⚡ electrical engineering

Explicit Ensemble Mean Clock Synchronization for Optimal Atomic Time Scale Generation

This paper introduces the Explicit Ensemble Mean (EEM) synchronization framework, which unifies time scale generation, clock synchronization, and oscillator frequency regulation under systems and control theory by decomposing atomic ensembles into observable and unobservable components, thereby demonstrating that standard Kalman filtering is a special case that optimizes long-term frequency stability and enables explainable timing systems.

Original authors: Takayuki Ishizaki, Takahiro Kawaguchi, Yuichiro Yano, Yuko Hanado

Published 2026-03-31

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Takayuki Ishizaki, Takahiro Kawaguchi, Yuichiro Yano, Yuko Hanado

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 the conductor of an orchestra, but instead of violins and flutes, you have a room full of atomic clocks. These clocks are incredibly precise, but like any human (or machine), they have tiny quirks. Some run a tiny bit fast, some drift slowly, and some have a "jittery" nervous system.

Your goal is to create one perfect master clock (called a "Time Scale") that is more stable and accurate than any single clock in the room. This master clock will be used to run the internet, GPS, and global finance.

For decades, scientists have tried to solve this by averaging the clocks together. But they faced two big problems:

The "Black Box" Problem: They used a complex math tool called Kalman Filtering (think of it as a super-smart guesser) to combine the clocks. It worked well, but nobody really understood why it worked or why it sometimes made the clock unstable in the long run. It was like driving a car with the hood closed; you know it goes, but you don't know how the engine works.
The "Short vs. Long" Dilemma: A clock that is perfect for measuring a split-second (short-term) might drift badly over a year (long-term). Existing methods could usually only optimize for one or the other, not both.

This paper introduces a new framework called Explicit Ensemble Mean (EEM) Synchronization. Here is how it works, explained simply:

1. The "Two-Part" Secret (The Analogy of the Choir)

The authors realized that the group of clocks can be split into two distinct groups, like a choir:

The "Relative" Group (Observable): This measures how much each singer is out of tune compared to the others. We can hear this easily.
The "Absolute" Group (Unobservable): This measures the average pitch of the entire choir. If the whole choir drifts up a note together, you can't hear it by listening to the singers relative to each other. It's "hidden" or "unobservable."

The Big Discovery: The old "guessing" method (Kalman Filtering) was actually trying to estimate this hidden "average pitch" all along, but it was doing it in a messy, unstable way. The new framework makes this "average pitch" explicit. It says, "Let's stop guessing the average; let's define exactly what average we want."

2. The "Smart Conductor" (The Control System)

In the old days, the system would just digitally "nudge" the clock numbers on a screen to make them match. It was like a conductor telling the violinist, "Pretend you played a C instead of a C-sharp."

This new paper proposes physically regulating the oscillator (the heart of the clock). It's like the conductor actually reaching out and tightening the violin string so it physically plays the right note. This is much harder to do, but it's more stable and real.

3. Solving the "Short vs. Long" Dilemma

Here is the magic trick the paper performs:

Short-Term: To keep the clock steady for the next few seconds, you want to trust the clocks that are least "jittery."
Long-Term: To keep the clock steady for the next year, you want to trust the clocks that don't "drift" over time.

Usually, you have to pick one strategy. This paper creates a hybrid strategy:

The "Fast" Mode: For the first few minutes, the system acts like a super-precise average of the best short-term clocks.
The "Slow" Mode: Every few minutes (or hours), the system gently "steers" the whole group toward the best long-term average.

Think of it like surfing.

In the short term, you ride the immediate wave (short-term stability).
But every now and then, you paddle to catch a bigger, more stable wave further out (long-term stability).
The new algorithm does this automatically, switching between the two modes seamlessly to give you the smoothest ride possible.

4. Why This Matters

No More "Black Boxes": The math is now transparent. We know exactly how the system decides which clocks to trust.
Resilience: If the GPS signal (which usually helps us keep time) goes down, this system can keep running perfectly on its own using just the atomic clocks in the room.
Better Time: It creates a "Time Scale" that is more stable than anything we've had before, which is crucial for the future of 6G networks, autonomous driving, and deep-space navigation.

Summary

The authors took a messy, "black box" way of combining atomic clocks and turned it into a clear, two-step dance:

Listen to the differences between clocks to keep them in sync.
Steer the whole group toward a specific "ideal average" that changes its strategy depending on whether you need stability for a second or for a year.

They proved mathematically that this new way is the "Goldilocks" solution: not too jittery, not too drifting, but just right for both the short and long term.

1. Problem Statement

The generation of atomic time scales (e.g., UTC) from an ensemble of atomic clocks is a cornerstone of metrology. Current methods typically rely on Kalman filtering to generate a "paper clock" (a digital time scale) by adjusting clock readings, rather than physically regulating the oscillators. However, existing approaches face several critical theoretical and practical limitations:

Indeterminate Operations: In standard Kalman filtering applied to clock ensembles, the error covariance matrix diverges over time, even though the Kalman gain converges. This implies the presence of "indeterminate" operations (diverging values annihilated by matrix multiplication), leading to potential numerical instability.
Suboptimal Stability: While standard Kalman filtering minimizes estimation error variance, it does not necessarily produce the optimal time scale in terms of Allan variance (the standard metric for frequency stability). Specifically, it often yields suboptimal short-term stability.
Lack of Unified Framework: There is no unified theoretical framework based on systems and control theory that simultaneously addresses time scale generation, clock synchronization, and the physical regulation of oscillator frequencies. Most algorithms only perform digital adjustments.
Unobservable State Mystery: The mechanism by which standard Kalman filtering implicitly computes the "ensemble mean" (the synchronization destination) has not been mathematically elucidated, particularly regarding its estimation of "unobservable" states.

2. Methodology

The authors propose a novel framework called Explicit Ensemble Mean (EEM) Synchronization, grounded in Systems and Control Theory. The methodology involves the following key steps:

A. Observable Canonical Decomposition

The authors apply an observable canonical decomposition to the standard ensemble atomic clock model. They decompose the system state $x[k]$ into two complementary components:

Observable Part ( $\xi_o$ ): Represents the synchronization error (relative differences between clocks).
Unobservable Part ( $\xi_{\bar{o}}$ ): Represents the synchronization destination (the ensemble mean).

By selecting a specific basis parameter $W$ defined by a user-specified weight vector $q$ , the unobservable state is explicitly defined as the weighted average of the clocks: $\xi_{\bar{o}}[k] = (I_2 \otimes q^T)x[k]$ .

B. Determinate Kalman Filtering

Using this decomposition, the authors reformulate the standard Kalman filtering algorithm.

They prove that the diverging error covariance associated with the unobservable subspace can be removed.
They derive an equivalent determinate form of the Kalman filter where the update equations for the observable and unobservable states are decoupled in a way that eliminates indeterminate operations.
This allows for the calculation of stationary Kalman gains without recursive covariance divergence.

C. Unified Control Strategy

The framework integrates state estimation with state feedback control:

Synchronization Control ( $\omega_o$ ): Stabilizes the observable state (relative errors) to zero, ensuring all clocks synchronize to the ensemble mean.
Collective Control ( $\omega_{\bar{o}}$ ): Regulates the unobservable state (the ensemble mean itself) by physically adjusting the oscillator frequencies.
Intermittent Feedback: To balance stability, the authors propose an intermittent control strategy where the collective control input is applied at specific intervals ( $m$ steps), allowing the system to behave differently in the short term versus the long term.

3. Key Contributions

Explicit Ensemble Mean (EEM) Framework:
The paper unifies time scale generation, synchronization, and physical oscillator regulation into a single control-theoretic framework. It explicitly defines the synchronization destination via user-specified weights.
Resolution of Indeterminate Operations:
The authors mathematically prove that standard Kalman filtering contains indeterminate operations due to unobservability. They provide an equivalent determinate formulation that removes these operations, ensuring numerical stability.
Reinterpretation of Kalman Filtering:
A counter-intuitive but crucial finding is that standard Kalman filtering is significant not for estimating the observable relative states, but for estimating the unobservable weighted average state. The paper proves that standard Kalman filtering is a special case of EEM synchronization that achieves the least Allan variance only in the long term (when the weight vector matches the inverse of the random walk noise covariance).
Optimal Weight Derivation:
The paper derives the optimal ensemble mean weights for:
- Short-term stability: Weighted by the inverse of white frequency noise variances ( $q_0$ ).
- Long-term stability: Weighted by the inverse of random walk frequency noise variances ( $q_\infty$ ).
  Since these weights differ, a static weight cannot optimize both simultaneously.
Balanced Stability Algorithm:
The authors design a novel algorithm that uses intermittent feedback control. By choosing a short-term optimal weight ( $q_0$ ) for the continuous dynamics and applying a collective control input periodically to steer the system toward the long-term optimal weight ( $q_\infty$ ), the system achieves superior stability in both time domains.

4. Results

The authors validated their framework using numerical simulations with an ensemble of 10 atomic clocks based on real hardware noise parameters.

Removal of Divergence: The determinate Kalman filter successfully eliminated the error covariance divergence observed in standard implementations.
Allan Variance Performance:
- Standard Kalman Filtering: Produced a time scale with excellent long-term stability but suboptimal short-term stability.
- EEM with Static Weights: Using $q_0$ improved short-term stability but degraded long-term stability; using $q_\infty$ did the opposite.
- Proposed Intermittent EEM: The new algorithm achieved the best of both worlds. The statistical Allan variance of the generated time scale followed the short-term optimal curve ( $q_0$ ) for small time intervals and smoothly transitioned to the long-term optimal curve ( $q_\infty$ ) for large intervals.
Physical Regulation: Unlike previous "paper clock" methods, the proposed method demonstrated the ability to physically regulate oscillator frequencies to achieve these results, not just digitally adjust readings.

5. Significance

This paper makes a fundamental contribution to time and frequency metrology:

Theoretical Clarity: It provides the first rigorous mathematical explanation of why and how Kalman filtering works for time scale generation, demystifying the "implicit" ensemble mean.
Explainable Timing Systems: By unifying estimation and control, the framework offers a transparent, "explainable" basis for designing timing systems, moving away from ad-hoc algorithms like KPW or AT1/AT2.
Resilient Infrastructure: The ability to physically regulate oscillators based on a unified control theory paradigm is critical for building resilient timing infrastructures that can operate robustly without GNSS signals (e.g., for cn-PRTC).
Future Applications: The framework opens new avenues for handling heterogeneous clock ensembles, third-order dynamics, and adaptive exclusion of faulty clocks, providing a solid foundation for next-generation atomic time scales.