Massachusetts Institute of Technology Department of Electrical Engineering and Computer Science 6.341:DISCRETE-TIME SIGNAL PROCESSING OpenCourse Ware 2006 Lecture 4 DT Processing of CT Signals CT Processing of DT Signals:Fractional Delay Reading:Sections 4.1-4.5 in Oppenheim,Schafer Buck(OSB). A typical discrete-time system used to process a continuous-time signal is shown in OSB Figure 4.15.T is the sampling/reconstruction period for the C/D and the D/C converters. When the input signal is bandlimited,the effective system in Figure 4.15(b)is equivalent to the continuous-time system shown in Figure 4.15(a). Ideal C/D Converter The relationship between the input and output signals in an ideal C/D converter as depicted in OSB Figure 4.1 is: Time Domain: xin]xe(nT) Frequency Domain: X(e“)=是∑2-X0(学-钟)》， where X(ej)and Xe(jn)are the DTFT and CTFT of x[n],ze(t).These relationships are illustrated in the figures below.See Section 4.2 of OSB for their derivations and more detailed descriptions of C/D converters. x(n) 012 Time/Frequency domain representation of C/D conversion
Massachusetts Institute of Technology Department of Electrical Engineering and Computer Science 6.341: Discrete-Time Signal Processing OpenCourseWare 2006 Lecture 4 DT Processing of CT Signals & CT Processing of DT Signals: Fractional Delay Reading: Sections 4.1 - 4.5 in Oppenheim, Schafer & Buck (OSB). A typical discrete-time system used to process a continuous-time signal is shown in OSB Figure 4.15. T is the sampling/reconstruction period for the C/D and the D/C converters. When the input signal is bandlimited, the effective system in Figure 4.15 (b) is equivalent to the continuous-time system shown in Figure 4.15 (a). Ideal C/D Converter The relationship between the input and output signals in an ideal C/D converter as depicted in OSB Figure 4.1 is: Time Domain: x[n] = xc(nT) Frequency Domain: X(ejω) = 1 T �∞ k=−∞ Xc(j( ω T − 2πk T )), where X(ejω) and Xc(jΩ) are the DTFT and CTFT of x[n], xc(t). These relationships are illustrated in the figures below. See Section 4.2 of OSB for their derivations and more detailed descriptions of C/D converters. Time/Frequency domain representation of C/D conversion 1
In the frequency domain,a C/D converter can be thought of in three stages:periodic repetition of the spectrum at 2n,normalization of frequency by and scaling of the amplitude by The CT frequency is related to the DT frequency w by =If xe(t)is not bandlimited to<maz<,repetition at 2r will cause aliasing.Note that the C/D converter here is an idealized sampler,representing the mathematical operations of sampling with a periodic impulse train followed by conversion from impulse train to a discrete-time sequence.It does not represent physical sampling circuits.In practice,the ideal C/D converter is approximated by analog-to-digital(A/D)converters,which will also quantize the input signal to a finite number of amplitude levels.OSB figure 4.2 illustrates C/D conversion of the same signal at two different sampling rates. Ideal D/C Converter An ideal D/C converter is depicted in OSB Figure 4.10.As discussed in Section 4.3 of OSB, the input/output relationship of this system is Time Domain: x(t)= ∑xnsinc((-n平) ∫TX(eu) D≤开 Frequency Domain: X(j2) 10 otherwise A D/C reconstructs the continuous signal by bandlimited interpolation.In the frequency do- main,this is equivalent to ideal low-pass filtering of the scaled spectrum.That is X,(j2)=H,(j)X(er), where ∫T≤票 ()=0otherwise. See OSB Figure 4.8 for the block diagram of such an ideal bandlimited filtering system,and OSB Figure 4.9 for the time-domain representation of bandlimited interpolation.Note that x[n]=x(nT)is not a definition of a D/C converter,because it does not specify the values of xr(t)for time instances t nT. In the frequency domain,a D/C can be interpreted as carrying out three operations in reverse of C/D conversion:bandlimiting to the base period-号≤w≤号=，normalization of frequency by T,and scaling of the amplitude by T.Similar to C/D converters,a D/C converter is also ideal,representing the mathematical operations of conversion from sequence to an impulse train,followed by idealized reconstruction. DT Processing of CT Signals OSB Figure 4.15 shows the cascade of a C/D converter,a discrete-time system,followed by a D/C converter.In this system,if ce(t)is bandlimited,i.e.Xe(js)=0 for <the overall 2
� In the frequency domain, a C/D converter can be thought of in three stages: periodic 1 repetition of the spectrum at 2π, normalization of frequency by T , and scaling of the amplitude by 1 . The CT frequency Ω is related to the DT frequency ω by Ω = ω . If xc(t) is not T T bandlimited to −Ωs < Ωmax < Ω 2 s 2 , repetition at 2π will cause aliasing. Note that the C/D converter here is an idealized sampler, representing the mathematical operations of sampling with a periodic impulse train followed by conversion from impulse train to a discrete-time sequence. It does not represent physical sampling circuits. In practice, the ideal C/D converter is approximated by analog-to-digital (A/D) converters, which will also quantize the input signal to a finite number of amplitude levels. OSB figure 4.2 illustrates C/D conversion of the same signal at two different sampling rates. Ideal D/C Converter An ideal D/C converter is depicted in OSB Figure 4.10. As discussed in Section 4.3 of OSB, the input/output relationship of this system is Time Domain: xr(t) = � n x[n]sinc�t−nT � T π � T X(ejω Frequency Domain: X ) Ω ≤ T r(jΩ) = 0 otherwise A D/C reconstructs the continuous signal by bandlimited interpolation. In the frequency domain, this is equivalent to ideal low-pass filtering of the scaled spectrum. That is Xr(jΩ) = Hr(jΩ)X(ejΩT ) , where π H T r(jΩ) = � T |Ω| ≤ 0 otherwise . See OSB Figure 4.8 for the block diagram of such an ideal bandlimited filtering system, and OSB Figure 4.9 for the time-domain representation of bandlimited interpolation. Note that x[n] = xr(nT) is not a definition of a D/C converter, because it does not specify the values of xr(t) for time instances t = nT. In the frequency domain, a D/C can be interpreted as carrying out three operations in π reverse of C/D conversion: bandlimiting to the base period −Ω 2 r ≤ ω ≤ Ω 2 r = T , normalization of frequency by T, and scaling of the amplitude by T. Similar to C/D converters, a D/C converter is also ideal, representing the mathematical operations of conversion from sequence to an impulse train, followed by idealized reconstruction. DT Processing of CT Signals OSB Figure 4.15 shows the cascade of a C/D converter, a discrete-time system, followed by a π D/C converter. In this system, if xc(t) is bandlimited, i.e. Xc(jΩ) = 0 for |Ω| ≤ T , the overall 2
system is linear and time invariant,and equivalent to a continuous time LTI system: Aef6n)=HGn)ior1m≤7 For example,Figure 4.13 shows the Fourier transform of signals through the overall system when the input is bandlimited.The discrete-time system in this case is an ideal lowpass filter with cutoff frequency we. Example: He(jw):Lowpass filter,cutoff frequency =/4 1/T:20kHz Hesf(j)Lowpass filter,cutoff frequency = CT Processing of DT Signals:Fractional Delay In addition to using discrete-time systems for processing continuous-time signals,the com- plementary situation is also possible,where a continuous time system is preceded by a D/C converter and followed by a C/D converter to process discrete-time signals,as depicted in OSB Figure 4.16.Section 4.5 of OSB analyzes such a system in detail.Given a CT system He(jn), the overall DT system behaves as: H(e)=H.0学)， <π. Equivalently,the overall system will be the same as a given H(e),if the continuous-time system satisfies He(j)=H(eiT),</T. An important example of such a system is discussed in OSB Example 4.9:consider a discrete- time system with the following frequency response: H(e)=e-jwA When△is an integer,this is a time delay by△units:ynl=r[n-△].When△is not an integer,however,such an interpretation is incorrect,since a discrete sequence can only be shifted by integer amounts.Instead,consider choosing He(jn)in OSB Figure 4.16 to be He(j)=H(eiT)e-jAT It then follows that yc(t)=xc(t-△T). With some computations,we see that the output of this system can be interpreted as ban- dlimited interpolation of the input discrete-time signal,followed by non-integer time delay and re-sampling.Summarizing using a direct convolution representation with -oo<n<oo: 州-店产。=。 00 .xinl hinl. k=0 3
system is linear and time invariant, and equivalent to a continuous time LTI system: π Heff(jΩ) = H(jΩ) for Ω| | ≤ . T For example, Figure 4.13 shows the Fourier transform of signals through the overall system when the input is bandlimited. The discrete-time system in this case is an ideal lowpass filter with cutoff frequency ωc. Example: Hc(jω) : Lowpass filter, cutoff frequency = π/4 1/T : 20kHz Heff (jΩ) : Lowpass filter, cutoff frequency = ? CT Processing of DT Signals: Fractional Delay In addition to using discrete-time systems for processing continuous-time signals, the complementary situation is also possible, where a continuous time system is preceded by a D/C converter and followed by a C/D converter to process discrete-time signals, as depicted in OSB Figure 4.16. Section 4.5 of OSB analyzes such a system in detail. Given a CT system Hc(jΩ), the overall DT system behaves as: H(ejω) = Hc(j ω ) , |ω| < π . T Equivalently, the overall system will be the same as a given H(ejω), if the continuous-time system satisfies Hc(jΩ) = H(ejΩT ) , |Ω| < π/T . An important example of such a system is discussed in OSB Example 4.9: consider a discretetime system with the following frequency response: H(ejω) = e−jωΔ . When Δ is an integer, this is a time delay by Δ units: y[n] = x[n − Δ]. When Δ is not an integer, however, such an interpretation is incorrect, since a discrete sequence can only be shifted by integer amounts. Instead, consider choosing Hc(jΩ) in OSB Figure 4.16 to be Hc(jΩ) = H(ejΩT ) = e−jΩΔT . It then follows that yc(t) = xc(t − ΔT) . With some computations, we see that the output of this system can be interpreted as bandlimited interpolation of the input discrete-time signal, followed by non-integer time delay and re-sampling. Summarizing using a direct convolution representation with −∞ < n < ∞: ∞ sin π(n − k − Δ) = x[n] sin π(n − Δ) y[n] = � x[k] π(n − k − Δ) ∗ π(n − Δ) = x[n] [ ∗ h n] . k=∞ 3