Scribd
Upload
56 views16 pages

Digital Signal Processing, Filter Structures: Dr. Alan Tan 28 August 2011

The document discusses different methods for realizing digital filters including direct and indirect realization structures. Direct realization structures include tapped delay line, direct form I, direct form II, and canonic form which directly implement the transfer function coefficients. Indirect or not-so-direct realization structures include cascade form which breaks the transfer function into cascaded substructures, and parallel form which represents the transfer function as a sum of rational functions. Cascade and parallel forms can realize higher-order systems in a more stable way compared to direct forms.

Uploaded by

鲁肃津
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
56 views16 pages

Digital Signal Processing, Filter Structures: Dr. Alan Tan 28 August 2011

The document discusses different methods for realizing digital filters including direct and indirect realization structures. Direct realization structures include tapped delay line, direct form I, direct form II, and canonic form which directly implement the transfer function coefficients. Indirect or not-so-direct realization structures include cascade form which breaks the transfer function into cascaded substructures, and parallel form which represents the transfer function as a sum of rational functions. Cascade and parallel forms can realize higher-order systems in a more stable way compared to direct forms.

Uploaded by

鲁肃津
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 16

Digital Signal Processing, Filter Structures

Dr. Alan Tan 28 August 2011


(Edition #3.3)

c 2011 Alan Tan

Lecture outline
Direct realization Tapped delay line, Direct form I, Direct form II, Canonic form Not-so-direct realization Cascade form, Parallel form

c 2011 Alan Tan

Page 2/16

Direct realization
For starters, suppose that the transfer function H(z) is rational, i.e., b0 + b1z 1 + + bq z q H(z) = 1 + a1z 1 + + apz p (1)

The direct form realization of (1) is a cascaded system of substructures Hb(z) and Ha(z), i.e., H(z) = Hb(z)Ha(z) where and Hb(z) = b0 + b1z 1 + + bq z q Ha(z) = (2)

1 (3) 1 + + a z p 1 + a1 z p The direct form is direct in that the transfer function coecients (i.e., the bs and as) are precisely those in actual realization.

c 2011 Alan Tan

Page 3/16

Direct realization (cont.)


Recall that H(z) = Y (z)/X(z) where X(z) and Y (z) are the z-transform representations of input x[n] and output y[n], respectively. From (2), we have Y (z) = X(z)(b0 + b1z 1 + + bq z q ) y[n] = b0x[n] + b1x[n 1] + + bq x[n q] A simple realization of (4) is the tapped delay line, as seen below.
...

(4)

Figure 1: The tapped delay line realization of (4). The structure consists of q memory units, q + 1 multipliers and q adders.

c 2011 Alan Tan

Page 4/16

Direct realization (cont.)


Repeating the procedure for (3), we obtain Y (z)(1 + a1z 1 + + apz p) = X(z) y[n] + a1y[n 1] + a2y[n 2] + + apy[n p] = x[n] or y[n] = x[n] a1y[n 1] a2y[n 2] apy[n p] The realization of (5) appears next. (5)

...

Figure 2: The tapped delay line realization of (5). The structure consists of p memory units, p multipliers and p adders.

c 2011 Alan Tan

Page 5/16

Direct realization (cont.)


Two realizations of (1) are permissible in the direct form, depending on the order of Hb(z) and Ha(z). The direct form I structure has Hb(z) before Ha(z), as illustrated in Figure 3.

Figure 3: The direct form I structure. The structure consists of p+q memory units, p + q + 1 multipliers and p + q adders.

c 2011 Alan Tan

...

...

Page 6/16

Direct realization (cont.)


Switching the order of Hb(z) and Ha(z) yields the non-standard direct form II structure.

...

Figure 4: The non-standard direct form II structure. The structure has as many memory units, multipliers and adders as direct form I.

c 2011 Alan Tan

...

Page 7/16

Direct realization (cont.)


The standard direct form II structure, also referred to as canonic form, combines the columns of memory units as one1. The structure is canonic in the sense that the fewest possible memory units are used.

Figure 5: The standard direct form II structure (illustrated for p = q ). This structure has fewer memory units (i.e., max(p, q)), but the same number of multipliers and adders as the ones before.
1

The direct form II referred to hereafter are implicitly assumed to be in canonic form.
Page 8/16

c 2011 Alan Tan

...

Direct realization (cont.)


Problem 1: Realize the second-order system 0.5(z 2 0.36) H(z) = 2 z + 0.1z 0.72 Solution: Note that 0.5 0.18z 2 H(z) = 1 + 0.1z 1 0.72z 2

(a)

(b)

Figure 6: Realization in (a) direct form I, and (b) direct form II of Problem 1.
c 2011 Alan Tan Page 9/16

Not-so-direct realization
Further factorization of the numerator and/or denominator polynomials of (1) produces a system of yet smaller substructures which, when connected in cascade, is identical to the original (1). Formally, the cascade form is described by H(z) = b0H1(z)H2(z) HK (z) where Hk (z) (for k = 1, 2, . . . , K) are the substructures and K is the number of substructures. These substructures are commonly rst- or second-order systems, and realized in direct form II.
...

Figure 7: Realization in cascade form.

c 2011 Alan Tan

Page 10/16

Not-so-direct realization (cont.)


Apparently, many dierent realizations of the same H(z) can be accomplished in cascade form depending on the ordering and pairing of the substructures. Ordering, here, refers to the order of Hk (z) while pairing means the grouping of the poles and zeros of H(z) to form Hk (z). Although these realizations are identical in theory, practical limitations (e.g., the propagation of roundo errors from one substructure to the next) may introduce deviations in the actual performance.

c 2011 Alan Tan

Page 11/16

Not-so-direct realization (cont.)


Alternatively, we may also formulate H(z) as a sum of rational functions, i.e., H(z) = c + H1(z) + H2(z) + + HK (z) where c is some constant. Such formulation leads to a realization of the parallel form. A straightforward realization of the parallel form is given by the partial fraction expansion of H(z).

Figure 8: Realization in parallel form.

c 2011 Alan Tan

...

Page 12/16

Not-so-direct realization (cont.)


Problem 2: Realize the system (z + 1)2 H(z) = 2 (z + 5z + 6)(z + 0.5) in (a) cascade form, and (b) parallel form. Solution: Factor H(z) as z+1 z+1 z 1 + z 2 1 + z 1 H(z) = H1(z)H2(z) = 2 = 1 + 6z 2 1 + 0.5z 1 z + 5z + 6 z + 0.5 1 + 5z

Figure 9: Realization in cascade form of Problem 2.


c 2011 Alan Tan Page 13/16

Not-so-direct realization (cont.)


Problem 2 (cont.) To realize H(z) in the parallel form, rewrite H(z)/z in the form of a partial fraction expansion, i.e., H(z) (z + 1)2 A0 A1 A2 A3 = = + + + z z(z + 2)(z + 3)(z + 0.5) z z + 2 z + 3 z + 0.5 with A0 = 1/3, A1 = 1/3, A2 = 8/15 and A3 = 2/15. This allows us, thereafter, to establish A1 A2 A3 H(z) = A0 + + + 1 + 2z 1 1 + 3z 1 1 + 0.5z 1

c 2011 Alan Tan

Page 14/16

Not-so-direct realization (cont.)


Problem 2 (cont.) The parallel form realization appears below.

Figure 10: Realization in parallel form of Problem 2.

c 2011 Alan Tan

Page 15/16

Not-so-direct realization (cont.)


Quantization of coecient values in real implementations generally introduces some loss of accuracy, and in the worst cases, severe errors. For instance, if the poles of H(z) are near the unit circle, the poles of the actual implementation may be displaced outside of the unit circle, resulting in an unstable system. These problems are usually more serious when higher-order systems are realized in the direct form. For such cases, the cascade and parallel realizations are preferred. The cascade form, in particular, is commonly recommended for the implementation of higher-order systems having closely clustered poles.

c 2011 Alan Tan

Page 16/16

You might also like