I nternat. J. Math. & Math. Sci. Vol. 3 No. 3 (1980) 535-547 535 STEADY STATE RESPONSE OF A NONLINEAR SYSTEM SUDHANGSHU B. KARMAKAR Western Electric Company Whippany, New Jersey 07981 U.S.A. (Received November i, 1979) ABSTRACT. This paper presents a method of the determination of the steady state response for a class of nonlinear systems. The response of a nonlinear system to a given input is first obtained in the form of a series solution in the multidimensional frequency domain. series solution will converge. Conditions are then determined for which this The conversion from multidimensions to a single dimension is then made by the method of association of variables, and thus an equivalent linear model of the nonlinear system is obtained. The steady state response is then found by any technique employed with linear system. KEYWORDS AND PHRASES. Nonlinear Assoon of vabl, Multidimensional frequency domain, transfer function. 1980 MATHEMATICS SUBJECT CLASSIFICATION CODES: i. 34-00, 34A45. INTRODUCTION. This paper considers a nonlinear system described by the equation of the form S.B. KARMAKAR 536 a3y3 + a5Y5 Ly + x(t) (i.i) L is a linear operator defined as L(s) a 3 a5; b n are constants, the response to it. bnsn + bn-lS x(t) n-i + + bo is the input to the nonlinear system and Many physical systems encountered in practice can be to above kind of equation. y(t) is reiated We shall investigate the conditions under which a bounded input to the system will produce a bounded output. This will lead to the stability conditions of the above type of nonlinear system. 2. ANALYSIS OF A NONLINEAR SYSTEM Equation (I.i) can be viewed as follows: Volterra [i] has shown that the response y(t) can be represented as some functional of the input x(t). The following functional expansion was suggested: I y(t) hl(t-l)X(l)dl + I[h2(t-l,t-2)X(Tl)X(2)didT2+. We assume a causal system, i.e., x(t) are [0,]. 0 for t < 0. (2.1) The limits of integrations Let Yn(t) I- -I hn(t-wl...t-Wn)X(Wl)X(W2)’’-X(n)dWld2--’dw n n-fold Then equation (2.1) can be written as y(t) hl(t- I) Z Yn (t) is the ist order Volterra Kernel and h Volterra Kernel. n _(t-l...t-w n (2.2) is the nth order 537 STEADY STATE RESPONSE OF A NONLINEAR SYSTEM 3. DETERMINATION OF THE VOLTERRA KERNELS By applying the technique or reversion of algebraic series [2] from (i.i) we get Y L-Ix- a3L-i( L-ix) 3 + L-I {3a32-i L (L-ix) 5 a5(L-ix) 5 + } (3.1) It is clearly understood that x and y are both functions of time t. Let us define L-lx where h(t) I h(t-Y)x()d (3.2) is the impulse response of the linear part of the system. Hence from (2.2) Yl ] Y3 -a3L-i h(t-w)x(Y)d[ -a3L -a -i (I 1 h(t-)x(w)d Il h(t-l)h(t-2)h(t-3 )x( Yl)X( 2)x( 3 didw2dr3 h(t-u)h(U-Tl)h(u-T2)h(u-T3)X(Tl)X(X2)x(3)dXldT2dx3du 3 Similarly )x( 1 )’" .x( -a5 I’ ’I 6 5)dl" .d 5dudv h(t-u)h(U-Tl)’’’h(u-x5)x(T1)’’’x(T5)dTl’’’dT5du" S.B. KARMAKAR 538 Hence by comparing (2.1) and (2.2) hl(t-T) h(t-T) h2(t--Tl,t--T2) 0 I h3 (t-l’t-2’t-3) =-a3 h(t-u)h(u-l)h(u-2)h(u-3)du h4(t-l...t-T 4) 0 h5(t-T1--.t-Ts) 3a I h(t-u)h(v-u)h(U-l)...h(u-w5)dudv a5 h(t-u)h(u-l)’’’h(u-5)du" Higher order kernels can thus be determined by expanding further the equation 4. (3.1). CONVERGENCY OF THE FUNCTIONAL EXPANSION Let maxlx(t) <_ X 0<t< HX where Ih(t)lct Y3 <- H < la31i]h(t-l)l Ix(m)l IIh(t-)llx(2)2 IIh(t- 3) Ix()1:3 IIh(t-u)Iu. or 4 Y3 < la31(Ilh(t)Idt), X3= la31H4X 3 539 STEADY STATE RESPONSE OF A NONLINEAR SYSTEM Similarly Let Hence Again by applying the technique of reversion of series that (4.1) is the solution of Y 3 it is easy to show (4.2) HX + H(Ia31Y3+Ia51Y 5) (4.2) la31 y3 la51 y5 (4.3) and X i Y . dY XI and Y YI corresponding to the value dX 0 < Y ! YI" Therefore in this above region a bounded output. YI can be determined from Let X 4 2 5HIasIYI + 3HIa31YI i 0 dY or + I then input will produce a bounded corresponding to -3Hla31 Hence if 0 < X < X 9H2a +20Hla51 S.B. KARMAKAR 540 considering only the positive value of YI’ YI YI is found from From (4.3) Therefore for x(t) < X1, the functional expansion (2.1) will be convergent. REMARKS: (1.1) With the assumed solution of equation hn(T 1 y(t) T 2 .Tn)X(t--TI) ..x(t--T n )d 1 ..d n n=l Y= cXnn n=l where I...IIh(T1 cn And for the time invariant system, C radius of convergence series (2.1) X1, n Tn) IdTl T2 d n is independent of time [3] which has the which in turn implies the convergence of the Volterra for maIx(t)l < x 0<t < REVIEW OF MULTIDIMENSIONAL TRANSFORM. 5. As an analogy to the linear system, the nth order transform of a nonlinear system [4] is defined to be H n (Sls 2" ..s n ): |...lh(l e . )exp[-SlXl+S2X2+...+s n T n n )]dx...dx n (5.1) 541 ’STEADY STATE RESPONSE OF A NONLINEAR SYSTEM Hn (sl’s2"’’sn) One method of determining is the harmonic input method [5]. When the system described by (i.i) is excited with a set of n unit amplitude exponentials at the noncommensurate frequencies s exp(slt) x(t) + s2...s n I exp(s2t) (5.2) + exp(s t) + n the output will contain exponential components of the form y(t exp m s ml !m2 n=l M I n l+m2 s 2 + +mn n (5.3) s while M under the summation indicates that for each n the sum is to be taken over all distinct values of m Thus to determine from (5.2) and (5.3), n < m where i < m_ < m_ +/- Hn(Sl,S 2...s n) n such that m_ + m_ + ...m +/- we substitute the values of x(t) and n n. y(t) respectively, in the system equation (I.i) and equate the coefficients of n! exp(sl+s2 +. .+Sn)t By substituting the values of y(t) and x(t) from (5.3) and (5.2), respectively, in (I.i) the first three nonzero transfer functions are found to be- Hl(S I) 1 (5 ) L(Sl -a 3 H3(Sl,S2,S 3) L(sI+s2+s3)L(Sl)L(s2)L(s3) a H5(Sl..-s 5) (5.5) 3 L(sI+s2+... s5)L(sI+s2+s 3)L(s 4)L(s 5) + a> L(sl+s2+...s5)L(sl)L(s2)...L(s5) (5.6) S.B. KARMAKAR 542 Let A mean associate variables Y(s) Then Sl,S 2.../s. HI(S)X(s) +A{H3(Sl,S2,s3)X(Sl)X(s2)X(s3)} +A{H5(Sl...s5)X(Sl)...X(s5)} + Now by applying the Final Value Theorem we get y(t) lira nonzero terms of 6. s+o (2.1) Subject to the convergency of sY(s) lira t (5.7) to represent it is often enough to consider the first two the system in the frequency domain. EXAMPLE. We consider a nonlinear system of the form (i.i) whose linear transfer function is given by i i H(s) L(s) s e s e n n +2 +o Let its impulse response be h(t). It is easy to see for series expansion (2.1) < s 2 +2 n 2 +n H is unbounded. > i it is easy to evaluate If 0 will have real roots. is a little more involved. < I h(t) - n If 0 < (n-i2 )t Let n’ 8 since < i the evaluation of H t sin Ilh(t)Idt If H is bounded, convergency of the Volterra series can further be investigated. For 0 < Hence the Volterra will be divergent, and the system will not have any bounded steady state response. 2 0,11h(t)Idt nl- 2 [7] 543 STEADY STATE RESPONSE OF A NONLINEAR SYSTEM henc e e h(t) -at sin 8t We consider the integral e- at Is in 8tldt H’ Now e ax sin e pxdx ax (a sin px-p cos 2 2 a a 2 +6 2 etc. lie I {e-m+e-2ml {e- 2m+ e- I o H +p ,m sin 8t < 0 in the intervals px) +e -m 3m + + ] m e +2e-m+2e-2m+ -e =2 2 a +6 [2 6 2 Now -coth(-u)= +6 coth u 2 o n- a2 + i l+e l-e -2u hence 13 H’ Therefore, H -m i -e 2 n +6 2 coth() coth C 2 aw 6 Henc e + S.B. KARM 544 As a numerical example we consider the following system" + 3y + tanh y + A cos mt By expanding tanh y and by retaining the first three terms we get + 4y + y + Here The above approximation is quite valid if y < i. i =--4 (6.2) A cos mt y n =2 Hence 2 -3x0.64984 Y1 9x(O lOO. +0.94432 XI x_+ hence 0.9717658 0.64984 i --- 64984)2 64984 0.9717658 YI (0"9717658) 3 Hence the system will have a steady response for lim s+0 s i As _2 15 + 2oo. 6498 5 (0"9717658)5 1.49539- 0.305888- 0.1155438 the linear part 1 2 1.07395 IA cos 0t < 1.07395. Response of 0. It is not necessary to determine the +s+4 s 2-----+co response of the nonlinear parts because being convergent, the other terms of the Volterra expansion can not exceed in absolute value of the linear term. the system will be stable for Ix(t)l < 11.07395 cos t I. If x(t) Hence A, then for A < 1.07309, the approximate value of the steady state response is given by 545 STEADY STATE RESPONSE OF A NONLINEAR SYSTEM lira y(t) Yl(t) t Sl/0 Y3(t) + lira t+ Y3 (t) Sl lira lira In order to determine lira t+ Yl (t) lira t t+ s 2 I 2nS I + m2n + A A i n Y3(Sl,S2,S3) we first obtain YB(s,se,s3) HB(s,se,sB)X(s)X(se)X(s3) From (5.5) 3 A3 L(sI+s2+s3)L(Sl)L(s2)L(s3 SlS2S 3 a Y3(Sl,S2,S3) Where 2 n 2 s’ + 2a s + n L(s) Now by the technique of association of variables [7,8,9] Y3(Sl,S2,S3 is translated in one frequency domain, thus we obtain Y3(s) i A I form 2 A2 i s +n ( 2+2s nS+n2) I A --are functions of 8 (s,,) n 2+2o2+oo2) n A1 but do not contain any factor of the Hence lira t+ Y3(t)=- lira s+o ( + lim s s+o sA3a3 s 2 +2n+n2) 2 + A8 i 4nS2 n +_ s+ n2) 546 S.B. KARMAKAR Jim t+ a3A38-- Y3(t) + 0 n The steady state response of the system is approximately given by Yss For =4, Yss If x(t) 6(t). i i 3.64 - 1 + 76-- Again by associating the variables lim Yss Since " i + A i i’ 2’ 3 s+o s S 2 2 +2 n s +n + lims+o l + we get B2 d nt cntain any factr f the ACKNOWLEDGMENT: [8] i frm-’s The author gratefully acknowledges the assistance offered to him by Prof. J. F. Barrett, Department of Mathematics, Eindhoven University of Technology, The Netherlands. REFERENCES i. Volterra, V., Theory of Functionals and of Integral and Integro Differential Equations, Dover Publications, New York, 1959. 2. Hodgman, C. D., Standard Mathematical Tables, Chemical Rubber Publishing Company, Cleveland, Ohio, 1959. 3. Barrett, J. F., The series-reversion method for solving forced nonlinear differential equations, Math. Balkanica, Vol. 4.9, 1974, pp. 43-60. STEADY STATE RESPONSE OF A NONLINEAR SYSTEM 547 4. Barrett, J. F., The Use of Functionals in the Analysis of Nonlinear Physical Systems, J. Electronics and Control 1__5, 567-615, 1963. 5. Kuo, Y. L., Frequency-Domain Analysis of Weakly Nonlinear Networks, Circuits and Systems, (IEEE) Vol. ii, No. 4, 2-8, August 1977. 6. Bedrosian, E., Rice, S. 0., Output Properties of Volterra Systems (Nonlinear Systems with Memory) Driven by Harmonic and Gaussian Inputs, Proc. IEEE 5_9 1688-1707, December 1971. 7. Dorf, R. C., Modern Control Systems, Addison-Wesley Publishing Company, Reading, Massachusetts, 197. 8. George, D. A., Continuous Nonlinear Systems MIT Research Laboratory of Electronics, Technical Report 355, July 24, 1959. 9. Lubbock, J. K., Bansal, V. S., Multidimensional Laplace transforms for solution of nonlinear equation, Proc. IEEE, Vol. 116 (12) 2075-2082, December 1969. I0. Koh, E. L., Association of variables in n-dimensional Laplace transform, Int. J. Systems Sci., Vol. 6 (2).