Ferkans — Interactive Telecom Tutor

From 2D to 3D: Multi-Baseline SAR Tomography

Standard SAR produces 2D images (range $\times$ azimuth). Tomographic SAR (TomoSAR) extends SAR to 3D by exploiting multiple parallel passes at different cross-track baselines. Each pass provides a different view of the same scene in the elevation direction — precisely the multi-view sensing geometry of Chapter 8, now applied vertically.

The elevation sensing is another instance of $\mathbf{y} = \mathbf{A}\mathbf{c} + \mathbf{w}$ , where $\mathbf{A}$ encodes the baseline-dependent phase.

Definition:
TomoSAR Signal Model

Consider $M$ SAR acquisitions with cross-track baselines $b_1, b_2, \ldots, b_M$ relative to a reference pass. For a fixed range-azimuth cell containing scatterers at elevations $z_1, z_2, \ldots$ , the $m$ -th acquisition measures:

$y_m = \sum_k \sigma_k \exp\!\left(-j\frac{4\pi b_m z_k}{\lambda R_0}\right) + w_m,$

where $\sigma_k$ is the complex reflectivity of the $k$ -th scatterer and $R_0$ is the slant range.

Discretizing the elevation axis on a grid $\{z_n\}_{n=1}^N$ , the observation vector $\mathbf{y} \in \mathbb{C}^M$ is:

$\mathbf{y} = \mathbf{A}_{\text{elev}} \, \boldsymbol{\gamma} + \mathbf{w},$

where $[\mathbf{A}_{\text{elev}}]_{m,n} = \exp(-j 4\pi b_m z_n / (\lambda R_0))$ is the elevation sensing matrix.

The elevation sensing matrix is a partial Fourier matrix with non-uniform samples (the baselines $b_m$ ). Its properties (coherence, conditioning) depend on the baseline distribution, exactly as analyzed in Chapter 8 for spatial sensing.

Theorem: TomoSAR Elevation Resolution

For a multi-baseline SAR system with maximum perpendicular baseline $b_{\max}$ , carrier wavelength $\lambda$ , and slant range $R_0$ , the elevation (Rayleigh) resolution is:

$\Delta z = \frac{\lambda R_0}{2 b_{\max}}.$

This is the direct analog of the SAR cross-range resolution formula with $b_{\max}$ playing the role of the synthetic aperture $L_{\text{sa}}$ .

Each baseline provides a spatial frequency sample in the elevation direction, analogous to how each azimuth position provides a spatial frequency sample in cross-range. More baselines with larger spacing fill more of the elevation frequency space, yielding finer resolution.

Proof

Elevation spatial frequency

The phase difference between passes $m$ and the reference is $\Delta\phi_m = 4\pi b_m z / (\lambda R_0)$ . The elevation spatial frequency is $\xi_m = 2b_m/(\lambda R_0)$ .

Bandwidth and resolution

The total elevation bandwidth is $\Delta\xi = 2b_{\max}/(\lambda R_0)$ . By the Fourier uncertainty principle, $\Delta z = 1/\Delta\xi = \lambda R_0/(2b_{\max})$ . $\blacksquare$

Example: TomoSAR for Urban 3D Mapping

TerraSAR-X operates at $f_0 = 9.65$ GHz ( $\lambda = 3.1$ cm) in LEO at $R_0 = 600$ km. A multi-pass campaign collects $M = 15$ acquisitions with maximum baseline $b_{\max} = 300$ m.

(a) Compute the elevation resolution.

(b) Can two scatterers at different floors of a building (floor spacing 3 m) be resolved?

Solution

Elevation resolution

$\Delta z = \lambda R_0 / (2 b_{\max}) = 0.031 \times 600{,}000 / (2 \times 300) = 31$ m.

Resolving building floors

With $\Delta z = 31$ m, the Rayleigh limit far exceeds the 3 m floor spacing. Two scatterers at different floors appear in the same resolution cell.

However, if the scatterers are sparse in elevation (e.g., only 2--3 dominant reflectors per cell), sparse recovery methods can resolve them well below the Rayleigh limit — this is precisely the compressed sensing advantage developed in Chapter 11 and applied in Section s05.

TomoSAR Elevation Profile Reconstruction

Demonstrates multi-baseline TomoSAR elevation sensing. The plot compares matched-filter (beamforming) and sparse (ISTA) recovery of multiple scatterers in the same range-azimuth cell.

Increase the number of baselines for better Rayleigh resolution. Add more scatterers to see when the matched filter fails but sparse recovery succeeds.

The dashed line shows the true scatterer locations.

Parameters

Number of baselines

M

8

SNR (dB)20

Number of scatterers3

Definition:
Differential TomoSAR (D-TomoSAR)

Differential TomoSAR extends TomoSAR by incorporating temporal baselines (repeat-pass acquisitions over time) alongside spatial baselines. This enables measurement of:

Structural deformation: Buildings settling, bridge flexing, dam displacement — at millimeter precision.
Temporal scattering changes: Seasonal vegetation changes, urban construction.

The signal model adds a linear deformation rate $d_k$ per scatterer:

$y_{m,t} = \sum_k \sigma_k \exp\!\left(-j\frac{4\pi b_m z_k}{\lambda R_0} - j\frac{4\pi d_k t}{\lambda}\right) + w_{m,t}.$

The joint estimation of elevation $z_k$ , reflectivity $\sigma_k$ , and deformation rate $d_k$ is a higher-dimensional sparse recovery problem.

⚠️Engineering Note

Practical Considerations for TomoSAR

Operational TomoSAR faces several challenges:

Baseline distribution: Non-uniform baselines from orbital perturbations create non-uniform Fourier sampling, requiring careful analysis of the sensing matrix conditioning.
Atmospheric phase screens: Tropospheric and ionospheric delays add unknown phase offsets between passes that must be calibrated.
Temporal decorrelation: Scatterers may change between passes (vegetation, construction), violating the static scene assumption.
Computational cost: Each range-azimuth cell requires an independent elevation inversion. For a megapixel SAR image, this means millions of small inverse problems.

Despite these challenges, TomoSAR has been demonstrated successfully for urban 3D mapping using TerraSAR-X, TanDEM-X, and Sentinel-1.

Tomographic SAR (TomoSAR)

Extension of SAR to 3D by using multiple parallel passes at different cross-track baselines. The elevation sensing matrix $\mathbf{A}_{\text{elev}}$ is a partial Fourier matrix indexed by the baseline positions, with elevation resolution $\Delta z = \lambda R_0/(2b_{\max})$ .

Related: Synthetic Aperture

Differential TomoSAR

TomoSAR extended with temporal baselines to jointly estimate elevation profiles and deformation rates at millimeter precision.

Common Mistake: Assuming Uniform Baselines in TomoSAR

Mistake:

Analyzing TomoSAR performance assuming uniformly spaced baselines. Real satellite orbits produce non-uniform baselines with gaps and clusters that degrade the elevation sensing matrix.

Correction:

Use the actual baseline distribution when designing the reconstruction algorithm. Non-uniform baselines may require compressed sensing or Bayesian methods rather than simple Fourier-based (beamforming) inversion. Condition number analysis of $\mathbf{A}_{\text{elev}}$ reveals the effective elevation resolution for each specific baseline configuration.

Key Takeaway

TomoSAR extends SAR from 2D to 3D by adding an elevation sensing dimension via multi-baseline acquisitions. The elevation sensing matrix is a partial Fourier matrix with resolution $\Delta z = \lambda R_0/(2b_{\max})$ . Sparse recovery is essential for resolving multiple scatterers in the same range-azimuth cell — the Rayleigh limit is far too coarse for building-level 3D mapping. Differential TomoSAR adds temporal monitoring at millimeter precision.

Tomographic SAR