Sound
Propagation in Ducts
Hongbin Ju
www.aeroacoustics.info
hju@math.fsu.edu
In this section we discuss sound
propagation in ducts. This is a typical case of acoustic waves in a confined
environment. In the wall bounded directions, acoustic waves bounce back and forth
to form standing wave patterns. In the axial direction acoustic waves propagate
freely. We will show how to solve acoustic equations for these boundary conditions.
The theory developed is very useful in applications such as the prediction and
control of aircraft engine noise.
Sound Propagation in Rectangular Duct
Assume a uniform
subsonic mean flow
in a rectangular duct with width d
and height h. The mean flow density is
, pressure , and velocity in +x-axis direction.
The scales used are: length L, velocity: sound speed , time , density: ambient density , pressure: , impedance: . The linear Euler
equations for the fluctuation quantities are:
,
(0-1)
,
(0-2)
,
(0-3)
. (0-4)
Variables
without bars are perturbation quantities. Isentropic process is assumed;
therefore density
can be calculated from pressure: , .
We use the trial
solution and the
method of separation of variables to
solve the equations. In the y- and z- directions there are wall boundaries.
In the x- direction the duct extends
to infinity. Assume the next form of solutions:
(1)
Substituting (1) into the linear Euler
equations, one obtains the following acoustic equations:
, , , (3)
when . The linear Euler equations with the isentropic process have two normal modes: acoustic mode (, phase speed not equal to flow velocity), and vortical mode (, phase speed equal to flow velocity). We are only interested in the acoustic modes; therefore and the acoustic equations in Eqs.(2&3) are chosen according to cht1.doc.
Rigid Wall Duct with Mean Flow
If the duct walls are rigid, the boundary conditions are
, at and ; (4)
, at and . (5)
In the flow direction the wave
amplitude must remain finite as .
We try the separation of
variable:
.
Substituting it into (2) leads
to:
.
The first term on the left
depends on y only. The second term
depends on z only. is constant with respect
to y and z. Therefore the following equations must be true:
, , .
(5-1)
and are constants. One can
use the trial solution method to solve the two ODEs in (5-1). The general
solution is:
. (6)
To satisfy boundary conditions
(4) and (5), we must have
, ; (7)
, . (8)
m and n can be any integers. Usually we use non-negative integers: , . Due to the wall confinements, wave numbers and can no longer be any real numbers. They must be multiples of and respectively. For this reason, and are called eigenvalues. With (7) and (8), solution (6) is:
It represents the standing waves in both the y and z directions. (Compared with the plane wave solutions discussed in cht1.doc.) Amplitude is to be determined by acoustic sources in the duct and the boundary conditions in the x direction. and are called the eigenfunctions. Each pair of (m,n) represents one acoustic mode with the solution (9) to Eq.(2). It is called a normal mode of the acoustic field.
According to Eq.(5-1), (7), and (8), eigenvalue is determined by the duct geometry:
. (10)
Eq.(2) implies only matters; therefore only the positive square root is chosen. is real and positive for the rigid wall duct. For each eigenvalue , there are two solutions of from Eq.(2):
They represent waves propagating in opposite directions in x. The direction of a wave is determined by group velocity. If , is real and the mode is cut-on. Taking the derivative with respect to in Eq.(11), we obtain the group velocity
[hj1] .
Since
, the wave with propagates in the -x direction. Noting
,
,
the wave with propagates in the +x direction.
If , we have
,
, (11-0)
.
i, not –i, is chosen for the square root so has positive imaginary
part and has negative imaginary
part. The imaginary part of provides exponential
attenuation factor in in Eq.(1), with which the
wave amplitude remains finite as . represents an evanescent or
cut-off mode.
The mode becomes cut-on if the frequency increases so that:
. (11-1)
is the cut-off frequency of mode (m,n). The cutoff ratio is defined as
(Hubbard1995, p.159):
. (11-2)
represents
propagating, or cut-on, modes. The
number of cut-on modes is limited at any frequency. A mean flow makes a mode easier
to cut on. A plane wave is always cut-on: , , , .
It is noted that the eigenvalues
and eigenfunctions in the y or z direction are solely determined by the
boundary conditions. They are independent of wave propagating directions. This is
not true when there is mean flow and the walls are not rigid, as will be discussed
later.
Substituting (9) into (1) and summing
up all possible modes, we obtain the general solution of acoustic pressure:
Acoustic velocities can be calculated
from Eq.(3). This solution represents a general sound field in a rectangular
rigid-wall duct. Any specific pressure field can be expanded to the sum of normal
modes of the rectangular duct. The amplitude of each mode is
determined by the sound sources and the boundary conditions in the x direction. Mode (m,n) is present in the
sound field, or excited by the source, if .
Given the total pressure field , how to calculate amplitude of each individual mode? It is noted that eigenfunctions are orthogonal:
The same is true for . One can calculate the jth component of pressure using the orthogonality. Applying (12-1) on (12) and integrate the equation on the duct section at x, we have:
Modes (m,n) are obtained by filtering out all other modes through the integration.
Depending on locations of sound sources and duct terminations, one can further
separate amplitudes and . For
example, if sound sources are at the left side of the section and at the right
hand side the duct extends to infinity (no reflection from this end), then
(12-3)
If waves propagate towards this
section from both sides, the integration (12-2) at two axial locations and are needed and and can be solved from the
linear equations:
.
For a rigid-wall duct, is real so can be calculated
by (11) for cut-on modes
and (11-0) for cut-off modes. In many practical problems such as in lined
ducts, is complex. Then branch
cuts must be inserted for and in (11-0) to make them single-valued. The branch cuts
must be chosen so that the imaginary part of
is not negative so the magnitude of the wave
is finite as .
We choose the branch cuts as
shown in Fig.1, with which the phase range of is . The
phase range for in (11-0) is so the amplitude of
the right-propagating wave with factor remains finite as . The phase range for is and amplitude of the left-propagating
wave remains finite as .
Fig.1, Branch cuts for and on the -plane[hj2] .
Sometimes we need to calculate eigenvalue from wave number :
, (13-1)
, .
Branch cuts for and are chosen as in Fig.2. The phase range of is . The
phase range of is so the real part of is always positive.
Fig.2, Branch cuts for and on the -plane.
Soft Wall Duct without Mean Flow
To simplify the problem, we assume no mean flow in the duct and the same impedance in the opposite walls. The boundary conditions are:
at , at ; (13)
at , at . (14)
and are admittance of the
soft walls.
We can still use the trial
solution Eq.(6). To facilitate the derivation, we rewrite Eq.(6) as:
(15)
by defining , , , and .
To satisfy boundary conditions
(13) in the y direction, we must have
; (16)
. (17)
From which we have,
. (18)
Eigenvalue can be solved from
this equation using the grid search/Newton iteration method.
Another way to solve equations (16) and (17) is to note:
. (19)
One possible solution is , then:
,
, (20)
which corresponds to the even symmetric solution in Morse&Ingard1968, p.504.
The other possible solution to Eq.(19) is , i.e.,
,
.
(21)
which
corresponds to the odd symmetric solution in Morse&Ingard1968, p.504.
Similarly can be solved.
Eigenvalue or is the same for the
wave propagating in +x or -x direction. However, eigenfunctions or are no longer
orthogonal as in the hard wall case.
If
there is a mean flow in the soft wall duct, eigenvalues
and are no longer
independent of the wave propagation direction in x. They are different for the different propagation directions.
This will be further discussed in the following section about the soft wall
circular duct.
Sound Propagation in Circular Duct
Scales used here are: length:
duct diameter D, velocity: sound
speed , time: , density: density , pressure: , impedance: .
The mean flow: and are constant; mean
velocity is in +x-axis direction and constant.
The Euler equations for the
fluctuation quantities are:
, (21-1)
, (21-2)
, (21-3)
.
(21-4)
(Density can be calculated
directly from pressure: , .)
In the radial direction there is
a wall boundary and in the circumferential direction there is a periodic
boundary. In the x direction the duct
extends to infinity. We assume the form of solutions:
. (22)
Substituting it into the Euler
equations, we have:
, (23)
, , . (24)
where . is the dispersion relationship for circular/annular ducts.
To solve from Eq.(23), assume the separation of variables:
. (25)
Substitute it into Eq.(23),
.
The left hand side of the equation
depends on r only. The right hand side
is only a function of . Then they must be equal to the same constant:
, (26)
. (27)
The general solution to Eq.(26) is:
. (28)
A, B and are to be determined by two boundary conditions in the direction.
Solutions to Eq.(27) are Bessel functions. It is either Bessel function of the first kind representing standing waves in the radial direction in a circular duct, or the second kind for standing waves in annular ducts, or Hankel functions and (the third kind of Bessel functions) representing propagating waves (outward/inward, no ducts). The general solution in the duct is:
. (29)
, are
respectively the th order first and second kinds of Bessel functions. They are
two independent solutions to Bessel equation (27). C, D and are determined by two
boundary conditions in the radial
direction.
For
a circular duct, the periodic boundary condition
applies in direction. The general
solution to Eq.(26) is:
,
(30)
where is an integer. In the
radial direction, the boundary condition at is that must be finite. is eliminated since it
is infinitive at . Therefore the general solution of is
.
(31)
The general solution of for a circular duct
is:
. (32)
is to be determined by the boundary condition at . ( is the radius of the duct.) All other variables can be computed from by Eq.(24).
As |m| increases, is small for small r and large for large r. Sound energy concentrates near the cylindrical duct wall for high |m| mode.
Rigid Wall Circular Duct
For a rigid wall duct, the boundary condition at is
.
(33)
To satisfy (33), one has the
eigenvalue:
, (34)
where is the nth root of . (See Appendix for how to find .) Since , eigenvalues are the same for : . And eigenfunctions satisfy: .
The
time domain solution of p is:
. (35)
and are the same for both directions in x. is orthogonal. (This is not true when there is mean flow and the duct wall is not rigid.) Wave number can be computed from Eq.(11) and the cut-on frequency from Eq.(11-1). For larger azimuthal mode (larger |m|), is larger, so is the cut-on frequency. Therefore there is less number of cut-on radial modes for larger |m|.
Soft Wall Circular Duct
Assume a plug flow in the duct. The tangential discontinuity at the disturbed surface leads to the continuity of displacement at the wall (cht21.doc) and the Myer’s impedance boundary condition (cht24.doc):
at . (35-1)
Plugging in Eq.(32) and in Eq.(24) into the Myer’s boundary condition, we have:
. (35-2)
It is not trivial to solve eigenvalue from this equation. The grid search/Newton’s iteration method can be used. With mean flow, is coupled with . Therefore, it is no longer independent of the wave propagation direction as in the no flow soft wall case or as in the solid wall with mean flow case. is different for waves propagating in +x and –x direction for soft wall duct with mean flow.
Sound Propagation
in Annular Duct
An annular duct has an inner circular wall
with radius and an outer circular
wall with radius . The equations are (21-1)~(21-4), with scales: length: outer
duct diameter ,
velocity: sound speed , time: , density: density , pressure: , impedance: . And the solution form is (22).
The
analysis of the periodic
boundary condition and the
general solution (30) for the circular duct still apply in the annular duct. In the radial
direction, is not in the physical
domain as in the circular duct. Therefore in Eq.(29) should be
kept in the general solution:
. (36)
C, D and are to be determined by two boundary conditions at and . With , sound energy doesn’t concentrate near the outer duct wall for high mode m as in a cylindrical duct.
Rigid Wall Annular Duct
For a rigid wall annular duct, the boundary conditions are
, at and . (37)
First of all, one should check if
is the eigenvalue. Since
, must be zero and . Since for , for , the only nontrivial solution to Eq.(36) is:
, . (38)
This represents a plane wave. The
plane wave in a circular or annular duct only propagates in the axial direction.
For all the cases other than and , we have . From boundary condition (37), the dispersion relation is
obtained from (36):
. (39)
The prime denotes the derivative with respect to the whole argument. It is not trivial to solve eigenvalue from Eq.(39). When the gap of annular duct is much smaller than , we can obtain an approximate solution of . We begin with the Bessel equation (27). By replacing by the medium radius , the Bessel equation becomes a second order ODE with constant coefficients:
. (39-1)
Its general solution is:
, .
and are both real, or a conjugate pair.
To satisfy boundary conditions (37),
,
i.e.,
Therefore the eigenvalue for an annular duct with small gap is
, n=0, 1, 2, ..... (40)
According to Eq.(11-2), for a cut-on mode in a narrow annular duct, n must be zero. Therefore
[hj4] .
For large m,
.
As , does the narrow annular approaches a rectangular duct? Compare Eq.(40) with Eq.(10). The radial wavenumber in the annular duct approaches the height mode, . The spinning mode wavenumber approaches the depth mode. However, there is an extra wavenumber in the annular duct. It comes from term in Eq.(39-1), which does not exist in the rectangular duct Eq.(5-1). If m is kept constant as , then . It is equivalent to regular duct with uniform mode shape in the depth direction. On the other hand, if remains constant as , then and . The annular duct approaches a rectangular duct. For the mode shape, as , , Eq.(36)
,
.
Compare it with Eq.(12), it approaches rectangular duct height mode shape.
Another asymptotic formula for the eigenvalues based on the WKB method was developed by Envia. (Envia1998, Eq.9)
One may solve Eq.(39) numerically using the Newton iteration method. Define function:
[HB5] .
For initial eigenvalue , the improved eigenvalue is , . This process is repeated until is small. A successful Newton method depends on a good initial eigenvalue . The next figure shows a typical . Separate -axis into intervals, ..., ,,.... If , is a good choice of . A FORTRAN90 code, modenumber_annularduct.f, using this method to compute eigenvalues in a solid wall annular duct is attached.
m=18, , .
One can show that . Therefore eigenvalues for m are also the eigenvalues for –m: . Eigenfunction for (-m,n) is the same as for (m,n).
A general solution of is then:
, m=n=0;
, other. (40-1)
Velocities are computed by Eq.(24). Wave number can be computed from Eq.(11) & (11-0).
The
time domain solution of p is:
Orthogonality:
According to Abramowitz&Stegun1964, Eq.(11.4.2), p.485,
Cut-on ratio
For annular ducts the cut-on ratio in
Eq.(11-2) can be rewritten as
,
, .
is the phase speed in
the circumferential direction. is called the cut-off
Mach number. It is roughly the ratio of radial wavenumber and the
circumferential wavenumber. For narrow ducts, . This is true for any ducts. For narrow ducts and large m, . For arbitrary annular ducts, depends on the hub/tip
ratio and m:
If for large m in narrow ducts without axial mean flow, the phase speed of the
mode in the circumferential direction must be supersonic to be cut-on.
In
the outlet of jet engines, the annular duct is often separated by
installations, as illustrated by Fig.3. The duct has a C-shape. It is here
called C-duct. The C-duct has an
inner duct with radius and an outer duct with
radius. The separation plates are at and .
Fig.3, C-duct.
The first major difference of
C-duct from the annular duct is the periodic boundary condition is no
longer applicable in the direction. Because of
the presence of the bifurcations, spinning duct modes are not possible.
Reflections at the bifurcations lead to the formation of a standing wave
pattern in the cross-section of the duct. Similar to the argument for Eq.(9) in
rectangular duct, the general solution of is:
, (42)
where
, . (43)
m is any nonnegative integer. may not be an integer
as in the circular or annular duct unless when . (When , the solution in a C duct is the same as that in an annular
duct.)
In the radial direction the
general solution is:
. (44)
The second major difference of
the C-duct from annular duct is the Bessel functions here may have non-integer
orders.
Rigid Wall C-Duct
For the rigid wall C-duct, the boundary conditions in radial direction are
, at and . (45)
The similar argument as for
Eq.(38) leads to the plane wave solution:
, .
(46)
If , the next dispersion relation can be derived from (44) and
(45):
.
(47)
Eigenvalue is to be determined by
this equation. The difficulty involved here is the Bessel functions with
non-integer order.
From the similar procedures for Eq.(40), when the gap of annular duct, , is small, the eigenvalue is approximated by:
, n=0,1,2,..... (48)
If the gap of the annular duct is not small, we can use (48) as an initial value in a Newton iteration method. The final results need to be reorganized.
The
time domain solution of p is:
. (49)
With solution of , all
other variables can be computed from Eq.(24).
Scales used will be: length: duct
radius R, velocity: sound speed , density: ambient density , pressure: , impedance: .
The mean flow: and () are constant; Shear mean velocity is in +x-axis direction.
Assume the next form of
solutions:
(50)
(Density can be computed directly
from pressure: .)
From Euler equations, one can
obtain the next equations for shear flow:
,
(51)
,
(52)
where . For a wall with impedance Z, the boundary condition is
, at .
(53)
Equations (51) and (52) and
boundary condition (53) form an eigenvalue problem. Only some special values of
can satisfy these
equations. These are eigenvalues, and the respective
functions and are eigenfunctions.
An important property of
eigenvalues of Eqs.(51)&(52) is given here. Suppose , , are a set of
eigenvalue and eigenfunctions of this problem, then , , form a set of
eigenvalue and eigenfunctions for Eqs.(51)&(52) satisfying the next B.C.:
, at .
(54)
Here
superscript * represents conjugate. Under some circumstances B.C.(53) is exactly the same as
B.C.(54), such as for hard wall, , or the impedance with only resistance, . Then , , are also the set of
eigenvalue and eigenfunctions for the original problem [Eqs.(51)&(52) with
B.C.(53)].
Newton
iteration method
can be used to solve the eigenvalues and eigenfunctions. For an initial guess
of the eigenvalue , integrate Eqs.(51)&(52) numerically from the outside of
the boundary layer to to get and at . and are functions of : , , which may not satisfy B.C. (53) at . Similarly we can have another solution and for initial eigenvalue
near . Suppose
the exact eigenvalue is , i.e.,
.
To the first order of Taylor
expansion, we have:
.
If we use , , we can get to have a better
approximation to the accurate eigenvalue. The process can be repeated until
certain accuracy is achieved. This is the Newton’s iteration method.
Choosing appropriate initial
value is critical for a
successful Newton’s iteration method. Eigenvalues of plug flow without boundary
layer, Eq.(34) and Eq.(11), are good choice of the initial eigenvalues. In most
of the time these choices work well. But in some cases Newton’s method fail
when using these initial eigenvalues.
Another way for choosing initial
eigenvalues is the grid search.
Compute function:
(55)
for a matrix in -plane, draw the contours of real(f)=0 and imag(f)=0 in the
-plane using a commercial software. The intersection points
in the -plane should be the solution of the eigenvalues. Read in
these data and use them as the initial eigenvalues in the Newton iteration.
As an example, let's find eigenvalues for mean flow with turbulent boundary layer in a circular rigid wall duct with radius: 62.33”. The velocity profile of the turbulent boundary layer is shown in Fig.4. Mach number: 0.452, boundary layer thickness: 1”,
Fig.4, Velocity profile of
turbulent boundary layer, Mach number: 0.452, boundary layer thickness:
1”.
The grid search method is used to
get the initial guess of eigenvalues. Since the wall is rigid, both and are the eigenvalues
according to the property discussed above. Therefore in Fig.5 only the up half
of the -plane is needed. Read in the data of the intersection points
of solid and dashed lines as the initial values in the Newton’s method.
Fig.5, Contours of real(f)=0 (solid lines) and imag(f)=0 (dashed lines) in the -plane
The next figures show the
results. Fig.6 shows the eigenvalues when frequency is just above the cut-on
frequency. Fig.7 is for frequency is well above the cut-on frequency. Fig.8 is
for frequency well below the cut-on frequency. In most of the cases, the effect
of shear flow is to move real parts of plug flow eigenvalues to right while the
imaginary parts keep nearly the same. This change is very small. Using plug
flow eigenvalues as initial values in Newton method works well. The only exception is when the
frequency is just above the cut-on frequency (Fig6). There are two cut-on modes
in the plug flow. The boundary layer makes these propagating modes into cut-off
modes. This explains why a cut-on mode wave was damped in experiments. In this
case the plug flow eigenvalues as the initial value fails.
Fig.6, Eigenvalues at 741Hz for plug flow (square) and shear flow with turbulent boundary layer (triangle). Mach number: 0.452, boundary layer thickness: 1”, duct radius: 62.33”, azimuthal mode number m=22, cut-on frequency for the 1st radial mode of the plug flow: 740.6Hz. [Real() in the right figure is rescaled to show the difference.]
Fig.7, Eigenvalues at 800Hz for plug flow (square) and shear flow with turbulent boundary layer (triangle). Mach number: 0.452, boundary layer thickness: 1”, duct radius: 62.33”, azimuthal mode number m=22, cut-on frequency for the 1st radial mode of the plug flow: 740.6Hz.
Fig.8, Eigenvalues at 730Hz for plug flow (square) and shear flow with turbulent boundary layer (triangle). Mach number: 0.452, boundary layer thickness: 1”, duct radius: 62.33”, azimuthal mode number m=22, cut-on frequency for the 1st radial mode of the plug flow: 740.6Hz. [Real() in the right figure is rescaled to show the difference.]
A numerical integration method
can also be used.
Group
velocity is defined for a wave packet. waves with frequency range (omega-domg,
omega+domg) make a wave packet, no matter how small domg is. Then the waves
propagate in group velocity. So as domg --> 0, the single wave also
propagates in group velocity.
Otherwise
it will be weird: a wave packet propagates at group velocity as long
as domg =/0. Whenever domg=0, it suddenly propagates at a different velocity.
In
nature there is no 'pure' wave. Any wave at one frequency has actually a small
range of frequency. Therefore the 'pure' wave propagates in group velocity.
In turbomachinery noise study we often define this cut-off Mach number:
.
For narrow ducts, . Actually this is true for any ducts.
For narrow ducts and large m,
.
For arbitrary annular ducts,
In turbomachinery noise the soure frequency is . The cut-on ratio
is the phase speed in the circumferetial direction.