Trustworthy Digital Representations of Analog Information - An Application-Guided Analysis of a Fundamental Theoretical Problem in Digital Twinning

Holger Boche, Yannik N. Böck, Ullrich J. Mönich, Frank H. P. FitzekAlgorithms 2023 · DOI

TL;DR

This paper tackles the fundamental challenge of accurately representing continuous analog signals on digital hardware, a critical issue for the development of reliable digital twins. The authors demonstrate that traditional discrete-time signal processing methods, while convenient, can lead to a loss of essential information, particularly concerning signal properties like norms. They propose and analyze continuous-time representations as a more trustworthy alternative for applications demanding high integrity, such as in future 6G communication systems.

What Problem Does This Paper Address?

The digital world is increasingly intertwined with the physical, leading to the rise of "digital twins" – virtual replicas of physical systems used for monitoring, control, and prediction. A core challenge in creating these twins lies in accurately capturing and processing analog information from the physical world using digital computers. While Shannon's sampling theorem provides a theoretical basis for reconstructing bandlimited analog signals from discrete samples, this paper argues that this theoretical reconstruction doesn't always translate to algorithmic trustworthiness. Specifically, it questions whether all relevant properties of an analog signal can be reliably computed from its discrete-time digital representation. This is crucial for applications like digital twinning in 6G, where provable performance and high integrity are paramount. The paper aims to formalize this problem and analyze the implications of different digital representations for signal processing and digital twinning.

Methods

The authors employ a rigorous approach rooted in computable analysis and Turing machine theory. They formalize a framework for "machine-readable languages" and "machine-readable descriptions" to represent analytic objects on digital hardware. This framework is then applied to sampling and interpolation theory. They introduce two types of digital descriptions for bandlimited analog signals:

  • Discrete-time descriptions (X*): These mirror the conventional approach in digital signal processing, relying on sampled values.
  • Continuous-time descriptions (F*): These aim to capture the signal's continuous-time behavior more directly.

The core of their analysis involves comparing the computability of essential signal properties, such as the signal's norm, using these two types of descriptions. They leverage concepts from Banach spaces and linear operators to mathematically define and analyze these representations.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • The "Trustworthiness" Gap: The paper highlights a critical gap between the theoretical reconstruction of analog signals (via Shannon's theorem) and the practical, algorithmic trustworthiness of their digital representations. Simply being able to reconstruct a signal doesn't guarantee that all its essential properties can be reliably computed.
  • Discrete vs. Continuous Representations: The authors demonstrate that while discrete-time representations are computationally convenient, they can lead to a loss of information, particularly concerning signal norms. Continuous-time representations, on the other hand, are shown to be more "sufficient" for capturing these properties.
  • Information Content of Representations: The paper establishes a formal hierarchy of information content between discrete and continuous representations, showing that continuous-time descriptions capture "more" information.
  • Implications for Digital Twinning: The findings have direct implications for digital twinning, suggesting that for applications requiring high integrity and provable performance (like in 6G), continuous-time representations might be essential for ensuring that critical signal properties are accurately captured and processed.

Key Results

  • The paper proves that the generalized Shannon equivalence (the theoretical ability to reconstruct a signal from its samples) does not hold true on an algorithmic level.
  • Discrete-time descriptions (X*) are insufficient for reliably computing certain relevant signal properties, specifically the signal's norm (||f||*).
  • Continuous-time descriptions (F*) are sufficient for reliably computing these same properties.
  • The relationship between these representations is formalized as F > X, meaning continuous-time descriptions contain more information than discrete-time ones.
  • The paper demonstrates that while the sampling operator (S) is computable, the interpolation operator (T) is not, highlighting a fundamental algorithmic limitation of discrete-time processing.

Limitations and Future Work

The authors acknowledge that their formalization of machine-readable languages and computable analysis is abstract but necessary for mathematical rigor. They suggest that their framework can be extended to analyze other relevant signal properties beyond norms, such as time concentration, time evaluation, and time derivatives. The work lays the groundwork for further investigation into the trustworthiness of digital representations in various critical infrastructure applications, particularly in the context of the upcoming 6G standard.

Summary updated

BibTeX
@article{boche2023trustworthy,
  author = {Holger Boche and Yannik N. Böck and Ullrich J. Mönich and Frank H. P. Fitzek},
  title = {{Trustworthy Digital Representations of Analog Information - An Application-Guided Analysis of a Fundamental Theoretical Problem in Digital Twinning}},
  journal = {Algorithms},
  year = {2023},
  doi = {10.3390/a16110514},
}
Figure