathbf{s},\mathbf{r}} is the sum of Tx and Rx contributions. β– \blacksquare

,

k-Space Coverage for MIMO Configurations

Visualises the k-space samples ΞΊij(fk)\boldsymbol{\kappa}_{ij}(f_k) for different MIMO configurations.

Each point represents one (Tx, Rx, frequency) measurement. Different colours indicate different Tx-Rx pairs; different marker sizes indicate different frequencies.

Observe how:

  • More frequencies extend the coverage radially (down-range resolution).
  • More Tx-Rx pairs extend the coverage angularly (cross-range resolution).
  • Co-located arrays fill a small angular sector; distributed arrays cover a wider angular range.
Parameters
3
4
4
28
1

Example: Resolution from k-Space Coverage

A multi-static MIMO system has 3 terminals arranged on a semicircle of radius 20 m, each with Nt=Nr=4N_t = N_r = 4 antennas. The system operates at f0=28f_0 = 28 GHz with W=2W = 2 GHz. Compute the imaging resolution.

Solution
Down-range resolution

From the bandwidth: Ξ”R=c/(2W)=3Γ—108/(2Γ—2Γ—109)=7.5\Delta R = c / (2W) = 3 \times 10^8 / (2 \times 2 \times 10^9) = 7.5 cm. This is independent of the array geometry.

Cross-range resolution

Three terminals on a semicircle span an angular aperture Ξ˜β‰ˆ180Β°\Theta \approx 180Β°. The cross-range resolution (from diffraction tomography) is: Ξ”xβŠ₯=Ξ»2sin⁑(Θ/2)=10.7Β mm2Γ—sin⁑(90Β°)=5.35Β mmβ‰ˆΞ»/2.\Delta x_\perp = \frac{\lambda}{2\sin(\Theta/2)} = \frac{10.7\text{ mm}}{2 \times \sin(90Β°)} = 5.35\text{ mm} \approx \lambda/2. This approaches the diffraction limit.

Per-terminal angular resolution

Each terminal's virtual array has 4Γ—4=164 \times 4 = 16 elements. At Ξ»/2\lambda/2 spacing, the per-terminal angular resolution is Ξ”ΞΈβ‰ˆΞ»/(16Γ—Ξ»/2)=1/8Β radβ‰ˆ7.2Β°\Delta\theta \approx \lambda/(16 \times \lambda/2) = 1/8 \text{ rad} \approx 7.2Β°. The multi-view geometry provides much finer cross-range resolution than any single terminal.

Definition:

Sensor Placement and k-Space Filling

The quality of the MIMO sensing matrix A\mathbf{A} depends on how well the k-space samples K\mathcal{K} fill the spatial frequency plane. Quantitative measures include:

Mutual coherence of A\mathbf{A}: ΞΌ(A)=max⁑pβ‰ q∣ApHAq∣βˆ₯Apβˆ₯βˆ₯Aqβˆ₯\mu(\mathbf{A}) = \max_{p \neq q} \frac{|\mathbf{A}_{p}^{H} \mathbf{A}_{q}|} {\|\mathbf{A}_{p}\| \|\mathbf{A}_{q}\|}

where Ap\mathbf{A}_{p} is the pp-th column of A\mathbf{A}. Low coherence (ΞΌβ‰ͺ1\mu \ll 1) is desirable for sparse recovery.

Condition number ΞΊ(A)\kappa(\mathbf{A}): determines noise amplification in least-squares reconstruction.

Optimal sensor placement seeks antenna positions {si,rj}\{\mathbf{s}_{i}, \mathbf{r}_{j}\} that minimise ΞΌ\mu or ΞΊ\kappa subject to physical constraints (platform geometry, minimum spacing, etc.).

For uniformly distributed sensors on a circle, the k-space coverage is nearly uniform --- this is the Fourier sampling geometry exploited in diffraction tomography (Chapter 14). Non-uniform distributions can achieve wider coverage but with gaps that increase sidelobes.

,

Quick Check

In a bistatic radar with Tx and Rx separated by baseline LL, what are the isorange contours?

Circles centred between Tx and Rx

Ellipses with foci at Tx and Rx

Hyperbolas with foci at Tx and Rx

Straight lines perpendicular to the baseline

Correction:
Ellipses with foci at Tx and Rx

The locus of constant Rb=βˆ₯pβˆ’sβˆ₯+βˆ₯pβˆ’rβˆ₯R_b = \|\mathbf{p} - \mathbf{s}\| + \|\mathbf{p} - \mathbf{r}\| is an ellipse with foci at the Tx and Rx positions.

Common Mistake: Ignoring k-Space Gaps

Mistake:

Assuming that adding more antennas always improves the image, without checking whether the new measurements fill k-space gaps or are redundant.

Correction:

What matters is not the number of measurements but their diversity in k-space. Two Tx-Rx pairs at similar bistatic angles and the same frequency produce nearly identical k-space samples --- adding no new information. Always verify k-space coverage before adding hardware. Diminishing returns set in when the k-space is already well-sampled.

🚨Critical Engineering Note

Synchronisation Requirements for Multi-Static MIMO

Distributed multi-static MIMO imaging requires tight synchronisation between nodes:

  • Time synchronisation: Timing error Ξ”t\Delta t introduces a range error Ξ”R=c Δt\Delta R = c\,\Delta t. For sub-centimetre accuracy at mm-wave: Ξ”t<30\Delta t < 30 ps.
  • Phase synchronisation: Phase error Δϕ\Delta\phi between local oscillators creates a k-space offset that blurs the image. For coherent imaging: Δϕ<5Β°\Delta\phi < 5Β° RMS.
  • Frequency synchronisation: Frequency offset Ξ”f0\Delta f_0 causes time-varying phase drift Ο•(t)=2πΔf0 t\phi(t) = 2\pi\Delta f_0\,t. Must be tracked and compensated.

GPS-disciplined oscillators achieve ∼10βˆ’12\sim 10^{-12} frequency stability (sub-Hz offset at mm-wave) but ∼10\sim 10 ns time accuracy --- insufficient for coherent imaging without additional over-the-air calibration.

Practical Constraints
  • β€’

    Time sync: < 30 ps for sub-cm range accuracy at mm-wave

  • β€’

    Phase sync: < 5 deg RMS for coherent image formation

  • β€’

    Frequency sync: sub-Hz offset at carrier (requires atomic reference or over-the-air calibration)

,
⚠️Engineering Note

Near-Field vs. Far-Field in MIMO Imaging

The virtual aperture analysis of Section 11.1 assumes far-field conditions: wavefronts are planar across the array. The far-field boundary is:

RFF=2D2Ξ»R_{\mathrm{FF}} = \frac{2D^2}{\lambda}

where DD is the physical aperture of the array. For a distributed MIMO system with D=20D = 20 m at f0=28f_0 = 28 GHz (Ξ»=10.7\lambda = 10.7 mm):

RFF=2Γ—4000.0107β‰ˆ75Β km.R_{\mathrm{FF}} = \frac{2 \times 400}{0.0107} \approx 75\text{ km}.

Most practical imaging scenarios (R<100R < 100 m) are in the near-field of the distributed array. The k-space analysis must account for spherical wavefronts rather than planar ones. For co-located arrays with D∼0.5D \sim 0.5 m, RFFβ‰ˆ47R_{\mathrm{FF}} \approx 47 m, so the far-field assumption is more often valid.

Practical Constraints
  • β€’

    Co-located MIMO: far-field valid for R > 2D^2/lambda (typically tens of metres at mm-wave)

  • β€’

    Distributed MIMO: almost always near-field; use spherical wavefront model

Bistatic Angle

The angle Ξ²\beta between the Tx-target and Rx-target directions, measured at the target. Ξ²=0\beta = 0 corresponds to monostatic (forward-looking); Ξ²=Ο€\beta = \pi corresponds to forward scatter.

Related: Bistatic Range and Isorange Contours

k-Space (Spatial Frequency Domain)

The Fourier-conjugate domain of physical space. Each radar measurement at a given (Tx, Rx, frequency) triple samples one point in k-space. The extent and density of k-space coverage determine image resolution and quality.

Related: Diffraction Tomography, Spatial Frequency

Key Takeaway

The multi-view, multi-frequency sensing geometry creates k-space samples ΞΊij(fk)=ΞΊ(fk)(e^tx+e^rx)\boldsymbol{\kappa}_{ij}(f_k) = \kappa(f_k)(\hat{\mathbf{e}}_{\mathrm{tx}} + \hat{\mathbf{e}}_{\mathrm{rx}}). Different frequencies provide radial diversity (range resolution); different view angles provide angular diversity (cross-range resolution). The quality of the sensing matrix A\mathbf{A} depends directly on how well these samples fill k-space.

Co-Located vs. Distributed MIMO RadarThe MIMO Sensing Matrix