⚡ electrical engineering

Applying the Spectral Method for Modeling Linear Filters: Bessel, Papoulis, and Legendre Filters

This paper introduces a spectral method-based technique for simulating continuous-time linear filters by representing signals as expansion coefficients and characterizing the filter via a two-dimensional nonstationary transfer function, which is demonstrated through the modeling of Bessel, Papoulis, and Legendre filters.

Original authors: Konstantin A. Rybakov

Published 2026-08-21

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Konstantin A. Rybakov

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

In the world of electronics and signal processing, a filter is a device that acts like a sieve for information. Just as a kitchen strainer separates pasta from boiling water, an electronic filter separates a useful signal, such as a voice or a radio broadcast, from unwanted background noise. These devices are fundamental to modern life, appearing in everything from the audio systems in our cars to the communication networks that carry our text messages. For decades, engineers have relied on a specific set of mathematical rules to design these filters, ensuring they let the right frequencies pass while blocking the rest. However, simulating how these filters behave on a computer has traditionally required a compromise: breaking down continuous, flowing time into tiny, discrete steps. This approach, while useful, introduces small errors and distortions, much like trying to draw a smooth curve using only a series of straight, jagged lines.

A researcher at the Moscow Aviation Institute has now proposed a different way to model these filters, one that avoids breaking time into steps entirely. Instead of forcing the signal into a grid of discrete moments, this new technique treats the signal as a continuous flow, analyzing it through a method called the spectral method. In this approach, signals are not viewed as a sequence of points in time, but as a collection of coefficients that describe their shape and behavior across a chosen set of mathematical building blocks. By using this continuous framework, the researcher can simulate how a filter transforms a signal without ever leaving the realm of continuous time. The study focuses on three specific families of filters known as Bessel, Papoulis, and Legendre filters. These are not the most common types found in everyday consumer electronics, but they possess unique properties that make them valuable for specialized tasks, such as preserving the shape of a signal in audio processing or achieving a sharp drop-off in noise without creating unwanted ripples.

The core of this work involves translating the mathematical descriptions of these three filter families into a new format that the spectral method can understand. For each type of filter, the researcher derived a specific set of rules, represented as a large, two-dimensional matrix, that dictates how the filter coefficients of an input signal are converted into the coefficients of an output signal. This matrix acts as a bridge, allowing the computer to calculate the filtered result directly. The study tested these new models by simulating a scenario where a clean, useful signal was mixed with a specific type of deterministic noise. The goal was to see how well each filter could strip away the noise while keeping the original signal intact. The results showed that all three filter families successfully reduced the noise, but they did so with different levels of precision and efficiency depending on their order, or complexity.

The simulations revealed distinct characteristics for each filter type. The Bessel filters, which are known for their smooth response, proved to be the least accurate in this specific test. They suppressed the noise less effectively than the others, and increasing the number of calculation steps did not significantly improve their performance. This suggests that for these filters, the limitation lies in the filter's own design rather than the method used to calculate it. In contrast, the Papoulis filters demonstrated high accuracy, though their performance varied depending on whether the filter order was even or odd. The results for these filters showed that the calculation method itself played a larger role in the final error, meaning that using more calculation steps could yield better results. The Legendre filters performed the best overall, particularly a specific variation of them, outperforming even the Papoulis filters in many cases. Like the Papoulis filters, their accuracy improved with more detailed calculations, indicating that their superior performance was limited only by the precision of the simulation.

One of the most significant findings of this work is that the new technique works without needing to convert the problem into a discrete, step-by-step format. This is a departure from standard practice, where engineers often use methods that approximate continuous signals with discrete points, a process that can cause frequency distortions or "aliasing." By staying in continuous time, the researcher demonstrated that it is possible to model these complex filters with high fidelity. The study also showed that this method is flexible enough to be applied to other, more advanced filter designs, such as modified Bessel filters and Halpern filters, suggesting a broader utility for the technique. While the research was conducted through computer simulation rather than physical hardware testing, the results provide a robust mathematical foundation for designing and analyzing these filters in a way that preserves the continuous nature of the signals they are meant to process.

The implications of this work extend to the design stage of signal processing systems. Engineers who need to create filters for applications requiring precise control over signal shape, such as in medical imaging or high-speed telecommunications, can now use this continuous-time modeling approach to predict how their designs will behave before building them. The study confirms that while different filters have different strengths, the spectral method provides a unified and accurate way to simulate them all. By avoiding the approximations inherent in discrete-time methods, this approach offers a clearer view of the filter's true capabilities, allowing for more precise engineering decisions. The research does not claim to replace existing methods for all applications, but it offers a powerful alternative for scenarios where maintaining the integrity of the continuous signal is paramount.

Technical Summary: Applying the Spectral Method for Modeling Linear Filters

Problem Statement
The paper addresses the problem of computer simulation for linear filters (specifically Bessel, Papoulis, and Legendre types) used in signal processing. Traditional simulation methods typically require a transition from continuous time to discrete time (e.g., via impulse invariance or bilinear transformation). The author notes that such transitions introduce inherent drawbacks, including frequency aliasing and frequency warping. The objective is to develop a modeling technique that operates entirely in continuous time, thereby avoiding these discretization artifacts, particularly useful during the filter design stage.

Methodology
The proposed technique utilizes the spectral method for analyzing linear control systems. Instead of solving differential equations directly in the time domain or converting to discrete time, the method represents input and output signals as ordered sets of expansion coefficients within a chosen basis (specifically a trigonometric basis for the numerical experiments).

Key components of the methodology include:

Spectral Representation: Signals $g(t)$ and $x(t)$ are expanded in an orthonormal basis $\{q(i, t)\}$ . The transformation from time-domain signals to their spectral images (infinite column matrices of coefficients) is denoted as $S$ .
Two-Dimensional Nonstationary Transfer Function ( $W$ ): The filter itself is modeled not by a standard transfer function $H(s)$ , but by a two-dimensional nonstationary transfer function $W$ . This matrix relates the input spectral coefficients to the output spectral coefficients via the algebraic relation $X = WG$ .
Direct Derivation: A core advantage of this approach is that $W$ $W$ can be derived directly from the filter's rational transfer function $H(s)$ $H (s)$ (or its corresponding differential equation) without first calculating the impulse response. The matrix $W$ $W$ is expressed as a function of the matrix $P$ $P$ (representing the differentiator) or $P^{-1}$ $P^{- 1}$ (representing the integrator).
- For a filter defined by $H(s) = M(s)/D(s)$ , the spectral matrix is $W(P) = D^{-1}(P)M(P)$ .
Basis Selection: While the description of the filter in the spectral domain is general, the numerical experiments employ a trigonometric basis. The paper provides explicit formulae for the elements of the matrices $P$ , $P^{-1}$ , as well as auxiliary matrices for time shifting ( $S_\tau$ ) and interval restriction ( $A_\theta$ ) required for error computation.

Key Contributions
The paper extends the application of the spectral method to three specific families of linear filters defined by orthogonal polynomials, which had not been previously modeled using this specific continuous-time spectral technique:

Bessel Filters: The paper derives the two-dimensional nonstationary transfer function for Bessel filters (and generalized Bessel filters) using reverse Bessel polynomials. It provides explicit recurrence relations and matrix formulations for these filters.
Papoulis Filters: The synthesis of Papoulis filters (optimal "L" filters) is detailed. This involves spectral factorization of polynomials derived from Legendre polynomials. The paper outlines the procedure for finding the necessary roots and constructing the corresponding matrix $W$ .
Legendre Filters: The method is applied to Legendre filters, which utilize modified associated Legendre polynomials. The paper details the synthesis process involving spectral factorization of equations based on these polynomials.

Additionally, the work briefly demonstrates how this technique can be extended to Modified Bessel, Halpern, and Ultraspherical filters. The paper also presents the necessary mathematical machinery, including the explicit elements of the nonstationary transfer functions for elementary blocks (differentiators, integrators, time shifts) relative to the trigonometric basis.

Results
The technique was validated through numerical experiments involving the filtering of a deterministic signal mixed with noise.

Setup: A useful signal $u(t) = \sin(5\pi t)$ was mixed with noise $v(t) = 0.1(\cos(85\pi t) + \sin(105\pi t))$ . The filters were tested with cutoff frequencies and various orders ( $n=2$ to $6$).
Error Analysis: The filtering quality was measured by the $L_2$ error between the filtered output and the original useful signal, accounting for phase delay.
Findings:
- Bessel Filters: Showed the lowest accuracy among the tested types, attributed to their slow amplitude roll-off. The error was dominated by the filter's inherent characteristics rather than the spectral method's truncation error.
- Papoulis Filters: Yielded high accuracy, with even-order filters generally outperforming odd-order ones. The error showed a dependence on the truncation order $L$ , indicating that both filter characteristics and spectral approximation errors influenced the results.
- Legendre Filters: Generally outperformed Papoulis filters, particularly when the parameter $m=2$ was used. Similar to Papoulis filters, higher orders required larger truncation orders to minimize spectral approximation errors.
- Convergence: Increasing the truncation order $L$ (matrix size) reduced the spectral method error, confirming the convergence of the spectral approximation.

Significance and Claims
The paper claims to propose a new technique for computer simulation of linear filters in continuous time. Its primary significance lies in:

Avoiding Discretization: It eliminates the need for transitioning to discrete time, thereby avoiding frequency aliasing and warping associated with methods like the bilinear transform.
Generalization: It successfully extends the spectral form of mathematical description—previously applied to Butterworth, Linkwitz–Riley, and Chebyshev filters—to Bessel, Papoulis, and Legendre filters.
Computational Framework: It provides a complete algebraic framework (matrix representations) for simulating these specific filter families, including the derivation of necessary nonstationary transfer functions and the handling of phase delays and error computation within the spectral domain.

The author concludes that this approach is particularly useful at the filter design stage and can be generalized to fractional-order filters, provided an appropriate basis (such as Legendre polynomials) is selected. The work is presented as a complement to existing literature on linear filter theory, offering a continuous-time alternative to standard discrete-time simulation methods.

Technical Summary: Applying the Spectral Method for Modeling Linear Filters

More like this