## A Set-Based Methodology for White Noise Modeling (1994)

Citations: | 1 - 0 self |

### BibTeX

@MISC{Paganini94aset-based,

author = {Fernando Paganini},

title = {A Set-Based Methodology for White Noise Modeling},

year = {1994}

}

### OpenURL

### Abstract

This paper provides a new framework for analyzing white noise disturbances in linear systems: rather than the usual stochastic approach, noise signals are described as elements in sets and their effect is analyzed from a worst-case perspective. The paper studies how these sets must be chosen in order to have adequate properties for system response in the worst-case, statistics consistent with the stochastic point of view, and simple descriptions that allow for tractable worst-case analysis. The methodology is demonstrated by considering its implications in two problems: rejection of white noise signals in the presence of system uncertainty, and worst-case system identification. 1 Introduction A general feature of mathematical models in engineering science is the presence of modeling errors, which arise due to poorly understood or highly unpredictable phenomena, or from simplifications deliberately introduced for the sake of model tractability. Essentially two approaches are available ...

### Citations

1493 | Probability inequalities for sums of bounded random variables
- Hoeffding
- 1963
(Show Context)
Citation Context ... of (r x (1); : : : r x (N \Gamma 1)) is defined on a space of increasing dimension, where no global averaging occurs. Our proof relies on a Hoeffding inequality for sums of bounded random variables, =-=[10]-=-: Theorem 7 (Hoeffding) Let z 0 ; : : : z N \Gamma1 be independent random variables, of meansand bounded by asz nsb, definesz = 1 N P N \Gamma1 n=0 z n . Then for ffl ? 0, P (z \Gammas? ffl)se \Gamma2... |

1191 |
System Identification - Theory for the User
- Ljung
- 1993
(Show Context)
Citation Context ...mbination of worst-case and average case analysis. Another example of the difficulty of combining these frameworks is the relation between robust 2 control and mainstream system identification (as in =-=[11]-=-), since the latter relies in the stochastic paradigm for noise. Recent efforts in pursuing this unification in the worst-case setting have once again used a pessimistic view of disturbances, resultin... |

377 |
Extrapolation, Interpolation, and Smoothing of Stationary Time Series
- Wiener
- 1949
(Show Context)
Citation Context ... bounded power signals and bounded energy (l 2 ) signals. 12 4.1 Bounded Power Signals There is a long historical tradition in a non-stochastic theory of white noise, going as far back as Wiener (see =-=[28]-=-), who considered ergodicity properties to build a spectral theory of stationary signals devoid of probability. For disturbance rejection problems, this approach was followed in Zhou et. al. [30], who... |

120 | The complex structured singular value - Packard, Doyle - 1993 |

70 |
S-procedure in nonlinear control theory
- Yakubovich
- 1971
(Show Context)
Citation Context ...ble computational properties (see [3, 15]). We now briefly explain how this result can be obtained from the Integral Quadratic Constraint (IQC) formulation, which originated in the work of Yakubovich =-=[29]-=-, and has recently been applied to this H1 problem by Megretski [14]. 19 Let z be the vector of all the inputs to system G, z = (z 1 ; : : : ; z F ; z F+1 ), where (z 1 ; : : : ; z F ) is the partitio... |

59 |
Chaotic Evolution and Strange Attractors
- Ruelle
- 1989
(Show Context)
Citation Context ...mes form high dimensional fluctuations (e.g. particle agitation). These have been classically modeled as stochastic processes, but could also be interpreted in the context of deterministic chaos (see =-=[23]-=-). In any event, stochastic noise is known to provide a good model, regardless of whether the probability measure is due to chance or is the ergodic measure of a chaotic system. Therefore, a natural r... |

46 |
Spectral Analysis and Its Applications. Holden-Day
- Jenkins, Watts
- 1968
(Show Context)
Citation Context ...y some form of local averaging that reduces the fluctuations (the variance). This smoothing is most commonly done by convolution of the periodogram with a window function; an abundant literature (see =-=[8]-=-) has studied shapes and properties of these windows. 9 0 50 100 150 200 250 0 1 2 3 4 5 6 Figure 2: Periodogram of a pseudorandom sequence In this paper we are interested in defining a set of typical... |

42 |
Stochastic Processes
- Bartlett
- 1966
(Show Context)
Citation Context ... 2 2 = r h (0) = 1 X t=0 h(t) 2 = 1 2 Z 2 0 s h (!)d! (2) For some of the frequency domain bounds obtained in this paper, we will further assume that s h (w) is a function of bounded variation (in BV =-=[0; 2]-=-). This means (see [22]) that TV (s h ) := sup P=fw 1 ;:::;w pg X i js h (! i+1 ) \Gamma s h (! i )j ! 1 (3) where the supremum is over partitions P = fw 1 ; : : : ; w p g of the interval [0; 2]. TV (... |

28 |
Control Oriented System Identification: A worst case/deterministic approach in 3-lm
- Helmicki, Jacobson, et al.
- 1991
(Show Context)
Citation Context ...forts in pursuing this unification in the worst-case setting have once again used a pessimistic view of disturbances, resulting in worst-case identification problems with weak consistency properties (=-=[9, 26]-=-) and high computational complexity ([6, 21]). In this paper we propose a new methodology for white noise modeling, aimed at resolving these difficulties. The starting point is the following question:... |

28 |
Robust stability with time-varying structured uncertainty
- Shamma
- 1994
(Show Context)
Citation Context ...tor on l 2 . For the system in Figure 4, for the case where \Delta is a structured, otherwise arbitrary linear operator on l 2 , the following necessary and sufficient condition for has been obtained =-=[24, 14]-=-: sup \Delta2B \Delta kuk 2 1 kyk 2 ! 1 () 9X 2 X : G(e j! ) 2 4 X 0 0 I 3 5 G(e j! ) \Gamma 2 4 X 0 0 I 3 5 ! 0 (39) In (39), X is the set of positive scaling matrices of the form X = diag[x 1 I q 1 ... |

26 |
Power distribution inequalities in optimization and robustness of uncertain systems
- Megretski, Treil
- 1993
(Show Context)
Citation Context ...tor on l 2 . For the system in Figure 4, for the case where \Delta is a structured, otherwise arbitrary linear operator on l 2 , the following necessary and sufficient condition for has been obtained =-=[24, 14]-=-: sup \Delta2B \Delta kuk 2 1 kyk 2 ! 1 () 9X 2 X : G(e j! ) 2 4 X 0 0 I 3 5 G(e j! ) \Gamma 2 4 X 0 0 I 3 5 ! 0 (39) In (39), X is the set of positive scaling matrices of the form X = diag[x 1 I q 1 ... |

14 | Tsitsiklis, “Optimal asymptotic identification under bounded disturbances
- Tse, Dahleh, et al.
- 1993
(Show Context)
Citation Context ...forts in pursuing this unification in the worst-case setting have once again used a pessimistic view of disturbances, resulting in worst-case identification problems with weak consistency properties (=-=[9, 26]-=-) and high computational complexity ([6, 21]). In this paper we propose a new methodology for white noise modeling, aimed at resolving these difficulties. The starting point is the following question:... |

13 |
On the time complexity of worst-case system identification
- Poolla, Tikku
- 1994
(Show Context)
Citation Context ...rst-case setting have once again used a pessimistic view of disturbances, resulting in worst-case identification problems with weak consistency properties ([9, 26]) and high computational complexity (=-=[6, 21]-=-). In this paper we propose a new methodology for white noise modeling, aimed at resolving these difficulties. The starting point is the following question: how does one decide whether a signal can be... |

11 |
Mixed H 2 and H1 performance objectives I: robust performance analysis
- Zhou, Glover, et al.
- 1994
(Show Context)
Citation Context ...(see [28]), who considered ergodicity properties to build a spectral theory of stationary signals devoid of probability. For disturbance rejection problems, this approach was followed in Zhou et. al. =-=[30]-=-, who considered the class of bounded power signals, defined by BP = ( x(t) : r x () = lim N!1 1 2N + 1 N X t=\GammaN x(t + )x(t) exists for eachs) (19) This class is well motivated since it includes ... |

10 |
Behavioral Approach to Robustness Analysis
- Paganini, D'Andrea, et al.
- 1994
(Show Context)
Citation Context ...erformance analysis and worst-case system identification. We will not attempt to present a full description of these directions in the limited space available here; they have been developed elsewhere =-=[18, 17, 19, 27]-=-, and journal versions are in preparation. In this paper our objective is to provide enough evidence that this methodology has useful implications. 5.1 Robust H 2 Analysis A problem which has received... |

10 |
The Robust H 2 Control Problem: A Worst-Case Design
- Stoorvogel
- 1993
(Show Context)
Citation Context ...s direction). The robust performance question one would really want to address in many practical cases is the effect of white noise over sets of systems (the "Robust H 2 " problem). Many aut=-=hors (see [25, 7]-=- and references therein) have addressed this problem in terms of a direct combination of the worst-case and stochastic frameworks, and have succeeded in obtaining upper bounds for system performance. ... |

7 |
Time Series: Data Analysis and Theory, McGraw-Hill
- Brillinger
- 1981
(Show Context)
Citation Context ... to a flat periodogram (the DFT of a delta-function correlogram) . The "raw" periodogram is typically very erratic, however, as demonstrated in Figure 2. This fact has long been recognized (=-=see, e.g. [2, 5]-=-) in the statistical spectral analysis literature; correspondingly, the standard methods for power spectrum estimation are based on smoothing the periodogram, by some form of local averaging that redu... |

6 |
Set Descriptions of White Noise and Worst Case Induced Norms
- Paganini
- 1993
(Show Context)
Citation Context ... a band centered at the autocorrelations of the filter. Therefore, such a band is a natural set description for colored noise, the output of a linear filter under white noise. 2. It can be shown (see =-=[16]-=-) that if fl ! 1 T , then for large enough N the upper bound in (12) is achieved. This is no longer true for large values of fl; for example, if fl = 1, there are no restrictions on the input signal, ... |

3 |
Theodosopoulos T.V., Tsitsiklis J., The sample complexity of worst-case identification of FIR linear systems
- Dahleh
- 1993
(Show Context)
Citation Context ...rst-case setting have once again used a pessimistic view of disturbances, resulting in worst-case identification problems with weak consistency properties ([9, 26]) and high computational complexity (=-=[6, 21]-=-). In this paper we propose a new methodology for white noise modeling, aimed at resolving these difficulties. The starting point is the following question: how does one decide whether a signal can be... |

3 |
S-Procedure in Optimal Non-Stochastic Filtering
- Megretski
- 1992
(Show Context)
Citation Context ... fl;T is described by a finite number of constraints flr u (0) \Sigma r u ()s0s= 1; : : : ; T (41) 20 which are quadratic on the signal u. In other words, they are IQCs (this was already suggested in =-=[13]-=-) corresponding to the quadratic forms oe +s(u) = flr u (0) + r u ( ), oe \Gammas(u) = flr u (0) \Gamma r u ( ). Using the same arguments as in the case of H1 , robust performance analysis over W fl;T... |

3 |
Necessary and Sufficient Conditions for Robust H 2 Performance
- Paganini
- 1995
(Show Context)
Citation Context ...erformance analysis and worst-case system identification. We will not attempt to present a full description of these directions in the limited space available here; they have been developed elsewhere =-=[18, 17, 19, 27]-=-, and journal versions are in preparation. In this paper our objective is to provide enough evidence that this methodology has useful implications. 5.1 Robust H 2 Analysis A problem which has received... |

2 |
Distribution of the circular serial coefficient for residuals from a fitted Fourier series
- Anderson, Anderson
- 1956
(Show Context)
Citation Context ...al for the stochastic results. Proof of Theorem 3: Part 1: For the case of a fixed time lags, the distribution of the autocorrelation r x () has been extensively studied in the statistical literature =-=[2, 1]-=-; exact expressions for the distribution of r x ()=r x (0) when x(t) is Gaussian are obtained in [1], and it follows that p N rx () rx(0) is asymptotically 26 normal N (0; 1). Since fl p N !1, and T i... |

2 |
Analysis of Robust H 2 Performance with Multipliers
- Feron
(Show Context)
Citation Context ...s direction). The robust performance question one would really want to address in many practical cases is the effect of white noise over sets of systems (the "Robust H 2 " problem). Many aut=-=hors (see [25, 7]-=- and references therein) have addressed this problem in terms of a direct combination of the worst-case and stochastic frameworks, and have succeeded in obtaining upper bounds for system performance. ... |

2 |
A counterexample in Power Signals Space, submitted to
- Mari
- 1994
(Show Context)
Citation Context ... non-stochastic framework for white noise signals, it is sometimes inconvenient due to its little mathematical structure. In particular, it is not a vector space (not being closed under addition, see =-=[12]-=-), so it is not a seminormed space. For this reason we will now consider white noise descriptions inside l 2 , which has the structure of a Hilbert space. At first, this seems unnatural, since white n... |

1 | Analysis of Implicit Uncertain Systems
- Paganini, Doyle
- 1994
(Show Context)
Citation Context ...erformance analysis and worst-case system identification. We will not attempt to present a full description of these directions in the limited space available here; they have been developed elsewhere =-=[18, 17, 19, 27]-=-, and journal versions are in preparation. In this paper our objective is to provide enough evidence that this methodology has useful implications. 5.1 Robust H 2 Analysis A problem which has received... |

1 |
Stoorvogel A.A., Mixed H 2 =H1 Control in a Stochastic Framework
- Peters
- 1994
(Show Context)
Citation Context ...the classical H 2 (LQG) problem. The main limitation to its applicability is that noise is rarely the prevailing source of uncertainty, and the others do not fit easily into a stochastic description (=-=[20] contains -=-some work in this direction). The robust performance question one would really want to address in many practical cases is the effect of white noise over sets of systems (the "Robust H 2 " pr... |

1 |
Classical system identification in a deterministic setting
- Venkatesh, Dahleh
- 1995
(Show Context)
Citation Context |