Documente Academic
Documente Profesional
Documente Cultură
net
Accurate model for the push-pull electrostatic loudspeaker.
Patrick De Visschere
University of Gent
Department of Electronics and Information Systems,
St. Pietersnieuwstraat 41, B-9000 Gent, Belgium.
pdv@elis.rug.ac.be
Abstract
A model is developed for the electrostatic loudspeaker, which incorporates
simultaneously the mechanical, the acoustical and the electrical behaviour
of the diaphragm. The non-linear model can be solved fairly accurate in
steady-state and is used for calculating the frequency- and level-dependent
sensitivity and distortion for an infinite strip push-pull electrostatic
loudspeaker. The results are compared with previously published and new
improved small-signal approximations.
1 Introduction.
The push-pull electrostatic loudspeaker (pp-ESLS) is known as a highquality loudspeaker. The basic characteristics of this transducer have been
studied by Hunt[1]. In this monograph, which was first published in 1954, a
one-dimensional model for the pp-ESLS (as well as for the single-sided
ESLS) is worked out. To handle the inherent non-linear behaviour of the
ESLS Hunt performed a harmonic analysis and by neglecting higher order
harmonics was able to calculate these harmonics consecutively. The radiation loading was incorporated by means of a mechanical impedance.
Although probably all relevant properties could be addressed in this way a
more accurate and more general model for the ESLS should at least be twodimensional in order to describe the vibration of the diaphragm properly.
Such an analysis must cope with two complications. Since the vibration of
the diaphragm is to a large extent determined by the air-reaction, the mechanical problem of the vibrating diaphragm must be solved together with
the radiation equation. This was done by Streng[2,3] for a circular membrane. In a subsequent paper the same author addressed the second complication[4], namely the dependence of the diaphragm-charge distribution on
the diaphragm-vibration. This effect is due to the finite diaphragm
resistivity and is responsible for a level-dependent sensitivity and for the
harmonic distortion. The analysis given was restricted to small-signals
however and focused mainly on the sensitivity question.
P.More
De Visschere
know-how on http://www.shackman.reromanus.net 1
P.More
De Visschere
know-how on http://www.shackman.reromanus.net 2
v+v
Eb = + a ,
r
(1)
(2)
where = 8.85 10-12 F/m is the dielectric permittivity of air. Using (1) we
find q in terms of the local deflection and the local potential :
q = 2
v r va
r2 2
(3)
(4)
v va r
(5).
r2 2
A more useful expression is found by eliminating the potential in (1) and (4)
using (3) :
Ef,b =
va
q
,
1
r
r 2
fe =
qva
q2
,
+
r 2 r
(6)
where the +signs holds for the frontside and the -signs for the backside.
These expressions have a straightforward interpretation. The first contribution to the fields and the force are due to the homogeneous field set up between the static electrodes by the signal voltages va and -va. The remaining
contributions are due to the image-charges induced by q in the static electrodes. These induced charges are easily found using Ramos theorem [5]
q
and are given by : q f ,b = 1 . With this the contributions to the
2
r
electric fields and to the electric force are easily found. The image force in
(6) corresponds with a negative stiffness for the diaphragm.
Due to the surface conductivity of the diaphragm, potential differences
will cause surface charge density variations according to :
P.More
De Visschere
know-how on http://www.shackman.reromanus.net 3
q
.
t
(7)
+ fe 2 p ,
2 = T 2 R
t
t
(8)
where p is the acoustical pressure at the frontside and where the term
involving R [kg/m2 /s = Rayl] is an optional damping term. Around the
perimeter of the diaphragm the boundary condition = 0 applies.
2.4 Acoustical model.
The pressure, which is anti-symmetrical with respect to the plane of the diaphragm, satisfies the wave equation :
2 p
1
p 2 2 = 0,
co t
2
(9)
2
p
,
2 = z
t
(10)
where z is the direction normal to the diaphragm (see Fig.1). In these equations o is the density of air (1.18 kg/m3) and c 0 is the wave velocity in air
(343 m/s).
3 The perfect pp-ESLS.
Ideally the conductivity of the diaphragm should be zero and the charge
2V
density should be fixed (and usually uniform) q = Q0 = 0 . In that
r
case, the electric force (see eq. (6)) becomes linear in the deflection and the
applied voltage :
Q02
Q0
.
v +
fe =
r a 2 r
(11)
P.More
De Visschere
know-how on http://www.shackman.reromanus.net 4
(12)
c 2
= c0 2 + 2e,
2
i
2 V02 a2
=
3r T
(13)
2
e
p1
=
m
z
2m =
1,
T
.
2 a 0 c02
(14)
A method for solving this problem, is published elsewhere [6]. We just want
to mention that we do not assume, as was done in [2], that the tangential
p
p
derivative of the pressure (
for an infinite strip or
for a circular
r
x
diaphragm) vanishes along the perimeter. Furthermore the pressure on the
diaphragm p1 and the diaphragm deflection 1 are obtained as
developments into suitable base-functions or modes :
p1 = Va
k=0
1,k k(x)
1 = Va
k=0
1,k
k(x) ,
(15)
where the k(x) are the eigenfunctions of the diaphragm. In what follows
numerical results will only be presented for the strip and the base-functions are then given by :
k(x) = T2 k + 1
1 x2 ,
(16)
P.More
De Visschere
know-how on http://www.shackman.reromanus.net 5
k = (2 k + 1) .
2
(17)
1 + j r T e j k r
cos ()
r
2 a3 2
p1(x) d x .
(18)
Using (15) and (16) it can then be shown that the mean pressure depends on
the first pressure coefficient P1,0 only, namely :
p1(r,) = 1 + j
V r T e j k r cos () P .
1,0
r
8 a a3 2
(19)
Apart from the 1/2 factor in (19) the frequency response of the perfect
pp-ESLS is thus given by P1,0 .
In a more realistic pp-ESLS the diaphragm will possess some conductivity
0, and then q will not longer be remain fixed but will instead adapt to
the vibration of the diaphragm. Mathematically this means that, in addition
to the previous equations, eq.(7) and its boundary condition, and relation (3)
must now be taken into account to. The main difficulty arises from the nonlinearity of relation (3) and (6) and a general solution seems unattainable.
However there are two possible simplifications we can make. In section 5
we again consider a sinusoidal regime and perform a harmonic development
of the quantities involved, as has been done by Hunt[1] for a simple onedimensional model. Since this is a steady-state model all information
regarding the temporal behaviour of the system is lost however. Therefore
we consider in section 4 a small-signal analysis, where that information is
retained.
First we consider the static stability of the diaphragm assuming a fixed di2 V
aphragm charge density Q0 = 0 0 . It is easily verified with eq. (8) and
r
(11) that the equilibrium position (va = 0, p = 0, = 0) is stable only if :
P.More
De Visschere
know-how on http://www.shackman.reromanus.net 6
e < min ,
(20)
va v
.
1 2
(21)
(22)
and now the third order term on the left q 1 2 can be neglected with respect to all other terms, leading to :
q = v + 2 + v a .
(23)
2
v = 2 + va
t
t
r
.
4 a c0
(24)
P.More
De Visschere
know-how on http://www.shackman.reromanus.net 7
2
v0(0) = 1 1 + Va Re 1
2
v0 = 0,
2
2
q0 = v0 + 1 1 + Va Re 1
2
(25)
Eventually the potential variation v0 will become zero again and the
charge density variation will become equal to :
q0 =
2
1
1 + Va Re (1) .
2
(26)
1 = Va H 1,0 cos ( x)
2
(27)
q0 = Va
k=0
0,k
(1 e
k t
) cos (k x) ,
(28)
8
Q0,1 = 30 H1,0 ,
(29)
These are shown in Fig.3 in full lines (for comparison the dashed curve
shows Q0,0 when the quadratic term in the deflection is neglected). Note that
Q0,0 vanishes for 0.1 . For this particular frequency the 2nd mode dominates the charge variation. The charge density variations build up with a
typical time constant for each mode. For the infinite strip the eigenvalues
are given by (17), and with -1 = 1000 M, the lowest time-constant equals
31.1 ms, corresponding with a cut-off frequency of 5.1 Hz. The cut-off frequencies for the higher modes are then 46 Hz, 128 Hz, 250 Hz, . For a
circular diaphragm (radius a) they are given by n = j0 n (n = 1, 2, ),
where j 0 n are the zeros of the Bessel function Jo(z) (j01 = 2.40, j02 = 5.52,
j 03 = 8.65, j 04 = 11.8,). For a circular diaphragm with diameter
2a = 50 cm (which has nearly the same area as the previous strip with a
height of 1 meter) the frequencies are 1.9 Hz, 10 Hz, 25 Hz, 46 Hz, . On
the strip with its poor aspect ratio charge does not have to travel so far in
order to reach equilibrium, which explains the faster response. This is also
the reason why the higher modes build-up much faster.
P.More
De Visschere
know-how on http://www.shackman.reromanus.net 8
2
1
v2 = 12 + Va 1
2
.
1 2
q2 = v2 + 1 + Va 1
2
v2 + j
(30)
With the low-frequency expression for the deflection (27) the solution is
easily found. For < < , this solution is given by :
2 2 H1,0
2
cos ( x) + H1,0 h(x)
q2 = j Va
2
2
4
.
h(x) = cos ( x) + exp (1 + j)
(31)
/ 2 (1 x )
The second contribution, which is due to the quadratic terms in (30), can
again not be neglected and has a somewhat peculiar behaviour ( h(x) is
almost everywhere equal to cos ( x) except near x = 1 , where it drops to
zero; this transition occurs more abrupt if becomes smaller). Note also
that with -1 = 1000 M ( = 1.9 10-3), and for = 0.05, q2 is of the order of 4 % at the maximum drive level. We can expect third harmonic distortion components for the deflection and the pressure being of the same order.
5 Steady-state analysis.
5.1 General theory
We consider a sinusoidal signal voltage :
e j t + e j t
.
va = Va
2
(32a)
Due to the non-linear terms in eq.(3) and (6) all relevant quantities (q, v, ,
fe , p) will contain higher order harmonic components. However the diaphragm charge density (q) and the diaphragm voltage (v) will only show
even harmonics :
1
u=2
j2n t
,
u2n e j2n t + u2n
e
(32b)
n=0
P.More
De Visschere
know-how on http://www.shackman.reromanus.net 9
u=2
j[2n + 1] t
u2n + 1 e j[2n + 1] t + u2n
,
+ 1e
(32c)
n=0
f1 = q0 Va +q0 1 ,
(33)
f3 = q2 Va +q0 3 + q0 q2 1 .
(34)
Va
( 1 + 1 ) +
2
1
1 q0 ,
2
(35)
2
1
1 2
1
Va 1 + 2 1 q2 + 2 1 q0 .
2
(36)
(37a)
P.More
De Visschere
know-how on http://www.shackman.reromanus.net10
(37b)
This should be compared with the small-signal result (26). The fundamental
of the electric force (33) can then be written as :
f1 = Va + 1 + f1 ,
(38a)
with :
f1 = q0 Va + 1 2 + q0 .
(38b)
The first two contributions in (38a) correspond with the perfect case
considered in section 3. Finally the first order equation (A2) reduces to :
2
j R + 2i 1 = p1 2e Va + f1 ,
(39)
which, apart from the non-linear force correction f1 , is equivalent with the
ideal equation (13) (eq. (14) remains unchanged). These equations can be
solved by iteration. We can still use developments (15), but should keep in
mind that the coefficients P1,k, H 1,k do now depend on Va , due to the nonlinearity. Fig.4 shows this dependence for the absolute value of the first
pressure coefficient for a number of frequencies. Below the first resonance
the response increases with the driving level, above the first resonance it decreases slightly. Note also that for = 0.05 run away occurs for an excitation beyond Va 0.75 . The coresponding DC charge variations are shown
in Fig.3 with the dash-dotted lines for two drive levels. As for the smallsignal approximations shown in the same figure, the DC charge variation is
expanded into the diaphragm modes :
2
q0 = Va
k=0
0,k
cos (k x) .
(40)
The discrepancy manifested in the upper frequency region in Fig.3b, between the accurate curves and the small-signal approximations is due to the
higher order deflection modes becoming important (see Fig.2b). These have
not been taken into account by the small-signal analysis.
P.More
De Visschere
know-how on http://www.shackman.reromanus.net11
(41a)
with :
f3 = q0 (q0 + 2) 3 +
1
V + q0 1
2 a
q2 .
(41b)
Substituting this in (A2) with n = 1, gives the equation for the third order
deflection :
j 3 R + 2i,3 3 = p3 2e f 3
2
c 2
= c0 23 + 2e,
(42a)
3 = 3
2
i,3
p3
= 3
m
z
+ 23 p3 = 0,
2
3 .
(42b)
The 2nd order charge density in (41b) is found from (36) and (A1, n = 1) :
1
1 2 1
q2 = j
q2 = v2 + 2 1 (Va + q0 1) ,
2
v2 ,
(43a)
(43b)
2
v2 +
v2
1
1 2 1
1 1 (Va + q0 1)
.
2
1
2
1 2 1
(43c)
P.More
De Visschere
know-how on http://www.shackman.reromanus.net13
P.More
De Visschere
know-how on http://www.shackman.reromanus.net14
References
[1]
[2]
J.H. Streng, Calculation of the surface pressure on a vibrating circular diaphragm in free space, J. Acoust. Soc. Am. 82 (2), August
1987, pp. 679-686.
[3]
[4]
[5]
[6]
[7]
P.More
De Visschere
know-how on http://www.shackman.reromanus.net15
2
v2n = n q2n ,
(A1a)
v0 = 1 .
(A1b)
The mechanical equation (8) and the acoustical equation (9) become :
c 2
2
j 2 n + 1 R + c0 22 n + 1 2 n + 1 = p2 n + 1 2e f2 n + 1 ,
(A2)
and :
+ 22 n + 1 p2 n + 1 = 0 ,
2
(A3a)
p2 n + 1
= 2n+1
m
z
2 n + 1 .
(A3b)
2
2n
(A4a)
+ q 2n + 1 .
f2n + 1 = q va
2n+1
(A4b)
The first contributions on the RHS are rather simple (in the remainder of
this appendix formulas are expressed with the original variables; it is easily
seen that normalised quantities obey the same relations) :
Va
( 2n 1 + 2n + 1)
2n 2
Va
va = ( 1 + 1)
0 4
va
(n 0)
(A5a)
P.More
De Visschere
know-how on http://www.shackman.reromanus.net16
(n 0)
(A5b)
The remaining contributions in(A4) however contain terms involving all the
harmonics of q and . This makes the resulting system of equations difficult
to solve. Fortunately higher order harmonics are supposed to be small with
respect to lower ones (|3| << |1|, |q2| << |q0| and so on). Following Hunt[1]
we retain only the dominant terms and (A5) then simplify :
va
2n
va =
0
Va
Va
4
2n 1
( 1 + 1
Va
q
2 2n
q va 1 = Va q0
q va 2 n + 1 =
(n 0)
(A6a)
(A6b)
(n 0)
The second contributions in the RHS of (A4) now become much simpler :
1
2 q 2n = 2
l=0
2 2l q2[n l] +
1
1 2
2
2n q0 + q2n 1 ,
4
2
(A7a)
with :
1
2 2n = 2
2 0 =
1
2
n1
2[n l] 1 2l + 1
(n 0)
l=0
(A7b)
and :
q2
n
1
1
2
q 2 2[n l] 2l + 1 ,
q
+
=
2n + 1 2 2n + 1 0 2
l=0
(A8a)
with :
1 n
q 2n = q2n q0 + q2[n l] q2l
2l = 0
2
q 2 0 = q0
(n 0)
(A8b)
P.More
De Visschere
know-how on http://www.shackman.reromanus.net17
Figure captions
Fig.1 Model for an idealised push-pull electrostatic loudspeaker.
Fig.2 The normalised response of an ideal strip pp-ESLS. Shown are the
absolute values of the first three modes for (a) the diaphragm
pressure, and for (b) the diaphragm deflection. Parameters values are
shown in Table 2.
Fig.3 The normalised (a) first and (b) second mode of the steady-state DC
charge density variation for sinusoidal excitation. Shown are two
small signal approximations (full lines and dashed lines) and exact
results for two drive voltage levels (dashed dotted lines).
Fig.4 The variation of the absolute value of the first mode of the
diaphragm pressure with the sinusoidal driving voltage level for
different frequencies.
Fig.5 Third harmonic distortion as a function of driving voltage level for
different frequencies and for a diaphragm resistivity of 1000 M.
Fig.6 The absolute value of the first, third and fifth mode of the third
harmonic diaphragm pressure normalised with the first mode of the
fundamental pressure as a function of frequency. Diaphragm
resistivity equals 1000 M.
Fig.7 Comparison between the steady-state (full line) and the immediate
(broken line) third harmonic distortion. The immediate value is
valid insofar as the DC charge density has not yet adapted to the
driving level. Diaphragm resistivity equals 1000 M.
P.More
De Visschere
know-how on http://www.shackman.reromanus.net18
Quantity normalised
with
voltage : v, va, Va
V0
charge density : q
Q0 =
2 V0
r
deflection :
cordinates : x,
time : t
a
2 c0
electric force : fe
2 V02
2r
electric field : E
V0
r
pressure : p
r T
2 a2
mechanical damping : R
T
a c0
P.More
De Visschere
know-how on http://www.shackman.reromanus.net19
a
T
10 cm
100 N/m
0.02 kg/m2
V0
2000 V
2.3 mm
50 Rayl
109
c/c0
0.206
0.763
0.06
1.89 10-3
P.More
De Visschere
know-how on http://www.shackman.reromanus.net20