You are on page 1of 3

1136

IEEE PHOTONICS TECHNOLOGY LETTERS, VOL. 10, NO. 8, AUGUST 1998

Efficient Architectures for Exactly


Realizing Optical Filters with
Optimum Bandpass Designs
C. K. Madsen

AbstractButterworth, Chebyshev, and elliptic bandpass filter


designs are optimal in the sense of band flatness or equiripple
characteristics. A new architecture using optical all-pass filters is
presented which can realize these designs exactly and efficiently
using either ring resonators or reflectors such as Bragg gratings
or thin-film interference filters. Design examples are given for a
seventh- and eighth-order elliptic filter, and the new architecture
is shown to be tolerant to loss. Previously, reflective filters could
only approximate optimal responses. An order of magnitude
improvement in transition width is demonstrated for an elliptic
filter compared to an optimized transmission response for an
individual thin-film filter.

Optimum bandpass filters can be realized as the sum or difference of two all-pass functions [2]. The frequency response
of each all-pass function is written as
and

and
leads to the following
The sum and difference of
expressions for the magnitude response of two new functions,
and , respectively,
(2)

Index Terms All-pass filters, bandpass filters, dielectric resonator filters, gratings, lattice filters, optical filters, optical waveguide filters.

PPLICATIONS for optical bandpass filters include


demultiplexing and add/drop in WDM systems. Basic
filtering techniques include multipath feed-forward interference using an array of waveguides or a lattice of
coupled MachZehnder interferometers (MZIs) and feedback
interference using thin-film filters, Bragg gratings, or ring
resonators. Filters having only feedforward interference paths
contain only zeros in their transfer functions; whereas, filters
with feedback paths contain both poles and zeros. Optimum
bandpass designs include the Butterworth, Chebyshev, and
elliptic filters, where the design is optimized with respect to
characteristics of the passband and stopband such as maximum
flatness or equiripple. Their transfer functions contain both
poles and zeros which must be located at particular values.
The only optical filter reported to date which can realize these
responses exactly is a lattice structure interleaving MZIs and
ring resonators [1]. A new optical architecture is presented
which allows the optimum filter designs to be implemented
exactly using thin films, Bragg gratings, or ring resonators as
the basic filtering elements. The new architecture is simpler
and more efficient than [1], requiring fewer couplers and phase
shifters, and easily accommodates reflective feedback paths.
Reflective feedback paths have a major advantage over rings
in that the period of the frequency response, the free-spectral
range (FSR), is not limited by the bend radius so broad-band
designs can be easily realized.
Manuscript received February 13, 1998; revised March 24, 1998.
The author is with the Lucent Technologies, Bell Laboratories Innovations,
Murray Hill, NJ 07974 USA.
Publisher Item Identifier S 1041-1135(98)05354-3.

(1)

(3)
When both all-pass functions have the same phase, their sum
is maximum. When their phases differ by , the sum is zero
but the difference is maximum. The resulting functions and
are power complementary, i.e.,
(4)
transFor filter synthesis, it is convenient to work with
with the following expressions are
forms. By replacing
and :
obtained for
(5)
(6)
and
share a common
Note that
, but have different numerator
polynomial
and
. The transform of an all-pass
a special symmetry between the numerator and
polynomials as shown in (7):

denominator
polynomials
function has
denominator

(7)

where represents a constant phase. The zero locations are


mirror images about the unit circle from the pole locations
which gives unity magnitude response for all frequencies.

10411135/98$10.00 1998 IEEE

MADSEN: EFFICIENT ARCHITECTURES FOR EXACTLY REALIZING OPTICAL FILTERS

1137

(a)

(a)

(b)

(b)
(c)
Fig. 1. Architectures: (a) schematic for a bandpass filter realized with two
all-pass filters; (b) an even order cascade ring implementation; and (c) an odd
order coupled cavity implementation using reflectors.

The optimal filter designs previously mentioned can be


decomposed into a linear combination of all-pass functions.
,
, and
have real coefficients and the numerator
has mirror image symmetry. In addition,
polynomial of
is power complementary to
whose coefficients
have even or odd symmetry depending on whether the filter
order is even or odd, respectively, where even symmetry refers
and odd symmetry to
. The details
to
has even or
of the synthesis method depend on whether
odd order. The linear combination of all-pass functions which
and
is expressed in (5) and (6). If
has
define
and
are complex conjugates and
even order, then
poles and zeros each.
have
I. OPTICAL ALL-PASS AND BANDPASS FILTERS
The sum and difference are easily implemented in an optical
filter using directional couplers. By setting the power coupling
50%, the sum and difference of the two all-pass
ratio
functions are realized in the cross and bar transmission ports
as shown in Fig. 1(a). The directional coupler through and
and
cross amplitude transmission are denoted by
, respectively. Because is imaginary, the optical
and
filter transfer functions are given by
Several optical all-pass filter architectures
are presented in [3]. A lossless ring resonator with a single
coupler is a single stage all-pass filter. Its transfer function in
the -domain is given by
(8)
, the ring circumference is
where
with radius , and the effective index is for the fundamental
. Bend loss limits the
mode. The FSR is given by FSR
smallest radius which can be used, and thus limits the largest

Fig. 2. (a) An eighth-order elliptic filter with a transition width of 0.0035

2 FSR. The coupling ratios and phases for a cascade implementation are
shown in the inset. (b) Transmission and reflection responses for an optimized
eighth-order all-pole filter which has a transition width of 0.035 2 FSR.
achievable FSR. An th-order all-pass filter can be realized by
cascading rings as shown in Fig. 1(b). The coupling ratios and
phases for each ring are calculated from the all-pass functions.
requires a coupling ratio
For example, to produce a pole at
and a phase of
. The rings are
of
nominally the same circumference with an incremental length
or index change added to each ring to realize the phase, i.e.,
. For example, heaters may be employed
on a section of the ring to locally change the index and create
the desired phase. The new design for optimum bandpass filters
couplers and
phase shifts. In contrast,
requires
the general waveguide filter architecture [1] requires
couplers and
phase shifters to realize an th-order filter.
A second ring resonator structure, known as a lattice filter,
where the rings are coupled to each other, can also realize an
all-pass function [3]. An analog to the lattice ring is a coupled
cavity structure [4] created with reflectors. To form an all-pass
function, the last reflector must have an amplitude reflectance
1. Fig. 1(c) shows a bandpass architecture
of unity, i.e.,
using two all-pass reflective lattice filters. The FSR is given
where
is the nominal cavity length.
by FSR
The cavity length or index is varied for each stage to create
. In the
transform description, the
the relative phase
amplitude reflectances replace the coupler transmittances, .
The partial reflectors must be wavelength independent over
the FSR so that the filter coefficients are constant. Thin
films with alternating layers of high and low index or Bragg
gratings can be used to realize the partial reflectors. In either
case, the index difference must be large enough to satisfy
the wavelength independence constraint. Partial reflectors in
waveguides should be realizable with fabrication methods used
to demonstrate photonic bandgaps [5].

1138

IEEE PHOTONICS TECHNOLOGY LETTERS, VOL. 10, NO. 8, AUGUST 1998

POLYNOMIALS FOR AN EIGHTH-

TABLE I
AND A SEVENTH-ORDER ELLIPTIC FILTER

Fig. 3. An eighth-order elliptic filter with 0.2-dB loss/feedback path length.

II. DESIGN EXAMPLES


An eighth-order elliptic filter is designed as a first example.
The design was chosen to have a 30-dB stopband rejection
and
and a cutoff of 0.1
FSR. The
for both
stopband rejection requirement for one response implies a
passband ripple of 0.004 dB for the other response. Typically,
or
, is given from a filter
only one response,
design program. The details for determining the other response
as well as decomposing them into the all-pass filters
and
are covered in [2]. The polynomials
,
,
are listed in Table I. The denominator polynomials
and
for the all-pass functions are both fourth order. The all-pass
functions are defined as
and
. The phase factors
and
are given in radians. Table I also contains the
polynomials for a seventh-order elliptic design. Now the
and
are real, so their roots occur
coefficients of
on the real axis or in complex conjugate pairs. Fig. 2(a) shows
the magnitude response for the transmission cross and bar
states for the eighth-order elliptic design. The normalized
frequency range of 0 to 1 corresponds to one FSR of the
and
, the coupling
frequency response. Given
ratios and partial reflectances are determined via (8) for the
cascade architectures or by step-down recursion relations [6]
for the lattice structures. The coupling ratios are shown for
the cascade implementation in the inset. The coupling ratios
and
while the phases are complex
are identical for
conjugates of each other. A schematic of a cascade ring
configuration is shown in Fig. 1(b) for an even ordered filter.
As seen for the single ring case in (8), there is a phase constant
,
equal to the sum of the individual phases, i.e.
associated with each all-pass transfer function which is not
and
equal to in general. In Fig. 1(b), the phase terms - are included in the upper and lower arms, respectively.
In Fig. 2(b), an all-pole transmission response is shown which
represents an optimized thin-film design with the same order,
cutoff frequency, and passband ripple. An all-pole response
cannot meet equiripple requirements in both the passband and
stopband. As an example, let the FSR 40 nm for both filters.

Then, the passband to stopband transition width is 1.4 nm for


the all-pole design compared to 0.14 nm for the elliptic design.
For WDM systems, filters with small transition widths are
important because they can accommodate narrower channel
spacings.
In the presence of loss, the all-pass property is broken.
for
,
This is easily verified in (8) by substituting
.
where loss in decibels for one feedback path is
With the eighth-order design, a loss of 0.2-dB/feedback path
was introduced. The resulting magnitude response is shown in
is reduced, but the
Fig. 3. The peak transmission of
is retained.
30 dB stopband rejection of
In summary, a new architecture was presented that allows
optimum bandpass filters to be exactly realized with thin films,
Bragg gratings, or ring resonators. Because reflective feedback
structures can be used, the FSR is not limited by bend radius,
as is the case with rings. The new architecture requires fewer
couplers and relative phase shifts per stage compared to a
general waveguide architecture, and it is robust to loss. In
addition, switching between the two power complementary
responses is easily achieved by introducing a -phase change
between the interferometer arms. Some applications include
demultiplexing, adding or dropping single channels as well as
filtering bands of channels for WDM systems.
REFERENCES
[1] K. Jinguji, Synthesis of coherent two-port optical delay-line circuit with
ring waveguides, J. Lightwave Technol., vol. 14, pp. 18821898, Aug.
1996.
[2] S. Mitra and J. Kaiser, Handbook for Digital Signal Processing. New
York: Wiley, 1993.
[3] C. K. Madsen and G. Lenz, Optical all-pass filters for phase response
design with applications for dispersion compensation, Photon. Technol.
Lett., vol. 10, pp. 994996, July 1998.
[4] E. Dowling and D. MacFarlane, Lightwave lattice filters for optically
multiplexed communication systems, J. Lightwave Technol., vol. 12,
pp. 471486, Mar. 1994.
[5] J. Foresi, P. Villeneuve, F. Ferrara, E. Thoen, G. Steinmeyer, S. Fan, J.
Joannopoulos, L. Kimerling, H. Smith, and E. Ippen, Photonic-bandgap
microcavities in optical waveguides, Nature, vol. 390, pp. 143145,
Nov. 1997.
[6] C. Madsen and J. Zhao, A general planar waveguide autoregressive
optical filter, J. Lightwave Technol., vol. 14, pp. 437447, Mar. 1996.

You might also like