102 Ch.

2 1 Discrete-TimeSignals and Systems

Now if ih,/

< 1 for all k, then

and hence

On the other hand, if one or more of the ih,l 2 1, h(n) is no longer absolutely summable, and consequently, the system is unstable. Therefore, a necessary and sufficient condition for the stability of a causal IIR system described by a linear constantcoefficient difference equation is that all roots of the characteristic polynomial be less than unity in magnitude. The reader may verify that this condition carries over to the case where the system has roots of multiplicity m.

2.5 IMPLEMENTATION OF DISCRETE-TIME SYSTEMS

Our treatment of discrete-time systems has been focused on the time-domain characterization and analysis of linear, time-invariant systems described by constantcoefficient linear difference equations. Additional analysis methods are developed in the next two chapters, where we characterize and analyze LTI systems in the frequency domain. Two other important topics that will be treated later are the design and implementation of these systems. System design and implementation are usually treated jointly rather than separately in practice. Often, the system design is driven by the method of implementation and by implementation constraints, such as cost, hardware limitations, size limitations, and power requirements. At this point, we have not as yet developed the necessary analysis and design tools to treat such complex issues. However, we have developed sufficient background to consider some basic implementation methods for either hardware or software realizations of LTI systems described by linear, constant-coefficient difference equations. This topic is treated in depth in Chapter 7.

2.5.1 Structures for the Realization of Linear Time-Invariant Systems

In this subsection we describe structures for the realization of systems described by linear constant-coefficient difference equations. Additional structures for these systems are introduced in Chapter 7. As a beginning, let us consider the first-order system

which is realized as in Fig. 2.33a. This realization uses separate delays (memory) for both the input and output signal samples and it is called a direct form I structure. Note that this system can be viewed as two linear time-invariant systems in cascade. The first is a nonrecursive, system described by the equation

whereas the second is a recursive system described by the equation

2.5 / Implementation of Discrete-Time Systems 103

As we have seen in Section 2.3.4, however, if we interchange the order of linear time-invariant systems in cascade, the overall system response remains the same. Thus if we interchange the order of the recursive and nonrecursive systems, we obtain an alternative structure for the realization of the system described by (2.5.1). The resulting system is shown in Fig. 2.33b. From this figure we obtain the two difference equations w(n) = -a,w(n - 1) + x(n) (2.5.4) y(n) = bow(n) + b,w(n - 1) (2.5.5) which provide an alternative algorithm for computing the output of the system described by the single difference equation given in (2.5.1). In other words, the two difference equations (2.5.4) and (2.5.5) are equivalent to the single difference equation (2.5.1). A close observation of Fig. 2.33b reveals that the two delay elements contain the same input w(n) and hence the same output w(n - 1). Consequently, these two elements can be merged into one delay, as shown in Fig. 2.33~. contrast to the direct In form I structure, this new realization requires only one delay for the auxiliary qi~antity w(n), and hence it is more efficient in terms of memory requirements. It is called the direct form 1 structure and it is used extensively in practical applications. 1 These structures can readily be generalized for the general linear time-invariant recursive system described by the difference equation

FIGURE 2.33 Steps in converting from the direct form I realization in (a) to the direct form II realization in (c).

104 Ch. 2 1 Discrete-Time Signals and Systems

Figure 2.34 illustrates the direct form I structure for this system. This structure requires M N delays and N + M + 1 multiplications. It can be viewed as the cascade of a nonrecursive system

and a recursive system

N

By reversing the order of these two systems as was previously done for the firstorder system, we obtain the direct form I1 structure shown in Fig. 2.35 for N > M. This structure is the cascade of a recursive system

followed by a nonrecursive system

We observe that if N 1 M, this structure requires a number of delays equal to the order N of the system. However, if M > N, the required memory is specified by M. Figure 2.35 can easily by modified to handle this case. Thus the direct form I1 structure requires M + N 1 multiplications and max{M, N] delays. Because it requires the minimum number of delays for the realization of the system described by (2.5.6), it is sometimes called a canonic form. A special case of (2.5.6) occurs if we set the system parameters a, = 0, k = 1, . . . ,N. Then the input-output relationship for the system reduces to

FIGURE 2.34 Direct form I structure of the system described by (2.5.6).

2.5 / Implementation of Discrete-Time Systems 105

w(n)

I
z-1

y ( n )

+I

w(n-1)

b1

z-1
-112

w ( n -2)

b2

2-1

-113

w(n - 3)

b3

-11~-2

w(n-M)

bM

(M=N-2)

- 1 1 ~1

-11~

FIGURE 2.35 Direct form II structure for the system described by (2.5.6).

which is a nonrecursive linear time-invariant system. This system views only the most 1 input signal samples and weights each sample by the appropriate coefrecent M ficient b, from the set {b,}, prior to addition. In other words, the system output is basically a weighted moving average of the input signal. For this reason it is sometimes called a moving average (MA) system. Such a system is an FIR system with an impulse response h(k) equal to the coefficients b,, that is,

h(k)

b,, 0,

O-(kIM otherwise

If we return to (2.5.6) and set M = 0, the general linear time-invariant system reduces to a "purely recursive" system described by the difference equation

In this case the system output is a weighted linear combination of N past outputs and the present input. Linear time-invariant systems described by a second-order difference equation are an important subclass of the more general systems described by (2.5.6) or (2.5.10) or (2.5.13). The reason for their importance will be explained later when we discuss quantization effects. Suffice it to say at this point that second-order systems are usually used as basic building blocks for realizing higher-order systems.

106 Ch. 2 / Discrete-Time Signals and Systems

The most general second-order system is described by the difference equation

which is obtained from (2.5.6) by setting N = 2 and M = 2. The direct form I1 structure for realizing this system is shown in Fig. 2.36a. If we set a , = a, = 0, then (2.5.14) reduces to y(n) = box(n) + blx(n - 1) b,x(n - 2) (2.5.15)

which is a special case of the FIR system described by (2.5.1 1). The structure for realizing this system is shown in Fig. 2.36b. Finally, if we set b, = b, = 0 in (2.5.14), we obtain the purely recursive second-order system described by the difference equation (2.5.16) y(n) = -a,y(n - 1) - a,y(n - 2) + box(n) which is a special case of (2.5.13). The structure for realizing this system is shown in Fig. 2 . 3 6 ~ .

2.5.2 Recursive and Nonrecursive Realizations of FIR Systems

We have already made the distinction between FIR and IIR systems, based on whether the impulse response h(n) of the system has a finite duration or an infinite duration.

FIGURE 2.36 Structures for the realization of second-order systems: (a) general second-order system; (b) FIR system; (c) "purely recursive system"