Sie sind auf Seite 1von 51

A FAMILY OF LINEAR-PHASE CROSSOVER NETWORKS 1801(I-5)

OF HIGH SLOPE DERIVED BY TIME DELAY

Stanley P. Lipshitz and John Vanderkooy


University of Waterloo
Waterloo, Ontario, Canada

Presented at
the 69th Convention
1981May 12-15 __/
LosAngeles
Thispreprint has been reproduced from the author'sadvance
manuscript, without editing,,corrections or consideration by
the Review Board. The AES takes no responsibility for the
contents.

Additional preprints may be obtained by sending request


and remittance to the Audio Engineering Society, 60 East
42nd Street, New York,New York10165USA.

All rights reserved.Reproduction of this preprint, or any


portion thereof, is not permitted without direct permission
from the Journal of the Audio Engineering Society.

AN AUDIOENGINEERING SOCIETY FI:tEffilNT


A Family of Linear-Phase Crossover Networks

of High Slope Derived by Time Delay

by

Stanley P. Lipshitz and John Vanderkooy


University of Waterloo, Waterloo, Ontario
Canada N2L 3Gl

Abstract

The design of linear-phase crossover networks has until now


necessitated the use of crossovers at least one of whose outputs suffers
from either frequency response ripple in the pass band or low rolloff
rate in the stop band. It may be desirabl%at least for experimental
purposes, to construct a high-quality loudspeaker which avoids the all-
pass phase characteristic exhibited by the best of conventional Butterworth
or Thiele-Linkwitz-Riley-derived high-slope crossovers, because such phase
distortion can subtly degrade some programme signals. This paper generalizes
a recent sugqestion and shows how, by introducting a suitable time delay
into the signal path,a family of compl_lentary constant-voltage crossovers
can be synthesized with high slopes in both low-pass and high-pass outputs.
Such crossovers are also applicable more widely; for example, in band-
splitting companders, limiters and siqnal processors of various kinds.
The performance of the new crossover design for different lowZpass aligmlents
is discussed in some detail.
O. Introduction

Conventional high-quality loudspeaker crossover networks are a


compromise between the conflicting requirements of (in decreasing order of
importance):

(a) flatness in the mag_nitude of the combined low- and high-pass outputs,

i.e. MS(m) = IHL(Jm) + HH(Jm)t - 1 _ in order to achieve flat acoustic


output on axis,

(b) adequately steep cutoff rates of the individual low- and high-pass

filters HL(J_) and Hli(J_o ) respectively in their stopbands, in order to


prevent degradation of performance due to undesirable characteristics of
the individual driver units outside their passbands,
(c) acceptable polar response for the combined output, taking into account
the physical separation oF the drivers; this normally makes in-phase low-
and high-pass outputs desirable [1], [2]:

_L(m)= _H(W) ,
at least through the crossover region, and
(d) acceptable phase response for the combined output, i.e. of

_S(_>)= arg [UL(j_) + HH(Jm)] , the most desirable characteristic being


phase lineariLy: _S(_J)= -_o_ , where = represents the equivalent
time delay of the combined network.

ttere,and in the sequel, we shall use HL(S) and HH(S ) to denote the
complex transfer functions of the low- and high-pass filters respectively
in terms of the complex frequency variable s = c_ + j_ , where m = 2_Tf
is the radian frequency. In polar form in the frequency domain we shall write:

H(j_)
= M(_)e
j_(m) , (1)

where M(m) is the magnitude and _(m) the phase of II(j_,_).


Subscripts
L or H will be used to denote low- and high-pass functions respectively,
and the subscript S will denote the combined system function.

The requirement (a) is necessitated by the extreme sensitivity


of human hearing to relatively small frequency response variations, and is
normally the primary requirement to be met by a hiqh-quality crossover network
(in conjunction,of course, with the actual responses of the individual drivers
--we shall have more to say about this later). The available crossover desiqns
have always necessitated some of the requirements (b), (c) and (d) being
forfeited. As regards the advantages of a linear-phase response, all the
evidence is not yet in, and the ultimate desirability of such a response
on general musical material is as yet unclear, but it is certain that on
suitable musically-related signals the type of all-pass phase characteristic
exhibited by most conventional crossovers can be subtly audible [3], [4].
In the circumstances, it seems worthwhile making known a new crossover

configuration, one particular case of which has recently been employed in


the construction of a laboratory standard loudspeaker [5], and which can

enable all of requirements (a), (b) and (d) to be met precisely, and (c) to
be ameliorated. This is our aim in the present paper. Firstly, however,
a brief survey of conventional crossover networks and how they meet requirements
(b) through (d) is in order. We shall not consider at all those networks
(such as second-order low- and high-pass Butterworth filters) which do not
meet the flat-magnitude requirement (a).

1. Conventional Crossover Networks

Of the elementary crossovers, only the first-order system of low-

and high-pass filters with crossover radian frequency m0 , namely

mO 1

HL: mO + s 1 + sn (2)
S
s n
NH= =1+ '
where sn : s/mO is the normalized complex frequency variable, giving

1 for the in-phaseconnection 1


1 ± !
Sn

NS = HL ! NH : n:
1+ s }(3)
1 - sn
I + sn
for the out-of-phase connection ]
3
can achieve linear-phase and flat-magnitude response (for the in-phase
connection). The out-of-phase connection yields a simple first-order all-
pass combined output. Unfortunately, the low- and high-pass outputs are
everywhere in phase quadrature (which degrades the polar behaviour near

mO) and have ultimate slopes of only 20 dB per decade (6 dB/octave) which
is inadequate even to limit excursion of the high-frequency driver below

its cutoff frequency. In Fig. 1 we plot the magnitudes ML(m n) and


MH(mn) and phases _L(mn) and _H(mn) of the HL and HH given by
Eq. (1) in terms of the normalized (radian) frequency m n = m/mo .

Looking at crossovers of higher slopes, one is immediately led


to the Butterworth alignments of higher order. For the odd-order Butterworth
alignments:

1
HL = B2r+l--
_
(4)
(_l)rsn 2r+l
HH = B2r+l(Sn)

of order (2r+l) , r = 0,1,2..... we do achieve magnitude flatness and


rolloff rates of (2r+l) × 20 dB/decade, but with an all-pass overall
phase characteristic of minimum order r [6], [7] and phase quadrature

between HL and HH . Here we have denoted by Bk(Sn) the normalized


kth-order Butterworth polynomials, given in Table 1 for k: 1 .....6 ,
the cases which we shall use for illustrative purposes in this paper.
These higher-order filters thus meet requirements (a) and (b),but (d)
is degraded progressively as the order increases and (c) is unchanged.
The standard third-order Butterworth filter pair, with inverted high-pass
for improved overall phase response [6], [7], has performance shown in
Fig. 2.

A recently-suggested alignment, whose origins go back to Thiele


[6] and Frater [1], is the Linkwitz-Riley filter [2] which has been
implemented in a number of modern high-performance monitoring loudspeakers
[8], [9]. These filters use low- and high-pass sections each composed
of two cascaded Butterworth filters; for example the 2r-th-order Linkwitz-

Riley, crossover is
4
1 1

HL = [Br(Sn)]2 _ L2r( _

(-1)rs2r (-1)rs2r (5)


HH ' n n ,
[Br(Sn)]2 --L2r(Sn )

r = 1,2..... where we have denoted by L2r(Sn) the squared Butterworth


r-th-order polynomial appearing in the denominators of the Linkwitz-Riley
designs. These polynomials of orders 2,4 and 6 are given in Table 2,
and in Fig. 3 we show the performance of the fourth-order Linkwitz-Riley
filter pair. These even-order filters have the desirable property that

HL and HH are everywhere in phase, and so are best from the polar point
of view (c), and of course, have cutoff slopes of 40r dB/decade, hut
produce an overall all-pass phase response [7]:

Br(-Sn )
HS = HL + HH = _
Br (6)

The concept of low- and high-pass hand-splitting filters of identical phase


response, as exemplified by the Linkwitz-Riley designs, has been generalized
[10] to alignments other than cascaded gutterworth. Of all the high-slope
crossovers which sacrifice the phase requirenent (d), the alignments (5)
appear to be the best overall compromise, and their popularity in high-
quality loudspeakers is deserved.

Looking now towards what can be achieved if one desires overall


phase-linearity, one finds the distributed-polynomial (quasi-r-th-order)
designs of Ashley and Kaminsky [11], the "filler driver" designs of Baekgaard
[12] and Wall [13], and the separated-polynomial constant-voltage designs
of Small [14] and Blinchikoff [15]. The name "constant-voltage" is derived
From the fact that all these Filters have combined low- and high-pass outputs
of unity:

HL(S : 1
n)+ HH(Sn) , (7)
and hence are all constant-magnitude and linear-phase overall. For any

given low-pass filter function HL(Sn), a complementary high-pass filter is


derived according to Eq. (7) by subtractinq HL(Sn) from unity; i.e. the
high-pass output is synthesized by subtracting the low-pass output from the
input signal. Unfortunately, either one or both of the low-pass and high-
pass cutoff slopes are reduced below that achievable by the denominator
polynomial alone.

For example, all the Small as3anmetricaldesigns [14], with

HL
=_) (8)
1 Br(Sn)- l

have a peaked high-pass with a rolloff rate of only 20 dB/decade irrespective


of the order r of the low-pass function, and so are not really suitable
for use with most drivers. Fig. 4(a) shows the responses of Eq. (8) with
r = 2. The extreme phase-difference between the low- and high-pass outputs
in the crossover region al so causes exaggerated sensitivity to phase errors
which results in exceptionally poor polar behaviour. The slope of the
derived high-pass filter can be increased, and at the same time the crossover
can be symmetrized [14] by separating the terms in a suitably-chosen
denominator polynomial between the two numerators. Fig. 4(b) shows a second-
order ssanmeLricalcrossover so derived_using a third-order denominator
polynomial:
l+as
tt
L= n
1 + asn + aSn2+ Sn3
a = 2 + /_ (9)
as 2 + s 3
= n n
IfH
1 + asn + aSn2 + Sn3

The phase difference at crossover is still large at 120 °, and the polar
response still significantly degraded [2].
An alternative procedure due to Ashley and Kaminsky [ll] distributes
the coefficients of the denominator polynomial between the two numerators,
thus producing symmetrical constant-voltage crossovers of reduced effective
order. For example, their "quasi-second-order" crossover

1 + Csn
HL=
1 + 2Csn + Sn2
(lO)
_sn + Sn2
HH- 2 '
1 + 2Csn + sn

with damping factor c = 0.5 displays 40 dB/decade slopes around crossover,


although these reduce to first-order further away. These are shown in

Fig. 5. The problems inherent in all constant-voltage designs are apparent.

The "filler driver" approach to constant-voltage crossovers [12],


[13] satisfies Eq. (7) by using additional loudspeakers to introduce the
terms missing in the numerator of the standard designs. For a second-order
crossover we thus have [12]

HL: 1
1 + V-2s
n + Sn2

V2sn
- HFD
: (ll)
1 + V_sn + Sn2

Sn2

HH = 1 +yr2s n + Sn2 '

where HFD denotes the transfer function of the filler driver, which rolls
off at only 20 dB/decade.

Something new is required in order to achieve phase linearity without


the reduced slopes inherent in the simple constant-voltage scheme. This is the
topic of the next section.
2. Delay-Derived High-Slope Filters

The idea which we now introduce seems to havefirst been used for loud-
speaker crossover networks by Tanaka et al. [5], although it has appeared earlier
[16]. The magnitude-peaking,rolloff rat_polar problems and phase difference
between low- and high-pass outputs of the constant-voltage filters are all
primarily due to the phase shift of the low-pass filter HL(S). When the
high-pass filter is derived by subtraction:
NH(S)
: 1 - Hk(S
) , (12)
the time-delay of the low-pass filter in its passband prevents a high rolloff
rate from being achieved by the complementary high-pass filter. This can be
ameliorated by the simple expedient of replacing 1 in Eq. (12) by the transfer
function e-Ts of a suitably-chosen pure time-delay T :

HH(S)
: e-Ts- HL(S) (13)
This corresponds to synthesizing the high-pass output by subtracting a given low-
pass output from a time-delayed version of the input signal. The constant-voltage
and time-delay-derived filter topologies are compared in Fig. 6. The two out-
put signals now combine to form a time-delayed version of the input signal:

HL(S)+ HH(S)= e-Ts` (14)


and thus afford perfect magnitude and linear-phase response.

Specifically,
the time-delay T shouldbe chosento equal the O-Hz
phase and group delays of the low-pass filter, i.e.

:- lim _k(W) d_L(0)


m+0 - d_ (15)

where the phase delay Tp(m) and group delay Tg(m) are definedrespectivelyby

and Tp(m)
= -_(m)
m (15)

Tg(m)
= -d_ (17)
This choice, as will shortly be seen, is necessary in order that it be possible to
make the cutoff slope of the derived high-pass filter steeper than that achievable
from the simple subtractive constant-voltage filter [Eq. (12)], which is the special
case of Eq. (13) when the delay T is chosen equal to zero. To clarify the
situation, let us assume that the low-pass filter is an all-pole designl)(which could
include, amongst others the standard Butterwort_ Bessel and Linkwitz-Riley alignments)
of order r :

HL(S
): 1 :1
1 + als+ a2s2+...+arsr _ (18)
k:O aksk
l) More generallow-passfilterscan also be considered.
where a0 = 1. Then it can be shown by Eqs. (16), (17')that its O-Hz delays
are given by

zp(O)
: Tg(O)
= a1 (19)

Alternatively, if we normalize the denominator polynomial in Eq.

(18) by setting

s _ s , where m 0 : (ar)-l/r , (20)


n mo

we can write Eq. (18) in the normalized form (which we shall use throughout
the sequel)
1
= 1 - (21)
HL(Sn) 1 + blsn + b2Sn2 +...+ br_lSN-1 + Snr r r
k=_obr Sn

where bo : br = 1. In terms of the normalized polynomial coefficients, we


find that Eq. (19) is replaced by

b1
up(O)
: Tg(O)
- mO (22)

while Eq. (13) reads

-mo_S n
Hli(Sn)
:e - HL(Sn) (23)

Hence, if we write, using Eq. (22),

=n= bl = _O=p(°): mO_g(O) (24)

for the normalized time delay of the low-pass filter (21), Eq. (23) becomes
in normalized form
-TnS n
HH(Sn)
:e - HL(Sn) (25)

We now manipulate Eq. (25) as follows:

-TnSn enn
flli(Sn)
:e 1 .....
1 I-bls n + b2Sn 2 + ...+ sn

-TS r
e n n i + blsn + +...+ sn ) - (1 + TnSn + n--_-.in
+...

1 + blsn + b2Sn 2 +.,.+ Sn r

9
Tn r-1
e-TnSn
I(bl - =n)Sn+ (b2 - -2--n2 r-1
)Sn2 +'"+ (br-1- (r--_!)Sn

n r Zn
+ (1- _.l)Sn -
T r _ k snk]
= k:r+l J(26)
1 + bls n + b2Sn2 +...+ br_lSn r-1 + Snr

where in the second step we have expanded the exponential using the Maclaurin
series. Ignoring the innocuous delay factor e- nSn , we deduce that for
greatest cutoff slope of the derived high-pass filter, as many consecutive

coefficients of powers of sn as possible should vanish, starting with the


lowest-power coefficient. To achieve this we have at our disposal the

coefficients b l, b2 .... , br_1 of the denominator polynomial, subject


only to stability and realizability restrictions. From Eq. (26) the desire
is thus to satisfy as many consecutive equations as possible from the following
list, starting with the first:

b1 = Tn

'_2 bl2
b2 : n
2 =_ 2

.; Tnk blk
bk= k! k! (27)

Tn r-1 blr-1

r blr
b =1= Tn : ____
r r! r!

orgenerally blk
bk - k! ' bl = 'tn: (r!)l/r , (k=2..... r-l), (28)

Polynomials satisfying this general criterion (28) will be called the normal-

ized General _nomialsof order r and denoted by Gr(Sn) in the sequel.


G2(Sn) through G4(Sn) are given in Table 3. These are the only stable

l0
General polynomials; all Gk(Sn) with k _ 5 can be shown [17, pp. 294-299]
to be unstable. It should be noted that G2(sn) _ B2(Sn) , and that, as
anticipated, the chosen time delay Tn should satisfy Eq. (24).

Referring back to Eq. (26) we see that the General polynomials

result in a derived high-pass filter HH(Sn) whose numerator begins with a

term in snr+l , and hence whose cutoff rate is (r+l) x 20 dB/decade, i.e.

20 dB/decade steeper than that of the low-pass filter HL(Sn) from which it
is derived. In fact the General polynomials give rise to the following
crossover filter transfer functions:

1
HL(Sn) :

k=0 . n

-e
-h 5 Z] sk
HH(Sn)
: k:r+l n (29)
r b k

k:O' n

with
Tn = b1 : (r!)1/r , (r:2,3,4) .

This is the advantage of these delay-derived filters over the standard constant-

voltage design. For, when we put Tn:O in Eq. (26) we see that the numerator

commences w'th the linear term in sn and hence never achieves a slope greater
than 20 dB/decade for any order r. Lest we think that the General filters

Gr(Sn) are the answer to all our dreams, we should first examine the magnitude
and phase responses of the resulting crossover outputs. These are shown in

Fig. 7 for G3 and G4 (G2 is the same as B2 and will be displayed shortly),
all higher-order Gk being unstable and thus unusable. G3 is quite usable
and has many attractive features. It displays only very slight peaking in
the passband; the phase difference at crossover is under 90° thus ameliorating
polar aberrations; and its low-and high-pass cutoff slopes are respectively 60
and 80 dB/decade. It is, of course_flat-magnitude and phase-linear, and the
-TS
influence of the time-delay e n n on the phase of the high-pass output can
be clearly seen (being a linear phase lag at high frequencies on a linear

frequency scale). G4 is not so attractive, and the response peaking and


extreme phase relationships at crossover appear to render it unacceptable
for our purposes.

Since the General polynomials yield only a single useful high-

order alignment (G3), we are led to ask what the results are like with
standard low-pass alignments like But_erworth, Bessel and Linkwitz-Riley

for HL(Sn). We can no longer satisfy all the conditions in Eq. (28), and
so the ultimate high-pass slope is reduced below that achieved by a similar-
order General alignment. Let us see how well we can do.

The Butterworth polynomials Bk(Sn) (see Table l) posse_the


desirable property that

bl
2 = 2b
2 (30)

for all orders r. The proof of this fact',is given in the Appendix. Table
1 will be seen to confirm this fact. This _omewhat unexpected result ensures
the satisfaction, not just of the first of equations (27) by a suitable choice

of Tn , but automatically also of the second of equations (27) for r _ 2 .


Thus we find from Eq. (26) that for the first-order Bl(Sn) low-pass (r=l),
the
high-pass numerator begins with the Sn2 term and so rolls off at 40
dB/decade, while for all higher-order low-pass Butterworth alignments Br(Sn),
3
the high-pass numerator begins with the sn term and so rolls of at
60 dB/decade. The actual transfer functions are found from Eqs. (21), (26)
to be, for r _ 2:

= 1
HL ( Sn ) r_Tsn--
B _

blk (31)
e

HH(Sn)
= k=3 ' k:r+l
Br(Sn)

with Tn = b1 In Fig. 8 (a-f) we show the magnitude and phase of

12
both low-pass and high-pass outputs of the delay-derived Butterworth cross-

over filters of orders 1 through 6. The B 3 case of Fig. 8(c) is that


adopted by Tanaka et al. [5] in their implementation of this crossover
concept. Note the expanded frequency axis in Fig. 8(a). The damped

rippling evident in the passband of H H , especially for the low-order


filters, is a consequence of the increasing phase lag contributed by the

time-delay Tn in conjunction with the decreasing magnitude and group-delay


of HL for _n > 1. The maximum high-pass slope of 60 dB/decade is also
evident. As would be expected from the rationale behind the delay-derived
crossover concept, all these filters exhibit less than 90 ° phase difference

between HL and HH at crossover, in contrast with the quadrature relation-


ship which characterizes all the odd-order standard Butterworth alignments.

Compare, for example, Fig. 1 with Fig. 8(a) for the B1 case, and Fig. 2
with Fig. 8(c) for the B 3 case. The even-order delay-derived filters are,
of course, sharper in their cutoff around crossover than the corresponding-
order Linkwitz-Riley cascaded Butterworth designs. Compare Fig. 3 with Fig.
8(d) for the fourth-order case. Perhaps most instructive is the comparison
shown in Fig. g of the three competing third-order filters. Fig. 9(a) shows

the standard B3 low- and high-pass filters (solid curves), while Fig. 9(b)
shows the delay-derived high-pass output (dash-dotted curve) and Fig. 9(c)
shows the Small asymmetrical constant-voltage high-pass output (dotted curve).
The latteFs unacceptable performance is evident. Note the peaking and reduced
rolloff rate. Returning to Fig. 8, the higher-order delay-derived Butterworth
filters satisfy our magnitude, slope and phase criteria (a), (b) and (d) very
well, but, although better than the standard odd-order Butterworth designs as
regards polar pattern (point (c)), the question arises as to whether we can
retain some of their benefits while improving the polar behaviour with some
other alignment. This is indeed possible using a Bessel alignment for

HL(Sn)-

Because of their maximally-flat-delay characteristic, the phase-


match between the time-delay e TnSn of Eq. (25) and the delay of the low-

pass filter function HL(Sn) can be expected to be best if we choose a


Bessel alignment for HL(Sn). This is indeed true. Of course, the more

13
droopy magnitude characteristic of the Bessel low-pass filters is apparent,
and so we improve polar pattern (point (c)) at some expense to point (b).

Let us denote the normalized Bessel polynomial of order k by Dk(Sn)


("D" for "delay"). These polynomials of orders 2 through 6 are given in
Table 4, and the magnitude and phase responses of the delay-derived filters
to which they give rise are shown in Fig, 10(a-e). Notice how, as predicted,
the phase-correspondence progressively improves with order and is much better
then the Butterworth filters. However, since condition (30) is not satisfied
by the Bessel polynomials, the high-pass sections only achieve a slope of
40 dB/decade. As a group the Bessel delay-derived crossovers nevertheless
have many attractive features.

Finally, let us complete our catalogue with the Linkwitz-Riley-

alignment delay-derived filters using L 2 , L4 and L 6 given by Table 2.


As is easily deduced from their definition as the square of the Butterworth

polynomials, all the L2k(Sn) except L2(sn) do indeed satisfy the


condition (30), and so while the L2-derived high-pass rolls off at
40 dB/decade, all the higher-order L2k-derived high-passes roll off
at 60 dB/decade. Their magnitude and phase behaviour is shown in Fig. 11 (a-c).
Compare Fig. 3 with Fig. ll(b). The former is an all-pass crossover design
with perfect polar behaviour, while the latter is not quite as good from
the polar point of view (although the phase difference at crossover is less
than 90° ) but is phase-linear and has quite good magnitude characteristics.
The sixth-order case shown in Fig. ll(c) is also attractive although the
high-pass slope is only 60 dB/decade.

Further low-pass alignment possibilities suggest themselves,


and further investigation in this direction might turn up an alignment which
combines the best features of the delay-derived filters with the good polar
behaviour of the standard Linkwitz-Riley crossovers. The simple all-pole
low-pass configuration of Eq. (18) will almost certainly need to be abandoned
if such is to be achieved.

In trying to assess the significance of the inter-driver phase


shift exhibited by the various crossover filters we have examined, some

14
kind of "figure of merit" would be desirable. A phase difference of 0° through
crossover is a goal to be aimed at, because this places the lobe of the
summed acoustic output on axis at all frequencies even though the drivers
are physically separated [2]. Of the filters described, only the standard
Linkwitz-Riley crossovers achieve this. All crossover designs with an
inter-driver phase difference through the crossover region produce a lobe
which tilts towards the lagging driver at crossover, and even though the
on-axis response may _ perfect, the presence of an off-axis lobe of increased
output is undesirable due to the colouration which it can impart to the
reverberant sound field. A measure of the severity of the polar lobe tilt
at crossover can be o_ained _ examining the rate of change of the magnitude
of the summed acoustic output with off-axis displacements. Clearly, this
derivative will be zero if the lobe remains on axis, and will increase with
increasing phase difference. At a rate dependent upon the filter cutoff
slopes, it will tend to zero as one leav_the crossover region. We shall
call this parameter the "lobing error" of the particular crossover filter.
Let us introduce an additional air-path normalized time delay _*n between
low-pass and high-pass outputs by putting

J+L(mn )

HL(m
n): ME(m
n)e (32)

NH(mn) : MH(mn)eJ[_H(mn)+T_mn]

A positive value of _ corresponds to a physical listener position


displacement towards the high-frequency driver, i.e. above axis for most
loudspeaker systems, which locate the tweeter above the _ofer. In tems
of our earl_r notation (1), we accordingly define the normalized lobin9

error E(mn) to be

[_Ms(mn)
]-- (33)
E(mn)
:L _ _:o

Using Eq. (32) this reduces (since Ms _ 1 on axis) to

15
n n

E(mn): 2Ms _*H ;_-_ T_:O _

mn 2 [(ML cos_L+MH cosq_H)2+(.qLsin_L+MHSi n_H )2 ]


2 ¢_H T_=O

- Wn _ + + cos(_L-_
2 B_H[ML2 MH2 2MLMH
H)]
T*:O

i.e.

E(m
n) = mnMLM
H sin(_L-_H) (34)

As expected, E(mn) equals zero if the two outputs are in phase,


and increases sinusoidally with phase difference to a maximum when they are
in quadrature. Thus our delay-derived odd-order Butterworth filters will
be better in this criterion than the standard Butterworth filters, and our
Bessel delay-derived filters will be extremely good, but of course not

quite as good as the standard Linkwitz-Riley filters which yield E(mn) s 0


at all frequencies. A positive value for E(mn) s_gnifies a lobe above
axis, and a negative value a lobe below axis. The magnitude of E(_n) is
a measure of the severity of the lobing. These features are apparent in
the curves of Fig. 12(aZe), where the lobing error is plotted as a function

of the normalized frequency mn for the delay-derived filters. In Figs.


12(a) and (b) we show the B_tterworth alignments B1 through B6 ; in
Figs. 12(c) and (d) we show the interesting third-order General alignment

G3 and the Bessel alignments D2 through D 6 ; finally Fig. 12(e) shows


the delay-derived Linkwitz-Riley alignments L 2 , L 4 and L6 . It is
believed that E(mn) is a meaningful measure of the severity of the polar
aberrations of the crossover design, and hence can be used to assess
crossovers for performance characteristic (c). It should be noted that
the delay normalization embodied in definition (33) employs the "nominal"

crossover frequency _0 which can differ significantly from the actual


crossover frequency at which the low- and high-pass magnitude curves

intersect. That is, the normalization is relative to m n : 1, and for

16
some aligmlents (particularly the Bessel and some of the low-order alignments)
the actual crossover frequency can deviate appreciably from mn = 1 . The
relative rankinqs are not affected by this factor. For comparison purposes
we show in Fig. 13(a) the lobing error curves of the standard Butterworth

(non-delayed) crossovers of odd order B 1 , B3 and B5 , and in Fig. 13(b)


the error curves of the constant-voltage designs of Fig, 4(a,b) and

Fig. 5.

3. Discussion

The delay-derived crossover concept of Fig. 6(b) can, of course,


be extended to multi-way systems. In Fig. 14 we show a 3-way crossover

of this type. Here _2 represents the O-Hz time-delay of the second


low-pass filter HL2(S) , and (_l+T2) that of the first low-pass filter
HLl(S ). Consequently the outputs are as follows:

low-pass _ ItL(S)= HLi(S)


-TlS
band-pass
_ HM(S)= e .ltL2(S
) - HLl(S
) (35)
-_]s -T s
high-pass_ HH(S)= e .[e 2 . HL2(S)] ·

and the combined output HS_ being the sum_is just a pure time-delay:

-(Tl+_2)S
Hs(S
)=e (36)

The requirmlent that (_l+T2) represent the time-delay of HLi(S) can


be seen to be necessary from a consideration of the sum of the band-pass
and high-pass outputs:

-(ll+r2)s
HM(S)+ HH(S)= e - HLl(S)

A further extension allows one to include in the crossover


design some of the features of the actual driver responses, so as to
enable the drivers to achieve target functions which improve upon their

17
inherent performance, and hence fonila system which is better than its

component parts. To be specific, let us suppose that TL(S) and TH(S)


are the "target functions" [9] of the low-frequency and high-frequency

acoustic outputs respectively, chosen so that their sum TS = T L + TH


represents the desired combined acoustic output of the system (magnitude
and phase), including band-limiting at the extreme top and bottom due

to the individual drivers' responses. Suppose that SL(S ) and SH(S)


represent the actual transfer functionsof the individual low- and high-
frequency drivers when mounted in the system, and let us introduce driver

response correction networks CL(S ) and CH(S) to adjust the drivers'


in-band responses for flatness (within realistic bounds), around the crossover
frequency, so that near crossover

CL(S)'SL(S)_ 1
(37)

CH(S)-SH(S)_ 1

CL(S ) and CH(S ) may contribute to the tailoring of the system's response
at each extreme of the frequency band, i.e. well away From the crossover

region. Obviously TL(S) and TH(S) must be realistically chosen so that


the actual driver responses do not differ grossly from the desired target

functions, i.e. so that CL(S) and CH(S) are achievable. In this


case the configuration of Fig. 6(b) can be elaborated to include driver
response correction as shown in Fig. 15(a). The crossover itself will \
contribute the total rolloff required by the target functions in the
transition region, i.e., near crossover

HL(S)_ TL(S)
(38)
-lS
e - HL(S)= TH(S)

ttence at all frequencies

TL(S)= ltL(S).CL(S).SL(S)
, (39)

18
and so the low-frequency acoustic output achieves its target. Provided

finally that in the crossover region TH(S) is chosen to match the delay-
derived high-pass output, i.e.

TH(S) : [e-Ts - Hk(S)].CH(S).SH(S)


, (40)

the overall crossover will indeed achieve a transfer function of e


at low
in the midband, and response as dictated by e-Ts.cL(S)'SL(S)
frequencies and by e-Ts.cH(S).SH(S) at high frequencies. We thus do
achieve phase linearity through the crossover region. Moreover, the

correction filters CL(S) and CH(S) can be moved through the summing
point and combined as shown in Fig. 15(b) if desired. This technique
enables driver resonance effects(for example the tweeter's fundamental
resonance below its operating range) to be accurately taken into account
in the overall design [18]. In addition, inter-driver time delays due
to physical positioning can be compensated for by including a delay

component into the correction filters CL(S) or CH(S) as appropriate.


All the techniques of digital analysis and design [19] can be marshalled
towards the accurate achievement of the desired system performance.

A further point worthy of discussion is the sensitivity of


the delay-derived crossovers to errors in the value of the time-delay

Tn . Any deviation of un from its design value will, of course, re-


introduce the cancelled terms from the numerator of the synthesized high-

pass output and so affect the ultimate rolloff rate of the high-pass filter
below the crossover frequency. For illustrative purposes we show in
Fig. 16 the effect on a delay-derived General third-order filter (a)_ of

decreasing Tn by 10% (b)j and increasing Tn by 10% (c). The effect


on the h_gh-pass cutoff is apparent. In Fig. 17 we show the effect of a

5% delay increase on the delay-derived B3 alignment (compare with Fig. 8(c),


but note the expanded frequency scale). The notch in the high-pass stop-
band is generated by the delay error. The combined output remains undegraded
of course.
A question which naturally arises is how to implement the time
delay. Analogue and digital delay lines spring immediately to mind, and

19
make the implementation trivial but expensive. In fact, given the availability
of a tapped delay linejother alternatives immediately spring to mind. For
example, a phaseless transversal low-pass filter could be implemented for

HL (i.e. with _L z O) so that no time delay at all would be required


in Fig. 6 and a transient-perfect crossover with high slopes would be given
simply by subtraction:

HH = 1 - HL

Thus a phaseless low-pass generates high-slope crossovers by the simple


constant-voltage method. Another point worth mentioning is that, with
the emerging availability of signals in digital form, the implementation of
an input signal delay, as required by these new crossover networks, involves
very little additional hardware, and makes them more attractive.

Assuming that a delay-line is not available, cascaded Bessel-


alignment all-pass filters can achieve the required delays, but of course
the wide (20-kHz) bandwidth necessitates the cascading of many such filters
in order to achieve the required delay. It seems possible [4] that further
research will indicate the need for phase-linearity to be primarily in the
midband region below a few kilohertz, and narrower bandwidth all-
pass filters could achieve the required phase linearity up to midband
with fewer stages required. If the all-pass delay network's phase response

is given by say _A(_) , the resulting crossover's combined output has


the same phase response:

eJSA (_)
mL(m
) + NH(W
)= (41)

Choosing _A(m) to be audibly innocuous might thus ease the design


complexity.

Finally, it should be mentioned that these new crossover ideas


are in no way limited to loudspeaker use only. Any band-splitting signal
processors,such as limiters, noise reducers, companders, etc. can benefit
from the use of this form of crossover design.

2O
4. Conclusions

A new family of high-slope linear-phase crossover topologies


has been described. Compared with previous linear-phase designs they
achieve much better stopband performance and greatly improved polar
response behaviour. Various alignments, including a new General alignment,
are possible, and selecting an alignment represents a trade-off between
cutoff slope and response lebing. A "lobing error" criterion is introduced
to enable this trade-off to be made on a more informed basis. With the

imminent availability of digital source material, linear-phase loudspeakers


may become desirable and their implementation more feasible. The new
crossover concept is also applicable to band-splitting filters in general.

5. Acknowledgment

The authors would like to acknowledge with thanks the financial


support of the Natural Sciences and Engineering Research Council of Canada.

21
6. References

[1] R.H. Frater, "_n-Phase Crossover Networks," 16th Int'l. Radio and

Elec. Eng_ Convention Digest (Melbourne, 1977 Aug.), pp. 250-252.

[2] S,H. Linkwitz, '_Active Crossover Networks for Noncoincident Drivers,"


J. Audio E_. Soc., vol. 24, pp. 2-8 (1976 Jan./Feb.),

[3] H. Suzuki, S. Morita and T. Shindo, "On the Perception of Phase


Distortion," J. Audio E_ Soc., vol. 28, pp. 570-574 (1980 Sept.).

[4] S.P. Lipshitz, M. Pocock and J. Vanderkooy, "Preliminary Results on


the Audibility of Midrange Phase Distortion in Audio Systems," Paper
presented at the 67th Convention of the Audio Engineering Society,
New York, i980 Oct. 31-Nov. 3 (Preprint #1714 (D-8)).

[5] M. Tanaka, H. Suzuki and M. Iwahara, "An Approach to the Standard


Sound Transducer," Paper presented at the 63rd Convention of the
Audio Engineerinq Society, Los Anqeles, 1979 May 15-18 (Preprint
_146D (D_ll)).

t6] A.N. Thiele, 'Uptlmum Passive Loudspeaker Dividing Networks," Proc.


i.R.E.E. Aust. vol. 36, pp. 220-224 (1975 Ju]y).

[7] P. Garde, "All-Pass Crossover Systems," J. Audio Eng. Soc., vol. 28,
pp. 575-584 (1980 Sept.).

[8] G.J. Adams, New Development _ in Lgudspeaker System Design - theB& W


801 Monitor (B & W Loudspeakers, Northing, U.K., 1979).

[9] "Crossover Filters - An Integral Part of Overall System EngineeringS'


KEFTOPICS, vol. 4 #2 (KEF Electronics Ltd., Tovil, Maidstone, U.K.,
1980).

[lO] R.A. Gabel, "On the Design of Complementary Filters," I.E.E.E. Int'l
Conference on Acoustics, S_peech and Siqnal Processing_cord),
Washington, D.C., 1979 April 2-4, pp. 805-808 (1979).

[11] J.R. Ashley and A.L. Kaminsky, "Active and Passive Filters as
Loudspeaker Crossover Networks," J. Audio Eng. Soc., vol. 19, pp.
494-502 (1971 June).

[12] E. Baekgaard, "A Novel Approach to Linear Phase Loudspeakers Using


Passive Crossover Networks," J. Audio En 9. Soc., vol. 25, pp. 284-294
(1977 May).
22
[13] P.K. Wall, "Active and Passive Loudspeaker Crossover Networks Without
Transient Distortion," Paper presented at the 50th Convention of the
Audio Engineering Society, London, 1975 (Paper 16).

[14] R.H. Small, "Constant-Voltage Crossover Network Design," Proc. I.R.E.E.


Aust., vol. 31, pp. 66-73 (1970 March). Reprinted in J. Audio En9, Soc.,
vol. 19, pp. 12-19 (1971 Jan.).'

[15] H.J. Blinchikoff, "Low-Transient High-Pass Filters," IEEE Trans. Circuit


Theory, vol. CT-17, pp. 663-667 (1970 Nov.).

[16] R.M. Golden, "Digital Filters," Chapter 12 in Modern Filter Theory and
Design, G.C. Temes and S.K. Mitra (eds.) (Wiley, New York, 1973),
pp. 520-522.

[17] F.F. Kuo, Network Analysis and Synthesis (2nd ed., Wiley, New York, 1966).

[18] W.M. Leach, Jr., "Loudspeaker Driver Phase Response: The Neglected
Factor in Crossover Network Design," J. Audio En 9. Soc., vol. 28, pp.
410-421 (1980 June).

[19] J.M. Bermanand L.R. Fincham, "The Application of Digital Techniques to


the Measurement of Loudspeakers," J. Audio En9. Soc., vol. 25, pp.
370-384 (1977 June).

[20] D.E. Johnson, J.R. Johnson and H.P. Moore, A Handbook of Active Filters
(Prentice-Hall, N.J., 1980).

23
Appendix

We wish to demonstrate that the Butterworth polynomials

Br(Sn):

Br(Sn): 1 + blsn + b2Sn


2 +'"+ br-1sr'l
n + Snr (Al)

satisfy Eq. (30): bl2 : 2b2 for all r _ 2 . The proof is accomplished
in stages by successively reducing Eq. (30) to a number of equivalent
conditions, the last of which follows from the mathematical relations

defining the zeros of Br(Sn) . Let us write (Al) in factorized form.


We have different forms in the even-(r = 2_) and odd-(r: 2_+1) order
cases:

B2_+l(Sn)
B2_(Sn) :
= (l+alSn+Sn2)(l+a2Sn+Sn
(l+Sn)(l+alSn+Sn2)(l+a2Sn+Sn
2)....(l+a_Sn+Sn2)
2) ....(l+a_Sn+Sn2) ,I(A2)

where the coefficients ak are given by [20]

a k = 2 sin L 2r ] (k = 1,2....._) (A3)

Lemma l: Eq. (30) is equivalent to the condition

ak2= , r _2 (A4)
k:l r-2 if r odd
_ Ir ifreven I

Proof: For r even, r = 2_ , Eq. (30) is equivalent to


2

ak : 2 + _ aka
i
k1 k,i=l
k < i

Expanding the sum on the left-hand side gives the first of Eq. (A4).

24
Similarly, for r odd, r = 2_+1 , Eq. (30) is equivalent to

ak + 1 = 2 + _, aka i + _' a k ,
k_l k,i=l k=l
k<i

and again expanding the left-hand side yields the second of Eq. (A4).

Lemma 2: Eq. (A4) is equivalent to the condition

--- = , r >2 (A5)


2 kJ1
_ cos [(2krl)_] I 10 if
if r
r odd
even

Proof: Using Eq. (A3):

ak2: _ 4 sin2 [(2____]___]


k:l k=l

= 29_ - 2 _ cos
k=l

from which the result follows by Eq. (A4).

Lemma 3: Eq. (AS) is true.

Proof: The sum of all the complex r-th roots of -1 is zero:

r
j(2k-1 )J,

_ e r =0
k=l

25
k:l co IC'
]0
But

Thus, for r even, r = 2_ ·

o: _ cos + _ cos
k=l k=_+l

: 2 X cos
k=l

by replacing the dummy summation variable k by r-k+l in the second


sum. This proves the first of Eq. (A5).

For r odd, r : 2_,+1:

0= y cos +cos_+ [ cos


k=l k=_+2

: -1 + 2 f. cos
k=l

by the same trick. This proves the second of Eq. (A5).

The desired result follows by combining Lemmas 1 through 3.

26
Table 1 Normalized Butterworth polynomials Bk(Sn) of orders 1 through 6

B1(Sn): 1 + sn

B2(Sn) = 1 + v_ sn + Sn2

B3(Sn) = 1 + 2sn + 2Sn2 + Sn3

B4(Sn)= 1 + 2.613sn + 3.414Sn2


+ 2.613Sn3+ sn4

B5(Sn) : 1 + 3,236sn + 5.236Sn2 + 5-236Sn3 + 3'236Sn4 + Sn5

B6(Sn) = 1 + 3.864sn + 7.464Sn2 + 9'142Sn3 + 7'464Sn4 + 3'864Sn5 + Sn6

Table 2 Normalized Linkwitz-Riley polynomials L2k(Sn) _ [Bk(Sn)]2 of

orders 2 through 6

L2(sn) = 1 + 2sn + Sn2

L4(Sn) = I + 2/2sn + 4Sn2 + 2/2Sn3 + Sn4

L6(sn) = 1 + 4sn + 8Sn2 + lOSn3 + 8Sn4 + 4Sn5 + Sn6

27
Table 3 Normalized General polynomials Gk(Sn) of orders 2 through 4

G2(sn) _ B2(Sn) = 1 + _s n + Sn2

G3(s
n) = 1 + 1.817s
n + 1.651Sn2 + Sn3

G4(Sn) : 1 + 2.213sn + 2.449Sn2 + 1.807Sn3 + Sn4

Gs(Sn) , G6(sn) ,etc. are unstable

Table 4 Normalized Bessel polynomials Dk(Sn) of orders 2 through 6

D2(Sn) = 1 + _s n + Sn2

D3(Sn) = 1 + 2.466sn + 2'433Sn2 + Sn3

D4(s
n) = 1 + 3.201s
n + 4.392Sn2 + 3.124Sn3 + sn4

D5(Sn) = 1 + 3-936sn + 6.886Sn2 + 6.777Sn3 + 3.811Sn4 + Sn5

D6(sn) : 1 + 4.672sn + 9-920Sn2 + 12'358Sn3 + 9-622Sn4 + 4.495Sn5 + Sn6

28
Figure Captions

Fig. 1 Magnitude and phase curves for the standard B 1 alignment


crossover of Eq. (2), as a function of the normalized radian

frequency _n = _/mO '

Fig. 2 Magnitude and phase curves for the standard Butterworth B3


alignment crossover of Eq. (4) with inverted high-pass polarity.

Fig. 3 Magnitude and phase curves for the standard Linkwitz-Riley L4


alignment crossover of Eq. (5). Note that both low- and high-
pass sections have identical phase response, and that consequently
the magnitude curves intersect at -6 dB.

Fig. 4 Magnitude and phase curves for (a) the constant-voltage B2


asymmetrical crossover of Eq. (8), and (b) the constant-voltage
s_nmetrical second-order crossover of Eq. (9) with a = 2 + /3 .

Fig. 5 Magnitude and phase curves for the constant-voltage quasi-second-


order crossover of Eq. (10) with c = 0.5. The rolloff slopes
start at 40 dB/decade but soon reduce to 20 dB/decade.

Fig. 6 The circuit topologies of (a) the constant-voltage crossover of


Eq. (12), and (b) the time-delay-derived crossover of Eq. (13).

FiB. 7 Magnitude and phase curves for the time-delay-derived crossovers

of General alignments (a) G3 , and (b) G4 . Notice that the


high-pass rolloff slopes are 20 dB/decade greater than the
corresponding low-pass slopes.

Fig. 8 Magnitude and phase curves for the delay-derived crossovers of

Butterworthalignments(a) B1 , (b) B2 , (c) B3 , (d) B4 ,

(e) B5, and (f) B6 . Notice that the high-pass slope is 40 dB/decade
for case (a) and 60 dB/decade for all the higher-order cases.

29
Fig. 9 A comparison of the magnitude responses of crossovers based upon

a B3 low-pass alignment: (a) the standard B3 crossover,


(b) the delay-derived B3 high-pass, and (c) the constant-
voltage asymmetrical B3 high-pass. The latter design high-
pass has a slope of only 20 dB/decade, the former two being
60 dB/decade.

Fi__g.lO Magnitude and phase curves for the delay-derived crossovers of

Besselaliqmlents (a) D2 , (b) D3 , (c) D4 , (d) D5 , and


(e) D6 . The high-pass slopes are only 40 dB/decade, but the
phase-match to beyond the crossover frequency is exceptional for
the higher orders.

Fig. ll Magnitude and phase curves for the delay-derived crossovers of

Linkwitz-Rileyalignments (a) L2 , (b) L4 , and (c) L6 .


Except for the first case whose high-pass rolls off at 40 dB/decade,
all these high-passes roll off at 60 dB/decade.

Fig. 12 The "lobing error" function E(mn) plotted against normalized


frequency w n . A positive value signifies a lobe above the
loudspeaker axis of severity proportional to the height, and a
negative value signifies a lobe below axis. The lobing error
function is defined in Eq. (34). We show the lobing error curves

for the delay-derived crossovers of alignments (a) B1 , B 2 , B3 ;


(b) B4 , B5 , 86 ; (c) G3 , D2 , D3 ; (d) D4 , D5 , D6 ; and
(e) L2 , L4 , L 6 . Note the exceptional performance of the higher-
order Bessel aligmnents.

Fig. 13 Plots of the lobing error E(wn) for (a) the standard odd-
order Butterworth alignments B 1 , B3 (with inverted high-pass),
and B5 ; (b) the constant-voltage B 2 asymmetrical, constant-
voltage symmetrical second-order (a = 2 + _), and constant-
voltage quasi-second-order (c = 0.5) crossovers.

3O
Fi9. 14 The circuit topology of a 3-way time-delay-derived linear-phase
crossover network satisfying Eq. (35).

Fig. 15 (a) The delay-derived crossover of Fig. 6(b) incorporating

correction filters C L and CH to tailor the drivers' behaviour


(see text). (b) The same circuit with the correction filters

CL and CH moved to before the summing point.

Fig. 16 Illustrating the effect of a delay error upon high-pass

performance of the General G3 delay-derived crossover:


(a) correct delay Tn , (b) delay of 0.gTn , and (c) delay
of 1.1T n The rolloff rate is reduced ultimately to
20 dB/decade.

Fig. 17 Showing the notch produced in the high-pass output by increasing

the time-delay of the delay-derived B3 crossover to 1.05 times


its design value. The rolloff rate below the notch is only
20 dB/decade.

31
i 80 °

Ltl Oo

- t 80 °

-o ,?f-

l- -zo ,J

.o'
-40
O.Ol O. I I lO 100

NORMAL I ZED FREIqUEN (Y


FiG, l,

-r'

- I AID'

-I3

I--

-tO

-20
0.1 I In

N_RMALI ZED FRE_U£N(_


FI_. 2

32
LBO° -_......._.
I,I
L_ FIo _
ax _
-1}30 °

_-_ 0
'13

ILl

}---

LO -tn
mr'

-2U
o.I I 10

NBRMRLIZED FRE_UEN£?
ELC.,_,_

tBB °

tJ'l
r:F_. 0 °

r'_

- t B_ °
,,f

'%1

I---

-7
LO -20 /'

\
-qO _
0.01 O.l I 10 100

NDRMRLIZED FREBUENCY
Fl6, q(a)

33
IBD°

Liq D·
-1-

-L_D o

-D
L.----J

t.d

'm

z -2.D
Lq
rr'"

-ND
D.Ot _ D.I I ID lDO

NDDMFIL I ZED Fi:(EI_LIENCY

t SD° i

rr' Do _
r'_

- t EtD° [

-'0

Ld

b--

7 -z?.D
I_g

..r-

-qD
D.DL fi. t t I0 LDO

EL_ NDRMFIL LZED FREDLIENCY

34
>_ _ N-P
OUTPUT
INPUT o,

t . HL(S) 0 L-P HL
OUTPUT
L-P FILTER

(A)

TIME-DELAY T

e'_ I 4-_ "° "'POUTPU


HH

INPUT 0

HL(S) 0 L-POUTPUT
HL
L-P FILTER

(B)

FIG, 6

35
LA
r-m- I'1'-,.,
-17'
In

i.i

l..Q

_.1 I ID

Fm, 7_) NDRMRLIZED FREBUENC¥

lflO......

-IBn'

,-o,

mz -113

-2fl /
O.I I ID

Fm, 7(e) NORMRL I ZED FREBUENCY

36
~ __0 °

t
It
!
D° ,.

O_

-lSD °

_-_ D
-U

Ld

--1

z
L_
-ID
n-

-2D
O.[ I ID

Fm,__(c) NDRMRLtZED FREf;UENCY

un Ej
0
O3
Z_S
n

- IBD °

¢--. D
-0

Ld

:--1

-ID
n-

-2D
D.I I _D

NDRMFIL I ZED FREDUENCY

38
- I BDo

z -ID

-2B
B.I I B

_,8(E) NBRMRLtZED FREBUEN(Y

LB_° _%_, ' .


Ld
Ln Q°
r-r"
-r-
r"l
- 1BDo

bi]

L,.J
-'1

7 -lB
r",-

-2D
D.I I D

Rm6,
8(F) NBRMRLI ZED FREF,
ilJENC¥
39
..'".

-a, J3 ......,......"" " .......

L._I
_,
,:_ ,...

z -2tm
L-_ "'
r'r-

./

-40
0.01 O,I I 10 100

NI3RMFIL 1ZED FREDUENCY


F_, 9

LP.
rr' I1
o .....
-r-

-- I BFI"

a
-'D

ILl
'-'1
b---

Z --10
rr-

-20
\
0.1 I 10

FIc,_(A_ NDRMRLIZED FREDUENCY

4O
r'r- .....

-iBF1 °

I,I

I--

-iD

-213--
13.1 I In

10(_)
FiG, NIFIRI"IFIL
I ZED FREBUEN£ I/

ri-'"' t Bn
un
_a:::: °n° '---------------% '"'-...._
- i BI-i °

r'----',,._.._,....Uba_.....,LUr.r.
-'_ _lob *'

-213
.I I 10

Ft6,10(C)_ NBRMRLl ZED FREBUENC_'

41
L,J '_%
LFi O°

- I BFI °

-tO

l
-20
B.I I 10

NBRMFIL 1ZED F REBLIENC'r'


FIG, lQ_I_)_

i,I

-r-

- 1BB °

ILl

---1
I---

7
LD
-10
m-'

-20
.I I I[3

FIG, lQ(E) NBRMRL I ZED F REBLIENC ¥

42
LE]r'I"

- I BD'

'"'El

'"'m
F.--

rT" /z

-- /

-2D
D.I { ID

NBRMRLIZEB FREDUENC_

uq
rm Do
'-r-
ri

-IBD °

,--..-, B
--.o,
Ld

I'--

zL_ -iD
rr"

-2FI
B.I I ID

NDRMRL
I ZED FREBLIEN<:
¥

43
iJ'l
n- Do
-r'
O.
I_O
° ___
-lBO °

_--_ D
r-n
-D

.Ld

I--

z -In
n-

-2D
D.I I ID

NDRMRL I ZED FREr_UENC Y'


FIG_

[]

.......'

-I
01 I 0

NDRMRLIZED FREDUENC_
Ez_JZ_
44
4-t

-I
D.I I ID

NDRMRLIZED FREBUENCY

4-1

B.I I ID

NI3RIdRL I ZED FREf_UENC¥


E__, 12(c)
45
4-1

Il:
r_
r-¢

Ld ._5

L_
Z

_n
El
._J

-I
D.I I ID

NORMRLI ZED FREDUEN{Y


FiG, ]2(DZ

4-I

rm

m-'
L_J
[]
in
z

r-n [6"",."
cz]
_J

-I
DI I D

NBRMRLIZED FREBUENC_

46
4-1

n,,
g3
m,,
tv'
I.,J
........ ST_ B_,
_, (4)
In ...... ''",. .,."°' .........................

..J

- I ' ST^m_,D13_

['1.[ I ID

NORMRLl ZED F RI:'_UENCY

4-1
(1) CONSTANT-VO..TAGE
ASYMMETRICAL2I_I)-OR_ER,
Fa. (8)
(II) CONSTANT.-VO..TAGE
S'fi_ETEICAL
21_]-'OF_ER,
EQ. (g)
',III) CONSTANT-VOLTAGE
QLIASI-_II>-OEI_R,
EQ, CiO)

Q_
r"1
n,,
r',-'
I.,..I

_Z "%'"'".

-,
O.l I ID

NDRMRLIZED FEE_I]ENC ¥

47
TIME-DELAY T2

HH
TIME-DELAY T 1 o H-P OUTPUT

INPUT 0

c)
L-P
HL OUTPUT
L-P FILTER l

FI6, 14

-_- _ _ TH(S)
INPUT 0

)) TL(s)
L-P FILTER DRIVER DRIVERS ACOUSTIC
CORRECTION OUTPUT
NETWORKS

(A)

"P°'°
I, IHL,s,cL,__
Ic sL s'l
(B)

FIG.
15 48
l BD °_ _"--'-_. "_ _ _'

- 1ED °

,-n

l--

r't-

-ND
D.DL D. I t D LDO

F G_ NDRHRL 1ZED FREI_UENCY

Ld

Un Do
-?-
o..

- l BD°

"D
u__._4

Ld

ZzI
k--

z -2D
L._

-ND
D.DI D, I I ID lDO

':-_GL__G_L_ NDRMRL l ZED FRISI_UENCY


49
LBn ° -- _ · ·

un
n- Do

- I BO o

-_B
O.Ot B.I t LO lBO

NDRMHLIZED FFIEBUENCY
FIG,_(c)

ul Bo
Iq ......................
[BB °
- t EIB"

n
I::l:]

z -2FI
EZ;

-HI] .....

n.nt n. t t in LnB
NFIRMFIL I 'ZED FREQUENCY
FI6, 17

5O

Das könnte Ihnen auch gefallen