Nonlinear Bandwidth and Bode Diagrams based on Scaled Relative Graphs

Julius P.J. Krebbekx, Roland Tóth, Amritam Das · 2025

Read Nonlinear Bandwidth and Bode Diagrams based on Scaled Relative Graphs by Julius P.J. Krebbekx, Roland Tóth, Amritam Das (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-01585.txt.

Introduction

Julius Krebbekx, Roland Tóth, and Amritam Das extend a classic chart type to nonlinear systems. Their 2025 paper uses scaled relative graphs, a graphic method for frequency-based analysis. Bode diagrams are standard for linear systems but do not fit systems whose response changes with input size. The authors narrow their graphs to set input types.

That step yields bounds on gain that vary with frequency for each input type. They treat pure tones, harmonics, and subharmonics apart. Applied to loop transfer and sensitivity, the method gives a bandwidth notion that matches the linear case. A DC motor with a stray nonlinearity serves as the test case. Computer runs confirm the results.

Catalog record

CREATORJulius P.J. Krebbekx, Roland Tóth, Amritam Das
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE45 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-01585

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Nonlinear Bandwidth and Bode Diagrams based on Scaled Relative Graphs Julius P.J. Krebbekx, Roland Tóth, Amritam Das Abstract Scaled Relative Graphs (SRGs) provide a novel graphical frequency domain method for the analysis of nonlinear systems. In this paper, we use the restriction of the SRG to particular input spaces to compute frequency-dependent gain bounds for incrementally stable nonlinear systems. This leads to a nonlinear (NL) generalization of the Bode diagram, where the sinusoidal, harmonic, and subharmonic inputs are considered separately. When applied to the analysis of the NL loop transfer and sensitivity, we define a notion of bandwidth for both the open-loop and closed-loop, compatible with the LTI definitions. We illustrate the power of our method on the analysis of a DC motor with a parasitic nonlinearity, verifying our results in simulations. I Introduction In the case of Linear Time Invariant (LTI) systems, graphical analysis using the Nyquist diagram [ 1 ] and the Bode diagram [ 2 ] is the cornerstone of control engineering. They are easy to use and allow for intuitive analysis and controller design methods. However, it is unclear how to generalize graphical frequency domain methods to nonlinear system analysis and controller design. When pushing the performance of dynamical systems through control design, the Nonlinear (NL) dynamics start to play a crucial role. Inspired by the success of analyzing performance through Bode diagrams in the LTI case, e.g. for mixed sensitivity shaping [ 3 ] , there have been multiple attempts to generalize the Bode diagram to the NL case. The describing function [ 4 ] , which is approximative, has been the first step in this direction, successfully applied in applications [ 4 , 5 , 6 ] . Most modern and non-approximative methods focus on sinusoidal inputs [ 7 , 6 ] , but other works also consider higher-order harmonics in the input [ 8 ] or specific periodic inputs [ 9 ] . The Scaled Relative Graph (SRG) [ 10 ] proposed in [ 11 ] is a new graphical method to analyze nonlinear feedback systems. It is an exact method, and it is intuitive because of its close connection to the Nyquist diagram. Moreover, SRG analysis can provide performance bounds in terms of (incremental) L_{2} -gain. Originally, it was applicable for Single-Input-Single-Output (SISO) systems only and has been recently extended to include unstable elements in the loop [ 12 ] . While practical stability analysis is an important outcome of the SRG framework, our aim is to take one step further and establish a frequency domain performance shaping tool for NL systems. Hence, in this paper, we develop novel non-approximative frequency domain tools for Time Invariant (TI) incrementally stable NL systems by leveraging the SRG framework. Compared to [ 13 ] , incremental stability replaces the assumption of a convergent system. The central idea is to compute the incremental gain for an input space of signals with a common period, where the SRG is the central computational tool. By evaluating these gains over a grid of frequencies, one obtains a Bode diagram for the NL system. The NL Bode diagram is used to define the NL bandwidth for the loop transfer and sensitivity, compatible with the LTI definition. Our method can reproduce most aspects of existing work, such as sinusoidal input Bode diagrams. Also, our results are more general since we are able to compute the gain for any frequency and arbitrary higher harmonics, but amplitude-dependent results, compared to, e.g. [ 13 ] , are not yet reflected in our approach. An entirely novel aspect of our work is the gain for signals with subharmonics , which allows low-frequency analysis of sensitivity functions, going beyond sinusoidal inputs. Finally, all our computations are based on modular interconnections of LTI and NL operators, instead of relying on state-space realizations, making the application of our method more accessible to engineers. This paper is structured as follows. In Section II , we present the required preliminaries. In Section III , we derive a method to analyze incrementally stable systems in the frequency domain, leading to our definition of the NL Bode plot. We show how SRGs are used to compute the NL Bode diagrams in Section IV , focusing on the loop transfer and the sensitivity. Finally, we apply our results to a practical design example in Section V and present our conclusions in Section VI . II Preliminaries II-A Notation and Conventions Let \mathbb{R} and \mathbb{C} denote the field real and complex numbers, respectively, with \mathbb{R}_{>0}=(0,\infty) , \mathbb{R}_{\geq 0}=[0,\infty) and \mathbb{C}_{\real>0}=\{a+jb\mid\,a\in\mathbb{R}_{>0},\,b\in\mathbb{R}\} , where j is the imaginary unit. We denote the complex conjugate of z=a+jb\in\mathbb{C} as \bar{z}=a-jb . Let \mathcal{L} denote a Hilbert space, with inner product \innerproduct{\cdot}{\cdot}_{\mathcal{L}}:\mathcal{L}\times\mathcal{L}\to\mathbb{C} and norm \left\lVert x\right\rVert_{\mathcal{L}}:=\sqrt{\innerproduct{x}{x}_{\mathcal{L}}} . For sets A,B\subset\mathbb{C} , the sum and product sets are defined as A+B:=\{a+b\mid a\in A,b\in B\} and AB:=\{ab\mid a\in A,b\in B\} , respectively. A disk in the complex plane is D_{r}(x)=\{z\in\mathbb{C}\mid|z-x|\leq r\} . Denote D_{[\alpha,\beta]} the disk in \mathbb{C} centered on \mathbb{R} which intersects \mathbb{R} in [\alpha,\beta] . The radius of a set \mathcal{C}\subset\mathbb{C} is defined by r_{\mathrm{min}}(\mathcal{C}):=\min_{r>0}:\mathcal{C}\subset D_{r}(0) . II-B Signals, Systems and Stability Denote a field \mathbb{F}\in\{\mathbb{R}^{n},\mathbb{C}^{n}\} , where n=1 is assumed, unless otherwise specified, since this work focuses on SISO continuous-time systems. A Hilbert space of particular interest is L_{2}(\mathbb{T},\mathbb{F}):=\{f:\mathbb{T}\to\mathbb{F}\mid\left\lVert f\right\rVert_{2}<\infty\} , where the norm is induced by the inner product \innerproduct{f}{g}:=\int_{\mathbb{T}}\bar{f}(t)g(t)\differential t , and \bar{f} denotes the complex conjugate, and \mathbb{T}\in\{\mathbb{R}_{\geq 0},[0,T]\} for any T>0 is the domain. For brevity, we denote L_{2}(\mathbb{R}_{\geq 0},\mathbb{F}) as L_{2}(\mathbb{F}) , L_{2}(\mathbb{R}_{\geq 0},\mathbb{R}) as L_{2} and, if the time domain is finite, L_{2}([0,T],\mathbb{R}) as L_{2}([0,T]) . For any T\in\mathbb{R}_{\geq 0} , define the truncation operator P_{T}:L_{2}(\mathbb{F})\to L_{2}(\mathbb{F}) as (P_{T}u)(t):=\begin{cases}u(t)&t\leq T,\\ 0&t>T.\end{cases} (1) The extension of L_{2}(\mathbb{F}) , see Ref. [ 14 ] , is defined as L_{2\mathrm{e}}(\mathbb{F}):=\{u:\mathbb{R}_{\geq 0}\to\mathbb{F}\mid\left\lVert P_{T}u\right\rVert_{2}<\infty\text{ for all }T\in\mathbb{R}_{\geq 0}\}. The space L_{2\mathrm{e}}(\mathbb{R}) , which we denote from now on as L_{2\mathrm{e}} , will be the most frequently used space of signals. Note that the extension is particularly useful since it includes periodic signals, which are otherwise excluded from L_{2} . Periodic signals v\in L_{2\mathrm{e}} can also be viewed as elements in L_{2}([0,T]) , where T is the period of the signal, i.e. v(t)=v(t+T) for all t\in[0,\infty) . Any u\in L_{2}([0,T]) can be written in Fourier series as u(t)=\sum_{k\in\mathbb{Z}}\hat{u}_{k}e^{2\pi jkt/T}, (2) where \hat{u}_{k}\in\mathbb{C} are the Fourier coefficients. One can define the Root-Mean-Square (RMS) norm of the signal as \left\lVert u\right\rVert_{\mathrm{RMS}}=\sqrt{\sum_{k\in\mathbb{Z}}|\hat{u}_{k}|^{2}} , which relates to the L_{2} -norm as \left\lVert u\right\rVert_{L_{2}([0,T])}=\sqrt{T}\left\lVert u\right\rVert_{\mathrm{RMS}} . Systems are modeled as operators R:L_{2\mathrm{e}}\to L_{2\mathrm{e}} . A system is said to be causal if it satisfies P_{T}(Ru)=P_{T}(R(P_{T}u)) , i.e., the output at time t is independent of the signal at times greater than t . Unless specified otherwise, we will always assume causality. Given an operator R on a L_{2} , the induced incremental norm of the operator is defined (similar to the notation in [ 15 ] ) as \Gamma(R):=\sup_{u_{1},u_{2}\in L_{2}}\frac{\left\lVert Ru_{1}-Ru_{2}\right\rVert_{2}}{\left\lVert u_{1}-u_{2}\right\rVert_{2}}. (3) The radius of the SRG corresponds to the incremental induced gain as r_{\mathrm{min}}(\operatorname{SRG}(R))=\Gamma(R) in terms of Eqs. ( 3 ). For causal systems, the induced incremental operator norm on L_{2} carries over to L_{2\mathrm{e}} since \left\lVert P_{T}(R(P_{T}u))\right\rVert_{2}\leq\left\lVert R(P_{T}u)\right\rVert_{2} and P_{T}u\in

Read the complete text