OUTLINE • Function spaces. • Input-output stability. • LTI Systems. • Small gain theorem. • Circle criterion. INPUT-OUTPUT STABILITY 1 M. SAMI FADALI EBME DEPT., UNR 2 FUNCTION SPACES _ SPACE • _ Space: Space of all piecewise continuous • Vector spaces with functions as “vectors” satisfying functions : Space of all • _ Spaces piecewise continuous functions satisfying _ ⁄ _ _ ∈ Finite _ norm of 3 • 4 EXTENDED SPACES SPACES _ • _2 Space: Space of all piecewise continuous satisfying functions . ∈ Truncation operator ⁄ : → , 0, _ Example • _1 Space: Space of all piecewise continuous satisfying functions . : 0 → , 0, , & ∈ _ 5 6 PROPERTIES OF INEQUALITY Extended Space : space of all functions whose truncations belong to the parent space . : Note → . . ∈ If , , ,∀ ∈ is not a normed space even though 1 1, ∈_ , is. 1 ∞ ∈_ then ∈_ • Linear operator I. II. 1 ,∀ , ,∀ ∈ _ ∈ ⁄ ,∀ ∈ 7 ⁄ 8 ASSUME EXAMPLE SATISFIES _1 Space of real-valued functions normed linear vector space of piecewise continuous functions . : → with norm . • is closed under the family of projections , i.e. if . ∈ ,then ∈ ,∀ ∈ , lim • . : → . . , • If 0 0, → • ∞ _1 non-decreasing function of ∈ ,i.e. . ∈ , ∈ ⇒ ∈ ,then ∈ iff lim ∞ use to verify _1 ∈ lim → → ∈ • _ Spaces satisfy the above assumptions. /2 ,∀ ∈ ∞ and ∉ 9 CAUSALITY 10 CAUSALITY AND PROJECTION • A mapping representing a physical system → satisfying : . . ,∀ ∈ ,∀ ∈ Present outputs do not depend on future inputs Explanation: perform two experiments 1. Apply an arbitrary input to obtain an output then truncate it. 2. Apply the truncated input to obtain an output then truncate it to get The outputs are the same for all physical systems but not for all mathematical functions. 11 • Express causality in terms of a projection operator • Causal operator , ∀ ∈ ,∀ ∈ • Property expressed in terms of the input and output and we do not need the notion of a state. • Assume the system is relaxed: zero initial conditions and the output is due to the input only. 12 INPUT-OUTPUT STABILITY GAIN • A system is input-output -stable if the output is in inputs in has a finite gain if and such that = bias term which allows including for systems where • If for the gain is Stability depends on the function space If the function space is obvious we may say that is input-output stable. ∀ ∈ ∀ ∈ , 13 EXAMPLE • LTI SYSTEMS _ static nonlinearity (Figure, page 163) for the gain is _ ∀ ∈_ ∀ ∈ , 14 • LTI system is input output stable if its impulse response is in the form _ Finite gain implies input-output stability but the converse is not true: counterexamples in Figure _ _ 15 _ • True for all the poles of the transfer function are in the open LHP 16 CLOSED-LOOP INPUT-OUTPUT STABILITY _2 GAIN • Space of square-integrable functions _ (excludes some important functions, e.g. sinusoids) _ ∀ ∈_ ∀ ∈ , • Feedback system: feedback interconnection of systems that satisfies the assumptions for all pairs of inputs _ ⁄ _ • Using Parseval’s Theorem, we can show that ∀ 17 VARIABLES 18 SMALL GAIN THEOREM input functions including commands, disturbances and noise. • output functions error signals • • Assume: no. of outputs of = no. inputs of • • Consider • If , then the feedback system is input-output stable. 19 20 PROOF • Assume ERROR for simplicity and show that for all pairs of inputs Eliminate (or ) and assume After truncation 21 OUTPUT 22 COMMENTS • The theorem does not address existence and uniqueness and stability is investigated separately assuming existence and uniqueness. So the output is also bounded if the error is bounded. 23 • The theorem provides a sufficient condition only and can be quite conservative. 24 EXAMPLE • LTI system with transfer function • saturation nonlinearity with slope APPLY SMALL GAIN THEOREM ∀ ∀ System is stable even if this condition is violated (sufficient condition only) ∀ 25 EXAMPLE • 26 SECTOR BOUND NONLINEARITY LTI system transfer function • constant gain • Small gain theorem gives the stability condition , i.e. • Closed-loop transfer function i belongs to sector • Closed-loop stability for • Small gain theorem is very conservative. 27 28 ABSOLUTE STABILITY • What are the conditions for the stability of a strictly stable linear subsystem and a nonilinearity belonging to sector ? is replaced by • Aizermann’s Conjecture: If a constant gain with the stable range of a subset of , then the closed-loop system with nonlinearity is asymptotically stable: FALSE: gives an overly optimistic estimate TYPES OF NONLINEARITIES 1. Single-valued and time-invariant : unique output for each input 2. With memory (e.g. hysteresis, backlash): output depends on the history of the input 3. General nonlinearities : time-varying and possibly with memory (e.g. hysteresis) 29 30 POPOV’S METHOD POPOV’S THEOREM Assume: open-loop stable Sector bound nonlinearity For any initial conditions, the system output is bounded and tends to zero as if the polar plot of , ∗ arbitrarily small for with imaginary axis poles for poles in the open LHP. • Use the modified frequency response ∗ 31 lies entirely to the right of the Popov line Slope , Intercept , and (depending on the nonlinearity) : if , if : and : and 32 EXAMPLE: POPOV CRITERION CIRCLE CRITERION 2 , • Sector bound nonlinearity 2 • For any initial conditions, the system output is if bounded and tends to zero as 2 2 , 2 4 4 Asymptote: vertical line with real axis intercept at 0.5 1⁄0.5 2 The closed-loop system is asymptotically stable with general nonlinearity in the sector 0,2 ∗ 2 2 4 Circle Criterion 0 Imaginary Axis -2 33 EXAMPLE: CIRCLE CRITERION 2 2 2 , 2 2 2 Circle Criterion 0 -2 Imaginary Axis -6 -8 34 CIRCLE CRITERION 2 4 4 Asymptote: vertical line with real axis intercept at 0.5 1⁄0.5 2 The closed-loop system is asymptotically stable with general nonlinearity in the sector 0,2 (same as Popov) -4 -6 -8 -4 35 • Gives conservative results. • Can use the Hurwitz bounds to get an optimistic stability assessment. • The actual stable sector will be somewhere between the conservative estimate of the circle criterion and the optimistic estimate of the Hurwitz bounds. 36