Dynamic Digital Holographic Wavelength Filtering - IEEE Xplore

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 16, NO. 7, JULY 1998

1259

Dynamic Digital Holographic Wavelength Filtering Michael C. Parker, Associate Member, IEEE, Member, OSA, Adam D. Cohen, Member, OSA, and Robert J. Mears, Associate Member, IEEE

Abstract— This paper describes the theory and results of a new generic technology for use in optical telecommunications and wavelength division multiplexing (WDM): dynamic digital holographic wavelength filtering. The enabling component is a polarization-insensitive ferroelectric liquid crystal (FLC) spatial light modulator (SLM) in conjunction with a highly wavelengthdispersive fixed diffractive element. The technology has been used to perform demultiplexing of single or multiple WDM signals, dynamic erbium-doped fiber amplifier (EDFA) gain equalization and channel management, and used to tune an erbium-doped fiber laser (EDFL) functioning as a high power, very narrow linewidth WDM source. Index Terms— Liquid crystal devices, optical equalizers, optical fiber amplifiers, optical fiber lasers, optical filters, photonic switching systems, spatial light modulators, wavelength division multiplexing.

I. INTRODUCTION

the AOTF suffers from high levels of crosstalk [13] and noise from a large variety of inherent sources: imperfect polarization splitting, interactions between RF surface acoustic waves (SAW’s), frequency shift of signal and time-dependent signal modulation due to RF SAW, and relatively high sidelobes outside the passband. In addition, the LiNbO3 substrate requires very highly controlled fabrication, since the birefringence tolerance is of the order of 10 5 over its entire length [13]. In practice, this limits the AOTF passband width to between 1–2 nm. The SAW velocity across the surface of the LiNbO3 limits the switching speed of the devices to about 10 s [14]. Other competing technologies include liquid crystal fiber Fabry–Perot etalons [15], [16], but they can only usefully filter one wavelength at a time, and tend to be polarization-sensitive, although polarization diverse operation has been reported [17], [18]. The same is true of semiconductor wavelength filters [19], [20].

T

HE development of the erbium-doped fiber amplifier (EDFA) [1] has opened up the possibility of almost limitless bandwidth data pipes, by using wavelength division multiplexing (WDM) [2], as well as allowing new optically transparent architectures, such as wavelength-routed networks [3]. These new systems and concepts require specialized functional components, such as tunable sources, receivers, switches and routers, reconfigurable optical amplifiers and wavelength converters. Optical telecommunications networks require components which are polarization-insensitive, optically transparent, have a low crosstalk and low loss, achieve high resolution tuning, are compact and operate at low powers. Technologies have been developed to demonstrate many of these components, but most do not satisfy all the desired requirements. II. TECHNOLOGY REVIEW The technology which has aroused the greatest interest as a potential generic WDM system technology is the acoustooptic tunable filter (AOTF) [4], [5]. Its distinguishing feature is multiple wavelength operation [6], as well polarization diverdB) [9] and compactness. sity [7], [8], potential low loss ( It has been successfully used to demonstrate WDM multiple wavelength demultiplexing [10] and EDFA gain equalization [11], and has been used within a tunable laser [12]. However, Manuscript received April 14, 1997; revised March 14, 1998. M. C. Parker was with the Cambridge University Engineering Department, Cambridge, CB2 1PZ U.K. He is now with Fujitsu Telecommunications Europe Ltd. Research, Colchester, CO1 1HH U.K. A. D. Cohen and R. J. Mears are with the Cambridge University Engineering Department, Cambridge, CB2 1PZ U.K. Publisher Item Identifier S 0733-8724(98)04819-1.

III. SLM TECHNOLOGY REVIEW A. Liquid Crystal Technology The large electrooptic response of liquid crystals (LC’s) has led to them being a highly developed technology for use in spatial light modulators (SLM’s). The two main types of liquid crystal used in SLM’s are twisted nematic LC [21], which allows continuous gray-scale modulation of the light and requires milliseconds to change state, and ferroelectric liquid crystal (FLC) [22] which switches state in times measured in microseconds [23]. The smectic C* FLC is normally used to achieve bistable modulation, while smectic A* FLC can be used for analogue modulation of the light. Liquid crystal SLM’s can also be used to either intensity-modulate [24] or phase-modulate [25], [26] the light. An advantage of the LC SLM technology is the economy and ease of manufacture. B. Holography in Telecommunications The use of holography in telecommunications to facilitate optical fiber-to-fiber switching and interconnection is increasingly attracting attention [27] because of the desirable features it has to offer: optical transparency, low crosstalk [28], polarization insensitivity [29], efficient use of available light and potential low loss, good wavelength insensitivity, and potential switching speeds as fast as 2 s [23]. Dynamic holographic interconnection using an FLC SLM acting as a reconfigurable computer generated hologram (CGH) has been suggested [26], [30] and successfully demonstrated by various groups [31]–[33]. Up to now, the wavelength-dispersion of

0733–8724/98$10.00  1998 IEEE

1260


Fig. 2. High resolution tunable wavelength filter using SLM and a fixed grating.

Fig. 1. Diffraction of light due to a regular grating.

CGH’s has been viewed as a problem, a form of chromatic aberration1—too small to be usefully exploited, however. This is due to the large SLM pixel sizes and their relatively few number, as well as the general wavelength insensitivity of the FLC cell. However, in combination with a fixed highspatial-frequency grating, the wavelength dispersion of an FLC SLM is amplified sufficiently to allow dynamic holograms to perform wavelength switching. Fig. 3. Polarization-insensitive 1.3 nm resolution wavelength filter.

IV. PRINCIPLE

OF

OPERATION

The operation of the tunable holographic wavelength filter is based on the wavelength-dispersive nature of gratings. Polychromatic light is angularly dispersed by a grating, since the different wavelengths are diffracted through different angles. This is demonstrated in Fig. 1, the basic equation [34] describing the angle of diffraction being (1) where is the angle of incidence of the collimated light, is the angle with which the diffracted light emerges from the (an integer) is the diffraction grating, is the wavelength, order, and is the spatial period of the grating. In practice, we are only interested in the positive first diffractive order, such . Changing the spatial period of the grating that causes the angle of diffraction to change. The use of an SLM displaying a phase pattern provides a programmable grating (i.e., a hologram) whose spatial period can be altered at will. In addition, holograms can be designed to have multiple spatial periods, to allow multiple wavelength tuning. A lens placed after the SLM converts the angular separation of wavelengths to a spatial separation, and a fixed spatial filter can then be used to arbitrarily select the desired wavelengths. Thus, different wavelengths can be selected and tuned to by changing the SLM spatial period . On its own, the SLM pixel pitch is too large for useful tuning to be obtained. For 1.55 m telecommunications WDM use, a wavelength demultiplexer needs to discriminate between different wavelengths separated by about 0.8 nm. For a compact filter this would require a pixel pitch of the order of the wavelength, which is too small for current SLM technology. However, a fixed diffraction grating 1 The wavelength-dispersion is still sufficiently small in the FLC SLM’s used in optical switching, for them to be considered effectively wavelengthinsensitive.

of this high spatial frequency used in conjunction with the SLM enables a compact high resolution filter. The SLM now operates to cause only slight changes in the angle of diffraction of the light. Fig. 2 shows the concept of using a relatively coarse SLM in conjunction with a highly dispersive fixed grating to achieve high resolution tuning. The figure shows (i.e., light being diffracted through a total angle of and total diffracted angle ). input angle The first angular deviation due to the hologram is given by and is variable since is the spatial period of the displayed hologram. The diffraction angle due to the fixed . The physical grating is a constant, and given by order of the two diffractive elements can be reversed. Fig. 3 architecture of the tunable wavelength filter shows the basic used to perform the experiments later described in Section VA. The FLC SLM and fixed grating combination is placed after a lens which collimates light from an input SM fiber. A second lens couples the first-order diffracted light back into an output SM fiber, which spatially filters the light. The filtered light has a passband FWHM of 2 nm, centered on a wavelength . For this relatively large (i.e., nonoptimized) architecture, the following expression [derived by linearizing and rearranging (4)] can be used to calculate the central wavelength of the passband (2)

mm is the displacement of the output SM fiber where mm is the focal length of from the optical axis, m is the fixed grating line-pair the lenses used and is equivalent to the fundamental width. The quantity of the hologram displayed on the SLM, spatial period m is the SLM pixel pitch, is where

PARKER et al.: DYNAMIC DIGITAL HOLOGRAPHIC WAVELENGTH FILTERING

1261

TABLE I

Fig. 4. Logarithmic plot of filter passband with FWHM

= 2.0 nm.

the number of pixels along one dimension of the SLM, and is any number between 0 and . The value of is determined by the displayed hologram, which can also be designed to have multiple spatial frequencies (i.e., multiple values of .) This allows the filter to have simultaneous multiple wavelength passbands—the relative transmissions of each passband can also be tailored by appropriate hologram design. V. TUNABLE WAVELENGTH FILTER A. Device Characterization Basic results of the holographic filter have already been reported [35], [36], albeit in cursory detail. The filter had a nm to tuning range of 82.4 nm ( nm) in steps averaging 1.29 nm, with a 3 dB passband of 2 nm. Fig. 4 shows the transmission profile at an arbitrary wavelength. It is very close to Gaussian in shape (on logarithmic scales it is almost parabolic), which is to be expected. However, beyond 3 nm either side of the central wavelength, the filter extremities depart from Gaussian behavior. This is because the actual shape is a Bessel function, which converges to zero more slowly than a Gaussian, and so has larger “tails” as illustrated in the diagram. The diagram shows that the filter has an SNR >30 dB, since the tails are still falling at 53 dB, and have not flattened out. However, this is only reached for wavelengths greater than 10 nm away from the central wavelength, owing to the convolution arising from the Gaussian coupling efficiency into the fiber end. In WDM systems, wavelength spacing is likely to be of the order of 0.8 to 4 nm [37]. This holographic filter achieves an isolation of approximately 20 dB at 3 nm from the central passband wavelength. The high 22.8 dB loss of the filter is accounted for in Table I. Reduction of the filter loss is discussed later in the paper. 1) Temporal Modulation of the Light: Since the FLC is not fully bistable it is necessary to periodically update all the pixels, with the frame being downloaded row by row [38]. A practical device, however, would make use of either a bistable FLC or an alternative addressing scheme, removing

Fig. 5. Temporal modulation of filtered light.

the need for the update process other than when changing between different holograms. A low amplitude 10 kHz ac signal is also continuously applied to the pixels during normal operation. The effect of the ac field is to both increase the tilt angle of the FLC (owing to the latter’s dielectric anistropy) and to “lock” the liquid crystals in their switched state. Both the update and constant ripple effects are manifested in the transmitted optical signal modulation, captured by a digital storage oscilloscope and displayed in Fig. 5. The loss of signal magnitude at the beginning of the update cycle can be attributed to hologram degradation due to FLC relaxation, leading to a loss of contrast before the SLM rows begin to be rewritten. The magnitude of the modulation during normal transmission is extremely small, approximately 0.035 dB, the effect being somewhat dependent on the SLM frequency setting. Such modulation would be well within the dynamic range of receivers. The 1 dB loss of the signal that occurs during the periodic frame update is undesirable, but new pixel addressing schemes currently under development for silicon backplane FLC SLM’s should eliminate the need for this process, even with an FLC material that is not fully bistable.

1262


Fig. 7. Reflective holographic filter in Littrow configuration.

Fig. 6. Transmission spectrum for three-wavelength filtering.

However, a fully bistable FLC used within the SLM would avoid all temporal modulation of the light and also allow failsafe operation of the device. In the event of a power failure, the SLM would still continue to diffract the light and the device still operate, albeit without reconfigurability. B. Multiple Wavelength Filtering Multiple wavelength filtering is one of the distinguishing features of holographic wavelength filtering and is important for WDM demultiplexing. This has also been demonstrated and an arbitrary result is shown in Fig. 6, where the filter has three passbands separated by 4 and 8 nm, respectively. The FWHM of each passband is still about 2 nm. The figure shows the triple passband transmission of ASE from an EDFA, with the loss for each passband averaging 30 dB. The higher loss is due to the available light being divided into three passbands, and the reduced diffraction efficiency of binary-phase holograms when they function to fan-out light. Multiple-phase holograms would have an improved diffraction efficiency, and an SLM with more pixels would have a greater SNR. The average hologram SNR or channel isolation is generally proportional to the number of hologram pixels , and inversely proportional to the number of filtered channels [39], such that SNR

(3)

Thus, the SNR performance of a hologram reduces as it is required to control more channels, but improves with more pixels. C. Wavelength Filter Design A practical, commercial holographic filter requires greater compactness and a higher resolution ( 0.8 nm). Of interest is to determine the minimum possible size of the holographic wavelength filter, for a given resolution. Reducing the physical size of the filter has the additional advantages of increased portability and stability. There are two generic architectures which need to be considered: architecture using a transmission SLM and • “linear” transmissive fixed grating, as seen in Fig. 3 and

Fig. 8. Wavelength filter with folded architecture.

• “folded” architecture employing two possible combinations of transmissive/reflective SLM and reflective/transmissive fixed grating, as seen in Figs. 7 and 8. The holographic filter of [35] is an example of a folded architecture with a transmissive SLM and reflective grating, while the filter of [36], already described in detail within this paper, is an example of a linear architecture. There are two fundamental parameters which determine the physical size of the lens or lenses used, of the filter: the focal length of the SLM, given by the and the aperture dimension and the number of product of the individual pixel pitch along one side of the SLM. A linear architecture pixels , filter will have a volume of the order of whereas a folded architecture will have a volume of , assuming the SLM has a square aperture. For the purposes of the analysis, we shall assume that current WDM telecommunications systems specifications2 require a tunable nm. The following filter to have a resolution calculations are performed at the standard erbium wavelength m, assumed to be situated at the center of the of filter tuning range. The spatial pitch of the fixed grating has for diffraction to occur without undesired to be greater than polarization effects. An arbitrary imposed limit for will be . that 1) Linear Architecture: A schematic diagram of the generic linear architecture for a holographic wavelength filter is shown in Fig. 3. The equation which describes the central wavelength to couple back into the output fiber can be deduced 2 Current draft ITU standards recommend WDM channel spacings of 100 GHz, which corresponds to about 0.8 nm channel spacings across the erbium window.


from the diffraction angles in the system (4) , defines the tuned The parameter , varying from 0 to wavelength. The middle of the tuning range, , corresponds . The filter tuning range is given to by solving (4) for the maximum and minimum tunable wavelengths, as is varied, such that (5) Reducing the spatial pitch of the fixed grating causes the angular diffraction to become large, so that full linearization [cf. (2)] of the equation is not possible. However, the angular due to the SLM does stay deviation small, so allowing a good approximation to the above equation (6) The filter 3 dB-passband width is determined by the output fiber which couples in a range of incident wavelengths across the finite width of its cleaved face. Light is coupled back into the fiber with a Gaussian-shaped variation in efficiency. Analysis [40] shows that the 3 dB-passband width is given to a close approximation by the wavelengths coupled in across . Using (6), the spread of coupled the fiber core diameter (which must be less than nm) wavelengths due to the spatial filtering of the fiber can be expressed as

1263

Equation (9) indicates that the minimum SLM aperture must be mm, for a focal length of the system mm. The tuning range of the filter can be calculated m, (which by assuming an SLM pixel pitch is technologically possible [41]), and a corresponding pixel . A larger pixel size, and smaller value number would reduce the tuning range. Using the minimum for , the SLM pixel size dimensions for the SLM aperture and the focal length , the displacement of the output fiber mm (i.e., wider than the SLM can be calculated as aperture.) The filter can then be tuned from nm to nm, yielding a tuning range of 86.9 nm. The minimum dimensions of the filter are 43.2 mm 11.4 mm 2.37 mm, giving a volume of 1.17k mm3 . 2) Folded Architecture: A folded reflective architecture is attractive since it requires only one lens and would appear to have the potential for greater compactness. A reflective architecture can be divided into two distinct classes. The first one, similar to a Littrow configuration is shown in Fig. 7. This arrangement was used in the experimental wavelength filter of [35]. Its main disadvantage is that it requires two passes through the SLM. For a binary-phase FLC device this increases the transmission loss by at least 4.38 dB. However, it has the advantage of reducing the off-axis angle of the diffracted light through the lens, avoiding numerical aperture problems, so enabling near diffraction-limited performance. The output fiber .A can be placed close to the optical axis, such that full analysis is unnecessary, but the basic equation describing the wavelength filtering of this configuration is given below, where is the angle which the fixed grating makes with the optical axis (10)

(7) , and that the 3 dB-passband width Assuming that is given by the wavelengths collected across the fiber-core m, then (7) can be closely of diameter approximated to yield the result that the lens focal length has a minimum value given by

mm

(8)

The collimated light passing through the SLM needs to illuminate the aperture of the SLM as much as possible, so that the full tuning capability of the SLM with minimum loss is achieved. Since the collimated light is Gaussian in intensity distribution, a sufficient condition is that the width of the beam coincides with the SLM aperture. From simple consideration of Gaussian optics we require

A similar analysis as for the linear architecture yields the result that the focal length has the same minimum given by mm. Since the ratio of to must be , mm. This means the SLM aperture needs to be that the minimum dimensions of the filter are 21.6 mm 2.37 mm 2.37 mm, with a volume of 0.121k mm . The tuning range of the filter can also be calculated, by assuming , which causes the displacement of the that mm. Letting the pixel pitch output fiber to be m, such that the pixel number , the filter nm to nm, can be tuned from with a total tuning range of 169.0 nm. The second reflective architecture utilizing a reflective SLM and transmissive fixed grating is shown in Fig. 8. Since there is only a single pass through the SLM, it has the same loss and resolution as the linear architecture. However, it has the disadvantage of a high off-axis angle through the lens. The expression which determines the filtered wavelength is given to a good approximation by (11)

(9)

Analysis reveals that the focal length of the lens must be at mm, with the minimum SLM aperture being least

1264


m, 149.4 nm; No Grating 53.2 mm 1.81 k mm3

Fig. 9. Folded-architecture wavelength filter without a fixed grating.

mm. Assuming a pixel pitch m, such , then the displacement of the that the pixel number is mm. The filter can be output fiber is calculated to be nm to nm, with tuned from a total tuning range of 149.4 nm. The minimum dimensions 18.2 mm 5.77 mm, giving a of the filter are 53.6 mm volume of 5.63k mm3 . 3) Architectures Without a Fixed Grating: The technology of SLM’s in the future may reach the point where pixels are sufficiently numerous and their size small enough that the fixed grating may be dispensed with altogether. An interesting exercise is to calculate the minimum pixel number and pixel pitch, as well as the overall size of such a wavelength filter, if nm. A folded architecture it is to have a resolution of in the Littrow configuration employing only an SLM to diffract the light is shown in Fig. 9. The governing equation defining the wavelength to be tuned is given by (12) Analysis shows that the minimum focal length must be mm with a minimum SLM aperture being mm, which corresponds to the number of pixels being at least , assuming that . Thus the minimum dimensions for a filter would be 53.2 mm 5.83 mm 5.83 mm, with a total volume of 1.81k mm . Current technology m, with and die sizes only allow pixel sizes of approximately 1000 pixels [41]. Thus, a significant advance in SLM technology would be required to make this type of wavelength filter possible. 4) Summary and Discussion: A summary of the minimum filter dimensions, volumes and tuning ranges for all the analyzed architectures is given as follows: 11.4 mm 2.37 mm 1.17k Linear 43.2 mm mm3 m, , Tuning Range 86.9 nm; 2.37 mm 2.37 mm Folded i) 21.6 mm 0.121 k mm3 m, , Tuning Range 169.0 nm; 18.2 mm 5.77 mm Folded ii) 53.6 mm 5.63 k mm3

, Tuning Range 5.83 mm

5.83 mm

m, , Tuning Range > 200 nm. The surprising conclusion to be drawn from these results is that the addition of a fixed grating not only provides a far higher resolution than could be obtained solely with existing SLM’s, but also reduces the physical size of the filter. Another counterintuitive result is that linear architectures tend to be almost as compact as those architectures based on reflective SLM’s. This is due to the numerical aperture of the lens which limits the degree to which the light can be angularly spread and still be efficiently coupled into the output fiber. Lenses with high numerical apertures will make small filters, but the lenses themselves are likely to be large, thus tending to actually increase the size of the reflective architecture. The analysis also shows that as the fixed grating pitch approaches , then the required focal length and SLM aperture becomes vanishingly small, even for a given resolution . However, this is at the expense of polarization sensitivity and other undesired side effects, such as spot-size distortion, which may became evident under these operating conditions. But it does offer the possibility of super-small, very high resolution holographic wavelength filter architectures. For polarization-insensitive operation, the analysis shows that the size of an architecture is proportional to the pitch of if the fixed the fixed grating (or the SLM pixel pitch grating is absent.) Both of these pitches are limited by, and at which the filter must be greater than the wavelength is operational. Thus, the factor which fundamentally limits the size of the filter is the wavelength of operation. Only by operating at shorter wavelengths can these generic wavelength filter designs be further reduced in physical size. D. Reducing the Filter Losses 1) Doubling Phase Modulation: The current loss in the filter due to the FLC material properties is 6.57 dB, on account . An FLC SLM can of the FLC switching angle , which only phase-modulate the incident light by 0 and of the light can be usefully means that only the fraction diffracted.3 Hence, the optimum value for the switching angle . A deep FLC phase modulation capability is is also desirable since a greater range of phase modulation is made possible for continuous phase holograms. One obvious solution to increase the total phase modulation capability is to cascade FLC SLM’s, but this is expensive in terms of hardware, increases the complexity due to alignment of the elements in the architecture, and increases overall size. A fold-back architecture which reflects the light back through the FLC SLM may appear to be a simple and effective method of increasing the phase modulation depth. However, such an architecture does not function as desired, because the two-phase modulations experienced by the light 3 This is an additional loss, independent of the diffraction efficiency of the displayed hologram.


Fig. 10. tion.

1265

FLC cell, quarter-wave plate and a mirror doubles phase modula-

during its two passes through the FLC cancel each other out. However, introducing a quarter-wave plate between the FLC cell and mirror, allows the two-phase modulations to add. This is illustrated in Fig. 10. A polarization-sensitive architecture using this configuration of FLC, quarter-wave plate and mirror, but placed between crossed polarizers, has already been proposed and experimentally verified [42]. Due to its polarization sensitivity, it was thought that the quarterwave plate had to be accurately aligned within the system. However, careful analysis shows this not to be the case, which means that a polarization-insensitive phase-doubled bistable-FLC SLM using an arbitrarily aligned quarter-wave plate is theoretically possible [43]. This configuration to double the phase modulation without crossed polarizers has been suggested and experimentally verified, however only for circularly polarized light [44]. The paper also suggests using a polymer cholesteric liquid crystal (PCLC) mirror in place of the quarter-wave plate/mirror combination, to act as a polarization-preserving mirror, but this makes the architecture polarization-sensitive. A similar concept using nematic liquid crystal, which has slightly different properties to FLC, has also been published [45]. An FLC SLM acting as a dynamic full -binary-phase hologram, will have at best a diffraction efficiency of 36.5%, equivalent to a loss of 4.38 dB. However, a double pass architecture with a quarter-wave plate can achieve this with an FLC switching angle of only 45 . 2) Blazed Fixed Grating: The fixed binary -phase grating used within the filter causes an additional loss of 4.38 dB. However, this loss can be substantially reduced by either using a multiple-phase fixed grating, a blazed fixed grating or refractive wavelength dispersive prism, which can all provide the required wavelength dispersion. 3) Sundry Improvements: Additional losses can be minimized by the application of anti-reflection coatings on all the surfaces. Large aperture lenses with large numerical apertures would increase the coupling efficiency between the fibers, but at the expense of size. Fusion splicings would avoid the FC/PC uniter losses.

Fig. 11. Light with a Gaussian intensity distribution incident on a grating.

greater than 340 Gb/s. A Gaussian intensity distribution of the light is assumed to be incident on the grating, as well as a of the light. This is illustrated in data stream modulation Fig. 11. The Gaussian intensity distribution of the light is defined as (13) At a point on the data front A–B before the grating, the . But intensity at a time is given by after the grating we need to include the temporal dispersion , so that along the data front A –B (14) is zero at the optical axis of the system and is where of the the speed of light. The intensity distribution . This light along A –B is also slightly different from is because we assume the Gaussian beam becomes narrower and more intense at its center when it is diffracted through an angle , so that (15) The total signal power given by the integration of

along the wave front A –B is along

(16) The total signal power is thus a time-domain convolution, which is solved easily in the frequency domain, where it simply becomes the product of the frequency spectra. Thus, is the frequency spectrum (i.e., Fourier transform) of the , such that , received time-varying signal and similarly with the following Fourier transform pairs: and . We can rewrite (16) as a frequency-domain product to give

E. Spectral Broadening of WDM Signal Spectral broadening or chirp of the WDM signal is an issue which needs to be considered. However, the analysis demonstrates that this only becomes a problem at channel rates

(17)

1266


The original frequency spectrum of the data signal was , but that has now been narrowed by the filter frequency to give us a received data power spectrum response . The frequency width by which the data frequency spectrum is narrowed is given by (18) If a minimum sized linear architecture is used for the filter then , = 1.19 mm, = 1.55 m giving us a = 170 GHz. This corresponds to frequency broadening of Gb/s, before a maximum allowable bit rate of temporal dispersion and intersymbol interference becomes a , we can rewrite (18) as problem. Since

Fig. 12. Eleven successively tuned lasing wavelengths.

(19) Using the results of previous analysis, such as (8) and (9) and where the focal length we can write: (20) Substituting the inequality (20) into (19), we find that the maximum bit rate is given by (21) Thus, we find that the maximum bit rate is ultimately limited of the filter, and the wavelength by the desired resolution and frequency of the carrier wave. A filter with a very will be both large and less able fine wavelength resolution to filter high bit rates. For a polarization-insensitive, low-loss filter, the maximum bit rate is also in effect independent of the geometry or architecture of the filter. The equation does show however, that as the fixed grating pitch tends toward , the maximum allowable bit rate increases, albeit at the expense of polarization sensitivity and distorted spot-size. Temporal dispersion of the signal may be compensated for by a suitably chirped fiber grating [46] placed after the filter. VI. TUNABLE FIBER LASER Tunable fiber lasers may potentially serve an important function in WDM telecommunications networks, acting as stable and pure laser sources. They have a very narrow linewidth, high output powers and large tuning ranges. We have already published the results of a tunable erbium-doped fiber laser [36], tuned using the holographic wavelength filter. Tuning over the range from 1528.6 to 1567.1 nm in steps of 1.3 nm has been achieved, with CW output powers of up to 13 dBm. The tuning range of 38.5 nm covered the whole of the erbium window, but only used a small section of the available 82.4 nm range of the wavelength filter. An arbitrary group of eleven successively tuned lasing modes4 is shown in Fig. 12. The figure shows that the lasing modes have at 4 The EASLM had a memory capacity for 11 holograms, so that the EDFL could only be tuned to 11 wavelengths at a time.

Fig. 13. Two competing lasing modes at 1556.0 and 1562.5 nm.

least 30 dB side-mode suppression. Each mode is on average 1.3 nm distant from its neighboring mode, which is in line with the expected performance of the wavelength filter. More power could be accessed by placing the output coupler after the EDFA. The slightly “ragged” appearance of some of the lasing peaks is due to mode-hopping, where nonuniformities in the gain spectrum excite other lasing modes lying within the filter pass band. However, the unidirectional architecture of the EDFL ensures that it remains single moded. The inherent EDFL 3 dB lasing linewidth was found to be of the order of 3 kHz, and the long term wavelength stability was about 0.1 nm. A. Multiple-Wavelength Lasing A hologram with a mixed spatial frequency has also been designed to allow the EDFL to simultaneously lase at 1562.5 nm and at 1556 nm. The filter had two passbands and the lasing spectral plot is shown in Fig. 13. Due to the gain medium being relatively homogeneous and dependent at the two wavelengths, mode competition means that the lasing mode powers fluctuated considerably. Fig. 13 is a “snapshot” of when the two modes were lasing equally for an instant. The power in each mode is also considerably lower than usual, since only half the EDFA power is available to each mode, the hologram has less than half the usual diffraction efficiency for each the two wavelengths, and a 10/90 coupler was used to output the lasing power. For successful stable multiple lasing to occur, each wavelength requires a more independent gain medium, so as to avoid mode competition.


1267

on channel #4 (due, say, to differing network origin of the signals). This system has the potential for incorporation into all-optical networks comprising multiple cascaded EDFA and holographic equalizer combinations. The effect of the equalizer would be to reduce the required dynamic range of receivers, and increase the transmission distance possible before inequalities in signal magnitudes require correction. The minimum gain of the system, corresponding to the channel that is the strongest following EDFA amplification and must be reduced to the level of the weakest, is given by (22)

Fig. 14.

Dynamic channel equalization (hologram 1).

is the EDFA gain at the channel wavelength in where is the dynamic range of equalization, is the question, is the passive number of channels to be equalized and filter loss at the channel wavelength. Clearly, there must arise a limit to the maximum number of channels, as both the term will diminish the minimum achievable gain, , is also and the range of EDFA gain to be equalized, i.e., likely to approach a value of 20 dB or more for a large number of channels spanning the complete erbium window. Also, as the number of channels increases, the proportion of the SLM’s 128 pixel columns corresponding to each wavelength channel must necessarily decrease, thereby reducing the efficiency of the hologram, manifested in a poorer SNR [cf., (3)]. Therefore, in order to equalize channels in a dense WDM system (potentially comprising 100 or more channels), a combination of passive, broad band fiber filtering, and active holographic SLM-based filtering may prove necessary. Ideally the spacing of the WDM channels would be reduced to the emerging ITU standard of 0.8 nm. This should be achievable in principle with the existing filter design, provided that the FWHM passband of the system be reduced. From (8), the passband FWHM is proportional to , so in order to keep the system as compact as possible, the preferred method of lowering the filter passband is likely to be a combination of a smaller grating pitch and the use of a tapered output fiber or other spatial filtering mechanism. VIII. FUTURE WORK

Fig. 15.

Dynamic channel equalization (hologram 2).

VII. DYNAMIC EQUALIZER The holographic wavelength filter can also be used for the active management of WDM channels. Preliminary experiments have successfully controlled five WDM channels spaced by 4 nm, reducing an input power variation of 8.5 dB down to 0.3 dB, while also providing gain of up to 3.3 dB [47]. In addition, an unused wavelength channel was “knocked” out and the EDFA amplified spontaneous emission (ASE) suppressed by at least 18 dB. The dynamic gain equalization is demonstrated in Figs. 14 and 15 by the SLM switching between two different displayed holograms, thereby managing the change in power/wavelength

A. Low-Loss Compact Space- Switch A holographic wavelength filter with a loss of 6–7 dB should be possible by careful design and use of improved optical materials. The currently unused extra dimension of the SLM can also be used to add functionality, such as in a space-wavelength switch. This would serve a very important function in dynamic wavelength-routed optical networks as a drop and insert node. Fig. 16 shows an “exploded” concept for a polarization-insensitive, optically transparent, compact, lowloss space-wavelength switch, utilizing all the ideas developed in this paper. The switch acts as a 3 3 fiber cross-connect, which can also perfectly shuffle wavelengths between the various fibers. It may be possible to use a GRIN lens instead of a bulk refractive lens, but the limited NA and aperture of a

1268


refractive index of a semiconductor such as silicon or galliumarsenide [49]. A disadvantage is that the Franz–Keldysh effect might be expected to cause a change in the local transmission of the semiconductor. However, the plasma effect occurs over time scales measured in nanoseconds, and so is at least three orders of magnitude faster than LC effects, it is generally polarization-insensitive and would also allow continuous phase-modulation.

Fig. 16.

Exploded 2f compact 3

Fig. 17.

Packaged 3

2 3 space-wavelength switch.

2 3 space-wavelength switch.

GRIN lens will tend to limit the number of fibers possible to interconnect. Fig. 17 shows how the packaged device might look, using a GRIN lens instead of a bulk refractive lens. B. Apodization It is becoming increasingly desirable that filters used in WDM networks have a rectangular passband [48]. A flat passband has a greater wavelength misalignment tolerance, and better cascadability. The relatively long “tails” of the Gaussian passband are undesirable, since accumulated crosstalk tends to become high. Possible means to reduce the long Gaussian tails of the current filter include changing the output fiber end, so as to alter the coupling characteristic. A concave fiber end or a tapered polished fiber end may be sufficient. A suitable phaseplate before the second lens, to filter out the higher angles, may lensing system incorporating also be a solution. Finally, a additional spatial filtering could also be considered. C. Other Technologies The maturing of the technology associated with FLC SLM’s has allowed it to be used to demonstrate the principle of digital holographic wavelength filtering. However, holographic wavelength filtering is based on a generic principle, which is not technology specific, and so can be exploited by other technologies as well. One technology which may become sufficiently developed in the future is phase-modulation using the plasma effect in semiconductors, where injection of charge causes local changes in the refractive index of the material. A pixellated transmissive semiconductor SLM with a controllable refractive index associated with each pixel, allows different optical path lengths, and thus spatial phasemodulation of the light. A change in carrier concentration of about 1024 m 3 will generally cause a 1–2% change in the

IX. CONCLUSIONS In this paper a new technique for high resolution wavelength filtering has been presented. Holographic wavelength tuning may find application in WDM telecommunications systems, where tunable sources, filters and receivers are required. The advantages of holographic tuning are as follows: • optical transparency; • polarization insensitivity; • digital, fast, low-power operation; • fail-safe operation and robustness; • fine resolution over a large wavelength range; • multiple wavelength operation; • low crosstalk. The technique can be used within a holographic spacewavelength switch, allowing arbitrary switching and shuffling of the wavelengths between the fibers. Dynamic gain equalization is an important issue which can also be addressed using holographic tuning. The holographically tunable EDFL may also find use as a source in a WDM network and as a local oscillator for coherent detection. REFERENCES [1] R. J. Mears, L. Reekie, I. M. Jauncey, and D. N. Payne, “Low-noise erbium-doped fiber amplifier at 1.54 m,” Electron. Lett., vol. 23, no. 19, pp. 1026–1028, 1987. [2] C. A. Brackett, “Dense wavelength division multiplexing networks: Principles and applications,” IEEE J. Select. Areas Commun., vol. 8, pp. 948–964, 1990. [3] G. R. Hill, “A wavelength routing approach to optical communication networks,” Brit. Telecommun. Technol. J., vol. 6, no. 3, pp. 24–31, 1988. [4] F. Heismann, L. Buhl, and R. Alferness, “Electro-optically tunable narrowband Ti : LiNbO3 wavelength filter,” Electron. Lett., vol. 23, pp. 572–574, 1987. [5] B. L. Heffner, D. A. Smith, J. E. Baran, A. Yi-Yan, and K. W. Cheung, “Integrated-optic acoustically tunable infra-red optical filter,” Electron. Lett., vol. 24, no. 25, pp. 1562–1563, 1988. [6] K. W. Cheung, D. A. Smith, J. E. Baran, and B. L. Heffner, “Multiple channel operation of an integrated acousto-optic tunable filter,” Electron. Lett., vol. 25, pp. 375–376, 1989. [7] W. Warzanskyj, F. Heismann, and R. C. Alferness, “Polarizationindependent electro-optically tunable narrow-band wavelength filter,” Appl. Phys. Lett., vol. 53, no. 1, pp. 13–15, 1988. [8] D. A. Smith, J. E. Baran, K. W. Cheung, and J. J. Johnson, “Polarizationindependent acoustically tunable optical filter,” Appl. Phys. Lett., vol. 56, no. 3, pp. 209–211, 1990. [9] H. Herrmann, P. M¨uller-Reich, V. Reimann, R. Ricken, H. Seibert, and W. Sohler, “Integrated optical, TE- and TM-pass, acoustically tunable, double-stage wavelength filters in LiNbO3 ,” Electron. Lett., vol. 28, no. 7, pp. 642–644, 1992. [10] A. d’Alessandro, D. A. Smith, and J. E Baran, “Multichannel operation of an integrated acousto-optic wavelength routing switch for WDM systems,” IEEE Photon. Technol. Lett., vol. 6, pp. 390–393, Mar. 1994. [11] S. F. Su, R. Olshansky, D. A. Smith, and J. E. Baran, “Flattening of erbium-doped fiber amplifier gain spectrum using an acousto-optic tunable filter,” Electron. Lett., vol. 29, no. 5, pp. 477–478, 1993.


1269

[12] D. A. Smith, M. W. Maeda, J. J. Johnson, J. S. Patel, M. A. Saifi, and A. von Lehman, “Acoustically tuned erbium-doped fiber ring laser,” Opt. Lett., vol. 16, no. 6, pp. 387–389, 1991. [13] J. L. Jackel, M. S. Goodman, J. E. Baran, W. J. Tomlinson, G.-K. Chang, M. Z. Iqbal, G. H. Song, K. Bala, C. A. Brackett, D. A. Smith, R. S. Chakravarthy, R. H. Hobbs, D. J. Fritz, R. W. Ade, and K. M. Kissa, “Acousto-optic tunable filters (AOTF’s) for multiwavelength optical cross-connects: Crosstalk considerations,” J. Lightwave Technol., vol. 14, pp. 1056–1066, June 1996. [14] C. A. Brackett, “Multiwavelength switching and interconnection networks,” in Photonics in Switching, Book II, J. Midwinter, Ed. New York: Academic, 1993, ch. 1. [15] J. S. Patel, M. A. Saifi, D. W. Berreman, C. Lin, N. Andreadakis, and S. D. Lee, “Electrically tunable optical filter for infrared wavelength using liquid crystals in a Fabry–Perot etalon,” Appl. Phys. Lett., vol. 57, no. 17, pp. 1718–1720, 1990. [16] K. Hirabayashi, H. Tsuda, and T. Kurokawa, “New structure of tunable wavelength-selective filters with a liquid crystal for FDM systems,” IEEE Photon. Technol. Lett., vol. 3, pp. 741–743, Aug. 1991. [17] J. S. Patel and M. W. Maeda, “Tunable polarization diversity liquidcrystal wavelength filter,” IEEE Photon. Technol. Lett., vol. 3, pp. 739–730, Aug. 1991. [18] K. Hirabayashi, Y. Ohiso, and T. Kurokawa, “Polarization-independent tunable wavelength-selective filter using liquid crystal,” IEEE Photon. Technol. Lett., vol. 3, pp. 1091–1093, Dec. 1991. [19] L. L. Buhl, R. C. Alferness, U. Koren, B. I. Miller, M. G. Young, T. L. Koch, C. A. Burrus, and G. Rayborn, “Grating assisted vertical coupler/filter for extended tuning range,” Electron. Lett., vol. 29, no. 1, pp. 81–82, 1993. [20] F. Heismann, L. L. Buhl, B. I. Miller, M. Newkirk, U. Koren, M. G. Young, and R. C. Alferness, “Polarization-independent wavelength filter using a grating-assisted vertical directional coupler in InP,” Appl. Phys. Lett., vol. 64, no. 18, pp. 2335–2337, 1994. [21] F. Mok, J. Diep, H. K. Liu, and D. Psaltis, “Real-time computergenerated hologram by means of liquid-crystal television spatial light modulator,” Opt. Lett., vol. 11, no. 11, pp. 748–750, 1986. [22] M. F. Bone, D. Coates, W. A. Crossland, P. Gunn, and P. W. Ross, “Ferroelectric liquid crystal display capable of video line address time,” Displays, pp. 115–118, July 1987. [23] M. A. Handschy, K. M. Johnson, G. Moddel, and L. A. PaganoStauffer, “Electro-optic applications of ferroelectric liquid crystals to optical computing,” Ferroelectr., vol. 85, pp. 279–289, 1988. [24] B. Bahadur, Liquid Crystals: Applications and Uses. World Scientific, 1990, vol. 1. [25] M. J. O’Callaghan and M. A. Handschy, “Diffractive ferroelectric liquidcrystal shutters for unpolarized light,” Opt. Lett., vol. 16, no. 10, pp. 770–772, 1991. [26] D. C. O’Brien and R. J. Mears, “Computer generated holograms optimized for illumination with partially coherent light using silicon backplane spatial light modulators as the recording device,” in Proc. SPIE, vol. 1505, pp. 32–37, 1991. [27] R. J. Mears, W. A. Crossland, M. P. Dames, J. R. Collington, M. C. Parker, S. T. Warr, T. D. Wilkinson, and A. B. Davey, “Telecommunications applications of ferroelectric liquid-crystal smart pixels (Invited Paper),” IEEE J. Select. Topics Quantum Electron., vol. 2, pp. 35–46, 1996. [28] M. J. Holmes, F. P. Payne, P. Dainty, T. J. Hall, and W. A. Crossland, “Low crosstalk devices for wavelength-routed networks,” in IEE Colloquium on Guided Wave Optical Signal Processing, Savoy Place, London, U.K., June 8, 1995, IEE Dig. no. 95-128. [29] S. T. Warr and R. J. Mears, “Polarization-insensitive operation of ferroelectric liquid-crystal devices,” Electron. Lett., vol. 31, pp. 714–716, 1995. [30] D. C. O’Brien, W. A. Crossland, and R. J. Mears, “A holographically routed optical crossbar: Theory and simulation,” Optic. Comput. Processing, vol. 1, no. 3, pp. 233–243, 1991. 4 free-space optical [31] H. Yamazaki and M. Yamaguchi, “4 switching using real-time binary phase-only holograms generated by a liquid-crystal display,” Opt. Lett., vol. 16, no. 18, pp. 1415–1417, 1991. [32] S. E. Broomfield, M. A. A. Neil, E. G. S. Paige, and G. G. Yang, “Programmable binary phase-only optical device based on ferroelectric liquid crystal SLM,” Electron. Lett., vol. 28, no. 1, pp. 26–28, 1992. [33] D. C. O’Brien, T. D. Wilkinson, R. J. Mears, and W. A. Crossland, “Generalized dynamic holographic interconnects using spatial light modulators,” in Proc. IEEE Conference at Palm Springs—Topical Meeting on SLM’s, 1993, vol. SWC2-4, pp. 192–195.

[34] M. Born and E. Wolf, Principles of Optics. New York: Pergamon, 1959, p. 403. [35] S. T. Warr, M. C. Parker, and R. J. Mears, “Optically transparent digitally tunable wavelength filter,” Electron. Lett., vol. 31, pp. 129–130, 1995. [36] M. C. Parker and R. J. Mears, “Digitally tunable filter and laser,” IEEE Photon. Technol. Lett., vol. 8, pp. 1007–1008, Aug. 1996. [37] C. A. Brackett, A. S. Acampora, J. Sweitzer, G. Tangonan, M. T. Smith, W. Lennon, K.-C. Wang, and R. H. Hobbs, “A scalable multiwavelength multihop optical network: A proposal for research on all-optical networks,” J. Lightwave Technol., vol. 11, nos. 5/6, pp. 736–753, 1993. [38] W. A. Crossland and A. B. Davey, “Addressing requirements for chiral smectic liquid crystal active backplane spatial light modulators,” Ferroelectr., vol. 149, pp. 361–374, 1993. [39] M. C. Parker, “Dynamic holograms for WDM,” Ph.D. dissertation, University of Cambridge, U.K., 1996, pp. 34–43. , “Dynamic holograms for WDM,” Ph.D. dissertation, University [40] of Cambridge, U.K., 1996, pp. 57–58. [41] J. R. Collington, M. P. Dames, W. A. Crossland, and R. W. A. Scarr, “Optically accessed electronic memory,” in Proc. 5th Int. Conf. FLC’s, 1996, vol. 181, no. 4, pp. 99–110. [42] M. A. A. Neil and E. G. S. Paige, “Improved transmission in a twolevel, phase-only, spatial light modulator,” Electron. Lett., vol. 30, no. 5, pp. 445–446, 1994. [43] M. C. Parker, “Dynamic holograms for WDM,” Ph.D. dissertation, University of Cambridge, U.K., 1996, pp. 16–23. [44] J. E. Stockley, G. D. Sharp, S. A. Serati, and K. M. Johnson, “Analog optical phase modulator based on chiral smectic and polymer cholesteric liquid crystals,” Opt. Lett., vol. 20, no. 23, pp. 2441–2443, 1995. [45] G. D. Love, “Liquid-crystal phase modulator for unpolarized light,” Appl. Opt., vol. 32, no. 13, pp. 2222–2223, 1993. [46] K. O. Hill, F. Bilodeau, B. Malo, T. Kitagawa, S. Th´eriault, D. C. Johnson, and J. Albert, “Chirped in-fiber Bragg gratings for compensation of optical-fiber dispersion,” Opt. Lett., vol. 19, no. 17, pp. 1314–1316, 1994. [47] M. C. Parker, A. D. Cohen, and R. J. Mears, “Dynamic holographic spectral equalization for WDM,” IEEE Photon. Technol. Lett., vol. 9, pp. 529–531, Apr. 1997. [48] D. A. Smith, H. Rashid, R. S. Chakravarthy, A. M. Agboatwalla, A. A. Patil, Z. Bao, N. Imam, S. W. Smith, J. E. Baran, J. L. Jackel, and J. Kallamn, “Acousto-optic switch with a near rectangular passband for WDM systems,” Electron. Lett., vol. 32, no. 6, pp. 542–543, 1996. [49] G. H. B. Thompson, Physics of Semiconductor Laser Devices. New York: Wiley, 1980, Appendix 4, pp. 535–537.

2

Michael C. Parker (A’97) was born in London, U.K., in 1971. He received the first class B.A. degree in electrical and information sciences from Cambridge University, U.K., in 1992. He received the Ph.D. degree in optical communications at Cambridge University, U.K., in 1996, where his research concerned the use of free-space ferroelectric liquid crystal spatial light modulators in holographic space/wavelength (WDM) switches. He spent a six-month sabbatical working for Carl Zeiss, Germany, developing an illumination measurement system for the objectives used in photolithographic steppers. Since 1997, he has been working for Fujitsu Telecommunications Europe Ltd., conducting research into WDM optical access networks and associated technologies, such as novel AWG designs. He has contributed to four patents and authored more than 20 publications. Dr. Parker is an Associate Member of the Institution of Electrical Engineers (IEE) and member of the Optical Society of America (OSA).

1270

Adam D. Cohen received the B.A. degree in electrical and information sciences at Pembroke College, Cambridge, U.K., in 1991. He returned to the optics field in 1994 and since then has been working toward the Ph.D. degree at Cambridge University Engineering Department. From 1991 to 1992, he was part of a team developing GaN-based blue-emitting diode lasers at Pioneer Electronic Corporation, Tsurugashimashi, Japan. From 1992 to 1994, he was at BT Laboratories, Martlesham Heath, U.K., working on SDH and ATM network management systems, including novel object-oriented data and control system architectures. His research concerns the application of ferroelectric liquid crystal spatial light modulators to dynamic multichannel and fine-tunable filters for WDM networks. In addition he has developed prototypes of, and extended designs for, fast spatial light modulators based on nonlinear optical organic polymers. He has authored 12 publications, contributed to one patent. Mr. Cohen is an Associate Member of the Institution of Electrical Engineers (IEE) and a member of the Optical Society of America (OSA).


Robert J. Mears (A’97) has 15 years of research experience in optoelectronics in both guided-wave, free-space optical systems, and EDFA’s. While at Southampton University, U.K., in the 1980’s, he developed rare-earth-doped fiber lasers and amplifiers and was the first to demonstrate the erbium-doped fiber amplifier. He has been a University Lecturer at Cambridge and Fellow of Pembroke College, U.K., since 1988. His research at Cambridge has developed silicon smart pixel architectures and devices for telecommunications and information processing, and use of ferroelectric liquid crystal spatial light modulators (FLC SLM’s) as dynamic holographic elements for space and wavelength switching. He has authored or coauthored more than 70 publications in optoelectronics and holds a number of patents. Dr. Mears is a recipient of the IEE Electronics Letters Premium.