⚡ electrical engineering

Synchronization and Localization in Ad-Hoc ICAS Networks Using a Two-Stage Kuramoto Method

This paper proposes a signal-agnostic, distributed two-stage Kuramoto method to achieve joint frequency-phase synchronization and accurate localization in peer-to-peer vehicular Integrated Communications and Sensing (ICAS) networks, while mitigating performance degradation caused by finite sampling frequencies.

Original authors: Dominik Neudert-Schulz, Thomas Dallmann

Published 2026-04-15

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Dominik Neudert-Schulz, Thomas Dallmann

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 a group of self-driving cars driving through a busy city. To avoid crashing and to talk to each other, they need two things to work perfectly:

Perfect Timing: They all need to speak at the exact same speed and rhythm (Synchronization).
Perfect Location: They need to know exactly where every other car is, even if GPS is blocked by tall buildings (Localization).

This paper presents a clever solution to get all these cars working together without needing a central boss (like a satellite or a tower) to tell them what to do. Instead, they act like a flock of birds or a school of fish, coordinating with each other.

Here is the breakdown of their solution using simple analogies:

1. The Problem: The "Out-of-Tune Orchestra"

Imagine a group of musicians (the cars) trying to play a song together.

The Issue: Each musician has their own metronome (clock), and they are all ticking at slightly different speeds. Some are rushing, some are dragging.
The Consequence: If they try to play together, it sounds like a mess. Furthermore, because they can't agree on the rhythm, they can't accurately guess how far apart they are from each other just by listening to the sound.

2. The Solution: The "Two-Stage Kuramoto Method"

The authors use a mathematical concept called the Kuramoto Model. Think of this as a "social network for clocks."

Stage 1: Finding the Common Beat (Frequency Sync)
Imagine the musicians are in a room. They listen to each other. If you hear someone playing too fast, you slow down a tiny bit. If you hear someone too slow, you speed up. Eventually, they all settle on the exact same tempo.
- The Paper's Twist: In the real world, our "ears" (digital sensors) aren't perfect. They take snapshots of the sound at specific intervals. If the snapshots are too slow, the musicians get confused and drift apart again. The authors added a "drift compensation" rule to keep them from wandering off, even with imperfect ears.
Stage 2: Getting in Step (Phase Sync)
Once they are playing at the same speed, they need to start the notes at the exact same moment. This is like a choir starting a song on "One, Two, Three!"
The paper's method ensures that not only do they play at the same speed, but they also hit the "downbeat" simultaneously.

3. The Magic Trick: Turning Sound into a Map (Localization)

This is the coolest part. Usually, to know where someone is, you need a map. But here, the cars figure out their location just by listening to the delay in the sound.

The Analogy: Imagine you are in a dark room with a friend. You clap your hands. Your friend hears the clap a split second later. You know that the longer the delay, the farther away they are.
The Paper's Innovation: The cars send out signals (like claps). By measuring exactly how long it takes for the signal to travel from Car A to Car B, they can calculate the distance.
The "Two-Stage" Advantage: Because the cars are now perfectly synchronized (Stage 1 & 2), they can measure these tiny time delays with extreme precision. They don't just know "Car B is far away"; they know "Car B is 30 meters to my left."

4. Why This Matters for Self-Driving Cars

No GPS Needed: In cities with tall skyscrapers, GPS signals often bounce off buildings or get blocked. This system lets cars find each other using only their own radios.
Signal Agnostic: It doesn't matter what kind of signal they use (whether it's a standard radio wave or a radar pulse). The method works like a universal translator for timing.
Robustness: The authors tested this with "imperfect" digital sensors (simulating real-world hardware). They found that without their special "drift compensation" trick, the system would fail. With the trick, it works beautifully.

Summary

Think of this paper as a recipe for a self-organizing traffic swarm.

Listen: Cars listen to each other's signals.
Adjust: They tweak their internal clocks to match the group (Synchronization).
Measure: They use the tiny time it takes for signals to travel to measure distances (Localization).
Stabilize: They use a special math trick to ignore the "noise" caused by imperfect digital sensors.

The result? A fleet of cars that can coordinate perfectly and map their surroundings, even if the GPS is dead and the hardware is cheap.

1. Problem Statement

The paper addresses the challenges of enabling Integrated Communications and Sensing (ICAS) in peer-to-peer (P2P) vehicular networks. In such decentralized environments, there is no central coordinating unit. To support high-precision radar applications (coherent signal processing) and autonomous driving, communicating entities (hosts) must achieve:

Precise Synchronization: Exact alignment in both frequency and phase across all hosts.
Accurate Localization: Reliable estimation of the relative positions of surrounding vehicles, especially in urban scenarios where Global Navigation Satellite Systems (GNSS) often fail.

Key Challenges Identified:

Existing consensus-based synchronization algorithms (e.g., standard Kuramoto models) often neglect propagation delays or do not contribute to position estimation.
Real-world implementations suffer from finite sampling frequencies, which introduce errors that severely degrade synchronization and localization performance if not mitigated.
There is a need for a signal-agnostic solution that works with various ICAS signal types (e.g., OFDM, FMCW, pulse radar).

2. Methodology

The authors propose a joint, distributed synchronization and localization scheme based on an extended Two-Stage Kuramoto Model.

A. Signal Model and Phase Definition

Signal Structure: Hosts transmit sequences of "base signals" ( $s_{p,n}$ ). The model is quasi-periodic, allowing signal duration ( $T_p$ ) and start times ( $\tau_p$ ) to vary.
Phase Definition: Instead of using the Hilbert transform (which is unstable for wideband signals), the authors define a transmit phase ( $\phi_T$ ) and receive phase ( $\phi_R$ ) based on the index of the base signal and the normalized time within that signal.
Phase Difference: The core metric is the phase difference ( $\psi_{p,q}$ ) between the signal sent by host $p$ and received by host $q$ . This difference inherently contains the propagation delay ( $\delta_{p,q}$ ).

B. Decomposition of Phase Differences

To separate synchronization from localization, the total phase difference matrix $\Psi$ is decomposed into two components:

$\Psi_\tau$ (Synchronization Component): Depends on the difference in base signal start times. This matrix is skew-symmetric ( $\Psi_\tau = -\Psi_\tau^T$ ).
$\Psi_\delta$ (Propagation Delay Component): Depends on the propagation delay. This matrix is symmetric ( $\Psi_\delta = \Psi_\delta^T$ ).

Using matrix algebra, the authors estimate these components from the observed total phase difference matrix $\Psi$ :
$\Psi_\tau \approx \frac{1}{2}(\Psi - \Psi^T)$
$\Psi_\delta \approx \frac{1}{2}(\Psi + \Psi^T)$

C. The Two-Stage Kuramoto Algorithm

The synchronization process is executed in two stages on each host:

First Stage (Frequency Consensus): Updates the base frequency ( $f_b$ ) using a consensus filter. A novel penalty term is added to prevent joint drift of signal durations caused by finite sampling rates. This term compares the average inverse signal duration against an "anchor frequency."
Second Stage (Phase/Start Time Synchronization): Updates the instantaneous frequency ( $f_q$ ) based on the skew-symmetric phase difference matrix ( $\Psi_\tau$ ). This forces the phase differences to converge to zero, aligning the start times of the base signals.

D. Localization

Once the propagation delay matrix ( $\Psi_\delta$ ) is isolated, the time delays are converted to distances. These distances are used to estimate the relative positions of all hosts using Multidimensional Scaling (MDS) and Procrustes analysis to align the estimated geometry with the true geometry.

3. Key Contributions

Joint Synchronization and Localization: The paper extends the Two-Stage Kuramoto method to simultaneously achieve frequency/phase synchronization and mutual localization in a fully distributed P2P network.
Signal Agnosticism: The algorithm relies on the ability to detect base signal start times and durations, making it applicable to a wide range of ICAS signals (communication-centric or radar-centric).
Mitigation of Sampling Effects: The authors identify that finite sampling frequencies cause frequency drift and instability. They propose a drift compensation mechanism (via the anchor frequency and penalty term in the first stage) that significantly stabilizes the system.
Matrix Decomposition Technique: A novel method to separate synchronization errors from propagation delays using the symmetry and skew-symmetry properties of the phase difference matrices.

4. Simulation Results

The authors validated the algorithm using simulations with four hosts in different scenarios:

Scenario A (Ideal/Infinite Sampling): The algorithm successfully converged. Instantaneous frequencies and start times aligned perfectly, and localization error dropped to near zero.
Scenario B (Finite Sampling, No Compensation): While frequencies converged somewhat, a joint drift occurred. The start times oscillated, and the localization estimates became unstable and inaccurate. This demonstrated the severe impact of ignoring sampling effects.
Scenario C (Finite Sampling, With Drift Compensation): By applying the proposed penalty term ( $\alpha_a = 0.05$ $α_{a} = 0.05$ ), the system successfully countered the drift.
- Frequency/Phase: Convergence was restored, closely matching the ideal case.
- Localization: The position estimates became stable and reliable, significantly outperforming Scenario B, though with a small residual error compared to the ideal case.

5. Significance

This work is significant for the development of autonomous vehicular networks and 6G ICAS systems.

It provides a practical solution for GNSS-denied environments (e.g., urban canyons) where vehicles must rely on mutual sensing for positioning.
It bridges the gap between theoretical synchronization models and real-world implementation constraints (specifically finite sampling rates).
By enabling coherent processing without a central infrastructure, it paves the way for robust, scalable, and high-precision distributed sensing networks.