MIT OpenCourseWare http://ocw.mit.edu 2.161 Signal Processing: Continuous and Discrete Fall 2008 For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.���� � � � � ���� MASSACHUSETTS INSTITUTE OF TECHNOLOGY DEPARTMENT OF MECHANICAL ENGINEERING 2.161 Signal Processing – Continuous and Discrete Convolution1 1 Convolution Consider a linear continuous-time LTI system with input u(t), and response y(t), as shown in Fig. 1. We assume that the system is initially at rest, that is all initial conditions are zero at time t =0, and examine the time-domain forced response y(t) to a continuous input waveform u(t). ����������������� ���� Figure 1: A linear system. In Fig. 2 an arbitrary continuous input function u(t) has been approximated by a staircase function ˜uT (t) ≈ u(t), consisting of a series of piecewise constant (zero order) sections each of an ������������ � � �� �� ���� ���� ������ � ���� ������������ ����� � � �� �� ���� ���� ������ � Figure 2: Staircase approximation to a continuous input function u(t). arbitrary fixed duration, T , where u˜T (t)= u(nT ) for nT ≤ t< (n +1)T (1) for all n. It can be seen from Fig. 2 that as the interval T is reduced, the approximation becomes more exact, and in the limit u(t) = lim u˜T (t). T →0 The staircase approximation ˜uT (t) may be considered to be a sum of non-overlapping delayed pulses pn(t), each with duration T but with a different amplitude u(nT ): ∞ D. Rowell, September 8, 2008 � u˜T (t)= pn(t) (2) n=−∞ where � u(nT ) nT ≤ t< (n +1)T pn(t)= (3)0 otherwise 1 1� 0 0 ) T t � � � n T ( n + 1 Each component pulse pn(t) may be written in terms of a delayed unit pulse δT (t), of width T and amplitude 1/T that is: pn(t)= u(nT )δT (t − nT )T (4) so that Eq. (2) may be written: ∞ u˜T (t)= u(nT )δT (t − nT )T. (5) n=−∞ We now assume that the system response to an input δT (t) is a known function, and is designated hT (t) as shown in Fig. 3. Then if the system is linear and time-invariant, the response to a delayed unit pulse, occurring at time nT is simply a delayed version of the pulse response: yn(t)= hT (t − nT ) (6) @ T ( t - n T ) y n ( t ) 1 / T @ T ( t - n T ) s y s t e m y n ( t ) n T ( n + 1 ) T t 00 Figure 3: System response to a unit pulse of duration T . The principle of superposition allows the total system response to ˜uT (t) to be written as the sum of the responses to all of the component weighted pulses in Eq. (5): ∞ y˜T (t)= u(nT )hT (t − nT )T (7) n=−∞ as shown in Fig. 4. � � �� �� ���� ���� � ������������������������� � � �� �� ���� ����� � ���������������� ���� �� Figure 4: System response to individual pulses in the staircase approximation to u(t). For causal systems the pulse response hT (t) is zero for time t< 0, and future components of the input do not contribute to the sum, so that the upper limit of the summation may be rewritten: N y˜T (t)= u(nT )hT (t − nT )T for NT ≤ t< (N +1)T. (8) n=−∞ 2� � Equation (8) expresses the system response to the staircase approximation of the input in terms of the system pulse response hT (t). If we now let the pulse width T become very small, and write nT = τ , T = dτ, and note that limT 0 δT (t) = δ(t), the summation becomes an integral:→ N T 0→ n=−∞ t y(t) = lim u(nT )hT (t − nT )T (9) u(τ )h(t − τ )dτ (10) = −∞ where h(t) is defined to be the system impulse response, h(t) = lim hT (t). (11)T 0→ Equation (10) is an important integral in the study of linear systems and is known as the convolution or superposition integral. It states that the system is entirely characterized by its response to an impulse function δ(t), in the sense that the forced response to any arbitrary input u(t) may be computed from knowledge of the impulse response alone. The convolution operation is often written using the symbol ⊗: � t y(t) = u(t) ⊗ h(t) = u(τ)h(t − τ)dτ. (12) −∞ Equation (12) is in the form of a linear operator, in that it transforms, or maps, an input function to an output function through a linear operation. The form of the integral in Eq. (10) is difficult to interpret because it contains the term h(t−τ) in which the variable of integration has been negated. The steps implicitly involved in computing the convolution integral may be demonstrated graphically as in Fig. 5, in which the impulse response h(τ) is reflected about the origin to create h(−τ), and then shifted to the right by t to form h(t − τ). The product u(t)h(t − τ ) is then evaluated and integrated to find the response. This graphical representation is useful for defining the limits necessary in the integration. For example, since for a physical system the impulse response h(t) is zero for all t < 0, the reflected and shifted impulse response h(t − τ ) will be zero for all time τ > t. The upper limit in the integral is then at most t. If in addition the input u(t) is time limited, that is u(t) ≡ 0 for t < t1 and t > t2, the limits are: yf (t) = ⎧⎪⎪⎪⎨ ⎪⎪⎪⎩ � t u(τ )h(t − τ )dτ for t < t2 t1 (13) � t2 u(τ)h(t − τ)dτ for t ≥ t2 t1 Example A simple RC first-order filter, shown in Fig. 6, is subjected to a very short unit impulsive voltage of duration ΔT = 0.001 seconds and magnitude 10 volts, and is observed to respond with a output vo(t) = 0.03e−3t . Find the response of the filter to a ramp in applied voltage V (t) = t for t > 0. Solution: The product of the impulsive force and its duration V ΔT = 0.01, and because of its brief duration, the pulse may be considered to be an impulse of strength 3multiplicationtttth(t)0time reversaltimeshiftingh(t -t)0th(-t)System impulse responseSystem inputu(t)0u(t)h(t-t)0tintegrationSystem responset0y(t)t1t111response at time t is defined by thearea under the curve.1Figure 5: Graphical demonstration of the convolution integral. 400 . 0 30 1 20 . 0 1 5t i m e (