Free-space optical delay interferometer with

tunable delay and phase

Jingshi Li,1,* Kai Worms,1 Ruediger Maestle,2 David Hillerkuss,1 Wolfgang Freude,1 and
Juerg Leuthold1
1
Institute of Photonics and Quantum Electronics, Karlsruhe Institute of Technology, Karlsruhe, D-76131, Germany
2
Agilent Technologies, Boeblingen, Germany
*[email protected]

Abstract: A free-space optical delay interferometer (DI) featuring a conti-

nuously tunable time delay, polarization insensitive operation with high ex-
tinction ratios and accurate phase and time delay monitoring scheme is re-
ported. The polarization dependence is actively mitigated by adjusting a bi-
refringent liquid-crystal device. The DI has been tested for reception of
D(m)PSK signals.
©2011 Optical Society of America
OCIS codes: (060.5060) Phase modulation; (120.3180) Interferometry.

References and links

1. R. A. Griffin, R. I. Johnstone, R. G. Walker, J. Hall, S. D. Wadsworth, K. Berry, A. C. Carter, M. J. Wale, J.
Hughes, P. A. Jerram, and N. J. Parsons, ―10 Gb/s optical differential quadrature phase shift key (DQPSK)
transmission using GaAs/AlGaAs integration,‖ in Proc. Optical Fiber Communication Conference (OFC'02),
(Anaheim, CA, USA, 2002), Postdeadline Paper FD6.
2. A. H. Gnauck, G. Raybon, S. Chandrasekhar, J. Leuthold, C. Doerr, L. Stulz, A. Agarwal, S. Banerjee, D. Grosz,
S. Hunsche, A. Kung, A. Marhelyuk, D. Maywar, M. Movassaghi, X. Liu, C. Xu, X. Wei, and D. M. Gill, ―2.5
Tb/s (64´42.7 Gb/s) transmission over 40´100 km NZDSF using RZ-DPSK format and all-Raman-amplified
spans,‖ in Proc. Optical Fiber Communication Conference (OFC'02), (Anaheim, CA, USA, 2002), Postdeadline
Paper PD FC2.
3. N. Kikuchi, K. Mandai, S. Sasaki, and K. Sekine, ―Proposal and first experimental demonstration of digital
incoherent optical field detector for chromatic dispersion compensation,‖ in Proc. of European Conference on
Optical Communication (ECOC’06), (Cannes, France, 2006), paper Th444.
4. X. Liu, S. Chandrasekhar, and A. Leven, ―Digital self-coherent detection,‖ Opt. Express 16(2), 792–803 (2008).
5. J. Li, K. Worms, P. Vorreau, D. Hillerkuss, A. Ludwig, R. Maestle, S. Schuele, U. Hollenbach, J. Mohr, W.
Freude, and J. Leuthold, ―Optical vector signal analyzer based on differential direct detection,‖ in Proc. of IEEE
Photonics Society(LEOS) 2009, (Belek-Antalya, Turkey, 2009), Paper TuA4.
6. J. Leuthold, B. Mikkelsen, G. Raybon, C. H. Joyner, J. L. Pleumeekers, B. I. Miller, K. Dreyer, and R. Behringer,
―All-optical wavelength conversion between 10 and 100 Gb/s with SOA delayed-interference configuration,‖
Opt. Quantum Electron. 33(7/10), 939–952 (2001).
7. S. Sygletos, R. Bonk, T. Vallaitis, A. Marculescu, P. Vorreau, J. Li, R. Brenot, F. Lelarge, G.-H. Duan, W.
Freude, and J. Leuthold, ―Filter assisted wavelength conversion with quantum-dot SOAs,‖ J. Lightwave Technol.
28(6), 882–897 (2010).
8. D. Hillerkuss, M. Winter, M. Teschke, A. Marculescu, J. Li, G. Sigurdsson, K. Worms, S. Ben Ezra, N. Narkiss,
W. Freude, and J. Leuthold, ―Simple all-optical FFT scheme enabling Tbit/s real-time signal processing,‖ Opt.
Express 18(9 Issue 9), 9324–9340 (2010).
9. G. Bosco and P. Poggiolini, ―On the joint effect of receiver impairments on direct-detection DQPSK systems,‖ J.
Lightwave Technol. 24(3), 1323–1333 (2006).
10. B. Mikkelsen, C. Rasmussen, P. Mamyshev, and F. Liu, ―Partial DPSK with excellent filter tolerance and OSNR
sensitivity,‖ Electron. Lett. 42(23), 1363–1364 (2006).
11. Y. Nasu, K. Hattori, T. Saida, Y. Hashizume, and Y. Sakamaki, ―Silica-based adaptive-delay DPSK demodulator
with a cascaded Mach-Zehnder interferometer configuration,‖ in Proc. European Conference on Optical Com-
munication (ECOC), 2010, (Torino, Italy, 2010) paper We.8.E.5.
12. L. Christen, Y. K. Lize, S. Nuccio, L. Paraschis, and A. E. Willner, ―Experimental demonstration of reduced
complexity 43-Gb/s RZ-DQPSK rate-tunable receiver,‖ IEEE Photon. Technol. Lett. 20(13), 1166–1168 (2008).
13. H. Kawakami, E. Yoshida, Y. Miyamoto, and M. Oguma, ―Analysing the penalty induced by PD of MZI in
DQPSK receiver using novel measuring technique,‖ Electron. Lett. 43(2), 121–122 (2007).
14. H. Kawakami, E. Yoshida, Y. Miyamoto, M. Oguma, and T. Itoh, ―Simple phase offset monitoring technique for
43 Gbit/s optical DQPSK receiver,‖ Electron. Lett. 44(6), 437–438 (2008).

#137432 - $15.00 USD Received 18 Feb 2011; accepted 26 May 2011; published 1 Jun 2011
(C) 2011 OSA 6 June 2011 / Vol. 19, No. 12 / OPTICS EXPRESS 11654
15. Z. Tao, A. Isomura, T. Hoshida, and J. C. Rasmussen, ―Dither-free, accurate, and robust phase offset monitor and
control method for optical DQPSK demodulator‖ in Proc. European Conference on Optical Communication
(ECOC), 2007, (Berlin, Germany, 2007) paper Mo.3.D.5.
16. L. Rovati, U. Minoni, M. Bonardi, and F. Docchio, ―Absolute distance measurement using comb-spectrum
interferometry,‖ J. Opt. 29(3), 121–127 (1998).
17. A. Cabral and J. Rebordao, ――Accuracy of frequency-sweeping interferometry for absolute distance metrology,‖
Opt. Eng. 46, 073602 (2007).
18. S. Schüle, U. Hollenbach, J. Mohr, J. Li, P. Vorreau, A. Efremov, J. Leuthold, and S. Schonhardt, ―Modular
integration of microactuators and micro-optical benches,‖ in Proc. of SPIE Conference on Micro-Optics, (Stras-
bourg, France, April 2008) vol. 6992, pp. 699202–699202–12 (2008).
19. J. Li, K. Worms, D. Hillerkuss, B. Richter, R. Maestle, W. Freude, and J. Leuthold, ―Tunable free space optical
delay interferometer for demodulation of differential phase shift keying signals,‖ in Proc. Optical Fiber Commu-
nication Conference (OFC 2010), (San Diego, Ca, USA, 2010), paper JWA24.
20. D. O. Caplan, M. L. Stevens, and J. J. Carney, ―High-sensitivity multi-channel single-interferometer DPSK
receiver,‖ Opt. Express 14(23), 10984–10989 (2006).
21. A. A. Freschi and J. Frejlich, ―Adjustable phase control in stabilized interferometry,‖ Opt. Lett. 20(6), 635–637
(1995).

1. Introduction
Delay interferometers (DIs) are needed to demodulate differential phase shift keying modula-
tion formats, e.g. DBPSK, DQPSK and higher-level formats like D8PSK (for short,
D(m)PSK). These formats have various advantages regarding noise, nonlinearities, and dis-
persion tolerance [1,2]. Recently, DIs have been used to detect the amplitude and optical
phase of coherent data signals [3–5]. The main advantage of such a ―self-coherent‖ detection
scheme [4] over a real coherent reception is that self-coherent detection does not require an
expensive, possibly wideband-tunable local oscillator. Moreover, DIs have also been used as
optical filters for all-optical wavelength conversion [6,7]. More recently, DI’s have been used
as ultrafast optical FFT processing elements enabling FFT processing of a 10 Tbit/s OFDM
signal [8].
While delay interferometers have become important for all of the above and many more
applications the fabrication of versatile delay interferometers is a challenge to this day. For the
above mentioned applications, a versatile DI should feature:
• Tunability in delay time and tunability for phase. In almost any of the above mentioned
applications a good reception quality, requires adaption of the time delay to the sym-
bol rate [9]. In again other instances, it is sometimes advantageous to deviate from
the one-symbol delay in order to mitigate transmission impairments caused by effects
such bandwidth-narrowing by concatenated filters [10]. A DI with a continuously
tunable delay would fulfill both requirements. Recently, a DI with adaptive delay has
been presented by using cascaded Mach-Zehnder interferometers [11], however this
only provides a discrete switching of three delays. A continuously tunable DI has
been proposed in [12], where the delay is introduced by passing two orthogonal pola-
rizations through a tunable differential group delay (DGD) element. This scheme,
however, requires that the input signal has equal power for both orthogonal polariza-
tions.
• Low polarization-dependent loss (PDL), and especially a low polarization-dependent
frequency shift (PDFS), which is of particular importance for demodulating
D(m)PSK signals [13]. This is normally achieved by carefully selecting the optical
coatings, which should perform for all polarizations alike.
• Accurate monitoring and control of the operating point defining phase and time delay.
Phase monitoring with a fixed time delay has been intensively discussed based on RF
power monitoring [14] or based on a correlation method [15]. Although the efficien-
cy of both methods has been demonstrated, the techniques are complicated, and they
are limited to certain data formats and bitrates. A pilot-tone driven lock-in algorithm
can be much more effective in fixing the DI operating point, especially for DQPSK

#137432 - $15.00 USD Received 18 Feb 2011; accepted 26 May 2011; published 1 Jun 2011
(C) 2011 OSA 6 June 2011 / Vol. 19, No. 12 / OPTICS EXPRESS 11655
 reception, when a DI pair needs be locked to the desired π/2 relative phase offset.
Sophisticated time delay control techniques can be found in the field of distance me-
trology [16,17]. A simple solution would be to use a known pilot tone, and to count
fringes when tuning the DI.
In this paper we present a polarization-insensitive, continuously tunable free-space optical
delay interferometer. Our approach operates over a broad wavelength range.
The paper is organized as follows: We first model an optical delay interferometer mathe-
matically. Then we present an appropriate free-space setup, which provides a calibrated, con-
tinuously tunable time delay from 0 ps to 100 ps. We achieve PDFS mitigation by employing
a liquid-crystal device to compensate the birefringence within the system. Delay and phase are
controlled with the help of a pilot tone. Experimentally, we demodulate DQPSK signals at
11.7 GBd, 28 GBd and 42.7 GBd, present eye diagrams and bit error ratios (BER), and com-
pare the outcome to the results measured with a commercially available DI with fixed delay.
In the last section we show that with two additional polarization beam splitters four logic DIs
can be folded into one single DI which serves as a tunable polarization division multiplexing
(PDM) receiver for polarization multiplexed D(m)PSK signals.
2. DI modeling
A schematic DI layout is shown in Fig. 1. It consists of two couplers and two optical paths of
different length. The upper path has a time delay of ΔT compared to the lower path. The input
electrical fields Ein,1 and Ein,2 are split by the first coupler (SI), then experience two different
optical paths, and are then combined at the second coupler (SII), where two output fields Eout,1
and Eout,2 are generated.

A T
Ein,2 Eout,2

SI B SII

Ein,1 Eout,1
Fig. 1. Schematic of an optical delay interferometer (DI). Inputs (Ein, 1, and Ein,2) are split by a
coupler (SI) into two paths with A and B as the respective power loss factors. A time delay ΔT
is introduced between the two paths. The signals interfere on the other coupler (SII) where two
outputs (Eout, 1 and Eout,2) are generated.

Following [6] we convolve (symbol *) the impulse response matrix of the DI with the
time-dependent column matrix of input electric fields Ein,1,2, and find the time-dependent
output electric fields Eout,1,2 (only one polarization is considered):

 Eout,2   Ein,2   a 1  sI  j asI exp  j12I 

E    S II T S I  *  , where S I   ,
 out,1   Ein,1   j as exp  j I  a 1  sI  
 I 21 
(1)
 A  t  T  0   b 1  sII  j bsII exp  j12II  
T  , and S II   .
 0 B  t    j bs exp  j II  b 1  sII  
 II 21 
The matrices SI and SII in Eq. (1) describe the two couplers, where a phase factor j provides
the ideal phase relations between the two output signals, which might be impaired by phase
offsets δθI12, δθI21, δθII12, and δθII21.The quantities asI, a(1  sI ), bsII and b(1  sII ) are the
power splitting ratios of the two couplers in the form of amplitude factors. If a = 1 and b = 1,
the couplers are lossless. The Dirac distributions in matrix T are the impulse responses of the

#137432 - $15.00 USD Received 18 Feb 2011; accepted 26 May 2011; published 1 Jun 2011
(C) 2011 OSA 6 June 2011 / Vol. 19, No. 12 / OPTICS EXPRESS 11656
―long‖ (upper) and the ―short‖ (lower) arms having a group delay difference ΔT, see Fig. 1.
Assuming a monochromatic optical signal with angular frequency ωc = 2π fc, a change of ΔT
introduces a phase offset ΔTωc in the upper arm. The quantities A, B < 1 are the power loss
factors in the two paths.
For a D(m)PSK signal demodulator with only one input to be used, i.e., Ein,1  0, Ein,2 = 0,
Eq. (1) simplifies to

Eout,1  t    AabsI sII    T  exp  j  21II  12I  


 Bab 1  sI 1  sII     Ein,1  t    d ,


(2)
Eout,2  t   j   AabsI 1  sII    T  exp  j I
12 


 BabsII 1  sI    exp  j12II   Ein,1  t    d .

We can also write Eq. (2) in the frequency domain. We define the Fourier transforms of Eout,1,
2(t) and Ein,1(t) by Êout,1,2(f) and Êin,1(f), respectively. The associated transfer functions are H1,
2(f) = | H1, 2(f)|exp[ jΦ1, 2(f)] = Êout,1, 2(f) / Êin,1(f).
The power transfer function then becomes

H1  f  = Bab   BabsI  BabsII   A  B  absI sII  2S cos  2πf T   21II  12I   ,

2

H 2  f  = BabsII  AbsI   A  B  absI sII  2S cos  2πf T  12II  12I  ,

2

where S =ab ABsI sII 1  sI 1  sII .

(3)
For the associated phases we find:

 AabsI sII sin  2πf T   21II  12I  

1 ( f )  tan 1  ,
 Bab(1  sI )(1  sII )  AabsI sII cos  2πf T   21  12  
II I
 
(4)
 AabsI (1  sII ) cos  2πf T  12I   BabsII (1  sI ) cos 12II  
 2 ( f )  tan 1  .
 AabsI (1  sII )sin  2πf T  12I   BabsII (1  sI ) sin 12II  
 
From the frequency dependence of the phase we derive the group delay τ1, 2 (f) = 1/(2π) dΦ1,
2(f) / d(f),

AabsI sII  S cos(2πf T   21II  12I )

 1  f  = T ,
H1  f 
2

(5)
AabsI   AabsI sII  S cos(2πf T  12II  12I ) 
 2  f  = T .
H2  f 
2

For an illustration of Eq. (5) we consider two extreme cases: When sI, sII = 0, a, b = 1, the
input field Ein,1 in Fig. 1 passes exclusively through the lower arm to the output 1. The squared
magnitude of the transfer function at the outputs is calculated to be |H1(f)|2 = B, |H2(f)|2 = 0,
and the output group delay is τ1 (f) = 0 according to the assumption Eq. (1). When sI, sII = 1, a,
b = 1, the input Ein,1 in Fig. 1 propagates through the upper arm to output 1 without coupling
into the lower arm. The squared magnitude of the transfer function at the outputs can be calcu-

#137432 - $15.00 USD Received 18 Feb 2011; accepted 26 May 2011; published 1 Jun 2011
(C) 2011 OSA 6 June 2011 / Vol. 19, No. 12 / OPTICS EXPRESS 11657
lated, |H1(f)|2 = A, |H2(f)|2 = 0, and the associated output group delay is τ1 (f) = ΔT according to
the assumption formulated in Eq. (1).
We describe an ideal DI (no loss, symmetric splitting) by the parameter set
1
A, B, a, b  1, sI , sII  , and 12I ,  21I , 12II ,  21II  0. (6)
2
We define f0 as a frequency where destructive interference occurs at output 1, and constructive
interference at output 2. We plot |H1,2(f)|2 as well as phases Φ1, 2(f) and group delays τ1, 2 (f) as
a function of the frequency offset from f0, blue curves in Fig. 2. The frequency offset is given
in units of the free spectral range FSR = 1 / ΔT. The two outputs have complementary power
transfer functions H1,2  f   1 2  1 cos  2πf T   . This leads to an infinite extinction ratio
2

ER = (maximum of output 1, 2) / (minimum of output 2, 1), Fig. 2(a). Whenever destructive

interference occurs and the transfer function |H1,2(f)| crosses zero, a π-phase shift is observed,
Fig. 2(b). These phase jumps result in δ-functions of the group delay,  T 2    f  N 2T  ,
N = 0, ± 1, ± 2, ± 3,…, while τ1, 2(f) = ΔT / 2 holds at all other frequencies, Fig. 2(c).
In practice, losses and splitting ratios in a DI are not ideal as defined in Eq. (6). As an ex-
ample we assume 2 dB of losses in one arm, i.e. A = 0.64, while all other parameters remain
as in Eq. (6). For this case, the quantities |H1,2(f)|2, Φ1, 2(f) and τ1, 2 (f) are plotted in Fig. 2 (red
curves). In contrast to the ideal case, the red curves have a finite ER = 20 dB, and an addition-
al loss of 0.085 dB. Furthermore, instead of the π-phase jumps, the phase slopes are reduced
near the frequency for destructive interference, which results in a finite negative group delay,
Fig. 2(c). This behavior is typical for any deviation of A, B, a, b, and sI, sII from the ideal
conditions defined in Eq. (6). Imperfect phase offsets 12I , 21I , 12II ,  21II will introduce a
frequency shift of the transfer functions.
(a) 0 (b) 4 (c)
Group Delay [T ]

0
|H1,2( f )|2 [dB]

Phase [rad]

2
-20
-2
0

-40 -2 -4
-0.6 -0.3 0 0.3 0.6 -0.6 -0.3 0 0.3 0.6 -0.6 -0.3 0 0.3 0.6
Freq. offset from f0 [1/ T ] Freq. offset from f0 [1/ T ] Freq. offset from f0 [1/ T ]
Fig. 2. Transfer function of the DI for different upper-arm power losses A = 1 and A = 0.64.
The case A = 1 denotes an ideal DI (blue). The case A = 0.64 denotes an imbalanced DI with
excess losses in the delayed arm (red). The three plots show (a) the power transfer functions
|H1,2 (f)|2, (b) the phase response Φ1, 2(f) and (c) the group delay τ1, 2 (f) for the constructive out-
put port (dashed lines) and the destructive port (straight lines), respectively.

In the following we use the general model of Eq. (1) for designing a practical implementa-
tion of a free-space optics delay interferometer.
3. DI implementation and polarization dependence
DIs typically can be implemented as asymmetric Mach-Zehnder interferometers or as Michel-
son Interferometers. The two configurations are different though.
A Mach-Zehnder delay interferometer configuration is displayed in Fig. 3(a) [18]. The in-
terferometer is formed by a beamsplitter (BS) and a reflector. The BS splits the collimated
input signal (red) in two paths. For changing the optical path length difference, the reflector is
mounted on a micro-actuator to create a variable path length with an associated delay ΔT.
However we notice that the upper, longer path usually has more loss than the lower path due

#137432 - $15.00 USD Received 18 Feb 2011; accepted 26 May 2011; published 1 Jun 2011
(C) 2011 OSA 6 June 2011 / Vol. 19, No. 12 / OPTICS EXPRESS 11658
to the additional reflector, so that the extinction ratio is degraded. Moreover, the minimum
time delay difference ΔT between the two paths is also limited by the fact that the upper opti-
cal path is always longer.
Conversely, the Michelson interferometer allows a zero optical path difference and mi-
nimal power difference between the signals in both arms, thus providing a (theoretically)
infinite large free spectral range (FSR) in combination with a high ER at ―Output 1, 2‖, see
Fig. 3(b) [19].
A prototype has been built as depicted in Fig. 3(c). The signals are coupled with fiber grin
lenses into and out of the DI. The dimensions of 85 × 45 × 25 mm 3 can be further decreased
by using the LIGA technology as in [18]. The actuator used in this setup has a mechanical
step size of about 2 nm, which corresponds to a phase offset of about 1°. The mechanical
tuning range amounts to 15 mm corresponding to a time delay tuning range of 100 ps. As the
actuator responds within milliseconds, the tuning can be performed extremely fast.

(a) (b) Actuator (c)

Reflector Actuator
T = 0
Actuator Reflectors Reflectors
Liquid
T φpol Crystal
Beamsplitter 2 Input 2
Input 1 Output 2 LC
Output 1
α 1
Beamsplitter α
BS
Input 2 Output 2 Input 1 h Output 1
v
Fig. 3. Practical implementation of DI (a) Schematic of the Mach-Zehnder-DI with two non-
polarizing beam splitters (BS) combined in one unit, and one reflector. (b) Schematic of the
Michelson DI with single non-polarizing BS and two reflectors. A liquid crystal compensates a
polarization dependent frequency shift. (c) Photograph of a prototype of the Michelson DI.

Now we consider an input with two orthogonal polarization states, v and h (―v‖ vertical
polarization, and ―h‖ horizontal polarization with respect to the incident plane). Because each
surface in the system has polarization-dependent transmittivity and reflectivity, the transfer
function of the DI is seriously affected by the input signal polarization state. While polariza-
tion dependent reflectivity can be minimized by carefully selected coatings, there remains a
polarization-dependent phase shift. This phase offset results in a polarization dependent fre-
quency shift (PDFS) of the DI transfer function.
To measure the PDFS of a DI, one could align the input signal parallel to one of two or-
thogonal (linear) polarizations, and calculate the difference between the respective transfer
functions. In practice, because a DI is usually coupled to a standard single mode fiber, the
input polarization state for the DI cannot be known. Therefore, we insert a polarizer directly
after the input lens and record the spectral response of the DI for horizontal polarization. Then
by rotating the polarizer we record the spectral response for a vertical polarization. The spec-
trally shifted responses are found at ―Output 1, 2‖ and are depicted in Fig. 4(a). Then with the
knowledge of the PDFS, we adjust a birefringent element (by voltage tuning a liquid crystal
(LC), Fig. 3(b)) for compensating the polarization-induced phase offset. When applying 1 V
to the LC, we observe the spectral response in Fig. 4(b), where PDFS has been obviously
mitigated as compared to Fig. 4(a).
The PDFS measurement in Fig. 4(c) shows how the birefringence can be undone by detun-
ing the liquid crystal. We have plotted Δφpol = 2π (PDFS / FSR) against the voltage applied to
the liquid crystal. The resolution of the plot is limited by the resolution of the equipment that
allowed us to sweep the wavelength with steps of 10 pm only. For a FSR of 200 GHz (1.6 nm)
we can resolve a phase difference of 2.2° (0.61% of FSR).

#137432 - $15.00 USD Received 18 Feb 2011; accepted 26 May 2011; published 1 Jun 2011
(C) 2011 OSA 6 June 2011 / Vol. 19, No. 12 / OPTICS EXPRESS 11659
 Once the birefringence is undone, it is undone for any delay ΔT as an additional free space
does not add to the birefringence. In Fig. 4(a), 4(b) one can notice that the ER = (maximum of
output 1, 2) / (minimum of output 1, 2) at ―Output 2‖ for both polarizations is larger than 30
dB. At ―Output 1‖ and for h-polarization the ER is 30 dB as well, however for the v-
polarization only 18 dB is measured. This difference is due to the fact transmittivities and
reflectivities of the BS coating are slightly polarization dependent. In our case output 2 has an
almost ideal extinction ratio because the two beams constructively interfering in the beam
splitter and being mapped onto output 2 undergo one reflection and one transmission each.
However, the two beams interfering in the BS and being mapped to output 1 undergo two
reflections and two transmissions, respectively, which leads to a power imbalance (splitting
ratio difference) for the slightest imbalance in the transmission and reflection coefficients of
the BS coating. As we discussed above, this will lead to the ER difference between the two
outputs at two orthogonal polarizations. As commercially available fixed-delay DIs have
typically ER > 18 dB, this prototype has comparable performance. A PDFS < (0.61% of FSR)
is also good enough for DQPSK demodulation [9,13].
(a) 0 (b) 0
Calibrated |H1,2( f )|2 [dB]

Calibrated |H1,2( f )|2 [dB]

-10 -10

-20 -20

-30 -30
PDFS
1553 1554 1555 1556 1557 1553 1554 1555 1556 1557
Wavelength [nm] Wavelength [nm]
(c)
20 11
Phase Offset [°]

PDFS [GHz]
10 5.5

0 0

-10 -5.5
0.6 0.8 1 1.2 1.4 1.6
Applied Voltage [V]
Fig. 4. Spectral response for vertical (blue) and horizontal (red) polarizations. (a) Large PDFS,
(b) LC is used to undo birefringence, (c) PDFS of DI for different voltages applied onto the LC
(the precision of the measured offset phase is limited by the resolution of the measurement
equipment). The plots show the output 2 (dashed lines) and output 1 (straight lines).

Polarization dependent loss (PDL) and differential group delay (DGD) are important as
well. Commercial measurement instruments, e.g. Agilent 86038B Photonic Dispersion and
Loss Analyzer, Agilent N7788BD Benchtop Optical Component Analyzer, usually provide
such data. It is straight forward since PDL and DGD can be derived from the ratio of the
transfer functions at orthogonal polarizations,

 H  f 2 
PDL  10log10  pol v 
2 ,

 H pol h  f  
 (7)

DGD   pol v  f    pol h  f  .

#137432 - $15.00 USD Received 18 Feb 2011; accepted 26 May 2011; published 1 Jun 2011
(C) 2011 OSA 6 June 2011 / Vol. 19, No. 12 / OPTICS EXPRESS 11660
To get an idea of the PDL and DGD to be expected in this device we tune our prototype to
minimize PDFS as was done for measuring the data in Fig. 4(c). We record the average
squared magnitude of the transfer function for all possible polarizations, find the PDL and
detect the DGD as a function of wavelength, Fig. 5(a) and 5(b) (FSR 42.7 GHz). The peaks
in PDL and DGD are due to the different transfer functions for the orthogonal polarizations.
When the birefringence control voltage deviates from the optimum operating point, double
peaks are observed in the PDL and DGD curves. We then introduce PDFS in a second expe-
riment and consequently find strong PDL and DGD as plotted in Fig. 5(c) and 5(d). The
squared magnitude of the transfer function is averaged over all possible polarizations, and
therefore the measured ER is small. Thus, one could minimize PDFS by minimizing the PDL
and DGD peaks at destructive interference, especially when at the single peaks in Fig. 5(a),
5(b) (upper traces) the PDFS is minimum. The ER of the average squared magnitude of the
transfer function for all possible polarizations is maximum when the PDFS is minimum.
(a) (b) 300
20 20
200

DGD [ps]
|H( f )|2 [dB]

PDL [dB]
0 0
100
-20 -20
0
-40 -40
1549.4 1549.8 1550.2 1549.4 1549.8 1550.2
Wavelength [nm] Wavelength [nm]
(c) (d) 500
20 20 400
|H( f )|2 [dB]

PDL [dB]

DGD [ps]

0 0 300
200
-20 -20
100
-40 -40 0
1549.7 1550.1 1550.5 1549.7 1550.1 1550.5
Wavelength [nm] Wavelength [nm]
Fig. 5. Measured average squared magnitude of the transfer function for all possible polariza-
tions (blue), of PDL (green), and of DGD (green) at DI output 1(solid line) and 2 (dashed line);
FSR 42.7 GHz. For minimized PDFS: (a) average loss and PDL. (b) DGD versus wavelength.
For large PDFS: (c) average squared magnitude of the transfer function and PDL. (d) DGD
versus wavelength.

So far we demonstrated a tunable DI with reduced PDFS. There remains the problem of
the inaccurate step size and the hysteresis of the actuator, so we need means for a stable and
precise setting of the time delay.
3. Time delay control
In practical application the absolute time delay needs to be controlled and mechanical instabil-
ities need to be mitigated. To this end we design an active feedback control circuit. In Fig.
6(a), the setup of the time delay and phase control is illustrated. A pilot tone at frequency fp
counter-propagates with respect to the communication signal at fc. Circulators are used for
separation of the pilot and signal tone. Because fp and fc can be widely different, appropriate
filters would also allow a co-propagating arrangement without circulators [20]. For each DI,
the pilot tone is detected with a low bandwidth photodiode, sampled for processing in the
digital domain and fed to the respective control circuit.

#137432 - $15.00 USD Received 18 Feb 2011; accepted 26 May 2011; published 1 Jun 2011
(C) 2011 OSA 6 June 2011 / Vol. 19, No. 12 / OPTICS EXPRESS 11661
 To set the absolute time delay accurately one starts off from a known time delay and then
counts the number of constructive and destructive interference fringes mmax and mmin, respec-
tively, when adjusting the delay. The associated change δT of the time delay ΔT can be de-
rived by counting the number of minima and maxima that a signal undergoes when detuning
the delay
mmax  mmin
T  . (8)
2 fp

(a) (b)
+ A/D Control D/A
~ fp
- 0
D/A nFM
Monitoring PD
T fc

Loss [dB]
Output 1
DI 1
FSR 
T
Input at fc Pilot tone fp
Output 2 fp f0 fc
Frequency [GHz]
Fig. 6. , Schematic of control circuit (a) Time delay and phase control setup of DI, (b) ideal
power transfer function between DI ―Input‖ and ―Output 2‖.

Since the direction of movement is known, we can estimate the time delay ΔT, and ΔT × c
(c is the speed of light) is expressed as multiple of the pilot tone wavelength. The remaining
delay in distance is less than the wavelength of the pilot tone, which effectively introduces a
phase offset to the signal. The remaining phase delay can be found by measuring the power
swing of the pilot tone while scanning the delay, and by evaluating the pilot tone power devia-
tion from the average power value at the end of the actuator travel.
For measuring the remaining phase delay more precisely [19] dithering the DI delay may
be used [20,21]. However, a mechanical DI dither introduces a deterioration of the transmitted
signal data quality. So we decided for dithering the pilot tone frequency fp harmonically ac-
cording to f (t) = fp + δνd sin (2πνFM t) using a frequency deviation δνd with a modulation fre-
quency νFM. Figure 6(b) shows the power transfer function and denotes its FSR, the DI’s oper-
ating frequency f0, and the signal carrier frequency fc. Assuming that the pilot tone is con-
nected to output 2 with f0 being the frequency with constructive interference in the output, the
frequency offset between pilot tone and DI is denoted Δ fp = fp  f0.
The phase monitoring photodiode in Fig. 6(a) produces an output current
I ac (t )   J1  2πn d T  sin  2πf p T  sin  2πn FM t 
(9)
 J 2  2πn d T  cos  2πf p T  cos  4πn FM t   ...

The symbols J1,2 stand for the Bessel functions of order 1 and 2. The amplitudes of sin(2πνFM
t) and cos(4πνFM t) can be extracted with a numerical lock-in scheme. ΔT is known with the
fringe-counting measurement. Therefore the terms J 1,2(2πδνd ΔT) can be estimated with good
accuracy, and we calculate Δ fp from the remaining trigonometric terms sin(2πΔ fp ΔT) and
cos(2πΔ fp ΔT). With Δ fp and pilot tone fp known, the frequency f0 of the transfer function’s
maximum is found, which has an offset Δ fc from the carrier frequency fc. Therefore, based on
the known offset Δ fc, a control loop can be set up to fine-tune the delay ΔT for the wanted
offset Δ fc, which fixes the phase difference in both arms of the DI. Using the same pilot tone,
two DIs in a IQ demodulator can be locked with a defined relative phase offset, i. e. π/2.

#137432 - $15.00 USD Received 18 Feb 2011; accepted 26 May 2011; published 1 Jun 2011
(C) 2011 OSA 6 June 2011 / Vol. 19, No. 12 / OPTICS EXPRESS 11662
4. Measurement and experiment result
To demonstrate the quality of the tunable delay, we perform experiments with a long a me-
dium and a short delay at 100 ps, 35.7 ps, and 10 ps. The average spectral responses over all
possible polarizations, together with PDL versus wavelength are shown in Fig. 7. The plots
show from left to right FSRs of 10 GHz (resolution limited), 28 GHz, and 100 GHz. No mat-
ter what the delay is the plots show an average ER of more than 20 dB with a minimum PDFS
at the two outputs. The PDL around the frequencies of constructive interference are close to
zero for all delays.
(a) FSR = 10 GHz (b) FSR = 28 GHz (c) FSR = 100 GHz
|H( f )|2 [dB] PDL [dB]

|H( f )|2 [dB] PDL [dB]

|H( f )|2 [dB] PDL [dB]

Resolution limited
10 10 10

0 0 0

-20 -20 -20

-40 -40 -40

1571 1573 1575 1571 1573 1575 1571 1573 1575
Wavelength [nm] Wavelength [nm] Wavelength [nm]
Fig. 7. Average spectral responses (blue) of the two outputs (solid line for output 1 and dashed
line for output 2) over all possible polarizations and PDL (green) at FSR = 10 GHz (a), 28 GHz
(b) and 100 GHz (c).

We then tested the performance across the spectral range. We performed three measure-
ments with a FSR = 42.7 GHz (delay of 23.4 ps) at different wavelengths, Fig. 8. Similar
performances have obtained in each case.
(a) (b) (c)
|H( f )|2 [dB] PDL [dB]

|H( f )|2 [dB] PDL [dB]

|H( f )|2 [dB] PDL [dB]

10 10 10

0 0 0

-20 -20 -20

-40 -40 -40

1526 1528 1530 1548 1550 1552 1571 1573 1575
Wavelength [nm] Wavelength [nm] Wavelength [nm]
Fig. 8. Average spectral responses (blue) of the two outputs (solid line for output 1 and dashed
line for output 2) over all possible polarizations and PDL at wavelength 1525 nm – 1530 nm
(a), 1545 nm – 1555 nm (b), and 1570 nm – 1575 nm (c) with FSR = 42.7 GHz.

The accuracy of the absolute delay control algorithms has been tested next. We first set the
DI at 700 GHz FSR. We then used a pilot tone at 1550.12 nm to perform the fringe count. The
DI is tuned from 700 GHz to 600 GHz and down till 20 GHz, then up to 800 GHz and finally
back to 700 GHz. The target value of each of the 40 measurements has been cross-examined
with an optical spectrum analyzer (OSA) by connecting the DI input to an ASE source. Figure
9(a) shows the relative deviation between the set FSR and the respective OSA measurement.
The maximum deviation is ± 0.08%, which is mostly limited by the repeatability of the OSA
measurement.
We then were interested in the accuracy of setting the DI to a particular phase. For this we
modulated the pilot tone with ν FM = 60 Hz and δνd = 2 GHz and used to scan the phase offset
over 360° at FSR = 40 GHz. Results at FSR = 80 GHz are also shown. In Fig. 9(a), the devia-
tions between the set phase offset and the cross-examinations with respect to an OSA are
shown. A maximum error of ~2° is found. A 6 hours phase deviation measurement is per-

#137432 - $15.00 USD Received 18 Feb 2011; accepted 26 May 2011; published 1 Jun 2011
(C) 2011 OSA 6 June 2011 / Vol. 19, No. 12 / OPTICS EXPRESS 11663
formed with the pilot tone being modulated with νFM = 30 Hz and δνd = 750 MHz. A phase
deviation < 2° and a standard deviation of 0.3° has been measured. The result is shown in Fig.
9(c).
(a) (b) FSR = 80 GHz (c)
FSR = 40 GHz Deviation from phase [deg]

Phase Deviation [°]

0.9 22

Phase Dev. [deg]

Difference [deg]
0.5 1
Deviation [‰]

11

FSR[THz]
0 0.6 0
00
-0.5 -1
0.3
-1-1
-1
-2 -2-2
0
0 10 20 30 40 0 90 180 270 360 00 1 22 3 44 5 66
Measurement No. Phase [deg] Time [hour]
Time [hour]
Fig. 9. , The plots (a) the accuracy for setting the absolute time delay (blue curve) when setting
the FSR to particular value (red curve) for a measurement cycle between 800 GHz and 20 GHz,
(b) the accuracy for setting a particular two delays with FSR = 40 GHz —•— red, FSR = 80
GHz —•— blue for a measurement cycle from 0 to 360 degrees and (c) absolute deviation from
set value over time when using the stabilization setup.

The outputs of the DI were then connected to a 50 GHz balanced detector. The balanced
receiver was built with a conventional photodiode combined with an inverted photodiode
through a RF combiner. Demodulation of NRZ-DQPSK signals was performed at 11.7 GBd,
28 GBd and 42.7 GBd with a DI delay of 1 symbol duration. Eye diagrams of the in-phase
channel captured with a DCA are shown in Fig. 10(a). BER measurements were also per-
formed with a PRBS sequence of 27 - 1 at symbol-rates of 28 GBd and 42.7 GBd as depicted
in Fig. 10(b). For the two orthogonal polarizations the required received power for fixed BER
is almost equal for both symbol-rates with the proposed DI. The performance is comparable
with a commercial available DI in the same measurement setup. In Fig. 10(c), similar received
power requirement for fixed BER is observed over a broad wavelength range.
(a) 11.7 GBd (b) (c)
0
3 3
-log10(BER)

-log10(BER)

28 GBd
5 5
0 42.7 GBd
7 7
42.7 GBd 9 28 GBd
9
0 11 11
-40 -35 -30 -25 -20 -40 -35 -30 -25 -20
Received Power (dBm) Received Power [dBm]

Fig. 10. Measurement results of the proposed DI at various bit rates and comparison with a
commercial DI (a) eye diagrams of NRZ-DQPSK I channel at 11.7 GBd, 28 GBd, and 42.7
GBd, (b) BER versus received power at 28 GBd (Δ green, for I and Q channels at polariza-
tion 1, blue for I and Q channels at polarization 2), and 42.7 GBd (black for I and Q
channels at polarization 1, , red for I and Q channels at polarization 2)for orthogonal pola-
rizations of the tunable DI and single polarization of the typical DI (◄ magenta for I and Q
channels), and (c) BER versus received power at 42.7 GBd at different wavelengths (black
for I and Q channels at 1545.56 nm,  red at 1550.12 nm, and Δ green at 1560.61 nm).

6. Polarization division differential detector

The tunable delay interferometer can be further extended into a polarization division differen-
tial detector. We show in a step-by step manner that the four DIs normally used for a polariza-

#137432 - $15.00 USD Received 18 Feb 2011; accepted 26 May 2011; published 1 Jun 2011
(C) 2011 OSA 6 June 2011 / Vol. 19, No. 12 / OPTICS EXPRESS 11664
tion diversity scheme with Inphase (I) and Quadrature (Q) phase detection can be simplified
into single DI configuration.
The general concept is first explained on a conventional polarization division differential
detector as shown in Fig. 11(a). Its optical front-end includes 4 DIs which have a time delay
ΔT that is optimized with respect to the symbol rate of the detected signal. The signal is first
split into the two polarization components Ex and Ey (which in general do not correspond to
the transmitter’s polarization components) by a polarization beam splitter (PBS). Then each
polarization is fed into 2 orthogonal DIs (π/2 relative phase offset) (I- and Q-DIs) and four
balanced detectors are used to detect the signal components.
(a) Basic configuration (c) Free-space micro-optical
T implementation
Ey Ix
Ex T
p/2 Qx T
T
PBS Iy
T
p/2 Qy

φPol
(b) I and Q DIs combined configuration
Ev E φPol
h
Input

l/4
T Ix PBS NPBS
Ey l/4 φPol
Ex φPolp/2 Qx
PBS
PBS T Iy
l/4 φPol
Qy

Fig. 11. , Schematics of polarization diversity self-coherent receivers (a) conventional configu-
ration with 4 DIs, (b) I and Q DIs combined configuration with 2 DIs, (c) free-space micro-
optical implementation where all elements are folded into 1 DI only.

The DI as described above can be simplified to combine the I- and Q-DIs in a single DI.
This novel concept is shown in Fig. 11(b). The scheme works as follows. First a PBS splits off
the two polarizations Ex and Ey. Next we would like to determine the I- and Q-phase compo-
nent of each of the polarizations. So we use quarter-wave plates (QWP) to convert the Ex and
Ey components into circular polarizations (which provides us horizontal and vertical fields Eh
and Ev. Please note that the real advantage of this scheme is that we have Eh and Ev fields of
equal power!). Each circular polarization is then fed into a separate DI. One branch of the DI
comprises the time delay and the other branch a birefringent element (e.g. a liquid crystal
ΔφPol). The delay may then be set for the h-component to provide the I-phase component. The
birefringent element may then be used to set the phase difference in the short branch to δφpol +
π/2 resulting in a π/2 relative phase offset for the v-component with respect to the h-
component. The v-component will thus then provide us the Q-phase term. Using PBSs at the
DI outputs the Eh and Ev fields can easily be separated and combined into the balanced detec-
tors. Thus we have reduced the number of required DIs to only 2 DIs in this newly proposed
configuration.
In Fig. 11(c), we propose an additional reduction of required components in a free space
optics implementation. Using a mirror, the two orthogonal polarizations are mapped into the
same DI configuration with two beams propagating in parallel sharing the same optical ele-
ments. The signals are split and combined with one single NPBS and reflected back by two
corner reflectors. One reflector is mounted on a movable actuator that introduces a time delay

#137432 - $15.00 USD Received 18 Feb 2011; accepted 26 May 2011; published 1 Jun 2011
(C) 2011 OSA 6 June 2011 / Vol. 19, No. 12 / OPTICS EXPRESS 11665
ΔT. In the other branch, the birefringent element is set to align the orthogonal relative phase
offset between the I and Q-components. Two PBSs (including the one at the input) are used to
separate the signals into h and v polarizations. The signals can be coupled into fibers or direct-
ly to photodiodes providing electrical signals for further processing. With this novel setup,
four logic DIs in a polarization diversity self-coherent detection scheme are folded into one
single and compact Michelson delay interferometer structure with one single actuator tuning
the time delay.
6. Conclusion
A tunable free-space optical delay interferometer has been modeled and presented. Its delay is
controlled with a pilot tone. The maximal time delay is up to 100 ps, which allows a reception
of symbol rates > 10GBd. Its PDFS is mitigated by utilizing an adjustable birefringent liquid
crystal. A more complex but very compact design for polarization multiplexed D(m)PSK
signal detection is also presented.
Acknowledgments
This work is supported by the Agilent University Relation Program, the European Commis-
sion’s Network of Excellence EuroFOS and Karlsruhe School of Optics and Photonics
(KSOP).

#137432 - $15.00 USD Received 18 Feb 2011; accepted 26 May 2011; published 1 Jun 2011
(C) 2011 OSA 6 June 2011 / Vol. 19, No. 12 / OPTICS EXPRESS 11666