⚡ electrical engineering

Real-time Range-Angle Estimation and Tag Localization for Multi-static Backscatter Systems

This paper proposes low-complexity algorithms for real-time range-angle estimation and tag localization in large-scale multi-static backscatter networks, demonstrating through a real-world testbed that these methods achieve high accuracy with up to 500-fold computational efficiency gains over existing brute-force approaches.

Original authors: Tara Esmaeilbeig, Kartik Patel, Traian E. Abrudan, John Kimionis, Eleftherios Kampianakis, Michael S. Eggleston

Published 2026-02-17

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Tara Esmaeilbeig, Kartik Patel, Traian E. Abrudan, John Kimionis, Eleftherios Kampianakis, Michael S. Eggleston

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are in a giant, echoey warehouse filled with hundreds of tiny, silent drones (the "tags"). You want to know exactly where every single drone is, but you can't ask them to shout out their location because they are too small and use almost no battery.

Instead, you have a team of four loudspeakers (the "transmitters") broadcasting a constant hum, and one super-sensitive microphone array (the "receiver") listening for the echoes. When a drone flies by, it catches the hum and reflects it back to the microphone, slightly changing the pitch. By listening to these echoes, you can figure out how far away the drone is and which direction it's coming from.

The Problem:
In a real warehouse, the sound bounces off walls, shelves, and other drones. It's a mess of echoes (multipath). To find the drones, you have to separate the "true" echo from the "fake" echoes.

The old way to do this is like trying to find a specific needle in a haystack by looking at every single piece of straw one by one. It works, but it takes forever. If you have 100 drones, the computer gets so tired it can't keep up, and you can't track them in real-time.

The Solution:
This paper introduces two new, super-smart ways to find the drones quickly, and two new ways to calculate their exact location without breaking a sweat.

1. The "Range-Angle" Detectives (Finding the Echo)

To locate a drone, you need two things:

Range: How far away is it? (Like measuring how long it takes for an echo to return).
Angle: Which direction is it coming from? (Like turning your head to hear where the sound is loudest).

The authors propose two new algorithms to find these numbers fast:

JRAC (The "Cluster Detective"): Imagine the echoes are like a crowd of people in a dark room. The old methods try to identify every single person individually, which is slow. JRAC turns on a spotlight and says, "Hey, look at that big group of people standing together!" It groups the messy echoes into "clusters" and picks the one closest to the microphone. It ignores the noise and finds the target instantly.
SRAE (The "Two-Step Dancer"): Instead of trying to find the distance and direction at the same time (which is hard), this method does it in two easy steps. First, it figures out how far away the sound is. Once it knows the distance, it only looks for the direction at that specific distance. It's like narrowing down a search from "the whole world" to "just this one street."

The Result: These methods are 40 times faster than the old ways, but just as accurate.

2. The "Map Makers" (Calculating the Location)

Once you know the distance and direction from your four loudspeakers, you have to draw lines on a map to see where they cross. That intersection is the drone's location.

The Old Way (Brute Force ML): Imagine trying to find a lost coin in a field by digging a hole every single inch of the field until you find it. It's guaranteed to work, but it takes days.
The New Way (Gradient Ascent): Imagine you are blindfolded on a hill and want to find the highest peak. Instead of digging everywhere, you take a step in the direction that feels "up." If it feels higher, you keep going. You take a few smart steps and quickly reach the top. This is much faster.
The Super-Fast Way (IRLS): This is like using a magic compass that instantly points to the center of gravity of all your clues. It uses a clever math trick to solve the puzzle in one go, without needing to take steps. It is 500 times faster than the "digging everywhere" method.

The Big Test

The researchers built a real test lab with 100 tags (drones) and 4 transmitters. It was a noisy, messy environment with metal walls causing lots of echoes.

Accuracy: Their new methods found the tags with a median error of just 3 meters (about 10 feet). That's pretty good for a room full of echoes!
Speed: They proved that you can track 100 devices in real-time. The old methods would have taken minutes to calculate one location; the new methods do it in milliseconds.

The Takeaway

This paper is like inventing a GPS for the "Internet of Things" that doesn't need batteries. By using clever math to ignore the noise and find the signal quickly, they made it possible to track thousands of tiny, battery-free devices in real-time. This is a huge step toward a future where your clothes, your furniture, and your tools can all talk to a network to tell you exactly where they are, without needing to be charged.

1. Problem Statement

The paper addresses the challenge of achieving real-time, scalable localization for thousands of devices in Multi-static Backscatter Networks (BNs) within the context of 6G Ambient IoT.

Context: Traditional monostatic backscatter systems suffer from limited range and scalability. Multi-static systems (multiple transmitters, multiple receivers) offer wide-area coverage but introduce significant computational complexity.
Core Challenge: Estimating key parameters (Range and Angle of Arrival - AoA) and fusing them into location estimates for hundreds of tags simultaneously requires algorithms that are both highly accurate and computationally efficient. Conventional methods like 2D FFT and 2D MUSIC (MUltiple SIgnal Classification) provide accuracy but incur prohibitive computational costs, making real-time deployment in dense networks impossible.
Specific Constraints: The system operates without external clock synchronization between transmitters (TX) and receivers (RX), relies on frequency-shifting tags to separate signals, and must function in heavy multipath environments.

2. Methodology

The authors propose a two-stage solution: Joint Range-Angle Estimation followed by Multi-static Localization.

A. System Model

Architecture: A network with $N_{TX}$ transmitters, $N_{RX}$ multi-antenna receivers, and frequency-shifting backscatter tags.
Signal Processing: Tags shift the carrier frequency to an adjacent band ( $F_{shift}$ ), allowing the receiver to process the direct path (TX-RX) and backscatter path (TX-Tag-RX) simultaneously without interference.
Parameters: The goal is to estimate the bistatic range ( $d_B$ ) and AoA ( $\theta, \phi$ ) of the Line-of-Sight (LoS) path.

B. Joint Range-Angle Estimation Algorithms

The paper introduces two low-complexity algorithms to replace standard 2D FFT and 2D MUSIC:

Joint Range-Angle Clustering (JRAC):
- Approach: A joint estimator that operates on a truncated range-angle heatmap.
- Process:
  1. Truncates the delay search space using 1D MUSIC.
  2. Generates a 2D heatmap for the reduced delay space and full angular space.
  3. Clusters dominant multipath components in the heatmap.
  4. Selects the cluster with the lowest delay to identify the LoS path.
- Benefit: Handles multipath robustly by treating clusters as units rather than individual peaks, reducing grid search complexity.
Stage-wise Range-Angle Estimation (SRAE):
- Approach: A decoupled, sequential estimation method.
- Process:
  1. Stage 1: Estimates delay (range) using 1D MUSIC.
  2. Stage 2: Uses the estimated delay to extract channel taps and estimates AoA using spatial MUSIC.
- Benefit: Significantly lower complexity than joint methods, though theoretically sub-optimal due to error propagation between stages.

C. Real-time Localization Algorithms

Once ranges and AoAs are estimated, two fusion algorithms are proposed to determine tag positions:

ML-based Gradient Ascent with Line Search:
- Formulates a Maximum Likelihood (ML) estimator assuming Gaussian range errors and von Mises-Fisher distributed angular errors.
- Solves the non-convex optimization problem using gradient ascent with multiple initializations (based on ellipse geometry) and line search to avoid local minima.
- Complexity: Higher than IRLS but avoids the exponential cost of brute-force grid search.
Iterative Re-weighted Least Squares (IRLS):
- Formulates a pseudo-linear problem combining range and angle measurements.
- Key Innovation: Introduces a scaling scheme to the pseudo-linear formulation. The authors observed that unscaled formulations artificially weight outliers (measurements with large residuals) due to the nature of the range equation. They scale the equations so the right-hand side is 1, ensuring robustness.
- Process: Iteratively updates weights based on residual errors and projects the solution onto a feasible cone to satisfy geometric constraints.
- Complexity: Extremely low; involves matrix inversion of fixed rank.

3. Key Contributions

Novel Architecture: A multi-static BN design that natively supports localization without external synchronization, utilizing frequency-shifting tags for interference-free dual-band processing.
Efficient Estimation Algorithms: Introduction of JRAC and SRAE, which achieve accuracy comparable to 2D FFT/MUSIC but with drastically reduced computational complexity.
Robust Localization Formulation: Development of an IRLS method specifically adapted for multi-static backscatter, featuring a novel scaling technique to mitigate outlier sensitivity in real-world multipath environments.
Large-Scale Validation: Experimental validation on a real-world testbed with 4 TXs, 1 RX, and 100 tags, demonstrating scalability and practicality.

4. Experimental Results

The authors evaluated their methods on a 20m $\times$ 7m indoor testbed with metal walls (heavy multipath) and two scenarios: a 4-tag setup (high SNR) and a 100-tag setup (dense, low SNR).

Estimation Accuracy:
- JRAC outperformed 2D FFT and 2D MUSIC in range estimation accuracy in real-world data due to its clustering mechanism.
- SRAE showed slightly lower accuracy than joint methods but remained competitive.
- AoA Estimation: MUSIC-based methods (including JRAC/SRAE steps) outperformed FFT in angle estimation, leveraging spatial diversity.
Computational Efficiency (Runtime):
- Estimation: JRAC and SRAE reduced runtime by up to 40 $\times$ compared to 2D MUSIC and 2D FFT.
  - Example: 2D MUSIC took ~10,380 ms; JRAC took ~475 ms; SRAE took ~55 ms.
- Localization: The proposed IRLS method achieved up to 500 $\times$ reduction in runtime compared to ML-based brute-force grid search.
  - Example: ML Grid Search took ~22,802 ms for 100 tags; IRLS took only ~51 ms.
Localization Accuracy:
- The proposed methods achieved a median localization error of 3 meters across all 100 tags in the sub-6GHz band (40 MHz bandwidth).
- Crucially, the low-complexity JRAC + IRLS combination achieved localization accuracy comparable to the high-complexity baselines (2D FFT + ML), proving that complexity reduction does not require accuracy sacrifice.

5. Significance

This work bridges the gap between theoretical backscatter localization and practical, large-scale deployment.

Scalability: By reducing computational complexity by orders of magnitude, the proposed algorithms make it feasible to localize thousands of IoT devices in real-time, a prerequisite for 6G Ambient IoT.
Practicality: The experimental validation in a harsh, multipath environment with unsynchronized hardware demonstrates that these methods are robust enough for real-world industrial and asset tracking applications.
Efficiency: The IRLS algorithm, in particular, offers a highly efficient alternative to ML-based search, enabling edge-computing devices to perform localization without heavy server offloading.

In conclusion, the paper demonstrates that Multi-static Backscatter Networks can effectively serve as integrated communication and sensing systems for next-generation IoT, provided that efficient algorithms like JRAC/SRAE and Scaled IRLS are employed to handle the computational load.