Ferkans — Interactive Telecom Tutor

Multi-Frequency Diffraction Tomography: Filling k-Space

A single frequency provides Ewald arcs of a fixed radius, limiting k-space coverage to a thin shell. By combining data from multiple frequencies, each contributing an Ewald sphere of different radius, we fill a much larger region of k-space. This section develops the composite k-space coverage geometry, analyzes range-dependent resolution, and presents the bandwidth synthesis technique.

Definition:
Multi-Frequency k-Space Coverage

For $N_v$ view angles $\{\theta_v\}$ and $N_f$ frequencies $\{f_n\}$ (wavenumbers $\{k_n = 2\pi f_n/c\}$ ), the composite coverage map is the union of all Ewald arcs:

$\mathcal{K} = \bigcup_{v=1}^{N_v} \bigcup_{n=1}^{N_f} \left\{\mathbf{K} = k_n(\hat{\mathbf{k}}_s - \hat{\mathbf{k}}_i^{(v)}) : \hat{\mathbf{k}}_s \in S^1 \right\}.$

Each Ewald circle has radius $k_n$ and center $-k_n\hat{\mathbf{k}}_i^{(v)}$ . Multi-frequency data produces concentric arcs (different radii, same center direction) for each view angle.

The spectral coverage ratio quantifies coverage quality:

$\eta = \frac{|\mathcal{K} \cap B_{2k_{\max}}|}{ |B_{2k_{\max}}|},$

where $B_{2k_{\max}}$ is the disk of radius $2k_{\max}$ .

Theorem: Range Resolution from Bandwidth

For a multi-frequency measurement with bandwidth $W = f_{\max} - f_{\min}$ , the range resolution of the diffraction tomography reconstruction is:

$\delta_r = \frac{c}{2W}.$

This is independent of the angular aperture and depends only on the total bandwidth.

Multi-frequency data fills the radial extent of k-space. The radial span is $\Delta K_r = 2(k_{\max} - k_{\min}) = 4\piW/c$ , and the resolution is $\pi / (\Delta K_r / 2) = c/(2W)$ -- the familiar radar range resolution formula.

Proof

Radial k-space extent

For a single view, the Ewald arcs at frequencies $f_{\min}$ and $f_{\max}$ have radii $k_{\min}$ and $k_{\max}$ . Along any radial direction, the accessible spatial frequencies span from approximately $k_{\min}$ (inner arc) to $2k_{\max}$ (outer arc, at backscatter). The radial extent contributed by frequency diversity alone is:

$\Delta K_r \approx 2(k_{\max} - k_{\min}) = \frac{4\piW}{c}.$

Resolution via inverse relationship

The spatial resolution is related to the k-space extent by $\delta_r = 2\pi / \Delta K_r = c / (2W)$ . $\quad\blacksquare$

Theorem: Cross-Range Resolution from Angular Aperture

For a single-frequency measurement at wavenumber $k_0$ with angular aperture $\Delta\theta$ (the total span of view angles), the cross-range resolution is:

$\delta_{\text{cr}} = \frac{\lambda}{4\sin(\Delta\theta / 2)}.$

For full $360^\circ$ coverage, $\delta_{\text{cr}} = \lambda/4$ (the diffraction limit).

Proof

Tangential k-space extent

For a fixed radial direction, the tangential k-space extent achieved by varying the view angle over $\Delta\theta$ is:

$\Delta K_t = 2k_0 \sin(\Delta\theta / 2).$

Resolution from tangential extent

The cross-range resolution is $\delta_{\text{cr}} = \pi / \Delta K_t = \pi / (2k_0\sin(\Delta\theta/2)) = \lambda/(4\sin(\Delta\theta/2))$ . For small $\Delta\theta$ : $\delta_{\text{cr}} \approx \lambda/(2\Delta\theta)$ . $\quad\blacksquare$

Multi-Frequency Coverage: Annular Region and the DC Hole

For a single view with $N_f$ frequencies spanning $[f_{\min}, f_{\max}]$ , the union of Ewald arcs fills an annular region in k-space between radii $\sim k_{\min}$ (inner) and $\sim 2k_{\max}$ (outer).

If the carrier frequency is high and the relative bandwidth $W/f_0$ is small, there is a "hole" near the Fourier origin. This missing DC component means the absolute scattering strength is ambiguous -- only contrast variations are imaged. Wider relative bandwidth fills more of the hole.

Frequency sampling requirement: To avoid range ambiguity, the frequency step must satisfy $\Delta f \leq c/(2D)$ where $D$ is the object diameter.

Definition:
Bandwidth Synthesis

Bandwidth synthesis combines multiple narrow-band measurements at different carrier frequencies to achieve the range resolution of a wideband system. If $N_f$ frequencies are uniformly spaced at $f_n = f_0 + n\Delta f$ ( $n = 0, \ldots, N_f - 1$ ), the effective bandwidth is:

$W_{\text{eff}} = (N_f - 1)\Delta f,$

giving range resolution $\delta_r = c/(2W_{\text{eff}})$ .

Advantage: Each individual measurement can use a narrow instantaneous bandwidth (simpler hardware), while the composite achieves the resolution of the full bandwidth.

Requirement: The frequency spacing $\Delta f$ must satisfy $\Delta f \leq c/(2D)$ to avoid range ambiguity (aliasing in range). The individual measurements must be phase-coherent across frequencies.

Multi-Frequency k-Space Coverage

Visualizes the k-space coverage map for multi-frequency, multi-view diffraction tomography.

Left: Ewald circles colored by frequency (blue = lowest, red = highest). Right: Coverage density map -- brighter regions indicate more measurements per Fourier bin.

Increase $N_f$ to fill the radial gaps (improve range resolution). Increase $N_v$ to fill the angular gaps (improve cross-range resolution). Observe the annular coverage and the DC hole at the center.

Parameters

Number of views

N_v

8

Number of frequencies

N_f

5

Bandwidth (GHz)4

Center frequency (GHz)10

Example: Narrowband vs. Wideband Coverage

Compare the k-space coverage and reconstruction quality for a 30 cm object at $f_0 = 10$ GHz:

Configuration	$N_v$	Bandwidth	$N_f$	$\eta$	$\delta_r$	$\delta_{\text{cr}}$
Narrowband, few views	4	100 MHz	1	0.08	1.5 m	1.5 cm
Narrowband, many views	36	100 MHz	1	0.35	1.5 m	1.5 cm
Wideband, few views	4	4 GHz	20	0.31	3.75 cm	1.5 cm
Wideband, many views	36	4 GHz	20	0.92	3.75 cm	1.5 cm

Explain the pattern.

Solution

Angular diversity controls cross-range

Cross-range resolution $\delta_{\text{cr}} = \lambda/(4\sin(\Delta\theta/2))$ depends only on the angular aperture, not bandwidth. Both narrowband and wideband systems achieve the same $\delta_{\text{cr}}$ for the same $N_v$ .

Frequency diversity controls range

Range resolution $\delta_r = c/(2W)$ depends only on bandwidth. The narrowband system ( $W = 100$ MHz) has $\delta_r = 1.5$ m -- useless for a 30 cm object. The wideband system ( $W = 4$ GHz) achieves $\delta_r = 3.75$ cm.

Both diversities are needed

High coverage ratio $\eta \to 1$ requires both angular and frequency diversity. The wideband, many-views configuration achieves $\eta = 0.92$ , approaching complete Fourier coverage.

Range-Dependent Resolution Variation

In multi-frequency DT, the k-space coverage is not uniform: high-frequency Ewald arcs have larger radius and contribute to higher spatial frequencies. This creates range-dependent resolution variation:

Points near the array (strong signal, large angular subtension) have dense k-space sampling and high resolution.
Points far from the array have sparser k-space sampling and lower resolution.

This effect is analogous to the range-dependent beam broadening in SAR (Chapter 9). Compensation strategies include: position-dependent regularization weights, spatially varying deconvolution, or adaptive gridding with finer k-space bins near the outer coverage boundary.

Comparison: Direct vs. Iterative Reconstruction

Property	Gridding + IFFT	SIRT/ART	CGLS	FISTA ( $\ell_1$ )
Speed	Very fast	Moderate	Moderate	Slow
Memory	$O(N)$	$O(N + M)$	$O(N + M)$	$O(N + M)$
Incomplete data	Streaks	Mild blurring	Mild blurring	Gap filling
Regularization	None	Implicit	Semi-convergence	Explicit $\ell_1$ /TV
Noise handling	Filter choice	Iteration stopping	Iteration stopping	Regularization $\lambda$
Best for	Well-sampled, fast preview	Moderate gaps	Moderate gaps	Sparse scenes, severe gaps

Historical Note: Wolf, Ewald, and the Sphere That Changed Imaging

1913--1982

The Ewald sphere construction was introduced by Paul Ewald in 1913 for X-ray crystallography -- he showed that each diffraction spot corresponds to a reciprocal lattice point intersected by a sphere of radius $1/\lambda$ . In 1969, Emil Wolf realized that the same geometric construction applies to diffraction tomography of continuous objects: scattered-field measurements fill Ewald spheres in the object's spatial frequency domain. This insight unified crystallography, acoustic imaging, and electromagnetic imaging under a single mathematical framework.

The multi-frequency extension was developed in the 1980s by Devaney and others, motivated by ultrasound and microwave medical imaging. Today, the same framework underlies MIMO radar imaging, ground-penetrating radar, and wireless sensing.

,

Common Mistake: Non-Uniform Fourier Density Biases Reconstruction

Mistake:

Even with good coverage ratio $\eta$ , the density of k-space samples is typically non-uniform:

Oversampled near the origin (low spatial frequencies).
Undersampled at high spatial frequencies.
Denser along certain directions (depending on view angles).

Naively inverting without density compensation produces images with blurred edges (suppressed high frequencies) and directional artifacts (anisotropic density).

Correction:

Apply the density compensation function (DCF) before inversion. Alternatively, use iterative methods (SIRT, CGLS) that implicitly handle non-uniform sampling through the normal equations.

Quick Check

A diffraction tomography system operates at center frequency 10 GHz with 4 GHz bandwidth. What is the range resolution?

$3.75$ cm

$1.5$ cm

$7.5$ cm

$\lambda/4 = 0.75$ cm

Correction:

3.75

cm

$\delta_r = c/(2W) = 3 \times 10^8 / (2 \times 4 \times 10^9) = 3.75$ cm.

🔧Engineering Note

Stepped-Frequency vs. Pulsed Wideband Systems

Stepped-frequency systems transmit narrowband signals at $N_f$ discrete frequencies and synthesize wideband resolution in post-processing. This is simpler than ultra-wideband (UWB) pulse systems: the RF frontend needs only narrowband bandwidth, ADC sampling rates are lower, and dynamic range is better.

Practical considerations:

Frequency coherence across steps requires a stable local oscillator (phase noise $< -100$ dBc/Hz at 1 kHz offset).
Scene motion during the frequency sweep causes range-Doppler coupling: targets that move during the sweep appear shifted in range. The sweep time must satisfy $T_{\text{sweep}} \ll \lambda / (2 v_{\max})$ where $v_{\max}$ is the maximum target velocity.
For static scenes (through-wall imaging, medical), stepped frequency is the standard approach.

Key Takeaway

Multi-frequency DT combines Ewald spheres of different radii to fill k-space in the radial direction. Range resolution depends on bandwidth ( $\delta_r = c/2W$ ); cross-range resolution depends on angular aperture ( $\delta_{\text{cr}} = \lambda/(4\sin(\Delta\theta/2))$ ). Both diversities are needed for high-quality reconstruction. Bandwidth synthesis achieves wideband resolution from multiple narrowband measurements. The DC hole near the Fourier origin limits the reconstruction of absolute scattering strength.

Multi-Frequency Diffraction Tomography

Multi-Frequency Diffraction Tomography: Filling k-Space

Definition: Multi-Frequency k-Space Coverage

Theorem: Range Resolution from Bandwidth

Radial k-space extent

Resolution via inverse relationship

Theorem: Cross-Range Resolution from Angular Aperture

Tangential k-space extent

Resolution from tangential extent

Multi-Frequency Coverage: Annular Region and the DC Hole

Definition: Bandwidth Synthesis

Multi-Frequency k-Space Coverage

Parameters

Example: Narrowband vs. Wideband Coverage

Angular diversity controls cross-range

Frequency diversity controls range

Both diversities are needed

Range-Dependent Resolution Variation

Comparison: Direct vs. Iterative Reconstruction

Historical Note: Wolf, Ewald, and the Sphere That Changed Imaging

Common Mistake: Non-Uniform Fourier Density Biases Reconstruction

Quick Check

Stepped-Frequency vs. Pulsed Wideband Systems

Key Takeaway

Definition:
Multi-Frequency k-Space Coverage

Definition:
Bandwidth Synthesis