Arxiv:1706.03886V1 [Physics.Optics]

Home , Medical physics

arXiv:1706.03886v1 [physics.optics] 13 Jun 2017 moigetral netdsraepamnpolaritons plasmon surface super- injected coherently NIMs externally by driven imposing compensation loss plasmon scheme achieves Π and surface [53] The with proposed. conceptualized was [52], was scheme Π or injection difficult media gain realize. with to compensation causality loss on makes based Kramers- rule which the a developed media Using they gain relations, Kronig limitation. of fundamental use the a However, involved that 50]. demonstrated [49, [51] popularity Stockman compen- gained to Amongst losses media intrinsic gain sate developed, [46–48]. com- were metamaterials that the schemes in towards the losses directed of were pensation indiscernible. efforts is are and new fields detector evanescent the Therefore, of of in- floor portion noise lens significant near the below perfect a the theoretical since 45] the in com- [44, of This losses performance [41–43]. material the significant is promises of region visible presence and ex- frared The limitation fundamental a ists. realization, em- their have powered metamaterials in developments recent although [40]. con- in the given in is imaging metamaterials super-resolution of [39]. text of object review an recent of informa- features A am- contain sub-wavelength the which can about fields (NIM) tion evanescent slab focus material a and index plify that (refractive) theorised negative Pendry of regained since has ground limit, imaging tremendous diffraction super-resolution Rayleigh’s allowing The thereby circumventing [1–38]. sens- of technologies other health, prospect and energy, defense, imaging, communications, ing, relevant computing, plasmonics applications and to and metamaterials of materials territory a optical the in novel to of led variety has wide regime electromagnetic sub-wavelength the ∗ ciepamnijcinshm o udffato imaging subdiffraction for scheme injection plasmon Active orsodn uhr [email protected] author: Corresponding eety e opnainshm,cle plasmon called scheme, compensation new a Recently, and nature in exist not does NIM perfect a However, at electrons and photons of interaction the Controlling asv otpoesn ceeb xedn h oscompe loss s the deeper extending cies. for by allows scheme passi here post-processing previous passive proposed wher the implementation imaging, active with limited the comparison sub-diffraction In for scheme active system. injection signal this imaging for the In source in auxiliary noise. external index convolved signal-dependent negative physically and non-ideal losses Pendry’s material for scheme compensation epeeta ciepyia mlmnaino h recentl the of implementation physical active an present We .INTRODUCTION I. ihgnTcnlgclUiest,10 onedD,Houg Dr, Townsend 1400 University, Technological Michigan nny Ghoshroy, Anindya 1 eateto lcrcladCmue niern , Engineering Computer and Electrical of Department 1 yt Adams, Wyatt ne a lens flat index 1 uZhang, Xu o h betsoni gr ,wihhsfu Gaus- four has which Consider 3, figure by resolved. in separated sians shown well object reasonably the still now is image sated mg,soni gr ,dmntae o h etr at feature the how demonstrates 4, figure in shown image, nojc ihsbwvlnt features. sub-wavelength with atten- object resolve the accurately an amplify thereby to and used processing components be Fourier post uated indeed opti- a can of process used demonstrated the type They al that any method. et. this to Adams, to equivalent applied [56] technique be In can modes. the and cal 55], general 54, or is [52, medium idea metamaterials originally gain was plasmonic scheme a for Π envisioned without the Although removed in effects. be losses non-linear could absorption NIM Therefore, the SPPs. local with (SPPs) I.1 lcrcfil antd qae [ squared magnitude field Electric 1. FIG. ini h betadiaepae o nojc ihfeature with object an for planes by image separated and object the in tion hc hw nojc ihtreGusa features figure Gaussian in three with illustrated object is by to an separated This prone shows also which is 1 [56]. it amplification losses, noise absorption for compensation filt inverse equivalent the with technique. resolved post-processing well reasonably is egh os spoieti h ore pcr beyond spectra Fourier the in prominent is Noise length. k k y o 2 lhuhti omo asv nes le provides filter inverse passive of form this Although |E| 0.5 1.5 2 = mlfiainadsprsino h noise the of suppression and amplification nivrefitrps-rcsigi used, is post-processing filter inverse an e 15- 0500511.5 1 0.5 0 -0.5 -1 -1.5 0 1 sto oee ihrsailfrequen- spatial higher even to nsation . mlmnain epooet s a use to propose we implementation, baeeghiaigfrbyn the beyond far imaging ubwavelength sse nfiue2 oee,tecompen- the However, 2. figure in seen as 5 1 a esi h rsneo realistic of presence the in lens flat nrdcdpamnijcinloss injection plasmon introduced y eipeettoso h plasmon the of implementations ve n Durdu and λ o tn I49119,USA 49931-1295, MI hton, λ / o with 4 / ,where 4, λ o / ihipretnegative imperfect with λ .TeFuirsetao h raw the of spectra Fourier The 4. .GüneyO. ¨ o 1 = λ µm y/ o λ stefe pc wave- space free the is h opnae image compensated The . 0 Compensated Image Original Object 1, ∗ m V − 1 ] 2 distribu- er s 2

6 106 Compensation Filter 10 Compensation Filter Original Object Original Object Compensated Image 3 103 10 Compensated Image Raw Image Raw Image 0 100 10 Amplitude -3 10-3 10Amplitude

-6 10-6 10 0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5 k /k k /k y 0 y 0

− − FIG. 2. Fourier spectra [Vm 1] of the three Gaussians and FIG. 4. Fourier spectra [Vm 1] of the four Gaussians and the the compensated image in figure 1, and the raw image ob- compensated image in figure 3, and the raw image along with tained without loss compensation. The compensation filter is the compensation filter. Notice that the feature at ky = 2 is ko the inverse of the transfer function. The compensated image not discernable under the noise and cannot be recovered well spectrum is obtained simply by multiplying the raw image with the compensation. spectrum with the compensation filter. Notice that the noise can be seen for high spatial frequencies which is amplified by the compensation. cation makes the Fourier components that are buried in the noise distinguishable. This allows for the recovery of 1.5 Original Object the previously inaccessible object features by adjusting Compensated Image the amount of compensation from the Π scheme. The technique presented in this paper is based on the 1 same negative index flat lens (NIFL) as in [56]. We use

2 the words ”passive” and ”active” to distinguish be-

|E| tween the compensation schemes applied in [56] and in 0.5 this work, respectively. Therefore, the inverse filter post processing used in [56] and figures 1-4 to emulate the physical compensation of losses can be called passive Π 0 scheme, since no external physical auxiliary is actively in- -1.5 -1 -0.5 0 0.5 1 1.5 volved as opposed to the active Π scheme here, where the y/λ 0 direct physical implementation using an external auxiliary source as originally envisioned in [52] is sought. The − FIG. 3. Electric field magnitude squared [Vm 1]2 for the active compensation scheme allows us to control noise four Gaussians, in the object and image planes, separated by amplification and hence extend the applicability of the Π λo/4 with λo = 1µm. The compensated image is very poorly scheme to higher spatial frequencies. resolved with one of the Gaussians missing.

II. THEORY ky = 2 is not distinguishable under the noise. The final ko compensated image, when subject to the same compen- We define the optical properties of the NIFL with sation scheme, is poorly resolved. We define a feature the relative permittivity and permeability expressed as as any spatial Fourier component that has substantial ′ ′′ ′ ′′ ′ ǫr′ = ǫ + iǫ and µr = µ + iµ , where ǫ = −1 and contribution to the shape of the object. It is clear that µ = −1. COMSOL Multiphysics, the finite element the Fourier components beyond ky = 2 have a significant ko method based software package that we use here, as- contribution to the four Gaussians and must be recovered sumes exp(jωt) time dependence. Therefore, the imag- from the image spectrum in order to accurately resolve inary parts of ǫr and µr are negative for passive media. the object. Therefore, noise presents a limitation which In this paper we have used 0.1 as the imaginary parts of must be overcome to make the Π scheme versatile. both ǫr and µr which is a reasonable value given currently In the present work, we demonstrate how the Π scheme fabricated metamaterial structures [57–59]. The geome- can be significantly improved with the use of a physical try used to numerically simulate the NIFL in COMSOL auxiliary source to recover high spatial frequency fea- is given in figure 5. The first step is characterizing the tures that are buried under the noise. We show that by NIFL with a transfer function. For a detailed discussion using a convolved auxiliary source we can amplify the on the geometry setup and transfer function calculations, object spectrum in the frequency domain. The amplifi- the reader is referred to [56]. Here, we present a brief 3 mathematical description of the compensation scheme. and σ controls the full width at half maximum (FWHM) −1 of the Gaussian. We convolve A(y)= F {A(ky)} with the object O(y) in the spatial domain and denote the new ′ object by O (y). This is expressed as

∞ ′ O (y)= O(y)A(y − α)dα. (5) Z −∞

We shall refer to this convolved object as the total object. Since convolution in the spatial domain is equivalent

to multiplication in the spatial frequency′ domain,′ the Fourier spectrum of the total object O (ky)= F{O (y)}, is related to the original object by

′ O (ky)= O(ky)+ O(ky)P (ky). (6)

The second term on the RHS will be referred to as the ”auxiliary source,” where Po in Eq. 4 defines its amplitude at the center frequency kc . Note that this term, −1 which is a convolution of O(y) with P (y)= F {P (ky)}, FIG. 5. The geometry built in COMSOL to perform numeri- represents amplification in the spatial frequency domain cal simulations (not to scale). OP and IP are the object and provided that P (ky) > 1. Even though the auxiliary image planes, respectively. Electric field is polarized along source is object dependent, as we will discuss later, the the z-axis (pointing out of plane). The object is defined as external field to generate auxiliary source does not re- an electric field distribution [Ez(y)] on the object plane. The quire prior knowledge about the object. Now, the Fourier operating wavelength is λo = 1µm and 2d = 0.5µm. Blue, transform of the fields in the object and image planes are white, and orange regions are the NIFL, air, and perfectly matched layer (PML), respectively. related to each other by the transfer function of the NIFL as defined by Eq. 1. Therefore, in response to the total object, the new field distribution in the image plane, ex- The spatial Fourier transforms of the electric fields ′ − ′ pressed as I (y)= F 1{I (k )}, is transformed as in the object and image planes are related by the pas- y sive transfer function, T (k ) of the imaging system, ′ ′ P y I (k )= T (k )O (k ), (7) which can be calculated with COMSOL. This is expressed y P y y mathematically as ′ where we can plug in the value of O (ky) from Eq. 6 to I(ky)= TP (ky)O(ky ). (1) obtain the convolved image

′ Here O(ky) = F{O(y)} and I(ky) = F{I(y)}, where I (ky)= TP (ky)O(ky)+ TP (ky)O(ky )P (ky). (8) O(y) and I(y) are the spatial distribution of the electric fields in the object and image planes, respectively, and F is the Fourier transform operator. According to The second term on the RHS of Eq. 8 is a measure of [56] the passive compensation is defined by the inverse the residual amplification which managed to propagate of the transfer function. Hence, the loss compensation is to the image plane. Therefore, by controlling Po, from achieved by multiplying the raw image spectrum in Eq. the object plane, we can tune the necessary amplification 1 with the inverse of the transfer function given by of high spatial frequency features to raise the desired fre-

−1 quency spectrum above the noise floor in the image plane. This process is illustrated in figure 6 for different auxil- CP (ky)= TP (ky) . (2) iary amplitudes. The new ”active” loss compensation scheme must con- ”active” For compensation we first define a mathemat- sider the extra power that is now available in the image ical expression given by spectrum. We distinguish the compensation scheme from

A(ky)=1+ P (ky), (3) Eq. 2 with the subscript ”A”. We start by deﬁning the active transfer function of the NIFL as

′ ky 2 ( − k ) I (ky) ko c P (ky)= Poexp − , (4) TA(ky)= . (9) 2σ2 O(ky) where Po is a constant. kc controls the center frequency The numerator of Eq. 9 is the image of the total object 2π of the Gaussian, ko = λ is the free space wave number which is given by Eq. 8. This transfer function is called 4

4 10 Original Object image. Therefore, the greater the auxiliary power, the Noise Added Raw Image Convolved Noise Added Raw Image (P = 400) lower is the required compensation through inverse ﬁlter o Convolved Noise Added Raw Image (P = 4000) within that region of spatial frequencies. It is interesting o Convolved Noise Added Raw Image (P = 40000) 0 o to note at this point the similarity of the active trans- 10 fer functions in ﬁgure 7 and those in [60]. In the latter, however, highly stringent conditions are imposed on the

Amplitude negative index lens to obtain such a transfer function. 10-4

III. NOISE CHARACTERIZATION

1 2 k /k 3 4 y 0 The active compensation scheme will be applied to

−1 an NIFL imaging system affected by noise where the FIG. 6. Fourier spectra [Vm ] of an arbitrary object il- noise process is a circular Gaussian random variable. Al- lustrating how the auxiliary amplitude can be tuned from though there are many different sources of noise, they can the object plane to raise the image spectrum above the noise be broadly classified into, ”signal-dependent” (SD) and floor by controlling the amplification. P (ky) is centered at ky ”signal-independent” (SI). The random nature of noise kc = = 3 with σ = 0.13. ko manifests itself in the form of an uncertainty in the level of the desired signal. This uncertainty is quantified by the standard deviation σ . The actual distortion can be ”active” because it considers the auxiliary to be a part n ′ thought of as a random selection from an infinite set of of the imaging system. Plugging in the value of I (ky) values and the selection process obeys a probability dis- from Eq. 8 into Eq. 9 we obtain the following expression tribution function. The standard deviation describes the for the active transfer function, range of values which have the greatest likelihood of be-

TA(ky)= TP (ky)+ TP (ky)P (ky). (10) ing selected. When the underlying signal is distorted by multiple independent sources of noise, each characterized 2 The active compensation ﬁlter is simply deﬁned as the by Gaussian distributions, then the variance (σn) of the inverse of the active transfer function and is expressed total noise is the sum of the variances of individual noise mathematically by sources [61].

−1 SD noise, as the name implies, is characterized by a σ that is intricately related to spatial (or temporal) CA(ky)= TP (ky)+ TP (ky)P (ky) . (11) n variations in the incoming signal intensity. The magnitude of the signal distortion therefore also increases with

10 the signal strength. Sources of SD noise in an imag- 10 Passive Transfer Function ing system can be present on the detector side or the Active Transfer Function Passive Compensation Filter transmission medium. For example, the statistical na- Active Compensation Filter 105 ture of photons manifests itself as noise which has Pois- sonian statistics. In radiographic detection equipment, such sources of noise are called quantum mottle or quan- 100 tum noise [62, 63]. Another source of SD noise origi-

Amplitude nates from roughness of the transmission media, which in sub-wavelength imaging systems can be for example, 10-5 surface roughness of the NIFL. One can think of surface irregularities as electromagnetic scatterers which ra- diate in different directions, distorting the propagating 0 1 2k /k 3 4 y 0 wave. Previous experiments on the impact of surface roughness [64–66] showed that increasing material losses FIG. 7. Comparisons of the passive transfer function TP (ky) in the NIFL improved the image resolution of the per- and compensation filter CP (ky) with the active transfer func- fect lens for relatively large surface roughness. Although tion TA(ky) and compensation filter CA(ky). P (ky) incor- this may seem counter-intuitive, it can be explained if porated into the active compensation filter is centered at roughness is modelled as a source of scattering. Adding ky kc = = 3 with σ = 0.13. ko material loss is equivalent to lowering the power transmission of the lens. This lowers the magnitude of the Figure 7 illustrates the active and passive transfer func- excitation field responsible for scattering effects and in tions and the corresponding loss compensation schemes. turn reduces the magnitude of the scattered field. If ma- The amount of active compensation drops within 2.5 < terial losses are kept constant, the scattering process will ky < 3. This indicates that in this region the auxil- be proportional to the intensity of illumination provided ko iary source is expected to provide compensation to the to the object. Therefore, such kind of noise is amplified 5 as the illumination intensity is increased. On the other 13 we can frame the equations for the noisy images as hand, the SI noise is quantified by a standard deviation which is not a function of the incoming signal. There- IN (y)= I(y)+ NSD(y)+ NSI (y) (14) fore, the random nature of the noise will be visible only when the incoming signal amplitude is comparable to the and distortions due to the SI noise. A good example of this ′ ′ ′ ′ I (y)= I (y)+ N (y)+ N (y) (15) is ”dark noise”, which affects a CCD sensor even in the N SD SI absence of illumination [67]. corresponding to the ideal images described by Eqs. 1 A well known model [68] used to describe the spatial and 8, respectively. The subscript N indicates the noisy distribution of a signal that has been distorted with both image. The standard deviations of the noise processes SD and SI sources of noise is are

r(y)= s(y)+ f(s(y))N (y)+ N (y), (12) −3 1 2 σn(SD) = 10 I(y), (16) where s(y) is the noiseless or ideal signal and r(y) is the ′ − ′ noisy version. N1(y) and N2(y) are two statistically in- σ = 10 3I (y) (17) dependent random noise processes with zero mean and n(SD) Gaussian probabilities with standard deviations σn1 and and

σn2, respectively. The noise processes N1(y) and N2(y) ′ −3 are signal independent. The signal dependent nature of σn(SI) = σn(SI) = 10 V/m. (18) noise is modelled by modulating N1(y) using the function f(s(y)). Generally, f(s(y)) is a non-linear function Eqs. 16 - 18 fully describe the random variables that are of the ideal signal itself which is chosen based on the sys- used to construct the SI and SD noise terms in Eqs. 14 tem which Eq. 12 is attempting to describe. For example, and 15. f(s(y)) is usually considered to be the photographic den- 1.5 sity which is unitless when modelling signal-dependent ﬁlm grain noise. Therefore, f(s(y))N1(y) represents the 1 eﬀective signal dependent noise term. We can re-write the expression in Eq. 12 as 0.5

y 0 E r(y)= s(y)+ NSD(y)+ NSI (y), (13) -0.5 Total Object where the standard deviation of NSD(y) is f(s(y))σn1 Original Object and the subscripts SD, SI distinguish between the sources -1 Noise Added Raw Image Noise Added Convolved Image of noise. The noise model of Eq. 12, referred to as the sig- Window Function -1.5 nal modulated noise model, is used for signal estimation -6 -4 -2 0 2 4 6 purposes with the Wiener filter. A detailed discussion on y ( µm ) this can be found in [68–71]. However, we will use Eq. −1 13 in this paper for mathematical convenience to analyse FIG. 8. Electric field [Vm ] distributions in the object and the relative contributions of SD and SI noise. image planes. The fields on the image planes are multiplied In the NIFL imaging system which we consider, the by the window function to reduce the errors in the Fourier ≈ 4 ideal signal s(y) will be the electric field distribution on transform. The total object has been scaled down by 10 . ′ the image plane, that is I(y) and I (y) for passive and active schemes, respectively. In [60], Chen, et. al adopted a 60dB signal to noise ratio (SNR) in their negative index lens considering an experimental imaging system detec- IV. RESULTS tor [72]. This corresponds to a SD standard deviation of 10−3I(y). In this work, we adopt the same standard for Having described how both SD and SI noise are added the SD noise. Additionally, we assume a SI noise process to the system, the next step is to evaluate the perfor- in the imaging system by adopting a spatially invariant mance of active-compensation and compare with the pas- standard deviation of 10−3[V/m]. This means that even sive version. We will attempt to image the previously ex- in the absence of illumination, there is a constant back- emplified object comprising four Gaussian features, sep- λo ground noise of the order of 1 mV/m in the detector. arated by 4 with λo = 1µm and compare the results Although the value of the SI noise is chosen arbitrarily, of passive and active compensation. Note that due to this does not limit the results discussed in this paper, the finite extent of the image plane, it is necessary to since the SI noise can be easily suppressed by additional multiply the electric fields with a window function to en- auxiliary power. sure that the field drops to zero where the image plane is By taking into consideration the SNR standard used abruptly terminated. Otherwise, errors are introduced in by Chen, et. al [60] and the mathematical form of Eq. the Fourier transform calculations. Windowing the field 6 distribution simply reduces these sources of error, which was then convolved with the object and the resulting to- will be then visible only in the higher spatial frequencies. tal field distribution on the object plane is shown by the Since these errors are very small compared with the am- blue line in figure 8. plitude of the SD and SI sources of noise, they do not Active compensation filter, defined by Eq. 11 and il- have a significant impact on the calculations. Increas- lustrated in figure 10, is then multiplied in the spatial ing the length of the image plane along the y-axis can frequency domain by the total image spectrum with the also reduce these errors, but because of computational added noise. The resulting compensated spectrum is the constraints this may not be desirable. red line of figure 10. The noise, which was visible in Figure 8 shows the spatial electric field distributions on the total image spectrum beyond ky = 4.5, is also am- ko the object and image planes. Noise was artificially added plified in this reconstruction process. However, in the to the fields on the image plane that were calculated with regions where the auxiliary source is sufficiently strong, COMSOL. The resultant noisy images are indicated by suppression of noise amplification is evident. The recon- the red and green lines in the figure. The Tukey (tapered structed spectrum perfectly coincides with the original cosine) window function was applied to the image plane object shown by the black curve in figure 10. The light only. The Fourier transforms of the images, with and blue line corresponds to the passively compensated image without added noise are shown in figure 9. The black obtained by multiplying Eq. 2 (i.e., dark blue line in fig- line, which corresponds to IN (ky) in Eq. 14, shows how ure 10) with the noise added raw image (i.e., black line in the added noise has clearly affected the raw image spec- figure 9). The advantage of the active compensation over trum beyond ky = 2, where all of the object features are ko its passive counterpart is therefore clearly evident from now completely buried under the noise and indiscernible.′ the reconstructed Fourier spectrum. After the loss com- k However, the blue line, which corresponds to IN (ky) in pensation process, the spectrum is truncated at y =4.7 ko Eq. 15, shows how these features can be recovered with because the simulated transfer function loses accuracy. the convolved auxiliary. We propose the following iterative process to apply the 102 auxiliary source. We then use active compensation filter Noise Added Raw Image Noise Added Convolved Image 0 to reconstruct the image spectrum. 10 Raw Image Convolved Image -2 1. Select an arbitrary kc in the region where the noise 10 Raw Image (not windowed) has substantially degraded the spectrum. Choose 10-4 a guess auxiliary amplitude by selecting P0. -6

Amplitude 10 2. Convolve the object with A(y) to obtain the total object. 10-8

-10 3. Measure the electric ﬁelds on the image plane cor- 10 responding to the total object. 0 1 2k /k 3 4 y 0

4. Re-scale P0 for the selected kc if necessary, to make −1 sure that adequate amplification is available in the FIG. 9. Amplitude of the Fourier transforms [Vm ] on a log scale. Red and green lines are the image spectra I(ky) and image plane and noise is not visible in the Fourier ′ I (ky) with no added noise, respectively. The blue line shows spectrum. ′ the image of the total object with added noise, IN (ky). Noise ky 5. Select another kc on the noise floor and ensure that is visible at > 4.5 due to the inadequate amplification. ko there is sufficient overlap between the adjacent aux- The apparent noise in the pink line are due to the numerical iliaries. errors introduced by the Fourier transform and shifts to higher ky values after applying the Tukey window as seen in the red ko 6. Repeat the processes in 1 − 5 by superimposing line. those multiple auxiliaries until the transfer function of the imaging system is reasonably accurate. We Figure 7 shows that the passive transfer function starts were restricted by the inaccuracy of the simulated ky ky to flatten beyond k = 4.5, even though the analytical passive transfer function TP (ky) beyond = 4.7 o ko which prevented us from going beyond. transfer function monotonically decreases (see figure 3 in [60]), inaccurate simulated transfer function TP (ky) indi- Note that in the above steps the selection of Po does cates that it is no longer possible to perform the required not require prior knowledge about the object. The blue compensation accurately (see Eq. 11). More precisely, and green lines of figure 9 show the total images with and the imaging system requires more compensation than the without added noise, respectively. They include four aux- transfer function predicts. The reconstructed spectrum iliary sources with the center frequencies kc = ky/ko = therefore starts to deviate from the original object when 2.8, 3.1, 3.6, 4.2 and P = 3000, 10000, 4 × 105, 106, ky > 4.5 as seen in the red plot of figure 10, indicat- o ko respectively, with the same σ = 0.35. The final A(y) ing inadequate compensation. This was one of the main 7

ky reasons why we were unable to image beyond = 5. image plane. A very narrow P (ky) in the spatial fre- ko quency domain translates to a wide field distribution in 10 10 Passive Compensation Filter the spatial domain. This created additional field oscilla- Active Compensation Filter tions towards the edges of the image plane increasing the Original Object errors in the Fourier transform calculations. Figure 11 105 Compensated Image (Passive) Compensated Image (Active) shows the amplitude squared of the reconstructed fields illustrating the improvement of the active over passive 100 compensation scheme. Amplitude

10-5 V. DISCUSSION

10-10 The active compensation scheme works, because the 0 1 2 k /k 3 4 y 0 convolved auxiliary source allows us to “selectively amplify” spatial frequency features of the object. This am- FIG. 10. Fourier spectra [Vm−1] of the reconstructed images plification cannot be achieved simply by the superpo- illustrating the difference between active and passive compen- sition of the object with an object independent auxil- sation. The passive compensation has significantly amplified iary source. This is illustrated in figure 12 where we set −1 the noise whereas the active one does not. O(ky)=1Vm in the second term of Eq. 6 and use the same P (k ) distributions in figure 9. The blue and green Additionally, reconstructing the object features suc- y lines correspond to the images of the object superimposed cessfully requires a strong amplification. The auxiliary with the object independent auxiliaries with and without amplitude necessary to produce this amplification is very added noise, respectively. The buried object spectrum at high and it starts to generate substantial electric field os- ky = 3 shown in figure 9 has gone undetected. cillations towards the edges of the image plane. Because ko the image plane is finite along the y-axis and the electric 2 field is abruptly cut at a point where it is non-zero, a 10 computational error is introduced in the spatial Fourier 100 transform. An artefact of this can be seen in figure 10 where the red plot shows that the feature at ky = 1 is 10-2 ko slightly shifted. The error is more prominent when the 10-4 intensity of illumination is increased. Extending the size of the image plane along the y-axis mitigates the error at 10-6 Amplitude the expense of computational or physical resources. 10-8 Noise Added Raw Image Noise Added Superimposed Image 1.5 Raw Image Original Object 10-10 Superimposed Image Compensated Image (Passive) Compensated Image (Active) 0 1 2k /k 3 4 y 0 1 −1 2 FIG. 12. Fourier spectra [Vm ] of the raw image super- |E| imposed with several object independent auxiliaries. Such superposition does not provide amplification and hence the 0.5 feature at ky = 3 is not recovered. Note that the black and ko red lines are the raw images with and without added noise, respectively, as also shown in figure 9. 0 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 y/λ The convolution process to construct the auxiliary 0 source that was described in this paper can be thought FIG. 11. Reconstructed images showing the difference be- of as a form of structured light illumination or wavefront tween active and passive compensation schemes. Note that engineering [73–80]. Along these lines, for example, a the passively compensated image has been scaled down by plasmonic lens imaging system was discussed recently in 7 10 . [73], where the authors described the fields on the image plane by Eq. 1 that contains an illumination func- Note that towards the tails of the amplification, or tion as a result of a phase shifting mask. Spatial fil- when ky > 4.5, the amplification is not strong enough ters based on hyperbolic metamaterials [81–83] may be ko to overcome the noise. Hence, there should be suffi- promising for the implementation of the proposed convo- cient overlap between the two adjacent auxiliaries. The lution. For example, an object illuminated with a high FWHM of P (ky), controlled by σ, can be selected arbi- intensity plane wave and projected on such spatial filters trarily. In our simulations we were limited by the finite can physically implement the convolved auxiliary source 8 corresponding to the second term in Eq. 6 and used in and analyze how each noise term contributes to the total step 2 of the iterative reconstruction process. Here, the distortion of the ideal image under the three illumination spatial filter needs to be engineered to have a transfer schemes. The linearity of the Fourier transform allows us function of the form similar to Eq. 4. The object (i.e., to plot each term in the equations separately and these such as an aperture based object illuminated by a plane are shown in figures 13 - 15 for the weak, strong, and wave) is to be placed on top of this additional metamate- structured illuminations, respectively. In the strong illu- rial layer. The field distribution at the exit of this layer mination case we have used a plane wave whose electric would be the convolution of the object field distribution field is 108 times stronger than the weak illumination. with the point spread function of the layer, hence lead- The green and black lines in figures 13 - 15 correspond ing to the auxiliary source term in Eq. 6 (i.e., second to the images with and without added noise, respectively. term). One way to engineer such a transfer function is The Fourier transforms of the SI and SD noise are the with the hyperbolic metamaterials which support high blue and gold lines, respectively, which add up to the spatial frequency modes. In [83], for example, the trans- total noise shown by the red line. mission coefficient for the transverse magnetic waves in a 2 hyperbolic medium was shown to have multiple peaks in 10 Raw Image the high spatial frequency region. The position of these Noise Added Raw Image 0 peaks can be tuned by changing the filling fraction or 10 Total Noise SD Noise the thickness of the hyperbolic medium. If one has en- -2 SI Noise gineered a metamaterial with a transfer function having 10 one transmission peak P0 around a certain spatial fre- 10-4 quency (i.e., center spatial frequency kc in Eq. 4) and -6 is zero everywhere else, the iterative process where P0 Amplitude 10 is re-scaled (i.e., to control amplification) is functionally -8 equivalent to re-scaling the amplitude of the plane wave 10 E0 illuminating the object. It is also worth mentioning 10-10 here that this physically means actively adjusting the 0 1 2 k /k 3 4 5 y 0 coherent plasmon injection rate in the imaging system to compensate the losses as conceptualized in [52]. On −1 FIG. 13. Fourier spectra [Vm ] of the raw images with and the other hand, controlling the center frequency will re- without added noise illustrating the contribution of the SD quire multiple or tunable metamaterial structures where and SI noise to the total distortion of the image under the the transfer function can be tuned to show transmit- weak illumination case. tance peaks at different center frequencies. Therefore, it would be advantageous to have a broad transmittance To compare the performance of the imaging system in a physical implementation as long as the noise amplifi- under the three illumination schemes, we will see how cation does not start to dominate. Another possible way closely the noise added images overlap with the images to construct the necessary transfer function may be with with no added noise. We should note that the spatial dis- the use of metasurfaces [84], which are ultrathin nanos- tribution of the SD noise will be spread out over multiple tructures fabricated at the interface of two media. The Fourier components [87] and therefore the random nature scattering properties of the sub-wavelength resonant con- of noise will not be visible in the spatial frequency do- stituents of the metasurfaces can be engineered to control main. This can be seen in the gold plots which are fairly the polarization, amplitude, phase, and other properties smooth compared to the blue lines. of light [74–77, 85, 86]. This can allow one to engineer an If we compare the SD noise spectra in figures 13 and 14 arbitrary field pattern from a given incident illumination we immediately conclude that as we increase the intensity [85, 86]. of the illumination, the SD noise is amplified through- In order to understand how the active compensation out the spectrum. However, a slight improvement to enhances the resolution limit of the NIFL we need a the noisy spectrum over the weak illumination is visi- deeper understanding of the effect of the noise on the ble within the region 2 < ky < 3. Additionally, if we ko ideal image spectrum. Eqs. 16 and 17 tell us that the sig- analyze figure 13, where the gold line intersects the black nal dependent noise is amplified proportionally with the line, we see that in the strong illumination case in figure illumination. But since the active compensation seems to 14, only the Fourier components until this intersection work so well, can we say that the noise is not amplified point are recovered. The intersection marks the spatial to the same extent as the signal? Then, would it be pos- frequency at which the ideal image (i.e., raw image with sible to achieve the same results by simply increasing the no added noise) I(ky) matches the Fourier transform of intensity of the plane wave illuminating the object? To the SD noise NSD(ky). Beyond this point, we can say address these questions we will consider below the “weak that the ideal image is completely buried under the SD illumination,” “structured illumination” and “strong il- noise alone. As we steadily increase the intensity of the lumination” cases. illumination, I(ky) and NSD(ky) increase by the same k We will take the Fourier transforms of Eqs. 14 and 15 proportion and therefore, the value of y where the two ko 9

10 10 Raw Image illumination schemes. Note that under the structured il- 108 Noise Added Raw Image lumination, the noise added convolved image closely fol- Total Noise 106 lows the ideal image until ky =4.5. This can be pushed SD Noise ko 104 SI Noise to even higher spatial frequencies if the transfer function 102 characterizing the lens is accurate. Also, note that the 100 structured illumination has successfully suppressed the 10-2 computational errors in the Fourier transform which are

Amplitude k 10-4 visible in the black line in figure 13 beyond y = 3. These ko 10-6 errors are amplified by a factor of 108 times in figure 14. 10-8 From the above discussion we conclude that by using 10-10 structured illumination the SD noise is not amplified but 0 1 2 k /k 3 4 5 y 0 redistributed when compared with the strong illumination. Therefore, it is possible to raise the high spatial FIG. 14. Fourier spectra [Vm−1] of the raw images with and frequency features of the object above the noise. This is without added noise illustrating the contribution of the SD the primary reason why structured illumination can ac- and SI noise to the total distortion of the image under the curately resolve the image while strong illumination fails. strong illumination case. The SD noise is amplified approx- 8 The technique is generally applicable to any arbitrary imately by a factor of 10 throughout the spectrum and the object with the use of any plasmonic or metamaterial lens contribution of the SI noise is very small. provided that accurate transfer function for the imaging system is available. A selective amplification process is 2 10 Raw Image used to recover specific object features by controlling P0 Noise Added Convolved Image near and beyond where the noise floor is reached in the 100 Total Noise SD Noise Fourier spectrum of the raw image. Therefore, no prior 10-2 SI Noise knowledge of the object is required. However, a necessary criterion is a sufficiently accurate transfer func- 10-4 tion in the region where the auxiliary is applied to cor- rectly estimate the required amount of amplification. It -6 Amplitude 10 should be noted that different objects may require dif-

-8 ferent auxiliaries, since the spatial frequency at which 10 the noise floor is reached may vary for different objects. 10-10 Therefore, it would be instrumental to have a tunabil- 0 1 2 k /k 3 4 5 ity mechanism for the versatility of the imaging system. y 0 Even though a single narrowband auxiliary would be still − FIG. 15. Fourier spectra [Vm 1] of the convolved images with sufficient to enhance the resolution of the raw image, fur- and without added noise illustrating the contribution of the ther enhancement in the resolution would demand either SD and SI noise to the total distortion of the image under the superimposing multiple narrowband auxiliaries or a sin- structured illumination case. The SD noise is approximately gle sufficiently broadband auxiliary within the range of at the same level as the weak illumination except that the accurate transfer function. Narrowband auxiliaries re- noise is redistributed. quire larger image plane and more post-processing while a single broadband auxiliary requires less post-processing and smaller image plane at the expense of possibly higher intersect does not change. We can therefore say that the noise amplification. Similarly, unnecessarily large ampli- improvement in the noisy spectrum in figure 14 is due to tude of a narrowband auxiliary may excessively amplify the signal rising above the SI noise which does not change the noise. Another likely limitation arises from increas- with the illumination intensity. Further increments in the ingly large power loss in the deep subwavelength regime, strength of the illumination will not improve the noisy which requires increasingly high amount of amplification image spectrum. to reconstruct extremely fine details of the object. This On the other hand, if we study the SD noise spectrum does not only reduce the efficiency but might also in- in figure 15, we can see that it is approximately at the troduce undesired non-linear and thermal effects in the same average level as figure 13. This is not surprising if optical materials, hence, limiting the resolution of the we compare the green and red plots of figure 8. We can imaging system. see that the spatial′ electric field distribution of the noise added images IN (y) and IN (y) are comparable. The only ′ VI. CONCLUSION difference is that IN (y) has high spatial frequency features. Since the standard deviation of the SD noise is proportional to the amplitude of the image, according In summary, we proposed an active implementation of to Eqs. 16 and 17, we can see why the SD noise is ap- the recently introduced plasmon injection scheme [52] to proximately the same in both the weak and structured significantly improve the resolution of Pendrys non-ideal 10 negative index flat lens beyond diffraction limit in the source will be the focus of our future research. Finally, presence of realistic material losses and SD noise. Simply we should note that we purposefully focused on imperfect by increasing the illumination intensity, it is not gener- negative index flat lens here that poses a highly strin- ally possible to efficiently reconstruct the image due to gent and conservative problem. However, in the shorter the noise amplification. However, in the proposed active term the proposed method can be relatively easily ap- implementation one can counter the adverse noise ampli- plied to experimentally available plasmonic superlenses fication effect by using a convolved auxiliary source which [1–3, 64, 66, 73] and hyperlenses [4–8]. Our findings also allows for a selective amplification of the high spatial fre- raises the hopes for reviving Pendrys early vision of per- quency features deep within the sub-wavelength regime. fect lens [39] by decoupling the loss and isotropy issues We have shown that this approach can be used to con- toward a practical realization [90–95]. trol the noise amplification while at the same time recover features buried within the noise, thus enabling ultra-high resolution imaging far beyond the previous passive im- VII. FUNDING INFORMATION plementations of the plasmon injection scheme [56, 88]. The convolution process to construct the auxiliary source Office of Naval Research (award N00014-15-1-2684). in the proposed active scheme may be realized physically by different methods, metasurfaces [74–78, 84–86] and hyperbolic metamaterials [81–83, 89] being the pri- VIII. ACKNOWLEDGEMENT mary candidates. A more detailed analysis on the design of such structures to implement the convolved auxiliary We thank Jeremy Bos at Michigan Technological Uni- versity for fruitful discussion on noise characterization.

[1] N. Fang, H. Lee, C. Sun, and X. Zhang, “Diffraction- ties,” Progress in Electromagnetic Research B 6, 203–217 limited optical imaging with a silver superlens,” Science (2013). 308, 534–537 (2005). [13] M. Sadatgol, M. Rahman, E. Forati, M. Levy, and D. O.¨ [2] T. Taubner, D. Korobkin, Y. Urzhumov, G. Shvets, and Güney, “Enhanced faraday rotation in hybrid magneto- R. Hillenbrand, “Near-field microscopy through a sic su- optical metamaterial structure of bismuth-substituted- perlens,” Science 313, 1595–1595 (2006). iron-garnet with embedded-gold-wires,” Journal of Ap- [3] X. Zhang and Z. Liu, “Superlenses to overcome the plied Physics 119, 103105 (2016). diffraction limit,” Nature materials 7, 435–441 (2008). [14] M. Choi, S. H. Lee, Y. Kim, S. B. Kang, J. Shin, M. H. [4] Z. Liu, H. Lee, Y. Xiong, C. Sun, and X. Zhang, “Far- Kwak, K.-Y. Kang, Y.-H. Lee, N. Park, and B. Min, “A field optical hyperlens magnifying sub-diffraction-limited terahertz metamaterial with unnaturally high refractive objects,” science 315, 1686–1686 (2007). index,” Nature 470, 369–373 (2011). [5] J. Rho, Z. Ye, Y. Xiong, X. Yin, Z. Liu, H. Choi, [15] W.-C. Chen, C. M. Bingham, K. M. Mak, N. W. Caira, G. Bartal, and X. Zhang, “Spherical hyperlens for two- and W. J. Padilla, “Extremely subwavelength planar dimensional sub-diffractional imaging at visible frequen- magnetic metamaterials,” Physical Review B 85, 201104 cies,” Nature communications 1, 143 (2010). (2012). [6] D. Lu and Z. Liu, “Hyperlenses and metalenses for far- [16] X. Zhang, E. Usi, S. K. Khan, M. Sadatgol, and D. O.¨ field super-resolution imaging,” Nature communications Guney, “Extremely sub-wavelength negative index meta- 3, 1205 (2012). material,” Progress In Electromagnetics Research 152, [7] J. Sun, M. I. Shalaev, and N. M. Litchinitser, “Experi- 95–104 (2015). mental demonstration of a non-resonant hyperlens in the [17] D. Schurig, J. Mock, B. Justice, S. A. Cummer, J. B. visible spectral range,” Nature communications 6, 7201 Pendry, A. Starr, and D. Smith, “Metamaterial electro- (2015). magnetic cloak at microwave frequencies,” Science 314, [8] H. Lee, Z. Liu, Y. Xiong, C. Sun, and X. Zhang, “Devel- 977–980 (2006). opment of optical hyperlens for imaging below the diffrac- [18] J. Gwamuri, D. Güney, and J. Pearce, “Advances in plas- tion limit,” Optics express 15, 15886–15891 (2007). monic light trapping in thin-film solar photovoltaic de- [9] J. Valentine, S. Zhang, T. Zentgraf, E. Ulin-Avila, D. A. vices,” Solar cell nanotechnology pp. 241–269 (2013). Genov, G. Bartal, and X. Zhang, “Three-dimensional op- [19] C. Rockstuhl, S. Fahr, and F. Lederer, “Absorption en- tical metamaterial with a negative refractive index,” na- hancement in solar cells by localized plasmon polaritons,” ture 455, 376–379 (2008). Journal of applied physics 104, 123102 (2008). [10] N. I. Landy, S. Sajuyigbe, J. Mock, D. Smith, and [20] A. Vora, J. Gwamuri, N. Pala, A. Kulkarni, J. M. Pearce, W. Padilla, “Perfect metamaterial absorber,” Physical and D. O.¨ Güney, “Exchanging ohmic losses in metama- 100 review letters , 207402 (2008). terial absorbers with useful optical absorption for photo- [11] V. V. Temnov, “Ultrafast acousto-magneto-plasmonics,” voltaics,” Scientific Reports 4, 4901 (2008). 6 Nature Photonics , 728–736 (2012). [21] A. Vora, J. Gwamuri, J. M. Pearce, P. L. Bergstrom, ¨ [12] M. I. Aslam and D. O. Güney, “On negative index and D. O.¨ Güney, “Multi-resonant silver nano-disk pat- metamaterial spacers and their unusual optical proper- terned thin film hydrogenated amorphous silicon solar 11

cells for staebler-wronski effect compensation,” Journal tric nanoplasmonic clusters,” Optics express 23, A682– of Applied Physics 116, 093103 (2014). A691 (2015). [22] I. Bulu, H. Caglayan, K. Aydin, and E. Ozbay, “Compact [39] J. B. Pendry, “Negative refraction makes a perfect lens,” size highly directive antennas based on the srr metama- Phys. Rev. Lett. 85, 3966–3969 (2000). terial medium,” New Journal of Physics 7, 223 (2005). [40] W. Adams, M. Sadatgol, and D. O.¨ Güney, “Review [23] H. Odabasi, F. Teixeira, and D. Guney, “Electrically of near-field optics and superlenses for sub-diffraction- small, complementary electric-field-coupled resonator an- limited nano-imaging,” AIP Advances 6, 100701 (2016). tennas,” Journal of Applied Physics 113, 084903 (2013). [41] S. Zhang, W. Fan, N. C. Panoiu, K. J. Malloy, R. M. [24] D. O.¨ Güney and D. A. Meyer, “Negative refraction gives Osgood, and S. R. J. Brueck, “Experimental demon- rise to the klein paradox,” Physical Review A 79, 063834 stration of near-infrared negative-index metamaterials,” (2009). Phys. Rev. Lett. 95, 137404 (2005). [25] I. I. Smolyaninov and E. E. Narimanov, “Metric signa- [42] A. N. Grigorenko, A. K. Geim, H. F. Gleeson, Y. Zhang, ture transitions in optical metamaterials,” Physical Re- A. A. Firsov, I. Y. Khrushchev, and J. Petrovic, view Letters 105, 067402 (2010). “Nanofabricated media with negative permeability at vis- [26] M. S. Tame, K. McEnery, S¸. Ozdemir,¨ J. Lee, S. Maier, ible frequencies,” Nature 438, 335–338 (2005). and M. Kim, “Quantum plasmonics,” Nature Physics 9, [43] G. Dolling, C. Enkrich, M. Wegener, C. M. Soukoulis, 329–340 (2013). and S. Linden, “Simultaneous negative phase and group [27] M. A. al Farooqui, J. Breeland, M. I. Aslam, M. Sa- velocity of light in a metamaterial,” Science 312, 892–894 datgol, S¸. K. Ozdemir,¨ M. Tame, L. Yang, and D. O.¨ (2006). Güney, “Quantum entanglement distillation with meta- [44] K. J. Webb, M. Yang, D. W. Ward, and K. A. Nelson, materials,” Optics express 23, 17941–17954 (2015). “Metrics for negative-refractive-index materials,” Phys. 70 [28] M. Asano, M. Bechu, M. Tame, S¸. K. Ozdemir,¨ R. Ikuta, Rev. E , 035602 (2004). ¨ [45] M.-C. Yang and K. J. Webb, “Poynting vector analysis D. O. Güney, T. Yamamoto, L. Yang, M. Wegener, and 30 N. Imoto, “Distillation of photon entanglement using of a superlens,” Opt. Lett. , 2382–2384 (2005). a plasmonic metamaterial,” Scientific reports 5, 18313 [46] M. P. Nezhad, K. Tetz, and Y. Fainman, “Gain as- sisted propagation of surface plasmon polaritons on pla- (2015). 12 [29] P. K. Jha, X. Ni, C. Wu, Y. Wang, and X. Zhang, nar metallic waveguides,” Opt. Express , 4072–4079 “Metasurface-enabled remote quantum interference,” (2004). Physical review letters 115, 025501 (2015). [47] A. K. Popov and V. M. Shalaev, “Compensating losses in negative-index metamaterials by optical parametric am- [30] R. A. Sperling, P. R. Gil, F. Zhang, M. Zanella, and 31 W. J. Parak, “Biological applications of gold nanoparti- plification,” Opt. Lett. , 2169–2171 (2006). cles,” Chemical Society Reviews 37, 1896–1908 (2008). [48] D. O. Güney, T. Koschny, and C. M. Soukoulis, “Reduc- ing ohmic losses in metamaterials by geometric tailor- [31] X. Huang, I. H. El-Sayed, W. Qian, and M. A. El-Sayed, 80 “Cancer cell imaging and photothermal therapy in the ing,” Phys. Rev. B , 125129 (2009). near-infrared region by using gold nanorods,” Journal of [49] S. Anantha Ramakrishna and J. B. Pendry, “Removal of 128 absorption and increase in resolution in a near-field lens the American Chemical Society , 2115–2120 (2006). 67 [32] S. Lal, S. E. Clare, and N. J. Halas, “Nanoshell-enabled via optical gain,” Phys. Rev. B , 201101 (2003). photothermal cancer therapy: impending clinical im- [50] Zhang, J., Jiang, H., Gralak, B., Enoch, S., Tayeb, G., pact,” Accounts of chemical research 41, 1842–1851 and Lequime, M., “Compensation of loss to approach 1 effective index by gain in metal-dielectric stacks,” Eur. (2008). 46 [33] R. Guo, L. Zhang, H. Qian, R. Li, X. Jiang, and B. Liu, Phys. J. Appl. Phys. , 32603 (2009). “Multifunctional nanocarriers for cell imaging, drug de- [51] M. I. Stockman, “Criterion for negative refraction with 26 low optical losses from a fundamental principle of causal- livery, and near-ir photothermal therapy,” Langmuir , 98 5428–5434 (2010). ity,” Phys. Rev. Lett. , 177404 (2007). ¨ [34] Z. J. Coppens, W. Li, D. G. Walker, and J. G. Valentine, [52] M. Sadatgol, K. Ozdemir, L. Yang, and D. O. Güney, “Probing and controlling photothermal heat generation “Plasmon injection to compensate and control losses in 115 in plasmonic nanostructures,” Nano Letters 13, 1023– negative index metamaterials,” Phys. Rev. Lett. , 1028 (2013). 035502 (2015). [35] M. D. Blankschien, L. A. Pretzer, R. Huschka, N. J. [53] M. I. Aslam and D. O. Güney, “Surface plasmon driven Halas, R. Gonzalez, and M. S. Wong, “Light-triggered scalable low-loss negative-index metamaterial in the vis- 84 biocatalysis using thermophilic enzyme–gold nanoparti- ible spectrum,” Phys. Rev. B , 195465 (2011). cle complexes,” ACS nano 7, 654–663 (2012). [54] D. O.¨ Güney, T. Koschny, and C. M. Soukoulis, “Sur- [36] Y.-S. Chen, W. Frey, S. Kim, K. Homan, P. Kruizinga, face plasmon driven electric and magnetic resonators for K. Sokolov, and S. Emelianov, “Enhanced thermal stabil- metamaterials,” Physical Review B 83, 045107 (2011). ity of silica-coated gold nanorods for photoacoustic imag- [55] M. I. Aslam and D. O.¨ Güney, “Dual-band, double- ing and image-guided therapy,” Optics express 18, 8867– negative, polarization-independent metamaterial for the 8878 (2010). visible spectrum,” JOSA B 29, 2839–2847 (2012). [37] S. P. Sherlock, S. M. Tabakman, L. Xie, and H. Dai, [56] W. Adams, M. Sadatgol, X. Zhang, and D. O. Güney, “Photothermally enhanced drug delivery by ultra-small “Bringing the perfect lens into focus by near-perfect com- multifunctional feco/graphitic-shell nanocrystals,” Acs pensation of losses without gain media,” New Journal of Nano 5, 1505 (2011). Physics 18, 125004 (2016). [38] A. Ahmadivand, N. Pala, and D. O.¨ Güney, “Enhance- [57] T. Xu, A. Agrawal, M. Abashin, K. J. Chau, and H. J. ment of photothermal heat generation by metallodielec- Lezec, “All-angle negative refraction and active flat lens- 12

ing of ultraviolet light,” Nature 497, 470–474 (2013). [76] Y. Zhao and A. Alù, “Manipulating light polarization [58] C. Garc´ıa-Meca, J. Hurtado, J. Mart´ı, A. Mart´ınez, with ultrathin plasmonic metasurfaces,” Phys. Rev. B W. Dickson, and A. V. Zayats, “Low-loss multilayered 84, 205428 (2011). metamaterial exhibiting a negative index of refraction [77] N. Yu, P. Genevet, M. A. Kats, F. Aieta, J.-P. Tetienne, at visible wavelengths,” Phys. Rev. Lett. 106, 067402 F. Capasso, and Z. Gaburro, “Light propagation with (2011). phase discontinuities: Generalized laws of reflection and [59] E. Verhagen, R. de Waele, L. Kuipers, and A. Polman, refraction,” Science 334, 333–337 (2011). “Three-dimensional negative index of refraction at opti- [78] Y. Xu, J. Sun, W. Walasik, and N. M. Litchinitser, cal frequencies by coupling plasmonic waveguides,” Phys. “Probing metamaterials with structured light,” Opt. Ex- Rev. Lett. 105, 223901 (2010). press 24, 26249–26254 (2016). [60] Y. Chen, Y.-C. Hsueh, M. Man, and K. J. Webb, “En- [79] D. R. Smith, D. Schurig, M. Rosenbluth, S. Schultz, S. A. hanced and tunable resolution from an imperfect negative Ramakrishna, and J. B. Pendry, “Limitations on subd- refractive index lens,” J. Opt. Soc. Am. B 33, 445–451 iffraction imaging with a negative refractive index slab,” (2016). Applied Physics Letters 82, 1506–1508 (2003). [61] R. D. Fiete and T. Tantalo, “Comparison of snr image [80] S. Cao, T. Wang, Q. Sun, B. Hu, and W. Yu, “Meta- quality metrics for remote sensing systems,” Optical En- nanocavity model for dynamic super-resolution fluores- gineering 40, 574–585 (2001). cent imaging based on the plasmonic structure illumina- [62] M. N. Wernick, O. Wirjadi, D. Chapman, Z. Zhong, N. P. tion microscopy method,” Opt. Express 25, 3863–3874 Galatsanos, Y. Yang, J. G. Brankov, O. Oltulu, M. A. (2017). Anastasio, and C. Muehleman, “Multiple-image radiog- [81] C. Rizza, A. Ciattoni, E. Spinozzi, and L. Columbo, raphy,” Physics in Medicine and Biology 48, 3875 (2003). “Terahertz active spatial filtering through optically tun- [63] G. T. Barnes, “Radiographic mottle: A comprehensive able hyperbolic metamaterials,” Opt. Lett. 37, 3345– theory,” Medical Physics 9, 656–667 (1982). 3347 (2012). [64] Z. Guo, Q. Huang, C. Wang, P. Gao, W. Zhang, Z. Zhao, [82] D. Schurig and D. R. Smith, “Spatial filtering using me- L. Yan, and X. Luo, “Negative and positive impact of dia with indefinite permittivity and permeability ten- roughness and loss on subwavelength imaging for super- sors,” Applied Physics Letters 82, 2215–2217 (2003). lens structures,” Plasmonics 9, 103–110 (2014). [83] B. Wood, J. Pendry, and D. Tsai, “Directed subwave- [65] H. Wang, J. Q. Bagley, L. Tsang, S. Huang, K.-H. Ding, length imaging using a layered metal-dielectric system,” and A. Ishimaru, “Image enhancement for flat and rough Physical Review B 74, 115116 (2006). film plasmon superlenses by adding loss,” J. Opt. Soc. [84] P. Genevet, F. Capasso, F. Aieta, M. Khorasaninejad, Am. B 28, 2499–2509 (2011). and R. Devlin, “Recent advances in planar optics: from [66] H. Liu, B. Wang, L. Ke, J. Deng, C. C. Choy, M. S. plasmonic to dielectric metasurfaces,” Optica 4, 139–152 Zhang, L. Shen, S. A. Maier, and J. H. Teng, “High con- (2017). trast superlens lithography engineered by loss reduction,” [85] C. Pfeiffer and A. Grbic, “Metamaterial huygens sur- Advanced Functional Materials 22, 3777–3783 (2012). faces: tailoring wave fronts with reflectionless sheets,” [67] G. C. Holst, “Ccd arrays, cameras, and displays,” (1998). Physical review letters 110, 197401 (2013). [68] J. F. Walkup and R. C. Choens, “Image process- [86] C. L. Holloway, E. F. Kuester, J. A. Gordon, J. O’Hara, ing in signal-dependent noise,” Optical Engineering 13, J. Booth, and D. R. Smith, “An overview of the theory 133258–133258 (1974). and applications of metasurfaces: The two-dimensional [69] J. J. Heine and M. Behera, “Aspects of signal-dependent equivalents of metamaterials,” IEEE Antennas and Prop- noise characterization,” J. Opt. Soc. Am. A 23, 806–815 agation Magazine 54, 10–35 (2012). (2006). [87] K. Hirakawa, “Fourier and filterbank analyses of signal- [70] R. Kasturi, J. F. Walkup, and T. F. Krile, “Image dependent noise,” in “2008 IEEE International Con- restoration by transformation of signal-dependent noise ference on Acoustics, Speech and Signal Processing,” to signal-independent noise,” Appl. Opt. 22, 3537–3542 (2008), pp. 3517–3520. (1983). [88] X. Zhang, W. Adams, M. Sadatgol, and D. O.¨ Güney, [71] G. K. Froehlich, J. F. Walkup, and T. F. Krile, “Esti- “Enhancing the resolution of hyperlens by the compen- mation in signal-dependent film-grain noise,” Appl. Opt. sation of losses without gain media,” Progress in Electro- 20, 3619–3626 (1981). magnetic Research C 70, 1–7 (2016). [72] M. Akiba, K. Tsujino, and M. Sasaki, “Ultrahigh- [89] X. Zhang, S. Debnath, and D. O.¨ Güney, “Hyperbolic sensitivity single-photon detection with linear-mode sil- metamaterial feasible for fabrication with direct laser icon avalanche photodiode,” Opt. Lett. 35, 2621–2623 writing processes,” JOSA B 32, 1013–1021 (2015). (2010). [90] C. M. Soukoulis and M. Wegener, “Optical metamate- [73] Z. Zhao, Y. Luo, N. Yao, W. Zhang, C. Wang, P. Gao, rialsmore bulky and less lossy,” Science 330, 1633–1634 C. Zhao, M. Pu, and X. Luo, “Modeling and experimen- (2010). tal study of plasmonic lens imaging with resolution en- [91] C. M. Soukoulis and M. Wegener, “Past achievements hanced methods,” Opt. Express 24, 27115–27126 (2016). and future challenges in the development of three- [74] A. V. Kildishev, A. Boltasseva, and V. M. Shalaev, “Pla- dimensional photonic metamaterials,” Nature Photonics nar photonics with metasurfaces,” Science 339 (2013). 5, 523–530 (2011). [75] A. Pors, M. G. Nielsen, R. L. Eriksen, and S. I. Bozhevol- [92] D. O.¨ Güney, T. Koschny, M. Kafesaki, and C. M. Souk- nyi, “Broadband focusing flat mirrors based on plas- oulis, “Connected bulk negative index photonic metama- 13 monic gradient metasurfaces,” Nano Letters , 829–834 terials,” Optics letters 34, 506–508 (2009). (2013). 13

[93] D. O.¨ G¨uney, T. Koschny, and C. M. Soukoulis, “Intra- medium,” IEEE Transactions on Antennas and Propa- connected three-dimensionally isotropic bulk negative in- gation 60, 3661–3669 (2012). dex photonic metamaterial,” Optics express 18, 12348– [95] S. Yang, X. Ni, B. Kante, J. Zhu, K. O’Brien, Y. Wang, 12353 (2010). and X. Zhang, “Experimental demonstration of opti- [94] S. M. Rudolph and A. Grbic, “A broadband cal metamaterials with isotropic negative index,” in three-dimensionally isotropic negative-refractive-index “Lasers and Electro-Optics (CLEO), 2016 Conference on,” (IEEE, 2016), pp. 1–2.