⚡ electrical engineering

Filtering Module on Satellite Tracking

This paper presents a mathematical framework for satellite state estimation using a linearized model and Kalman filter variants, demonstrating that a proposed micro-Kalman filter ( $\mu$ KF) achieves performance nearly identical to classical and advanced filters with low, bounded estimation errors.

Original authors: Moh Kamalul Wafi

Published 2026-04-16

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Moh Kamalul Wafi

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 trying to track a tiny, fast-moving drone flying through a thick fog. You have a camera on the ground, but the camera is a bit shaky, and the fog makes the image blurry. You know the laws of physics that govern how the drone should move, but you can't see it perfectly.

This paper is about building the smartest possible "guessing machine" to figure out exactly where that drone is, even when your eyes (sensors) are lying to you a little bit.

Here is the breakdown of the paper using simple analogies:

1. The Problem: The Shaky Camera

The author, Moh Kamalul Waﬁ, is looking at satellites orbiting Earth.

The Reality: Satellites move in predictable circles.
The Problem: When we try to measure their position from the ground, our instruments make mistakes. It's like trying to measure the distance to a car with a ruler that stretches and shrinks randomly.
The Goal: We need to filter out the "noise" (the shaky measurements) to find the true path of the satellite.

2. The Classic Solution: The "Standard GPS"

For decades, engineers have used a tool called the Kalman Filter.

The Analogy: Think of the Kalman Filter as a very experienced coach. The coach has a notebook with the satellite's predicted path. Every time the satellite sends a new (blurry) signal, the coach looks at the signal, compares it to the prediction, and says, "Okay, the signal is a bit off, but the prediction is pretty solid. Let's adjust the guess just a little bit."
How it works: It constantly updates its "confidence" in its own guess. If the sensors are noisy, it trusts the math more. If the sensors are clear, it trusts the sensors more.

3. The New Idea: The "Micro-Kalman Filter" (µKF)

The author proposes a new version called the Micro-Kalman Filter.

The Analogy: Imagine the Standard Coach (Classic KF) keeps a notebook of uncertainties (e.g., "I'm 80% sure"). The Micro-Kalman Filter is like a coach who keeps a notebook of certainties (e.g., "I am 20% unsure").
Why change? Mathematically, working with "certainty" (called the Information Form) is sometimes easier for computers to handle, especially if you want to split the work among many small computers (like a team of drones talking to each other).
The Big Question: Does this new "certainty-based" coach do a better job than the old "uncertainty-based" coach?

4. The Experiment: The Race

The author ran a massive simulation (a video game scenario) to test this.

The Setup: He created a virtual satellite orbiting Earth. He added two types of "fog" (noise) to the measurements:
1. Light Fog: Small measurement errors.
2. Heavy Fog: Large measurement errors.
The Racers: He pitted the Micro-Kalman Filter against:
- The Classic Kalman Filter.
- The Extended Kalman Filter (EKF).
- The Unscented Kalman Filter (UKF).
- An Adaptive Kalman Filter.

5. The Results: A Dead Heat

The results were surprising but reassuring:

The Winner: There was no winner because everyone tied.
The Finding: The Micro-Kalman Filter performed exactly the same as the Classic Kalman Filter and all the other fancy versions.
The Takeaway: The new "Micro" method is just as accurate as the old standard. It doesn't make mistakes, and it doesn't miss the target.

6. Why Does This Matter?

You might ask, "If they are the same, why write a paper?"

The Metaphor: Imagine you have a car that gets 30 miles per gallon (the Classic Filter). You invent a new engine design (the Micro Filter) that also gets 30 miles per gallon, but it's built differently.
The Benefit: Even though the mileage is the same, the new engine might be easier to fix, easier to share with other cars, or easier to fit into a smaller space.
The Future: The author suggests that while the "Micro" filter is great for simple, straight-line problems (like our satellite), the real challenge is using these smart guessing machines for non-linear problems (like a drone doing flips) or for networks where many satellites talk to each other to figure out where they all are.

Summary

This paper is a "proof of concept." It says: "We built a new, slightly different version of the world's most famous tracking tool. We tested it in a storm, and it performed just as perfectly as the original. Now we can use this new version for more complex tasks in the future."

It's a victory for efficiency and flexibility in the world of space tracking.

1. Problem Statement

The paper addresses the challenge of satellite state estimation in the context of space exploration and communication systems. Accurate tracking of a satellite's orbital motion is critical for system performance but is complicated by:

Nonlinear Dynamics: Satellite motion is governed by Newton's laws under an inverse-square gravitational field.
Measurement Uncertainty: Ground-based observations are subject to noise, specifically affecting range (radial) and angular measurements.
Computational Constraints: Real-time implementation requires efficient algorithms, particularly for systems where repeated matrix inversions (as in standard Kalman Filters) may be computationally burdensome.

The objective is to develop a mathematical framework for estimating the satellite's state (position and velocity) using a linearized model and to evaluate the performance of a proposed Micro-Kalman Filter (µKF) against classical and advanced filtering variants.

2. Methodology

A. Mathematical Modeling

The authors formulate a linearized state-space model based on small deviations from a nominal circular orbit.

State Vector ( $x$ ): Defined as deviations from the nominal trajectory:
- $x_1$ : Radial deviation ( $r - R$ )
- $x_2$ : Radial velocity ( $\dot{r}$ )
- $x_3$ : Scaled angular deviation ( $R(\theta - \omega t)$ )
- $x_4$ : Scaled angular velocity deviation ( $R(\dot{\theta} - \omega)$ )
Dynamics: The nonlinear equations of motion are linearized around the nominal orbit ( $r=R, \theta=\omega t$ ). The system is treated as a Linear Time-Invariant (LTI) system with process noise neglected in the dynamics (uncertainty introduced solely via measurement noise).
Measurement Model: Two noisy observations are collected at discrete time steps:
1. Radial deviation ( $y_1$ )
2. Scaled angular deviation ( $y_3$ )
- Measurement noise ( $v$ ) is modeled as zero-mean Gaussian with a diagonal covariance matrix $\Sigma_v = \text{diag}(\phi, \psi)$ .

B. Filtering Approaches

The paper compares several estimation strategies:

Classical Kalman Filter (KF): Uses the standard covariance-based recursive prediction and update steps. It also considers a Steady-State formulation where the Kalman gain is computed offline by solving the Discrete Algebraic Riccati Equation (DARE) to reduce online computational load.
Micro-Kalman Filter (µKF): The core proposal of the paper. This is an Information Filter formulation.
- Instead of propagating the error covariance matrix ( $P$ ), it operates on the information matrix (inverse covariance).
- The update step is performed in the information domain before state propagation.
- It utilizes transformed measurements ( $z_k = H^T \Sigma_v^{-1} y_k$ ) and information weights.
- This structure is designed to be more suitable for distributed estimation and avoids explicit Kalman gain inversion in the update step.
Comparative Benchmarks: The study also simulates the Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF), and an Adaptive Kalman Filter to provide a comprehensive performance baseline.

C. Simulation Setup

Parameters: Simulations run for $N=1000$ steps with a sampling interval $h=0.01$ .
Noise Configurations: Two scenarios are tested:
- Type 1: Baseline noise variances ( $\phi=0.1, \psi=0.5$ ).
- Type 2: Increased angular noise variance to test robustness.
Metric: Performance is evaluated using the Mean Square Estimation Error (MSEE) averaged over multiple iterations.

3. Key Contributions

Unified Framework: The paper presents a unified mathematical framework comparing the classical covariance-based KF with the information-based µKF for satellite tracking.
µKF Validation: It demonstrates that the µKF, despite its structural differences (operating in the information domain), achieves estimation accuracy nearly identical to the classical KF for linear Gaussian systems.
Steady-State Analysis: The paper details the derivation of steady-state gain matrices via the DARE, highlighting the potential for reduced computational complexity in time-invariant systems.
Comprehensive Benchmarking: The study includes a direct comparison with EKF, UKF, and Adaptive KF, confirming that for linearized satellite dynamics, the simpler linear approaches (KF and µKF) are sufficient and optimal.

4. Results

Accuracy: The µKF successfully tracks all four state variables (radial position, radial velocity, angular position, angular velocity) with small, bounded errors.
Equivalence: Under linear Gaussian assumptions, the MSEE of the µKF is virtually indistinguishable from the classical KF. The difference in MSEE between the two methods is negligible (on the order of $10^{-4}$ ).
Noise Robustness: Even when measurement noise variance is increased (Noise Type 2), the µKF maintains performance comparable to the classical KF.
Linearity vs. Nonlinearity: The EKF and UKF showed performance similar to the KF, confirming that the linearization of the satellite dynamics was sufficient and that complex nonlinear approximations were unnecessary for this specific model.
Tables & Figures:
- Tables 1 and 2 show specific MSEE values for different states under both noise types.
- Table 3 confirms the averaged MSEE ( $\Xi$ ) for both KF and µKF are extremely close.
- Figures 1 and 2 visualize the state trajectories and error convergence, showing rapid stabilization.

5. Significance and Conclusion

The paper concludes that the Micro-Kalman Filter (µKF) is a viable and effective alternative to the classical Kalman Filter for satellite tracking in linear Gaussian systems.

Theoretical Value: It validates the equivalence of the information filter and covariance filter formulations in this context.
Practical Value: The µKF offers a structurally different approach that is particularly advantageous for distributed estimation (Distributed Kalman Filtering) and scenarios where information fusion is required, without sacrificing accuracy.
Future Directions: The author suggests extending this framework to nonlinear systems (using exact linearization or unscented filtering), incorporating adaptive mechanisms for time-varying noise, and applying the method to networked multi-agent systems and integrated estimation-control problems.

In summary, this work provides a rigorous mathematical and numerical confirmation that information-based filtering (µKF) is a robust, accurate, and computationally efficient method for satellite state estimation, offering a strong foundation for future distributed space systems.