Journal of Sound and Vibration (1990) 141(l), 103-125

SIMPLIFIED METHODS TO CALCULATE

THE ATTENUATION OF SILENCERS

W. FROMMHOLD AND F. P. MECHEL

Fraunhofer-lnstitut ftir Bauphysik, Bereich Akustik, Nobelstrafie 12, D-7000 Stuttgart 80,
West Germany

(Received 6 October 1988, and in revised form 6 November 1989)

The fundamentals for the calculation of attenuation in silencer ducts have been known
for many years, but reliable approximate computation methods suitable for engineering
purposes are still needed. Graphical methods using conformal mapping exist only for the
case of locally reacting linings and are not suited for machine calculation. Iterative methods
of approximation need a great amount of calculation, are sensitive to the choice of start
solutions and have the unpleasant property of jumping sometimes into a solution for a
higher duct mode. The finite element method also needs a great amount of calculation
which does not seem to be justified for simple silencer geometry. Explicit approximating
formulae are generally too restricted in their validity range for engineering purposes. An
exception to this is the expansion of tangent or Bessel functions by the method of continued
fractions. This method of approximation is discussed in detail in the paper. In the case
of locally reacting linings an explicit calculation of the attenuation curve is possible. In
the case of bulk reacting linings the method is combined with a simple prediction scheme
and then allows a non-iterative straightforward calculation of the attenuation curve. This
procedure is also used to evaluate the effect of flow on attenuation. The proposed
calculation methods are tested numerically and by measurements.

1. THEORY
First a rectangular symmetrical silencer duct as shown in Figure l(a) is considered. The
duct walls are lined only on two sides; at z = 0 and B a hard wall is assumed. The
attenuation Dh in decibels in the x-direction normalized to half of the height of the free
duct is wanted. This attenuation is obtained from the propagation constant r by

Dh = 8.69r’ (dB), r = r’+ jr”. (1)

(0) (b)

Figure 1. Shape of silencer ducts. (a) Rectangular with lining on two opposite sides; (b) round.
103
0022-460X/90/106103+23 $03.00/O 0 1990 Academic Press Limited
104 W. FROMMHOLD AND F. P. MECHEL

The decrease Dl of the sound pressure level for a silencer duct of length L is then
approximately
Dl= Dh(L/h). (2)

The consideration is restricted to symmetrical modes in the free cross-sectional area of

the duct and therefore a pressure distribution of the form

p(x, Y, 2, t) =po exp (-TX) cos (E~.Y)cos (v) exp (jmt) (3)
is assumed. Inserting equation (3) into the wave equation yields
l-2=~;.+~:-k;, ko= w/co. (4)
To determine eY and E, the boundary condition of matching the field admittance to the
wall admittance G,,, at the surface of the absorber lining and of the rigid duct wall is to
be satisfied. With u, = 0 at z = 0 and B it follows that
E2.n= nzr/ B, n,=0,1,2 ,.... (5)
For E,, the solution depends on the type of lining.

1.1. RECTANGULAR SILENCER WITH LOCALLY REACTING LINING

It is assumed that there is no sound propagation in x-direction inside the lining

(resonators, porous lining with hard partitions or with high flow resistivity). With the
particle velocity

uY= (j/ koZo) Q/~Y (6)

the boundary condition leads to the defining equation

a$ tan ( E$) = jk,hZ,G,, Zo = poco. (7)

With the abbreviations
U = k,hZ,G, and E = ( Eyh)2, (8)

equation (7) is rewritten as

D tan (v%) = j U. (9)
(For a porous lining with hard partitions G, = l/[Z, coth (TJ)] is valid.) From the
solution E of this equation, then

Th=JE+(E,h)2-(koh)2, Real (Th)aO, (10)

is obtained.
Graphical solutions of equations (7) and (9), obtained by using conformal mapping
of the complex quantity E in the complex U-plane for the fundamental mode have been
given in the literature several times. Here also maps for the first two higher modes have
been calculated and are presented in Figure 2. These maps are a useful tool for the
verification of approximation techniques, especially insofar as the critical branch points
and the matching of the numerical solutions to the correct duct modes are concerned.

1.2. RECTANGULAR SILENCER WITH BULK REACTING LINING

For this type of lining the wave propagation inside the absorber has to be taken into
account. The absorber linings are layers of thickness d, made of a homogeneous material
which is described by its characteristic propagation constant r, and characteristic wave
 CALCULATION OF ATTENUATION IN SILENCERS
105

0
/I II /
106 W. FROMMHOLD AND F. P. MECHEL

impedance 2,. For simplicity, a plane wave in the z-direction is presumed. The sound
field in the absorber material can be written as

~~(x,~,t)=~~exp(-Tx)cosh[e,(y-h-d)]exp(jor). (11)
This satisfies the boundary condition at the back side of the absorber. Inserting this into
the wave equation in the absorber material leads to

P=r’,-e;. (12)
Two boundary conditions at the surface of the absorber are now to be satisfied: (a)
equality of the field admittance and the wall admittance in the y-direction; (b) equality
of the axial propagation constant r inside and outside the absorber. This leads to the
defining equation

eh tan (sh) = (jk,Z,/r,Z,)e,h tan (s,d). (13)

By combining this with equations (4) and (12), and using the abbreviations of equation
(8) and A = d/h, equation (13) can be rewritten as

~tan(~)=(j~Z,/T,Z,)J(T,h)‘+(~h)‘-Etanh(AJ(T,h)’+(~h)‘-E). (14)

The solution for E is usually obtained by an iterative procedure. A stepwise approximation

starting from the hard-wall conditions is described in reference [5].
1.3. ROUND SILENCER WITH LOCALLY REACTING LINING

In a round silencer, as shown in Figure l(b), equations (1) and (2) remain valid.
According to reference [7] the wave equation is written in cylindrical co-ordinates. Its
solutions are Bessel functions J,. Therefore the pressure distribution is written as
p(x, r, t) = p. exp (-TX) cos (me) J,( .c,r) exp (jmt), (19
with the mode number rn = 0, 1,2 . . . . Inserting this into the wave equation yields
r’=&-k;. (16)

Consideration here is restricted to the axially symmetric mode m = 0. Then, from the
radial particle velocity, the field admittance

G, = V,/P = (j~lkoZo)J~(~r)IJO(~r)
= ~~~l~o~o~~-J~~~~~llJo~~~~(17)
is obtained. Satisfying the same boundary condition as in section 1.1 leads to the defining
equation
DJ,(G)/J,(a) = j U, (18)
with the abbreviations of equation (8). (In the case of a porous lining with hard partitions
j U corresponds to the right side of equation (20) (to follow) with a = jT,h.)

1.4. ROUND SILENCER WITH BULK REACTING LINING

According to reference [7], for the fundamental mode the pressure distribution in the
absorber is written as

pat& r, t)=poexp (--TX) Jo(&,r)-~~:h,=d,~~Yo(~~r)] exp (jut), (19)

1 &cl
Y, being Neumann functions. Satisfying the boundary conditions as in section 1.2 leads
 CALCULATION OF ATTENUATION IN SILENCERS 107

to the defining equation

oJ~(a) j&G
m=r.z,= Jr(a)Y,((I +A)=) -Y,(a)J,{(l
J,(a)Y,((I+A)JEa)-Y,(a)J,{(l+n)C}’
+A)a}
(20)

with E, = E -[(I’ah)2+(bh)2] and A = d/h.

1.5. EFFECT OF MEAN FLOW ON ATTENUATION

The influence of mean flow on the attenuation of silencers is known to be a rather

complicated matter. On the other hand, the simplified calculation methods presented in
this paper are proposed mainly for ventilation ducts with small mean velocity U. It is
known that for small Mach numbers M = U/c the effect of flow on attenuation is very
small and often can be neglected. Hence a rough evaluation seems to be sufficient.
For a locally reacting lining it is assumed that (a) the flow is uniform over the free
cross-sectional area of the duct, and (b) the wall admittance G, of the liner is not
influenced by the flow. Then the wave equation can be written as
Ap - (jk,+ Ma/t~x)‘p = 0. (21)
Equation (10) becomes
Th=jk,,h[-M+~1-(1-M2)E/(lcoh)2]/(1-M2). (22)
The sign of the root is taken so as to make Real {Th} L 0. The sound wave travels in the
positive x-direction. The Mach number A4 is positive for a flow in the positive x-direction
and is negative for a flow in the negative x-direction.
Equation (6) for the particle velocity becomes due to the convective superposition of
the flow,

u,. = {j/Z&(1 +jMr/k,)) aplay, (23)

and similarly for the radial velocity u, in the case of a round duct. The boundary condition
at the absorber surface formulated for fitting the particle velocity leads to an additional
factor w on the right side of equation (9): one has

L?tan(fi)=jUw (24)
in the case of a rectangular duct, and, in place of equation (18),

e~,(*)/J,(fl) = j UW (25)
in the case of a round duct, with
w=[1-M~1-(1-M2)E/(k,h)2]/(1-M2). (26)
In more recent publications [4-61 the boundary condition is not formulated for particle
velocity, but for particle displacement. This leads to a factor w2 in equations (24) and
(25) in place of w.
In the case of a bulk reacting lining the formulation of boundary conditions including
flow in the linear material is difficult. But in consideration of the fact that generally bulk
reacting linings are used only at flow velocities U < 20 m/s and that for reasons of stability
the specific flow resistivity of the porous material is higher than 10 kNs/m4, it may be
permissible for a rough evaluation to assume that (a) the flow velocity inside the liner is
zero, (b) the axial flow velocity is uniform in the free area of the duct, but drops rapidly
to zero in a thin boundary layer at the surface of the absorber, and (c) the characteristic
108 W. FROMMHOLD AND F. P. MECHEL

impedance and the propagation constant of the porous material are not affected by flow.
Then equation (23) is used for the boundary condition in the y-direction, whereas the
other conditions remain unchanged. This leads again to an additional factor w according
to equation (26) on the right side of equation (14) (respectively of equation (20) in the
case of a round duct)-or to a factor w2, if the displacement admittance is used.

2. PERMISSIBLE ERROR LIMITS AND ACCURACY TEST FOR

APPROXIMATE SOLUTIONS

Only the case of rectangular silencers with locally reacting linings is considered here.
Before approximate solutions for equation (9) are examined in the following, it is useful
to define the permissible error of E’ and E” in a way that the error for the attenuation
Dh is certain to remain below a given limit. Only the fundamental mode is considered.
From
Dh = 8.7 Real (JE-(k,l;jl), (27)
error analysis leads to an expression of the form
d~h/~h =~(F,~~E’/E’I+F,~AE”/E”I). (28)
The multiplier F2 is always smaller than 2, whereas the multiplier F, remains smaller
than 1 for typical porous linings. Only in the case of slightly damped resonator linings
may F, exceed the value of 1 at (k,h)* = E’.
This means that in general, for an upper error limit of 10% in Dh a relative error in
E’ as well as in E” of the same magnitude is permissible.
Approximate solutions of equation (9) by a suitable expansion of the tangent are tested
in the following way. First a grid of exact E-values in the complex plane is presupposed.
Then the corresponding U-values are calculated from equation (9), and serve as input
to the approximate formula, with result E,. The values E, thus obtained are compared
with the presupposed exact values E by drawing lines with constant EL or Eg in the
E-plane. The approximation is the more accurate, the more closely these lines reproduce
the presupposed grid.
A polynomial in E of degree N is explicitly solvable up to N = 4. For the accuracy
test among the N solutions the one closest to the presupposed value of E always is
considered.

3. EXPLICIT SOLUTIONS FOR SILENCERS WITH LOCALLY REACTING LINING

3.1. RECTANGULAR DUCT

3.1.1. The continuedfractions expansion of fi tan (a)

The expansion of z tan (z) by the method of continued fractions proposed in reference
[l] has proved to be a powerful means because the expansion is valid even around and
beyond the poles of the tangent. The expansion formula is [2]

tan(z)=Eg$.... (29)

By adding successive fractions, this leads stepwise to the following polynomial approxima-
tions of equation (9):
E =jU, 3E/(3-E)=jU, E(15-E)/(15-6E)=jU, (30-32)
E(105-lOE)/(105-45E+E*)=jU. (33)
 CALCULATION OF AlTENUATION IN SILENCERS 109

In the following only the polynomial approximations with the same order N in the
numerator and denominator polynomials are considered, and the equations are solved
for E:
E =3jU/(3+jU), (34)
E =(105+45jU*~ll 025+5250jU++605(jU)2)/(20+2jU), (35)
E3+C2E2+C1E+CO=0, (36)
C2=(-1260-21OjIJ)/(21+jU), Cl =(10395+4725jlJ)/(21+jlJ),

CO=(-10395jU)/(21+jU),
E4+C3E3+C2E2+C1E+CO=0, (37)
C3=(-6930_63OjU)/(36+jU), C2=(270270+51975jU)/(36+jIJ),

Cl=(-2027025_945945jIJ)/(36+jU), CO=2 027 025jU/(36+jU).

Figure 3(a) shows the accuracy test described above for equation (34), and Figure 4(a)
for equation (35).

Figure 3. Accuracy test for the continued fractions expansion of order 1. (a) Equation (34); (b) equation (41).

Equation (35) already represents a very close approximation for the range of the most
interesting mode 0, except in the immediate environment of the first branch point. The
accuracy test for equations (36) and (37) is shown in Figure 5. The branch point is now
also reproduced correctly in the range of mode 0. In addition, equation (36) covers the
technically interesting part of mode 1, and equation (37) even the whole range of mode
1, again except for the neighbourhood of the second branch point.
 110 W. FROMMHOLD AND F. P. MECHEL

v-5 0 5 10 15
E,’
Figure 4. Accuracy test for the continued fractions expansion of order 2. (a) Equation (35); (b) equation (42).

(a)

I I I I I I I
__,__T__r-_,__i__i__,__
I I 1 I I I I

0 10 20 30 40 50 60

10

_l ’ 0 10 20 30 40 50 60

Figure 5. Accuracy test for the continued fractions expansions of order 3 and 4. (a) Equation (36); (b)
equation (37).
 CALCULATION OF ATTENUATION IN SILENCERS 111

A special problem is the assignment of the N solutions of the polynomial with order
N to the physical modes or to the Riemann sheets in Figure 2. Only the linear approxima-
tion, equation (34), with its small validity range is not in doubt. Figure 6 shows the
separating line between the two lowest solutions of equations (35), (36) and (37),
respectively, in the E-plane of mode 0. Certainly it is easier to calculate both solutions
and to decide for that solution with the smaller attenuation than to map the separating
line into the U-plane.

Equation (351-c

-4 -3 -2 -1 01 2 3 4 5
Ed
Figure 6. Separation line for the two lowest solutions of equations c35)- -(37) in the E-plane.

3.1.2. Improvement of low order approximations

The easily solvable linear and quadratic equations (34) and (35) do not cover the whole
area of interest in the E-plane. Therefore, they need to be modified in a way that, still
remaining below a certain limit of error, their validity range is extended and shifted to
the technically more interesting domain with high attenuation.
The procedure is as follows. Only the form of the approximating formula j U,(E) for
the exact function j U(E) = V@ tan (fi) is taken. The coefficients of the expansion series
are then replaced by free parameters PO. . . P,, and these are defined in such a way that
in a given validity range, G, the squared error sum Q is minimized:

Q(f’o . IjU(E)-jU,(E,P,...P,)I’dEHrnin (38)

This minimization cannot be carried out analytically because j U, does not have the form
of simple power series. The integral in equation (38) has to be calculated numerically
for an n-dimensional parameter space and then a search for the minimal Q has to be
carried 01:’ To keep this procedure feasible it should be known which parameters are to
be varied. For this purpose the continued fractions, equations (31) and (33), are written
in product form as
K17(E-N,)/n(E--P,)=jU, (39)
where the NV represent zeroes, the P, respresent poles and K is a multiplier chosen so
that for very small values of E equation (30) is approximated.
In Table 1 the zeroes and poles of the approximate equations (31) and (33) are compared
to the analytical values according to equation (9). It is evident that only the respective
highest pole and zero differ from the exact values. Thus the variation can be restricted
to these two parameters and K.
Improvement of the linear approximation. Equation (31) is written as

-KP,E/(E-P,)=jU,. (40)
112 W. FROMMHOLDAND F. P. MECHEL
TABLE 1
Zeroes and poles for the exuct equation (9) and for lower
continued fractions expansions

N, N2 4 p2

Exact 0 9.87 2.467 22.21

Equation (31) 0 - 3.0 -
Equation (33) 0 10.5 2.469 42.53

With K = 1 and P, = 3 the original approximate formula is shown in Figure 3(a). The
important parameter for the extension of the validity range is the respective highest pole.
From Figure 2 it might be deduced that this pole has to be complex. For a given K first
the error sum Q( Pi, P:) is calculated by applying a survey strategy for Qmin*Then the
calculation is carried out in a dense grid of P-values in the neighbourhood of the minimum.
To begin, a target area -2 < G’ < 1,0 < G” < 3 is selected, because all linings behave at
low frequencies as springs and therefore start with negative E’. In Figure 7 is shown the
Q-distribution for two values of K as contour plots in the P-plane. The absolute minimum
is obtained by drawing the relative minima in the P-plane versus the K-axis according
to Figure 8.

P’
7 2.8 2.9 3.0 3.1
I 1 I I I

K = I.0

-0.4

h -0.5

-0.6

-0.7
Figure 7. Contour plot of squared error sum Q(K, Pi, P;) according to equation (38) for the linear continued
fractions expansion equation (40).

To answer the question about the appropriate choice of the target area G the exact
solution E has been calculated for the usual range of parameters A = 0.5 and 2 and
r = Ed/pc = 1 and 8, and has been plotted in Figure 9(a). The characteristic absorber
constants r, and 2, have been calculated by using the formulae in reference [3].
From the engineering point of view the domain of high frequencies is of little interest
because attenuation there is small and the assumption of a plane wave is no longer
justified in general. Therefore, according to Figure 9(a), the target area is chosen as
-1.5~ G’< +1.5,0< G”<3. For this value of G the minimum of Q has been found at
 CALCULATION OF ATTENUATION IN SILENCERS 113

500

x
0 400-

x
,:2 ?;
0
300- \ x

/
x\
x\ XHX
x,,-XR
1 I I I I1 I / 1 I
096 090 100 I.02 104 I.06 l-06 110 1 2
K

Figure 8. Squared error sum Q,,(K, P,,,) according to equation (38) for the linear continued fractions
approximation equation (40) for the bottom points in the P-plane versus K.

fh (kHr cm)
Figure 9. (a) Exact solution E(kh) for a locally reacting lining; A = 0.5 and 2, &II G 5. -, I = Ed/pc = I;
- - -, r = 8. (b) Attenuation Dh for r = 1, A = 0.5 and 2. -, Exact solution; - - -, approximate solution from
equation (41).

P = 2*74- jO.52 and K = 1.05. Then, from equation (40), the approximate formula is
deduced:
E = (2.74 - jOmS2)jU/[2.88 - jO.55) + j U]. (41)
The accuracy test for equation (41) is shown in Figure 3(b). Compared with the original
approximate formula as shown in Figure 3(a), the validity range has been almost doubled.
114 W. FROMMHOLD AND F. P. MECHEL

In Figure 9(b) the attenuation Dh for the exact solution E is compared with Dh for the
approximate solution from equation (41) for the most critical curves in Figure 9(a) with
r = 1. It becomes evident that even for a relatively small specific flow resistivity and/or
thickness d of the absorber the attenuation curve can be calculated accurately far beyond
the frequency of the attenuation maximum. This is a remarkable result for a linear
approximate formula.
Improvement of the quadratic approximation. The variation of parameters is limited to
the complex pole P2 and the zero Nz. For P, the exact value 2.467 according to Table 1
is used.
The target area G for the optimization is taken around the first branch point with
2 < G’ < 53 < G” < 6, because the transition area from mode 0 to mode 1 is of some
interest from the technical point of view and equation (35) is not sufficient in that area.
The quadratic solution thus improved is
E =(78*94-j543)+jU(34*47-j2*2)*<
1.2 (42)
(16.1-jl.l1)+2jU ’

where
~=J(6203-j857)+(2887.3-j372)jU+(867.4-j13O)(jU)*. (43)
It is tested in Figure 4(b). Now the accuracy is sufficiently good also in the neighbourhood
of the first branch point and in the transition area to mode 1.
Thus equation (42) meets all requirements for the calcuation of silencers with locally
reacting linings. This is demonstrated in Figure 10, for which equation (42) has been
applied to calculate the attenuation of a silencer lined with Helmholtz resonators. This
lining, described in reference [8], has been designed to reach the branch point and thus
the theoretical maximum attenuation Dh = 19 dB. The approximate formula reaches a
value of 18.5 dB. The calculated attenuation is in reasonable agreement with measurements
[8]. The precision of the description of the Helmholtz resonators could not be checked.
The fact that the theoretical maximum attenuation is not reached in the experiment could
be caused by higher losses in the sharp edged necks than assumed in reference [8].
Calculated curves with higher values of the resistance 2’ than the presumed Z&,, are in
good agreement with measurements.

3.1.3. Explicit solutions for higher modes

The continued fractions expansion is also applicable to the case of higher modes in
the y-direction. The argument z in equation (29) then becomes a- nr, with the mode
number n. For the linear approximation the general equation (40) takes the form
KPG(G-nr)/(P-(a-n~)~)=jU,. (44)
The solution for E leads to a quadratic equation,
C2E2+C1E+CO=0, (45)
with

C2 = (j U+ KP)‘, C~=~~U(~U+KP)[(~T)~-P]-(~T)~(KP+~~U)~,

CO = (j U)‘( ( n*)2 - P)‘.

For the mode n = 1 the parameters P and K have been varied as described above in
order to obtain the best approximation in a target area 8 < G’< 20,O < G”< 20. The
results are
P = 2.9 - jO_5, K = 1.08. (46)
 CALCULATION OF AI-I-ENUATION IN SILENCERS 115

20
1

18

16

14

12

E
z 10
a 2zopt
/

O_A&
0.6 0.0

Figure 10. Attenuation of a duct lined on one side with Helmholtz resonators designed to reach the
theoretical maximum attenuation [8]; h = 0.05 m. -, Calculated according to equations (42) and (43); 00 0 0.
measured [8].

These values are not necessarily the optimum for higher modes, but the accuracy test in
Figure 11 reveals that this explicit solution is sufficient for at least the first three modes
and for a porous lining with a flow resistivity not too small.

ny- 1 ny=2 fly=3

40

3c

-z 2c

1c

Figure Il. Accuracy test for the explicit linear approximation to higher modes, equations (45) and (46).
-, Exact solution for Zd/pc = 4, d/h = 1 and 2, kh c 15.
116 W. FROMMHOLD AND F. P. MECHEL

For the quadratic approximation of tan (z) it is more convenient to use z = &h- nr.
Then the quartic equation
(Eh)4+C3(&h)3+C2(&h)*+C1(&h)+C0=0, (47)
with
M=nr, C3= M(30+4jU)/(-lo-jU),

C2=[105-30M2-jU(6M2-45)]/(-lo-jU),

Cl = M[lOM*-105+jU(4M2-90)]/(-lo-jU),

CO=[-jU(M*(M*-45)+105)]/(-lo-jU)
is obtained. The solutions E1...4are given by squaring ~h,...~. The validity range shown in
Figure 12 is sufficient also for resonator linings.

ny= 1 ny=2 fly=3

Figure 12. Accuracy test for the explicit quadratic approximation to higher modes, equation (47).

3.2. ROUND DUCT

The continued fractions expansion for J,(z)/JO(z) in equation (18) is [2]

J,(z)
-=---. z z* z* .. (48)
Jo(z) 2- 4- 6- ’

It is shown in reference [7] that the quadratic approximate formula

E =96+3OjU*J9216+2304jU+912(jU)*
(49)
12+jU
obtained from equations (18) and (48) is sufficient for all types of linings except for the
immediate neighbourhood of the first branch point.

4. A STRAIGHTFORWARD PROCEDURE FOR SILENCERS WITH BULK

REACTING LINING

4.1. PROCEDURE FOR THE RECTANGULAR DUCT

The experience gained in the previous section suggests applying the continued fractions
expansion up to degree two to both tangent functions in the defining equation (14), thus
 CALCULATION OF ATTENUATION IN SILENCERS 117
obtaining a quartic equation in E. But solutions of this kind do not meet all requirements
because the argument of the hyperbolic tangent on the right side of equation (14) reaches
very large values. An investigation of this argument, using the iterative scheme of Muller
[9] for the solution E, has shown that E is always smaller than (T,h)*+ (kJt)*. Thus the
value of the square root is not affected noticeably if the unknown solution E is replaced
by a suitably predicted value Ep. Then the right side of equation (14) is dependent only
on the characteristics of the liner, geometric data and frequency, and the problem can
be solved in the same manner as for a locally reacting lining.
To obtain reliable predictions the procedure is started at very low frequencies, where
an explicit solution of equation (14) is available. First the following abbreviations are
introduced:

u = j kO-Glr,-?, , V=(T,h)‘+(k,h)2, A =dlh. (50)

Then equation (14) is rewritten as

JZ tan (G) = LJm tan (Am). (51)
Now the lowest approximation tan (z) = z is used for the left side and the highest possible
approximation for a quadratic equation in E for the right side [2], tanh (z) =
z( 15 + z’)/( 15 + 6z2), of equation (14). The thus obtained quadratic equation
C2E2+ClE+CO=0 (52)
with
C2 = A2(6+AU), Cl = -15-A(6AV+ U(l5+2A’V)), CO= AUV(lS+A*V),
meets all requirements for low frequencies. Now from these start-values a prediction for
E at the next value of the frequency parameter is made by using the principles of linear
predictive coding (LPC). For this purpose a two-dimensional correlation analysis has
been carried out. Integrating up to k,h = 6 yields correlation factors close to one for
adjacent points on the E-curve. This means in terms of LPC that Ep( i) = E( i - 1). On
the other hand, if the correlation between adjacent points is so high, any extrapolation
should be more effective than the LPC-scheme for stochastic “signals”. Therefore a very
simple extrapolation from the two last points on the E-curve is used:
E,(i)=E(i-l)+[E(i-1)-E(i-2)]=2E(i-l)-E(i-2). (53)
In Figure 13(a) the relative error of Ep is plotted versus bh for equally spaced frequency
steps. At low frequencies E grows rapidly with k,h and therefore the error is considerable.
To overcome this deficiency a non-linear frequency spacing is chosen so that at low koh
the points on the E-curve are closer to each other. It is shown in Figure 13(b) that a
simple logarithmic frequency scaling is sufficient to reduce the error drastically. Moreover,
the error is now distributed rather regularly on k,h.
If this prediction error is only a few percent, then the overall error in the solution E,
E =(105+45Q*Jll 025+5250Q+1605Q2)/(20+2Q), (54)
with
Q = Um tanh (Am), (55)
is negligible. This is proved in Figure 14. Here the solution E from equations (54) and
(55), with Ep from equation (53) used at frequency steps of l/12 octave, is compared to
the “exact” solution from the iterative Muller scheme stopped at 1E(i) - E (i - l)]/] E (i)l =
10e4. For every discrete value of k,,h a line is drawn between the two solutions: i.e.,
118 W. FROMMHOLD AND F. P. MECHEL

6-

4-

0 1 2 3 4 5 6
kh
Figure 13. Relative error IE - E,I/IEI of the prediction E,, equation (53); d/h = 1, Zd/pc = Z/4/8. (a) 100
equally spaced steps in k&; (b) 99 steps of l/12 octave.

between the exact solution and the approximate solution. These lines appear as points
in Figure 14, so both solutions are very close to each other.
The calculation method described is compared to experiments in Figure 15. The insertion
loss of a homogeneous porous plate has been measured in a quasi-reflection-free termin-
ated duct drive with white noise from four loudspeakers in line. The absorber has been
mounted with its surface in the wall plane of the duct and has been replaced by a hard
plate for reference measurement. The sound pressure has been averaged over the width
of the duct in both measurements in order to reduce the effect of higher modes in the
z-direction. Thus the insertion loss will be close to the propagation loss Dh for a plane
wave, which is calculated. Indeed, the measurement results agree quite well with the
calculation except for high frequencies, where the presence of higher modes in the incident
sound field cannot be neglected.
The calculation method described above has been tested also for industrial splitter-type
silencers, and again good agreement with the measured insertion loss has been found in
the medium frequency range where the attenuation maximum is found.
Finally, it is to be mentioned that the proposed method is not applicable for porous
materials with low flow resistivity and/or small thickness d. The reason for this is illustrated
in Figure 16. The solution E calculated with Muller’s iterative scheme is shown. If starting
from the low frequency end, the mode zero is pursued for both tangent functions in
equation (14). Starting from the high frequency end gives the same result independent
of the direction of the frequency variation only in the case of higher values of the parameter
r. For low values of r a higher order solution for E is obtained with an attenuation less
than that of the fundamental mode. That means in the transition region two different
modes (with different phase velocity) are propagating and the behaviour of the silencer
duct cannot be described by a one-mode value Dh.
The calculation procedure described above has been tested also for these cases. The
result is similar to the curve for r = 1 in Figure 16: the procedure works correctly at low
 CALCULATION OF ATTENUATION IN SILENCERS 119

?,-
(a)
r=S

1 -

C)_
! I-
(b)

2,_

LIJ

4 I-(cl / I

i ,_
; . . . .,,
.’ ,: ..,;.., ‘,
1 : ,’ ; :
‘,.
;’ .’ ,..’ ‘..,
: .., .. .,
;’ !’
( .‘,‘
i/

CI_
4 3
E’
Figure 14. Accuracy test for the calculation procedure of equations (53)-(55) with starting solutions from
equation (52), frequency steps l/12 octave. At every frequency point a line is drawn between the approximate
solution and a high accuracy iterative solution; r = Ed/PC. (a) d/h=0.5; (b) d/h=l; (c) dlh=2.

40 ‘- 0.5m +I

8
2 30 i

1;;
0.1 0.315 1 3.15 3
f (kHd
Figure 15. Insertion loss of a bulk reacting lining; 3 = 42 kNs/m4, d = 0.042 m, h = 0.058 m, B = 0.5 m.
-, Calculated by using equations (52)-(55); 0 0 0, measured.
120 W. FROMMHOLD AND F. P. MECHEL

Figure 16. Solution E for equation (14) obtained with an iterative procedure starting from the low and high
frequency ends; A = dl h = 0.5, r = Ed/PC.

frequencies and jumps into the higher order solution after some uncontrolied excursions
in the transition region. In Figure 17 this is confirmed by experiment. The simplified
calculation is close to the two solutions of an iterative procedure, but both calculation
methods agree with experiment only at low frequencies. A lower limit for the application
of the simplified method for bulk reacting lining is therefore given by r = 1.2 for A = 0.5,
r=1*4forA=l,andr=2forA=2.

60

0
0.1 0,315 1 3.15 10
f CkHr)

Figure 17. Attenuation of a silencer with bulk reacting lining and low Row resistivity; E = 12 kNs/m4,
d = 0.032 m, h = 0.068 m, L = 0.5 m, B = 0.5 m. -, Iterative procedure; -, proposed straightforward
procedure; 0 0 0, measured (same configuration as in Figure 15).

4.2. PROCEDURE FOR THE ROUND DUCT

To solve equation (20) an iterative procedure was used in reference [7], starting with
the solution E from equation (49) for a locally reacting lining and inserting this solution
into the right side of equation (20). This procedure is rather time-consuming due to the
expense for the calculation of cylinder functions J, and Y, and due to the slow convergence
in the region of maximum attenuation [7].
Here, in a first step, the iteration is replaced by the prediction method as described
above. No reliable explicit solution at very low frequencies has been found. Therefore
the procedure is started at the high frequency end, where the solutions E are nearly
 CALCULATION OF ATTENUATION IN SILENCERS 121
independent of frequency and parameters r and A (compare with Figure 14). It is sufficient
to use a fixed averaged value, which has been determined by using the iterative method
in reference [7]:
EP=5*3+j1*18 at k&-15, EP-5*3+jl*44 at k,h-11.5. (56)
Inserting this Ep on the right side of equation (20) yields the first two solutions, which
serve as input for the first prediction. With the abbreviations from equation (50) the right
side of equation (20) is written as
Q = U~~J,(~)Y,Rl+A)~l-Y,(~)J,[(l+n)~]
r P
J,(JE7;---?iT)Y,[(l+A)m]-Y,(w)J,[(l+A)v]’ (57)
where the index r stands for round duct. Then equation (49) is solved with Qr instead
of j U. The thus obtained two solutions E serve as input to the first prediction according
to equation (53). This procedure works well for frequency steps of l/12 octave and for
values of the parameter r greater than r ==2.5 for A =0.5, r=3 for A = 1, and r-4.5 for
A = 2. For lower values of r the procedure becomes unstable. The reasons are the same
as described in section 4.1. But it is possible to use the predictive scheme also in the case
of lower values of r in the following way. First a set of solutions is calculated for the
minimal value of r given above and stored in a reference file E,. Then in a second run
with Ep = E,(i) the solutions for the wanted lower value of r are obtained.
The accuracy of this procedure has been tested by reproducing the attenuation curves
given in reference [7] with the non-iterative predictive scheme. The results are shown in
Figure 18. The dotted reference curves are masked by the solid lines for the straightforward
procedure, thus indicating good accuracy. Only at r = 2 and A = 1 and 2 is a deviation
visible.
The next step is to avoid the extensive calculations needed for the cylinder functions
in equation (57). In reference [7] it has already been mentioned that the use of
Qr = Um tanh (Am), (58)

valid for the slit-shaped duct (index s) instead of QI (equation (57)), gives at least a
rough evaluation for the round duct. Now a numerical comparison of Q, and QS has
been carried out for the range of parameters A = 0.5 . . .2 and r = 2.. .8. For high
frequencies the results are close together but they differ considerably for low frequencies.
Therefore some corrections C’ and C” are necessary, matching
Q,-C’Q:+jC”Q:. (59)
As a first approximation these corrections have been determined in respect to their
dependence on A and on a parameter F = A 1-1 including frequency and Tad:
C’=(1+O~5A){l-0~9exp[-(0~l/A)(F-5)2]}, (60)
C”= 1+0*9A/[l +(F/1d45)2’5]. (61)
The choice of a real parameter F is justified for low frequencies; in the region of the
maximum attenuation more sophisticated approximations, especially for C’, might be
found.
The accuracy test for the straightforward procedure with Q = C’Q: + jC”Q: according
to equations (58), (60) and (61) instead of j U in equation (49) is shown as dashed curves
in Figure 18. The agreement with the exact curves is sufficiently good for engineering
purposes. Only in the region of maximum attenuation is a slight deviation visible. This
is of little importance because in technical silencers such high attenuations of Dl z 8Ll h
seldom are reached due to acoustical by-passes.
122 W. FROMMHOLD AND F. P. MECHEL

o.,- 1 10 1 )O
fh(kHz cm)
Figure 18. Attenuation of a round silencer with a bulk reacting lining. . . . ., Exact iterative solution according
to reference [7], equation (20); -, straightforward procedure using the exact expression Q,, equations (49),
(53), (56) and (57); - - -, straightforward procedure using the approximated expression CQ,, equations (49),
(53), (56) and (58)-(61); 000, measured [7]. (a) r=2; (b) r=4; (c) r=8.

Compared to the exact solution as described in reference [7] the straightforward

procedure with the corrected expressions for the slit-shaped duct reduces computation
time by a factor of about 170 and simplifies programming considerably.

5. A STRAIGHTFORWARD PROCEDURE FOR SILENCERS WITH MEAN FLOW

5.1. RECTANGULAR SILENCER WITH LOCALLY REACTING LINING

To solve equations (24) and (26) an iterative procedure has been proposed in reference
[lo], starting with the solution E for the duct without flow in equation (26) and then
using explicit formulae from section 3. But iteration at every frequency point is superfluous,
if the prediction method developed in the previous section is applied to the solution of
equations (24) and (26) for E with
Q=jU[1-M~1-(l-M2)Ep/(koh)2]/(1-M2). (62)
 CALCULATION OF ATTENUATION IN SILENCERS 123
This Q then is used in equation (54). It is easy to obtain reliable starting solutions at
very low frequencies by using equation (30) and a “one step iteration”
E==jU[1-MJ1-(1-M2)jU/(k,h)2]/(1-M2). (63)
For the first two frequency points (about 20 Hz) equation (63) is used. From these values
the first prediction is made by using equation (53). This very simple and quick procedure
has been compared to the iterative procedure as mentioned above, stopping it at 1E(i) -
E(i - l)]/]E(i)l = 10P4. For a low flow resistivity E=5kNs/m4 and d=0*2m and h=
0.1 m, the relative error in Db reaches only 1% at U = *50 m/s, whereas for E = 15 kNs/m4
even at U = *lo0 m/s the error does not exceed 0.5%.
This simple straightforward procedure has been applied to the case of a silencer with
a locally reactive lining covered with a perforate sheet, described in reference [ll]. The
perforate sheet is treated as an additional series impedance in the calculation of G, in
equation (9). If the influence of mean flow is taken into account by a factor w in equation
(24), agreement with measurements is rather good, as shown in Figure 19. A factor of
w2 in equation (24), however, yields a much too high attenuation at U = -100 m/s.

zi
0 50 c U =-lo0
v/
m/s ,?O
7+/<
0 + 50

I I
0.5 1
f (kHz)

Figure 19. Attenuation of a silencer with locally reacting lining at mean flow velocities U; E = 11 kNs/m4,
d =0.075 m, h =0.035 m, L= 1 m. -, Calculated by using equations (53). (54) and (62); .& 0, V, +, 0,
measured [ 111.

5.2. RECTANGULAR SILENCER WITH BULK REACTING LINING

More tentatively, the straightforward procedure as described in section 4 has been
extended simply by a factor w as suggested in section 1.5. Equation (55) now is written
as
Q=U~tanh(A~)[l-M~l-(1-M2)Er,/(~h)2]/(l-M2). (64)
To obtain the two starting solutions needed for the first prediction, two values E, and
E2 are calculated from equation (52) at very low frequencies and then inserted into
equation (26), thus obtaining w, and w2. Finally the original solutions are corrected to
ET= E,w,, E; = E,w,. (65)
From these values E* the first prediction is made. The procedure is similar to the case
of a locally reacting lining and it works sufficiently well for small Mach numbers M.
124 W. FROMMHOLD AND F. P. MECHEL

The procedure was tested experimentally for a splitter type silencer consisting of three
splitters in a test duct of free area 1.2 x O-5 m*. The splitters had a thickness of 2d = O-2 m,
and were spaced 2h = O-2 m from each other and h = O-1 m from the wall of the duct.
The length of the silencer was 2 m. The duct opens into a reverberation room where the
averaged sound pressure level was measured with and without the silencer. The experi-
mental and numerical results are shown in Figure 20. The measured insertion loss does
not correspond exactly to the calculated attenuation. This is mainly due to the presence
of higher modes in the incident sound field produced by the array of loudspeakers and

50

40

2 3o

20

10

0
0.1 0,315 1 3.15
f(kHd

Figure 20. Attenuation of a splitter type silencer with a bulk reacting lining at low mean flow velocities, U;
H = 15 kNs/m4, d = 0.1 m, h = 0.1 m, L= 2 m, B = 0.5 m. -, Calculated by using equations (53), (54) and
(64) and factor w; . ., calculated by using factor w*; + + +, measured U = -10 m/s; 0 0 0, measured .!I = 0 m/s;
x x x, measured U = +20 m/s.

to higher modes produced by the splitters themselves. But here only the effect of mean
flow on the attenuation was of interest. The interrelation of measured data corresponds
well to the interrelation of calculated ones. Also in this case the use of a factor w (dashed
lines) seems to be closer to measurements than the use of w* (dotted lines). A compromise
could be found simply in the following way:
w-[l-KMJl-(1-Mz)E,l(k,h)*]/(1-M2). (66)
Here K is a free parameter between 1 and 2 matched to experimental data.

6. CONCLUSIONS

A short theory for the attenuation of a plane wave in rectangular and round silencer
ducts has been given for both locally and bulk reacting linings. The effect of mean flow
has been evaluated.
In the case of locally reacting linings the transcendental defining equation is transformed
into an algebraic one by the method of continued fractions expansion. It is shown that
an explicit quadratic solution can meet the requirements from a technical point of view.
For the case of bulk reacting linings a straightforward procedure is given by substituting
for the unknown solution in the expression describing the wave propagation inside the
 CALCULATION OF ATTENUATION IN SILENCERS 125

absorber a simply predicted value. Thus the problem is reduced to the case of locally
reacting linings and solved explicitly. The procedure starts at the low or high frequency
and where simple explicit solutions are available, and uses non-linear frequency steps to
keep the prediction error small.
Both simplified methods have been tested theoretically, and were compared to experi-
ments and provided good results.
The effect of mean flow can also be evaluated by a combination of the prediction
method with the explicit solution from continued fractions expansion. This procedure
has only been tested experimentally, and seems to be sufficient for low Mach numbers.
All of the calculation methods proposed in this paper need only complex algebra, are
easy to program and run for only a few seconds on a personal computer.

ACKNOWLEDGMENT
The support of this work by Bayerisches Landesamt fur Umweltschutz, Munchen, is
gratefully acknowledged.

REFERENCES
1. F. P. MECHEL 1976Acusticu 34,289-305. Explizite Niiherungsfonneln fiir die Schalldampfung
in rechteckigen Absorberkanilen.
2. W. B. JONES and W. J.THORN 1980 Continued Fractions. London: Addison-Wesley.
3. F. P. MECHEL 1976 Acustica 35, 210-213. Ausweitung der Absorberformel von Delany und
Bazley zu tiefen Frequenzen.
4. A. CUMMINGS 1976 Journal of Sound and Vibration 49,9-34. Sound attenuation in ducts lined
on two opposite walls with porous material, with some applications to splitters.
5. W. EVERSMAN 1975 Journal of Sound and Vibration 41, 252-255. Computation of axial and
transverse wave numbers for uniform two-dimensional ducts with flow using a numerical
integration scheme.
6. M. L. MUNJAL 1987 Acoustics of Ducts and Mujlers. New York: John Wiley.
7. F. P. MECHEL 1976 Acustica 35, 179-189. Die Berechnung runder Schalldampfer.
8. U. KURZE 1965 Acustica 15, 139-150. Untersuchungen an Kammerdampfern.
9. J. STOER 1976 Einfiihrmg in die Numerische Mathematik I. Berlin - Heidelberg - New York:
Springer-Verlag.
10. F. P. MECHEL 1986 Proc.Fortschritte der Akustik-DAGA ‘86, 701-704. Algorithmen der
Schalldlmpfer-Berechnung.
11. F. P. MECHEL and P. MERTENS 1963Acusticu 13, 154-165. Schallausbreitung in absorbierend
ausgekleideten Striimungskanllen bei hohen Windgeschwindigkeiten.