You are on page 1of 15

A study of localization eects in one dimensional disordered optical systems

Girish Kulkarni under the supervision of Dr. Sushil Mujumdar, Tata Institute of Fundamental Research July 11, 2013
Abstract Investigations of one dimensional disordered optical systems are presented in three parts. In the rst part, the statistical variation of transmittance uctuations with the system length is numerically analyzed. In the second part, we compute the statistics of nearest neighbor spectral spacings of high transmittance modes which clearly reveal Wigner surmise-like level repulsion at small system sizes and a gradual crossover to Poissonian statistics at large system lengths. In both cases, a strong universality for dierent disorder strengths is observed in the variation of statistics with respect to the parameter L/ , where L is the system length and is the localization length. Lastly, we experimentally realize a 1D disordered system in a single mode ber with randomly spaced Bragg gratings and develop a perturbative measurement technique to map the mode intensity prole in the ber.

Acknowledgments
I would like to express my sincere gratitude towards my project advisor Sushil Mujumdar for his patient guidance and support. I thank him for giving me the opportunity to work with him and ensuring a very enjoyable learning experience in his laboratory. His advice on several matters, both professional and personal, have helped me tremendously and for his encouragement and motivation, I will always be indebted. I am grateful to Sandip Ghosh for kindly lending us the tunable diode laser for our experiments. I am thankful to our collaborators Prof. S. Asokan and Guru Prasad from the Indian Institute of Science, Bangalore for helping us with the fabrication of gratings for our experiments. I would also like to express my gratitude to Mr. Chogale, Mr. Katarkar, Mr. Sinha and others from the TIFR workshop for their help and technical support. I sincerely thank Kabir Ramola and Rahul Dandekar for useful discussions and suggestions. I am thankful to all my labmates - Ravitej, Anjani, Swapnesh, Sreeman, Randhir, Sadiq and others for a very stimulating work environment. I have been fortunate to have several insightful and engaging discussions with them on work and a variety of other subjects. I apologize for the times I may have annoyed them with my carelessness in the lab and I am grateful to them for helping me imbibe a sense of discipline in work and experiments. I thank all my friends - Arjun, Mangesh, Ishan, Harshant, Prathyush, Amit, Arunachal, Nilakash, Krishnendu and others for all the good times together. I am indebted to Rajeev for helping me on various fronts like an elder brother. I am thankful to my teachers at TIFR from whom I have gathered the knowledge and inspiration to pursue physics. I also gratefully acknowledge the experience I gained with M. Krishnamurthys group at TIFR. I am also thankful to Arnab Bhattacharyya, Shiraz Minwalla, A. Gopakumar and other faculty members who have kindly helped me on several occassions. My warm thanks to my friend Mugdha, with whom I had the pleasure and fortune of spending some really amazing times together. I am grateful for her encouraging words that always came when I needed them most and nudged me on when I almost gave up. Finally, I thank my family and friends for their love and support.

Introduction
Over several decades, wave transport in disordered media has remained a topic of great interest and applicability. Its study has led to our understanding of a plethora of physical phenomena, right from the most common eects like resistance in impure crystals to some very intriguing eects in disordered systems like Anderson localization [1]. It is now well understood that resistance to electron transport in crystals occurs due to impurities or crystal defects. In weakly scattering systems, transport is satisfactorily explained by the diusion model, which essentially sums the probabilities of wave propagation between two points in the system. As a result, for a metallic wire, the conductance decreases inversely as the length of the wire as predicted by diusion. However, according to the quantum theory we must really sum the complex probability amplitudes for wave propagation along possible paths, and hence there is also a contribution from the interference terms. As one gradually increases the scattering impurities or the disorder, this interference of multiply scattered waves leads to the localized regime where the conductance decreases exponentially with length. This gradual crossover from a metallic phase to an insulating regime is called the disorder induced metal-insulator transition or the localization transition[2]. Above a certain threshold of disorder, these multiply scattered waves can destructively interfere causing the electron transport to come to a complete halt leading to a vanishing conductance. This phenomenon of spatial connement of waves in a disordered medium is known as Anderson localization. While the eect was predicted for electron transport using quantum mechanical considerations, it has also been found to occur for light transport in strongly scattering optical media and can be understood classically using the wave theory of light [3]. As a result, several interesting phenomena studied in disordered electronic systems have directly analogous eects for light in disordered optical media. Since the modeling of 3D disordered systems is rather complicated, one often resorts to studies with lower dimensional systems as paradigms and gradually carry over to three dimensions. In this study, we analyze the transport of light through a 1D disordered system consisting of transversely innite dielectric slabs of randomly varying widths and refractive indices. A convenient method for this purpose is the transfer matrix formalism, which essentially implements Maxwells boundary conditions at each interface and associates a 22 matrix for each dielectric layer [4] relating the amplitudes of the forward and backward propagating waves at its two ends. The eective transfer matrix for the entire multilayer is then obtained as the ordered product of the individual transfer matrices of the consisting dielectric slabs. Using this method, it is possible to obtain the transmittance and the spatial eld prole across a system for an incident wave of any frequency. In 1D disordered systems, there is no diusion regime and every state is inherently localized [4]. The characteristic length over which the eld amplitude decays exponentially is known as the localL ization length which is dened as = <lnT > , where L is the system length and < lnT > denotes the ensemble average of the logarithm of the transmittance over several disordered congurations for xed system size. For a given incident frequency, a system with small size (L/ 1) is associated with a high mean transmittance and is said to be in the metallic phase. As the system size is increased, a localization transition occurs at L/ 1 and the system crosses over into the insulating phase at large system size (L/ 1) with a very low mean transmittance. In this article, we present a systematic study of 1D disordered systems and their statistical behavior near the localization transition. We rstly characterize the scaling of the transmittance uctuations with system length, both analytically and numerically. We then consider the phenomenon of resonant transmission of modes in the localized regime and study the peculiar intensity proles associated with such modes. We also numerically characterize the statistics of the nearest neighbor spacings between such high transmittance modes and discuss their scaling across the localization transition [5]. We realize a 1D disordered system experimentally in a single mode ber with randomly spaced Bragg gratings. We develop a perturbative measurement technique to map the mode intensity prole across such an array of gratings and with promising numerical conrmations, we proceed towards experimental measurements of the same.

I: Transmittance uctuations
In disordered systems, dierent congurations give rise to dierent cumulative interference eects for a given monochromatic wave. As a result, uctuations are observed in the transmittances for the incident wave over dierent system realizations. Such uctuations have hitherto been characterized for electron transport in systems like disordered wires and their scaling with system length has been shown to be universal [6]. In this section, we translate these studies to optically disordered systems and treat the scaling of transmittance statistics near the localization transition. The principal idea behind the scaling theory of transmission eigenvalues for disordered wires is that these eigenvalues exhibit a Brownian motion as the length of the wire is increased [7]. The resulting Fokker-Planck equation for their distribution is known as the Dorokhov-Mello-Pereyra-Kumar (DMPK) equation, l 2 P (1 , 2 , ..., n , L) P (1 , 2 , ..., n , L) = n (1 + n )J L (N + 2 ) n=1 n n J
N

(1)

where P (1 , 2 , ..., n , L) denotes the probability distribution function, n are related to the transmission eigenvalues Tn as n = (1 Tn )/Tn , L is the system length, l is the elastic mean free path, N N J = i=1 j =i+1 |j i | is the interaction Jacobian, N is the number of propagating transverse modes at the Fermi level (referred to as channels), and is the symmetry index {1, 2, 4} (depending on the presence or absence of time reversal symmetry and/or spin-rotational symmetry). In order to apply this formalism to our system under consideration, we note that for a multilayer of 1D dielectric slabs, there is only a single channel for light propagation, i.e N = 1 and as a result the Jacobian J 1. The localization length is related to the elastic mean free path as (N + 2 )l/2 and for N = 1, we get l. Hence, eq.(1) for our system reduces to the DMPK equation for a linear 1D chain, P (, L) = (1 + ) P (, L) (2) L We note that this equation is independent of the symmetry index . We perform a transformation of variables on this equation as u = (1 + 2) and obtain an equation for the distribution P (u, L) as, 1 2 P (u, L) = (u 1) P (u, L) L u u (3)

We solve [8] this equation subject to the initial conditions P (u, 0) = (u 1) (since a system of zero length corresponds to ballistic transport, i.e T = 1 and since T = 2/(u + 1), we get u = 1) and obtain, eL/4 P (u, L) = 2 2 (L/ )3/2
u

xex

/4 L

cosh(x) cosh(u)

dx

(4)

This distribution corresponds to a lognormal distribution in the transmittance T. We now calculate the scaling of the mean of the transmittance with system size for this distribution.

< T >=
1

2 u+1

P (u, L)du

(5)

which may be simplied by changing the order of integration and performing the integral over u to obtain, 2 4 x 2 e x < T >= eL/ dx (6) cosh( (L/ )x) 0 Clearly, these statistics exhibit a universal behavior in the parameter L/ independent of any other system-specic variations. We now numerically calculate the transmittance distributions for 1D disordered lattices using the transfer matrix formalism and check the analytically obtained scaling equations for the distributions.

We consider the transport of monochromatic waves (of wavelength 0.5 m) through a system of dielectric layers of randomly varying (uniformly distributed) refractive indices n and widths (between 50 nm to 400 nm). The transmittances for the xed wavelength across several realizations of such systems are calculated using the transfer matrix method and the scaling of their statistics with length is studied. We depict below the distributions of transmittances across systems with system sizes of 0.5 and 4 . Clearly, these systems exhibit lognormal distributions with the mean decreasing with increasing system size.

Figure 1: Transmittance distribution for a system length of 0.5

Figure 2: Transmittance distribution for a system length of 4

We now calculate the means and the variances of these distributions for increasing system lengths. In order to probe universality in the parameter L/ , we repeat the calculations for dierent scattering strengths of the dielectric layers (n = 1 to 2 for system A, n = 1 to 1.7 for system B and n = 1 to 1.5 for system C) corresponding to systems with dierent localization lengths. We also compare the numerically calculated scaling of the mean transmittance with the scaling predicted analytically by eq.(6).

Figure 3: Universal scaling of the mean transmittance with the parameter L/

Figure 4: Universal scaling of the variance of the transmittance with the parameter L/

As may be seen in Fig.3 and Fig.4, a strong universality is exhibited in the variation of these quantities with respect to the parameter L/ . The exponential decay of the mean with increasing system length, the familiar signature of localization in disordered systems [4], is exactly as predicted by eq.(6) corresponding to a linear 1D chain. Also of signicant interest is the peculiar trend of the variance which exhibits a maximum near the localization transition (L/ 1) and monotonically decays thereafter. Therefore, the statistics of transmittances in 1D disordered optical systems are found to exhibit a universal scaling with system length, exactly akin to those for electron transport through a linear 1D chain. While the exponential decay of mean transmittance with system length is also observed in nonlocalized absorptive systems, the scaling of the higher moments of the transmittance may serve as clear statistical signatures of localization [9]. Therefore, the peculiar scaling of the variance as observed in this study may prove to be of importance in experiments pertaining to localization.

II. Nearest neighbor level spacings


Electronic disordered systems are typically modeled by large dimensional random matrices [7] whose eigenvalues correspond to the eigen energy levels of the systems. The statistics of the nearest neighbor spacings [12] between these eigenlevels has been a subject of signicant interest. It is found that the 2 statistics resemble a Wigner-surmise (ax ebx , where is the symmetry index as dened earlier, a and b are constants) at small system sizes (metallic phase) and cross over [11] to a Poisson distribution at large system lengths (insulator phase). This reects the fact that the eigenlevels have a propensity to be spectrally far apart at small system sizes, but as the system length is increased, they tend to come closer. This is physically understood to occur due to a strong correlation of electron wavefunctions at small system sizes leading to spectral level repulsion between eigenlevels in the metallic phase. As the system size is increased, the correlation between wavefunctions fades away causing the energy levels to come closer till nally in the deep insulating phase, the nearest levels have a maximum propensity to be only innitesimally separated. In this study, we draw a direct analogy between the eigenvalue spectra of electronic disordered systems and the transmission spectra of optically disordered lattices. Since the eigen energy levels of a disordered system are typically associated with high conductances for electron transport [13], they would analogously correspond to high transmittance modes in the optical spectra. We conduct a systematic numerical study of the statistics of the nearest neighbor spacings between high transmittance modes and characterize their scaling near the localization transition. The occurrence of such high transmittance modes has been understood as a resonant tunnelling through optical potential double barriers [14] formed perchance in the bulk of the system. Due to the multiple backscattering inside this eective cavity, these modes are often associated with an enhanced intensity prole inside the cavity. To demonstrate instances of the formation of such enhanced localized modes, we depict the spatial eld prole of a typical high transmittance mode alongwith the refractive index variation for two system sizes, namely 1.2 and 4.8 .

Figure 5: Spatial eld and refractive index prole for a high T mode for L = 1.2

Figure 6: Spatial eld and refractive index prole for a high T mode for L = 4.8

In Fig. 5 corresponding a system near the localization transition (L = 1.2 ), we observe that the interference of forward and backscattered waves leads to an enhancement of eld amplitude at around 7 m from the origin. In larger systems i.e in the insulating phase, while one typically expects the eld amplitude to decay exponentially from the incident end [4], when the mode tunnels resonantly through a double potential barrier (or an eective cavity) formed in the system, the multiple backscattering in the cavity can lead to a strong enhancement of eld amplitude as may be observed in Fig. 6.

Thus, each resonant frequency forms its own eective cavity in the system at a spatial location which can be estimated and hence such modes have the possibility of light amplication with gain. Furthermore, owing to the high quality factors associated, these modes have potential applications in random lasers [15]. We now compute the transmission spectra for three dierent system sizes, namely L = 0.5 (in the metallic phase), L = 1.2 (near the localization transition) and L = 4.8 (in the deep insulating phase) and employ a peak-nding algorithm to determine the spectral separation between high transmittance modes (T > 0.9). In Fig. 7, we illustrate the transmission spectra for the three sizes and the corresponding histograms of the mean-normalized level spacings are depicted in Fig. 8.

Figure 7: Typical transmission spectra for system sizes 0.5 , 1.2 and 4.8 from top to bottom

Figure 8: Corresponding nearest neighbor spacings statistics for the same system sizes

We observe that for L = 0.5 i.e a system in the metallic phase, the high transmittance modes are wide and spectrally well separated from each other. The corresponding nearest neighbor spacings statistics clearly demonstrate an enhanced propensity of the levels to maintain a nite spectral separation. In this sense, we observe level repulsion between such modes at small system sizes. As the system length is increased, we note that for L = 1.2 the resonant modes become narrower and their spectral separation decreases. In other words, they begin to have a larger propensity to be spectrally near-apart and the level repulsion recedes away until in the deep localized regime for L = 4.8 , the levels have a maximum tendency to be only innitesimally separated and the spacings show a Poissonian distribution. In our attempts to characterize the functional behavior of these statistical distributions, we found that none of the generalized gamma distributions [16] t our statistics satisfactorily. Hence we choose to characterize our distributions by doing a piece-wise t i.e tting the repulsion part and the tail independently as shown in Fig. 9.

Figure 9: Piecewise tting of the nearest neighbor spacing distributions An important feature to notice is that our systems exhibit a nonlinear level repulsion near the origin. This is in distinction with a linear level repulsion observed in electronic disordered systems with time reversal symmetry (for which = 1 leading to the Wigner surmise for the Gaussian orthogonal 2 ensemble sebx , which is linear for s 0). As a result, we choose to t the repulsion part with a power law x . Furthermore, in contrast to a Gaussian tail in electronic systems, we nd that the tail of our distributions have a Poissonian nature even at short system lengths and hence we t it by an exponential decay ex . As before, we check for universality in the scaling of these distributions with L/ . However, since the localization length is a parameter dependent on the ensemble-averaged transmittance, we must take into account its variation with the incident frequency. We also study the variance of the spacings to verify its asymptotic approach to a Poisson distribution.

Figure 10: Typical spectral variation of localization length for a system

Figure 11: Universal scaling of the variance of the normalized nearest neighbor spacing distributions

We note that the localization length does not change much over the spectral range under consideration as shown in Fig. 10 and hence we calculate at an intermediate wavelength of 0.6 m and use it for our scaling analyses. As seen in Fig. 11, the variance of these mean-normalized distributions too scales universally and asymptotically approaches unity or the mean of the distribution, which is the signature of a Poisson distribution.

We now study the scaling of the repulsion and tail exponents and with respect to the parameter L/ for systems with three dierent disorder strengths.

Figure 12: Scaling of the level repulsion exponent for the three system types

Figure 13: Scaling of the tail exponent for the three system types

As may be inferred, a strong universality is seen in the scaling of the repulsion and tail exponents of these distributions with respect to the parameter L/ . We note that the repulsion part is typically characterized by a small number of data points and hence the tting has a lower precision. On the other hand, the exponential tail is more accurately characterized due to the large number of data points, enabling a very good t. Thus, the nearest neighbor spacings statistics of high transmittance modes in 1D disordered optical systems has been characterized to show a universal behavior independent of the disorder strength. They exhibit a clear level repulsion for small system sizes which eventually fades away as the system length is increased and the statistics become Poissonian in nature. This eect is analogous to the classic Wigner-surmise to Poissonian crossover observed in the nearest neighbor spacings statistics of electronic disordered systems near the metal-insulator transition. Some important dierences are the nonlinear nature of the level repulsion parts and the Poissonian nature of the tails of these distributions for small system sizes. While in disordered wires, level repulsion is understood to arise due to a correlation of electronic wavefunctions, in optical systems it could be understood in terms of cavity formations[14]. For small system sizes, resonant cavities for only a few modes are expected and hence a nite separation is observed. When the system length is increased, the number of potential cavity formations increases leading to several more high transmittance modes which are spectrally closely packed. We hope to pursue a rigorous theoretical treatment of this understanding in future.

III. Single mode bers with Bragg gratings


Although light propagation through a multilayer of dielectric slabs can in principle be considered as one dimensional, in practice there is always some nite scattering in the transverse directions and as a result, eects peculiar to 1D systems are obscured. However, light propagation through an optical ber can, to a very good approximation be considered as one-dimensional in nature. Thus, we now try to realize a 1D disordered system using optical bers and attempt to directly observe 1D localization eects in such systems. While one may introduce disorder inside an optical ber by the use of impurities, such scatterers can also lead to absorption and leakage. A more suitable choice of scatterer is the ber Bragg grating[15], which essentially consists of a periodic modulation of refractive index causing an enhanced reectivity at a desired wavelength. Thus, if we conduct our studies at a xed wavelength, a ber with randomly spaced Bragg gratings would correspond to a 1D non-absorptive disordered system. We now describe briey the principle of a Bragg grating and the fabrication technique employed by us.

Figure 14: Schematic of the principle of a Bragg grating (image: National Instruments)

Figure 15: Schematic of the UV phase mask setup used for Bragg grating fabrication

As depicted in Fig. 14, a Bragg grating consists of very small refractive index modulation which forms a reector for a given (Bragg) wavelength B = 2nef f , where nef f is the eective waveguide index and is the spacing period. Such periodic index modulations can be produced using an ultraviolet (UV) phase-mask as shown in Fig. 15. The m = 1 diraction modes of a UV beam from the phase mask are interfered on the ber, which due to its photosensitivity undergoes a permanent index change due to the periodic interference pattern[15]. However, this method involves stripping the jacket of the ber rendering it very brittle. We perform numerical simulations of typical transmittance spectra for a single Bragg grating of dierent reectivities and the behavior of the ensemble averaged transmittance with number of scatterers (or system size).

Figure 16: Single grating spectrum for three dierent reectivities

Figure 17: Ensemble averaged transmittance with the number of gratings

10

We observe in Fig. 16 and Fig.17 that the ensemble averaged transmittance at the Bragg wavelength for a random array of gratings decays exponentially with number of gratings at a rate that increases with increasing strength of the scatterers. We now present experimental data obtained from grating arrays fabricated by us. The fabrication of gratings was done by the phase mask technique using a 240 nm UV excimer laser and interfering the rst two diraction modes on a UV cured photosensitive ber (range 15001600 nm) which produced an index modulation of the order of 104 and a spatial period of around 250 nm. Each grating had a bandwidth of about 0.8 nm and a reectivity of about 3 4 % and the measurements were made using an optical spectrum analyzer and a broadband source.

Figure 18: Transmittance decay with number of gratings fabricated

Figure 19: Transmission spectrum of a 6-grating random array from a broadband source

In Fig. 18, we note the time-averaged transmittance variation with the number of gratings shows an exponential decay. A monotonic decay with no resonant increase in transmittance was probably due to averaging of interference eects leading to uncorrelated reectors. While such studies [19] have been done for large arrays (consisting of over 50 gratings of less than 0.5 % reectivity each), we conduct studies on arrays of 8-10 gratings acting as correlated reectors with interference eects present. The transmission spectrum across such an array of 6 gratings is depicted in Fig. 19. We observe a clear dip at around 1526.5 nm where the gratings overlap most and some side-lobe dips in the neighborhood. We suspect this could be due to the spectral misalignment (upto 0.4 nm) between the gratings caused due to variations in ber pre-stress during fabrication. However, when the interference eects are manifestly present, it is even possible to observe an increase in the transmittance with the addition of gratings. This may be clearly seen in Fig. 20, which shows the numerically obtained trend of transmittance in a ber with the addition of gratings. Such a rise in transmittance due to constructive interference of waves is often associated with locally enhanced intensity proles in the ber as depicted in Fig. 21.

Figure 20: Numerically calculated transmittance with addition of gratings in a ber clearly showing a rise due to interference eects 11

Figure 21: Simulated intensity prole across the same system at 7 gratings (marked in red) clearly showing mode enhancement

We observe in Fig. 20, that as we add gratings (of individual reectivities of 40 % each) at random spatial intervals (varying between 200-1400 m) on a ber, in this particular instance the transmittance decreases till 5 gratings and then undergoes an increase to about 80 % when two more gratings are added. The local eld enhancement due to superposition of multiply backscattered waves is clearly seen in Fig. 21 which is a plot of the intensity prole for the same system with seven gratings. It is to be noted that in our calculated intensity proles, we simply use the grating matrix resulting in a zero spatial extent for the gratings (marked in red). We now develop a perturbative mode prole measurement scheme to directly observe such localized modes experimentally. We note that a single mode ber with Bragg gratings is essentially like a cavity resonator with several scattering optical potentials. It is known that local cavity perturbations produce a transmittance change that is proportional to the mode intensity at the perturbation point [20]. We use this principle to implement a perturbative technique to directly map the spatial intensity prole in a ber with randomly spaced Bragg gratings. In order to verify this numerically, we rst calculate the transmittance for a ber with gratings and also compute the spatial intensity prole produced. We then introduce a perturbation numerically as a local dissipation produced by a small imaginary part in the refractive index of a thin dielectric layer. This dissipative layer is moved across the system in small steps and the resulting percentage change in the transmittance relative to the unperturbed system is computed. A strong correlation between this transmittance change prole with the unperturbed system intensity prole would indicate the success of this measurement scheme.

Figure 22: Intensity prole across an unperturbed system of six randomly spaced gratings

Figure 23: Corresponding perturbative transmittance change prole for the same system

Figure 24: Intensity prole across an unperturbed system of six randomly spaced gratings

Figure 25: Corresponding perturbative transmittance change prole for the same system

In Fig. 22 and Fig. 24, we depict the numerically calculated spatial intensity proles of two distinct systems of randomly spaced Bragg gratings. In Fig. 23 and Fig. 25, we respectively depict the corresponding transmittance-change proles obtained by numerically implementing a point-by-point translation of the local dissipation layer. We notice that there is a strong correlation between the relative transmittance-change prole and the intensity prole of the unperturbed system. There is clearly a direct dependence of the perturbative change in transmittance on the mode intensity at the perturbation point. With these encouraging results, we now attempt to experimentally implement such a measurement scheme.

12

In this regard, we note that local bends in a single mode optical ber are known to introduce dissipative losses. Hence, a local perturbation may be applied by introducing a small bend in the ber. Such a perturbation has to be repeatable with the same bending stress so that the magnitude of the dissipation in uniform for all measurements across the ber. Furthermore, it must also be small and elastic so that the perturbation does not damage the ber and produce a permanent alteration of the mode prole within.

Figure 26: Schematic of the perturbation device depicting the local bend produced

Figure 27: A picture of the perturber used for mode prole mapping across the ber

For this purpose, we fabricate a perturbation device whose schematic is depicted in Fig. 26. This device consists of a groove through which the ber is inserted by means of which the device may be translated freely across the ber. A screw may be lowered onto the ber to produce a local bend and the depth of this screw is noted by means of a circular dial to ensure precise and repeatable control of the perturbation. This device was tested to produce elastic and repeatable measurements on a pure ber. However, for a ber with gratings, since the ber is stripped of its outer jacket at the locations of the gratings during fabrication, care must be taken while translating the perturber across gratings. Breakage of gratings can lead to splicing losses and permanent mode prole changes. We encounter a further signicant feature with bers with several Bragg gratings. We have observed that the output power across a ber with several randomly spaced Bragg gratings uctuates with every measurement. We depict in Fig. 28, the measurements across a six-grating array obtained from a tunable diode laser source at the Bragg wavelength 1526.8 nm on an optical spectrum analyzer. The uctuations are found to persist even after averaging over several data points, i.e they are clearly nonconvergent. Ths could be a consequence of mechanical vibrations, atmospheric changes etc which cause random strain changes in the ber that consequently alter the period of index modulation of the grating eventually leading to random variations in the center wavelengths of the gratings. We fabricated ber stretching devices as shown in Fig. 29 to control the stress in each individual grating and hence obtaining control over its spectral position.

Figure 28: Experimental observations clearly show the nonconvergent nature of the random power uctuations

Figure 29: A ber stretching device designed to spectrally align each individual grating by producing stress in the ber

It is possible to put this claim to test numerically by allowing for very small spectral variations in 13

the individual gratings by a random variation in their index modulation period. We consider gratings of reectivity of about 4 % each and are allowed to have a random spectral variation of 0.3 nm around the center wavelength 1526.8 nm. We depict the spectra of full precision center grating and the maximum allowed variation allowed in Fig. 30. We now simulate the transmission spectra for three realizations of a system of 6 gratings with random spectral shifts within the allowed range of 0.3 nm and depict them in Fig. 31.

Figure 30: Spectra of single gratings with the maximum spectral variation allowed

Figure 31: Transmission spectra for three congurations of random varying 6-grating array

As may be inferred from Fig. 31, the transmission spectra can vary erratically around the center wavelength even with very small jitter in the individual grating wavelengths and hence could result in power uctuations that we observe. For better sweep rates and faster spectrum acqusition, we have integrated an InGaAs linear array (Hamamatsu 9214-512S) on our spectrometer. We now turn our attention to the statistics of these uctuations, as obtained both numerically and experimentally.

Figure 32: Histogram of transmittances observed experimentally across a system of six gratings

Figure 33: Histogram of transmittances calculated numerically across a system of six gratings

In Fig. 32, we depict the histogram of the transmittances measured experimentally across a sixgrating ber normalized to the input power from a tunable diode laser source. We observe that the transmittance uctuates over a wide range (from 0.3 to 0.7). In Fig. 33, we depict the histogram of transmittances calculated numerically across a six-grating array by allowing a spectral variation of 0.3 nm in each of the individual gratings. It may once again be inferred that the uctuations in the transmittances can be very large even for very small random spectral variations in each grating. As a result, our aim is now to decipher the eect of the perturbation device in the mode prole measurement scheme over and above these random power uctuations. We have attempted to isolate the ber from atmospheric uctuations and mechanical vibrations by pasting the ber onto the optical table, but the uctuations are still found to persist. Since the uctuations are proving to be inevitable, we are now studying the eect of the perturber on the statistics of the transmittance uctuations. Due to the brittle nature of the gratings, we are currently making eorts in developing a grating fabrication setup in our laboratory. With this knowledge and experience in hand, we now plan to pursue a direct observation and measurement of localized modes in single mode bers with Bragg gratings.

14

References
[1] Absence of diusion in certain random lattices, PW Anderson - Physical Review, 109, 14921505 (1958) [2] Fifty years of Anderson localization, A Lagendijk, B van Tiggelen, DS Wiersma - Phys. Today, (2009) [3] Localization of light in a disordered medium, DS Wiersma, P Bartolini, A Lagendijk, R Righini - Nature, 390, 671-673 (1997) [4] Symmetry and transport of waves in one-dimensional disordered systems, JB Pendry - Advances in Physics, Taylor & Francis, (1994) [5] Level repulsions in high transmittance modes in 1D random lattices, G. Kulkarni and S. Mujumdar, Photonics 2012 Proceedings, (2012) [6] Universal conductance uctuations in metals, PA Lee, AD Stone - Phys. Rev. Lett, 55, 16221625 (1985) [7] Random-matrix theory of quantum transport, CWJ Beenakker - Reviews of Modern Physics, 69, 731808 (1997) [8] Wave propagation in a one-dimensional random medium, GC Papanicolaou - SIAM Journal on Applied Mathematics, SIAM, (1971) [9] Statistical signatures of photon localization, A. A. Chabanov, M. Stoytchev & A. Z. Genack - Nature, 404, 850-853 (2001) [10] Statistics of energy levels and eigenfunctions in disordered systems, AD Mirlin - Physics Reports, Elsevier (2000) [11] Asymptotics of universal probability of neighboring level spacings at the Anderson transition, I. Zharakeshev & B. Kramer - Phy. Rev. Lett, 79, 717720 (1997) [12] Statistics of spectra of disordered systems near the metal-insulator transition, BI Shklovskii, B Shapiro, BR Sears, P Lambrianides and H. B. Shore, Phys. Rev. B, 47, 1148711490 (1993) [13] Quantum Mechanics, E. Merzbacher , Third Edition, Wiley & Sons [14] Resonances in one-dimensional disordered systems: localization of energy and resonant transmission, KY Bliokh, YP Bliokh, VD Freilikher - Jou. Opt. Soc. Am. B, Vol. 21, No. 1 (2004) [15] The physics and applications of random lasers, DS Wiersma - Nature Physics 4, 359 367, (2008) [16] Nearest-neighbour spacing distributions of the -Hermite ensemble of random matrices, G Le Ca er, C Male, R Delannay - Physica A: Statistical Mechanics and its applications - Elsevier (2000) [17] Cavity formation and light propagation in partially ordered and completely random onedimensional systems, SH Chang, H Cao, ST Ho - IEEE Journal of Quantum Electronics Vol. 39 Issue. 2 (2003) [18] Fiber Bragg Gratings, R. Kashyap Academic Press (1999) [19] Localization of light in a random-grating array in a single-mode ber, O Shapira, B Fischer - Jou. Opt. Soc. Am. B, Vol. 22, No. 12, (2005) [20] Near-eld imaging and frequency tuning of a high-Q photonic crystal membrane microcavity, Sushil Mujumdar, A. F. Koenderink, T. Suenner, B. C. Buchler, M. Kamp, A. Forchel, and V. Sandoghdar, Optics Express, 15, 17214 (2007)

15

You might also like