This tutorial paper provides a comprehensive look at how the shift toward extremely large antenna arrays (massive MIMO) and higher frequencies (mmWave/THz) is fundamentally changing wireless localization and sensing (L&S). By moving from the traditional "far-field" assumption to a "near-field" reality, researchers can now estimate a target's distance and full velocity vector using only a single base station and narrowband signals.
TL;DR
As antenna arrays grow larger, the region where we must treat wireless waves as spherical rather than planar—the "near-field"—expands to hundreds of meters. This paper explains how this shift allows systems to estimate a target's distance and transverse velocity in the space domain, unlocking high-precision sensing that was previously impossible without wide bandwidth or multiple nodes.
What Problem Does This Paper Address?
For decades, array signal processing has relied on the far-field assumption: the idea that a signal source is so far away that its wavefront appears flat (planar) by the time it hits the antenna array. In this regime, an array can only determine the direction (angle) of a signal, not its distance. To find distance, you traditionally needed either a very wide bandwidth (to measure time-of-arrival) or multiple base stations (triangulation).
However, as we move toward 6G, base stations are using larger apertures ($D$) and higher frequencies (shorter wavelengths, $\lambda$). This pushes the Fraunhofer distance ($2D^2/\lambda$)—the boundary of the near-field—outward. For a 1-meter array at 30 GHz, the near-field extends to 200 meters. In this region, the wavefront is noticeably curved (spherical). The authors tackle the challenge of modeling these curved waves and developing algorithms to exploit this curvature for better localization and sensing.
Methods
The paper transitions from fundamental electromagnetic theory to practical signal processing algorithms:
- Signal Modeling: The authors replace the linear phase shifts of far-field models with an exact Euclidean distance model for each antenna element. This captures the "spatial non-stationarity" of the signal.
- Subspace Fitting: They adapt classical algorithms like MUSIC (MUltiple SIgnal Classification) and DML (Deterministic Maximum Likelihood) for the near-field. While far-field MUSIC searches only for an angle ($\theta$), near-field MUSIC performs a 2D search for both angle and distance ($r$).
- Complexity Reduction: To avoid the "curse of dimensionality" in multi-target scenarios, they discuss "modified MUSIC" which uses the Fresnel approximation and symmetric array structures to decouple the angle and distance searches into two 1D problems.
- Tracking Moving Targets: For targets in motion, the paper describes using an Extended Kalman Filter (EKF). Because near-field effects cause a non-uniform Doppler shift across the array, the EKF can track not just radial velocity (moving toward/away) but also transverse velocity (moving across the field of view).
Datasets
As a tutorial and review paper, it does not introduce a new physical dataset. Instead, it uses numerical simulations to validate the discussed theories. Typical parameters used in their examples include:
- Array: 128 to 257-antenna Uniform Linear Arrays (ULA).
- Frequency: 28 GHz (mmWave).
- Scenarios: 5-target environments with varying Signal-to-Noise Ratios (SNR).
Key Insights and Ideas
- Distance from Space, Not Time: In the near-field, distance information is embedded in the spatial phase profile across the array. This means you can estimate distance even with a narrowband signal, which is impossible in the far-field.
- The Transverse Velocity "Gift": In the far-field, all antennas see the same Doppler shift. In the near-field, different antennas see slightly different angles to the target, creating a "Doppler gradient." This allows the system to resolve the target's full 2D velocity vector.
- The "Effective" Near-Field is Smaller: A major "aha" moment is the distinction between the physical near-field (Fraunhofer distance) and the effective near-field for L&S. The authors show that for practical distance resolution, the effective region is actually about one-tenth of the Fraunhofer distance ($d_T \approx 0.1 d_{FA}$).
- Resolution vs. Accuracy: While accuracy (Cramér-Rao Bound) improves with SNR, resolution (the ability to tell two targets apart) is fundamentally limited by the array aperture.
Key Results
- MUSIC vs. DML: Simulations show that while DML is theoretically more accurate, MUSIC is more robust when using "single-target approximations" to find multiple targets, as it better mitigates interference between closely spaced objects.
- Tracking Performance: The EKF-based approach successfully tracks targets even with imperfect knowledge of the source signal, demonstrating that near-field L&S is viable for real-time applications.
- Distance Sensitivity: The Cramér-Rao Bound (CRB) for distance estimation increases quadratically (or worse) as the target moves further away, confirming that near-field benefits degrade quickly as you approach the far-field boundary.
Limitations and Future Work
- Computational Complexity: Searching a 4K-dimensional space for $K$ targets is impractical. While decoupling methods exist, they often rely on approximations (like Fresnel) that fail if the target is too close.
- Hardware Constraints: Implementing fully digital arrays with hundreds of antennas is expensive and power-hungry. The authors suggest hybrid analog-digital or low-resolution (one-bit) ADCs as necessary future research areas.
- Environment Modeling: Most current models assume a simple line-of-sight (LoS) path. Real-world terrestrial L&S must deal with multipath scattering, which significantly complicates the signal model.
- Wideband Integration: While near-field works for narrowband, combining it with wideband (OFDM) signals could provide even better "delay-space" joint processing, but this introduces "spatial-wideband" effects (beam squint) that need addressing.
Summary updated