betaFeedback

WorksPhysics and Astronomy

Layer-Peeling Inversion of Temporal Refractive Index Profiles from Noisy Data

Abstract

We introduce a direct method for reconstructing the temporal evolution of a medium from electromagnetic field measurements. A sequence of temporal refractive-index changes splits an incident wave into forward- and backward-propagating components, producing a spatial response of finite extent. We show that this finite support, together with causality, enables sequential reconstruction of the temporal interfaces through a time-domain extension of layer peeling. For continuously varying media, we introduce a time-adjusted Goupillaud medium and, through space-time duality, establish the corresponding momentum-conservation and admissibility conditions. The method remains robust under incomplete and noisy measurements, providing a general framework for the metrology, characterization, and design of time-varying electromagnetic systems.

Introduction

The interaction of electromagnetic (EM) waves with time-varying media has been studied for decades [1, 2, 3, 4, 5]. Such media have been shown to support a variety of unconventional wave phenomena, including nonreciprocity, temporal bandgaps, harmonic generation, and photonic time-crystal effects [6, 7, 8, 9, 10, 11]. In recent years, technological advances have made it increasingly feasible to move from theoretical concepts toward practical implementations and to harness temporal modulation for the manipulation of EM waves [12, 13, 14, 15, 16]. This progress has been enabled, in large part, by the ability to rapidly modulate electromagnetic devices and metasurfaces on time scales sufficiently short to substantially affect high-frequency electromagnetic fields [17, 18, 19].

In parallel, inverse scattering has long played an important role in electromagnetics, providing a framework for inferring the properties of electromagnetic systems from field measurements [20, 21, 22]. Closely related inverse-design approaches address the complementary problem of determining the system properties required to produce a prescribed field response [23, 24].

The convergence of these two research directions has led to growing interest in inverse problems for time-varying media, and particularly in determining the temporal variation of material properties required to realize a desired wave response [25, 26, 27, 28]. A powerful conceptual tool for addressing such problems is space-time duality [29], through which models and techniques developed for spatially varying media can be adapted to systems that vary in time. This correspondence has enabled the temporal counterparts of numerous established concepts and methods, including scattering matrices [25, 30], Fresnel coefficients [31], layered-media formulations [32, 33, 34], finite-difference time-domain methods [35], transmission-line and input-impedance descriptions [27], and quarter-wave transformers [26].

Inverse design, however, assumes that the desired response is prescribed and seeks a temporal modulation that realizes it [25, 36, 37]. The corresponding metrology problem is fundamentally different: the temporal evolution of the medium is unknown and must be inferred from measured electromagnetic fields. In practice, these measurements are inevitably incomplete and contaminated by noise, making direct application of inverse-design methodologies difficult. In this work, we show that causality, together with the finite support of the spatial impulse response, provides sufficient structure to enable robust reconstruction of temporary varying material properties from such measurements. Specifically, we develop a time-domain extension of layer peeling that sequentially reconstructs the temporal interfaces and extend the formulation to continuously varying media through a time-adjusted Goupillaud model. The resulting framework enables accurate reconstruction even from incomplete and noisy data, providing a direct approach to the metrology and characterization of time-varying electromagnetic systems.

Description of the system

We consider an infinite and homogeneous, non-dispersive medium, with permittivity ϵ0\epsilon_0 and permeability μ0\mu_0. Throughout the analysis, the permeability remains constant, so the material can be characterized by a single parameter - the refractive index nn. Inside the medium, a TEM wave E0+,H0+E_0^+, H_0^+ with a finite width in space is propagating to the right (in the positive xx direction). At time t=0t = 0, the refractive index of the medium is switched instantaneously, to n1n_1. Due to the continuity of magnetic flux and electric displacement before and after switching, reflections will be caused. The reflection and transmission coefficients are given by [32]:

R1=12n0n1(n0n1−1)T1=12n0n1(n0n1+1).(1)R_1=\frac{1}{2}\frac{n_0}{n_1}\left(\frac{n_0}{n_1}-1\right)\qquad T_1=\frac{1}{2}\frac{n_0}{n_1}\left(\frac{n_0}{n_1}+1\right). \tag*{(1)}

For simplicity of discussion, from now on only the electric field will be addressed. After switching the medium properties from n0n_0 to n1n_1, a right traveling wave and a left traveling wave will be propagating in the medium: E0(x,t)=TE0+(x−c1t)+RE0+(x+c1t)E_0(x,t)=TE_0^{+}(x-c_1t)+RE_0^{+}(x+c_1t), where c1=c0/n1c_1=c_0/n_1 is the speed of light after switching (c0c_0 is the speed of light in vacuum). More generally, we may assume a superposition of right and left traveling waves is propagating in the medium when switching of the properties occurs, and the waves are allowed to propagate for a period of Δt1\Delta t_1 after the change. The relationship between the waves before and after this procedure (a temporal layer) is given by the following matrix equation [25]:

\begin{bmatrix} E_1^{+} \\ E_1^{-} \end{bmatrix} = \begin{bmatrix} -\Delta x_1 & 0 \\ 0 & \Delta x_1 \end{bmatrix} \begin{bmatrix} T_1 & R_1 \\ R_1 & T_1 \end{bmatrix} \begin{bmatrix} E_0^{+} \\ E_0^{-} \end{bmatrix} \triangleq Q_1 \begin{bmatrix} E_0^{+} \\ E_0^{-} \end{bmatrix}, \tag*{(2)}\end{bmatrix}

where the Δx1\Delta x_1 operator is a shift operator in space: Δx1E=E(x+Δx1)=E(x+c1Δt1)\Delta x_1 E=E(x+\Delta x_1)=E(x+c_1\Delta t_1).

Now let’s consider that this process is repeated NN times, with unknown values {ni}i=1N\{n_i\}_{i=1}^{N} and propagation times {Δti}i=1N\{\Delta t_i\}_{i=1}^{N}. In the following, we outline a method to solve the inverse scattering problem using this scheme, i.e. reconstruct the refractivity values and time durations, by knowledge of the input signal at time t=0t=0 and the output signal (at time t=∑i=1NΔtit=\sum_{i=1}^{N}\Delta t_i), as well as the state of the medium at the end - nNn_N. Without loss of generality, we will assume that at t=0t=0 only a right-traveling wave exists in the medium. Observing Eq. (2), the overall field after the series of switching events is: EN=QNQN−1…Q1E0E_N=Q_NQ_{N-1}\ldots Q_1E_0, where Ei=[Ei+,Ei−]T\boldsymbol{E}_i=[E_i^{+},E_i^{-}]^{\mathsf{T}}, and QiQ_i is the transfer operator of switching ii, representing the switching from ni−1n_{i-1} to nin_i and propagation over time Δti\Delta t_i. This idea is conveyed in the spatial impulse response of this system, which will be finite for any finite NN (FIR). Since the system is linear and space invariant, the output response can be represented as a convolution of the input field with a spatial impulse response h(x)h(x): EN+=h∗E0+E_N^{+}=h*E_0^{+}. Here hh is the impulse response for transmission, but the same is true for reflection. Figure 1(a)+(Figure 1b) shows the evolution of the temporal (IIR) and spatial (FIR) impulse responses for structures with spatial and temporal layers, respectively. The total number of propagation paths is 2N2^N, distributed between the forward- and backward-propagating responses; generically, each contains 2N−12^{N-1} path contributions. We will show in this paper, that the impulse response finiteness, together with causality, are the fundamental concepts that allow the reconstruction of the medium temporal variation from field measurements in NN steps.

Evolution of the impulse response in layered systems

Figure 1. Evolution of the impulse response in layered systems. (a) In space, causality indicates that reflections from superficial layers will appear before reflections from deeper ones. This idea renders that there is a certain duration Δt\Delta t at the beginning of the reflected signal through which there is no signal overlap, and the response is related to the incident wave only through the properties of the first interface. This is the main layer peeling concept that allows recovering the layer parameters. (b) In time, each property switching separates the signal into two waves (right and left traveling), where waves that are further away from x=0x = 0 underwent less direction inversions. Causality indicates there is a certain length Δx\Delta x at the end of the reflected/transmitted signal through which there is no signal overlap. In both cases, this part of the signal is related to the incident wave through multiplication by a single constant, which allows the recovery of the first layer (in space) or the last layer (in time). (c) A space-time illustration of the temporal layer model in discussion, with relevant notations for the analysis. For layer ii, the refractive index of the medium is nin_i, for duration Δti\Delta t_i. The distance each wave has propagated in space during this time is Δxi\Delta x_i (d) An illustration of the temporal layer peeling algorithm. Left - the non-overlapping region 2ΔxN2\Delta x_N can be found from the supports of EN+E_N^+ and EN−E_N^-. Center - the signals are shifted in space and the properties of the last switching are found. Right - the signals are shifted back in time through inversion QN−1Q_N^{-1}, thus reducing their supports, and the process can be repeated.

The deconvolution process

The layer peeling method is a well known method to directly and efficiently recover the properties of a layered material from the reflection response [20]. The key idea underlying this method is causality, which implies the response of deeper layers is delayed relative to the response of the more superficial ones. Thus, for example, if the reflection is measured as a function of time at the entrance to a transmission line, there exists a certain time t1t_1, that for all t<t1t<t_1 the measured reflection is caused only by the first interface. This fact allows reconstruction of the line’s properties. Then, we may ‘peel’ this layer (once its properties are known) and apply the same principle on the second layer, and so on. Even though there is an infinite number of reflections, the peeling allows to recover the material properties in NN steps, NN being the number of interfaces in the line. Here we apply similarly, yet somewhat different considerations, to recover the properties of the material in time.

Returning to the medium undergoing temporal switching, based on the discussion above it is clear that the right and left traveling waves after NN switching events - EN+(x)E_N^{+}(x) and EN−(x)E_N^{-}(x), will be a superposition of the input field E0+(x)E_0^{+}(x) (i.e. the field at time t=0t=0), shifted to various locations in space, equivalent to convolution with h(x)h(x). To portray a clear picture of the properties of EN+(x)E_N^{+}(x) and EN−(x)E_N^{-}(x) that will reveal the medium parameters in earlier times, let us consider EN−1+(x)E_{N-1}^{+}(x). In this state (as in any other), there exists a wave component in EN−1+(x)E_{N-1}^{+}(x) that was never reflected, and has always propagated to the right. Of all the wave components, this component has traveled the longest distance to the right, which can be expressed as: xN=∑i=1NΔxi=∑i=1Nc1Δtix_N=\sum_{i=1}^{N}\Delta x_i=\sum_{i=1}^{N}c_1\Delta t_i. Δxi\Delta x_i denotes the distance each of the waves traveled between switching events ii and i+1i+1. See an illustration of these parameters in Figure 1(c), for a system with N=2N=2. Any other component (right traveling or left traveling), should be to the left of this wave. We note here that we are only considering the right side of the output response, however the same is true for the left side of the output response. Now the parameters are switched one more time, from nN−1n_{N-1} to nNn_N, and the waves are allowed to propagate for time ΔtN\Delta t_N. Now the pulse in discussion was separated into two components, as per Eq. (2). These arguments, as well as the fact that the input pulse is of finite length, lead to two important takeaways: (1) There exists a point in space - xfx_f such that EN+(x)=0 ∀x>xfE_N^{+}(x)=0\ \forall x>x_f. (2) There exists a point in space to the left of xfx_f (that we will denote by xpx_p), such that to its right, only one replica of the input wave exists (there is a region with no signal overlap). The first takeaway stems directly from the fact that the system is FIR, while the second stems from causality. Therefore, in the vicinity of the right edge of the signal (denoted by a gray square in Figure 1(d), the signal can be expressed as:

EN+(x)={AN+E0+(x−xN),xp<x<xf,0,x>xf,(3)E_N^{+}(x)= \begin{cases} A_N^{+}E_0^{+}(x-x_N), & x_p<x<x_f,\\ 0, & x>x_f, \end{cases} \tag*{(3)}
EN−(x)={AN−E0+(x−xN+2ΔxN),xp<x+2ΔxN<xf,0,x+2ΔxN>xf.(4)E_N^{-}(x)= \begin{cases} A_N^{-}E_0^{+}(x-x_N+2\Delta x_N), & x_p<x+2\Delta x_N<x_f,\\ 0, & x+2\Delta x_N>x_f. \end{cases} \tag*{(4)}

In the equations above, AN+=∏i=1NTiA_N^{+}=\prod_{i=1}^{N}T_i is the amplitude of the rightmost right-traveling wave and AN−=∏i=1N−1TiRNA_N^{-}=\prod_{i=1}^{N-1}T_iR_N is the amplitude of the rightmost left-traveling wave. By knowledge of the supports of EN+E_N^{+} and EN−E_N^{-} we may find ΔxN\Delta x_N, and shift the signals accordingly, to the state in which they were ΔtN\Delta t_N time units ago, right after switching. Then, using Eq. (3) together with Eq. (1), the refractive index before switching - nN−1n_{N-1}, is recovered. The last step is to shift the signals back in time, to the state in which they were before the switching, and then repeat the process. This can be done easily by inverting Eq. (2):

\begin{bmatrix} E_{N-1}^{+}(x)\\ E_{N-1}^{-}(x) \end{bmatrix} = \frac{1}{T_N^2-R_N^2} \begin{bmatrix} T_N & -R_N\\ -R_N & T_N \end{bmatrix} \begin{bmatrix} E_N^{+}(x+\Delta x_N)\\ E_N^{-}(x-\Delta x_N) \end{bmatrix}. \tag*{(5)}\end{bmatrix}

The last step reduces the number of nonzero terms of h(x)h(x) by a factor of 2, since the new overlap between the wave components will cancel out half of them. This allows the process to be over by NN steps. Because this is in-fact a shift of the waves back in time, the takeaways that were mentioned above are valid for EN−1+E_{N-1}^{+} and EN−1−E_{N-1}^{-}, as they were for EN+E_N^{+} and EN−E_N^{-}, and the process can be repeated. To summarize the algorithm: (1) Find the supports of EN+E_N^{+} and EN−E_N^{-} (on the right side of the xx axis), denoted by xf+x_f^{+} and xf−x_f^{-}. (2) Find the spatial shift between the waves: ΔxN=12(xf+−xf−)\Delta x_N=\frac{1}{2}(x_f^{+}-x_f^{-}). This promptly gives ΔtN\Delta t_N. (3) Shift the signals EN+E_N^{+} and EN−E_N^{-} and recover nN−1n_{N-1} using Eqs. (1) and (3) in the vicinity of xf+x_f^{+} and xf−x_f^{-}. (4) Shift the signals back in time using Eq. (2) and return to (1). The process is terminated when the reflected wave vanishes.

To demonstrate the applicability of the general layer-peeling procedure to abrupt temporal variations, we next consider a refractive-index profile consisting of several successive discontinuities whose temporal locations are not known a priori. In this case, we use the general formulation described above, rather than the equal-traveling-distance discretization introduced in § 3. At each peeling step, the propagation distance associated with the last temporal interval is inferred from the relative locations of the edges of the forward- and backward-propagating signals, after which the corresponding temporal interface is recovered and removed. Figure 2 shows an example in which a sequence of sharp changes in n(t)n(t) is accurately reconstructed from the measured fields. A finite-support trapezoidal incident waveform is used so that the relevant signal edges can be identified accurately. This example also illustrates an important limitation of the general procedure: errors in determining the signal edges accumulate through successive peeling steps. Consequently, reconstruction of abrupt discontinuities at unknown temporal locations becomes increasingly sensitive to noise and edge-localization errors as the number of temporal interfaces increases.

Reconstruction of a temporal refractive-index profile and corresponding fields

Figure 2. Reconstruction of a temporal refractive-index profile containing abrupt discontinuities using the general layer-peeling procedure of Sec. III. (a) Ground-truth (labeled GT in the legend) refractive-index profile n(t)n(t) and its reconstruction. The temporal locations of the discontinuities are not assumed to be known a priori, but are recovered sequentially from the measured fields. (b) Corresponding forward- and backward-propagating fields obtained using a finite-support trapezoidal incident waveform. The sharply defined signal edges facilitate accurate estimation of the propagation distance at each peeling step. Such an accuracy is not required for the method suggested in § 4. There, a smooth Gaussian signal is used for demonstration.

Equal traveling distance layers

The algorithm detailed in the previous section is exact for a piecewise-constant temporal profile and can approximate continuously varying profiles through sufficiently fine temporal discretization, for any input signal with finite support. For the sake of concreteness, and since our ultimate goal is the practical metrology of continuously varying time-dependent media, we shall assume from now on that the distance traveled by the waves between two consecutive switching events is constant, i.e. Δxi=Δx ∀i\Delta x_i=\Delta x\ \forall i. This choice is inspired by analogous problems in space, in which the traveling time in each layer is assumed constant (Goupillaud medium) [38]. The constant traveling distance allows mathematically compact expressions for the wave functions, which comes to our aid when recovering the medium parameters, and is very suitable for continuous switching. This is because the degree of freedom in choosing Δti\Delta t_i allows Δxi\Delta x_i to be constant. For a constant traveling distance between switching events, the transfer matrix between temporal layers ii and i+1i+1 for a harmonic wave with wavenumber kk obtains the form:

\begin{bmatrix} E_i^{+}\\ E_i^{-} \end{bmatrix} = \begin{bmatrix} T_i e^{-jk\Delta x} & R_i e^{-jk\Delta x}\\ R_i e^{jk\Delta x} & T_i e^{jk\Delta x} \end{bmatrix} \begin{bmatrix} E_{i-1}^{+}\\ E_{i-1}^{-} \end{bmatrix}, \tag*{(6)}\end{bmatrix}

Due to the spatial homogeneity, the wavenumber kk is conserved in time [37]. If we denote: z=e2jkΔxz=e^{2jk\Delta x} and R~i=12(ni−1ni−1)\widetilde{R}_i=\frac{1}{2}\left(\frac{n_{i-1}}{n_i}-1\right), this equation becomes:

\begin{bmatrix} E_i^{+}\\ E_i^{-} \end{bmatrix} = \frac{n_{i-1}}{n_i}z^{1/2} \begin{bmatrix} (1+\widetilde{R}_i)z^{-1} & \widetilde{R}_i z^{-1}\\ \widetilde{R}_i & 1+\widetilde{R}_i \end{bmatrix} \begin{bmatrix} E_{i-1}^{+}\\ E_{i-1}^{-} \end{bmatrix}, \tag*{(7)}\end{bmatrix}

Here we harness the zz-transform notation from digital signal processing to indicate a constant spatial shift, or a phase shift that is linear with kk. The field waves after NN switching events are obtained by multiplication of the transfer matrices in Eq. (7), resulting in polynomials in zz. The overall transfer matrix of NN events would be:

\begin{bmatrix} E_N^{+}\\ E_N^{-} \end{bmatrix} = \frac{n_0}{n_N}z^{N/2} \begin{bmatrix} A(z) & B(z)\\ C(z) & D(z) \end{bmatrix} \begin{bmatrix} E_0^{+}\\ E_0^{-} \end{bmatrix}, \tag*{(8)}\end{bmatrix}

where,

\begin{bmatrix} A(z) & B(z)\\ C(z) & D(z) \end{bmatrix} = \prod_{i=1}^{N} \begin{bmatrix} (1+\widetilde{R}_i)z^{-1} & \widetilde{R}_i z^{-1}\\ \widetilde{R}_i & 1+\widetilde{R}_i \end{bmatrix}. \tag*{(9)}\end{bmatrix}

The product is ordered so that the matrix corresponding to i=1i=1 is rightmost. The full transmission and reflection coefficients would be:

Ttot(k)=n0nNzN/2A(e2jkΔx)(10)T_{\mathrm{tot}}(k)=\frac{n_0}{n_N}z^{N/2}A(e^{2\mathrm{j}k\Delta x}) \tag*{(10)}
Γtot(k)=n0nNzN/2C(e2jkΔx).(11)\Gamma_{\mathrm{tot}}(k)=\frac{n_0}{n_N}z^{N/2}C(e^{2\mathrm{j}k\Delta x}). \tag*{(11)}

Here, A(z)A(z) is of degree NN and C(z)C(z) is of degree N−1N-1. The form of Eq. (9) is familiar from scattering theory in space [39]. The transfer matrices in space and time can be related, thus providing conditions of admissibility. For the real, positive refractive indices considered here, space-time duality renders that AA has to be a maximum phase system, i.e., all the zeros of AA must lie outside of the unit circle. Also, the conserved quantity in space invariant systems is momentum (the temporal equivalent of energy conservation) [40]. This renders that on the unit circle:

nNn0∣A(ejθ)∣2−nNn0∣C(ejθ)∣2=1.(12)\frac{n_N}{n_0}|A(e^{\mathrm{j}\theta})|^2-\frac{n_N}{n_0}|C(e^{\mathrm{j}\theta})|^2=1. \tag*{(12)}

See App. A and App. § 7 for derivations of these conditions. As shown in Eqs. (9) and (9a), the transmission

and reflection responses are represented through the zz-transform of their impulse responses. Therefore, the coefficients may be calculated through the spectrum of TtotT_{\mathrm{tot}} and Γtot\Gamma_{\mathrm{tot}}, or directly from the spatial representation of the signal using an optimization approach. While for design problems the former may be a more direct method, for metrology problems both methods are applicable. We have chosen the latter.

For the inversion process, we start by assuming MM switching events, where the signal travels a distance of Δx\Delta x between two consecutive events. The distance Δx\Delta x may be chosen arbitrarily, although should be sufficiently small. However, the overall distance MΔxM\Delta x should correspond to the actual distance traveled by the signals, which may be estimated by the support of the measured right traveling signal Eout+E_{\mathrm{out}}^+. Then, assuming we have the input signal E0+E_0^+ and the output signals Eout+E_{\mathrm{out}}^+ and Eout−E_{\mathrm{out}}^- as vectors, sampled at LL points, we use the input signal to construct the shift matrices SA,SC∈RL×MS^A,S^C\in\mathbb{R}^{L\times M}:

SijA=E0+(xi+(M−2j)Δx),(13)S^A_{ij}=E_0^+\left(x_i+(M-2j)\Delta x\right), \tag*{(13)}
SijC=E0+(xi+(M−2j+2)Δx).(14)S^C_{ij}=E_0^+\left(x_i+(M-2j+2)\Delta x\right). \tag*{(14)}

That is, each column of SAS^{A} and SCS^{C} is a shifted version of the input Eout±E_{\mathrm{out}}^{\pm}. Since Eq. (9a) contains the common factor n0/nMn_{0}/n_{M}, we define the normalized output vectors Eout±=(nM/n0)Eout±\mathbf{E}_{\mathrm{out}}^{\pm}=(n_{M}/n_{0})\mathbf{E}_{\mathrm{out}}^{\pm}. Then we solve:

a^=arg⁡min⁡a∥Eout+−SAa∥22(15)\hat{\mathbf{a}}=\underset{\mathbf{a}}{\arg\min}\left\|\mathbf{E}_{\mathrm{out}}^{+}-S^{A}\mathbf{a}\right\|_{2}^{2} \tag*{(15)}
c^=arg⁡min⁡c∥Eout−−SCc∥22(16)\hat{\mathbf{c}}=\underset{\mathbf{c}}{\arg\min}\left\|\mathbf{E}_{\mathrm{out}}^{-}-S^{C}\mathbf{c}\right\|_{2}^{2} \tag*{(16)}

and find the coefficient vectors a\mathbf{a} and c\mathbf{c}. These vectors represent the polynomials A(z)A(z) and C(z)C(z) as follows: A(z)=∑i=1Maiz−iA(z)=\sum_{i=1}^{M}a_i z^{-i}, C(z)=∑i=0M−1ciz−iC(z)=\sum_{i=0}^{M-1}c_i z^{-i}. From Eq. (9) we note that: aM=∏i=1M(1+R~i)a_M=\prod_{i=1}^{M}(1+\widetilde{R}_i) and cM−1=R~M∏i=1M−1(1+R~i)c_{M-1}=\widetilde{R}_M\prod_{i=1}^{M-1}(1+\widetilde{R}_i), which renders:

R~M=cM−1aM−cM−1.(17)\widetilde{R}_M=\frac{c_{M-1}}{a_M-c_{M-1}}. \tag*{(17)}

This result is consistent with the main idea of the general algorithm, that the properties of the last switching are ’hidden’ in the part of the wave that has propagated the longest distance (or accumulated the largest phase). After recovering R~M\widetilde{R}_M, the temporal layer is peeled off by inversion of Eq. (7), resulting in new polynomials for A(z)A(z) and C(z)C(z), of degrees M−1M-1 and M−2M-2, respectively. Performing this process MM times will recover the properties of all layers.

Results

This section demonstrates the suggested algorithm in action, for continuous modulation of the material properties. First, an arbitrary continuous modulation of n(t)n(t) was simulated with N=300N=300. The resulting coefficients of A(z)A(z) and C(z)C(z) were calculated, and the measured signals Eout+(x)E_{\mathrm{out}}^{+}(x) and Eout−(x)E_{\mathrm{out}}^{-}(x) were generated numerically using these coefficients, assuming a Gaussian pulse input with spatial width (std) σx\sigma_x. Then, the output was used to reconstruct the properties of the layers, first by applying the least-squares (LS) method to solve Eqs. (12a) and (16), and then recovering the values of nn. An important key in the LS problem is the choice of the number of layers (MM) for reconstruction. As shown in Figure 3(left, blue curve), when no noise is present, and the algorithm assumes a number of layers which is comparable to the number of layers in the simulation (M=N=300M=N=300), the results match the profile very well. Physically, this means that the layer thickness (in time) is small compared to the typical time of property variation, or, equivalently, that the assumption that n(t)n(t) is constant throughout the layer duration Δti\Delta t_i is a good approximation of the continuous time variation. However, when the layer

Reconstruction of refractive-index variation in time from clean and noisy data

Figure 3. Reconstruction of refractive-index variation in time from clean and noisy data (SNR = 15 dB). (r) denotes reconstruction with regularization.

thickness is large compared to the actual variation time (N=300N=300, M=30M=30), the LS problem results in a large and discontinuous variation of the properties due to the need to fit a signal comprised of 300 shifted Gaussians using 30 shifted Gaussians. To overcome this issue, a regularization term was added to the LS problem:

a^=arg⁡min⁡a{∥Eout+−SAa∥22+λ∥Da∥22}(18)\hat{\mathbf{a}}=\underset{\mathbf{a}}{\arg\min}\left\{\left\|\mathbf{E}_{\mathrm{out}}^{+}-S^{A}\mathbf{a}\right\|_{2}^{2}+\lambda\left\|D\mathbf{a}\right\|_{2}^{2}\right\} \tag*{(18)}
c^=arg⁡min⁡c{∥Eout−−SCc∥22+λ∥Dc∥22}(19)\hat{\mathbf{c}}=\underset{\mathbf{c}}{\arg\min}\left\{\left\|\mathbf{E}_{\mathrm{out}}^{-}-S^{C}\mathbf{c}\right\|_{2}^{2}+\lambda\left\|D\mathbf{c}\right\|_{2}^{2}\right\} \tag*{(19)}

where DD is a second order difference matrix and λ\lambda is the regularization parameter (we used λ=0.001\lambda=0.001). The second-order difference penalty introduces a smoothness prior appropriate for the continuously varying profiles considered here and stabilizes the deconvolution when neighboring shifted replicas of the incident field become nearly linearly dependent. It therefore suppresses rapidly varying coefficient sequences that are weakly constrained by the measurements, reducing spurious oscillations in the reconstructed profile. This significantly improves the reconstruction of the properties, as can be seen in Figure 3 (left, red curve).

With the inclusion of additive white Gaussian noise (AWGN) to the model, an additional issue must be addressed. We define the unitless parameter ξ=σx/Δx\xi= \sigma_x / \Delta x, which has a major effect on the recovery quality. In Figure 3 (right, blue curve), the pulse width σx\sigma_x is large compared to the layer thickness Δx\Delta x (ξ=50\xi= 50). In this case, reconstruction is not possible. This is because the columns of SAS^{A} and SCS^{C} are nearly dependent and the LS problem in Eqs. (12) becomes ill-conditioned. This can be solved by decreasing the number of layers (increasing Δx\Delta x - red curve) or by using a narrower input pulse (decreasing σx\sigma_x - orange curve).

To further examine reconstruction from incomplete measurements, we consider finite-rate spatial sampling of the output fields. Since the reflected signal is not strictly bandlimited, no finite sampling rate provides a complete representation of its spatial spectrum. We therefore define an effective bandwidth using the half-power bandwidth of the reflected signal, and denote the corresponding Nyquist sampling rate by fNf_N. Figure 4 shows reconstruction results, including measurement noise, for sampling rates ranging from 0.5fN0.5f_N to 10fN10f_N. Reasonable reconstruction is obtained over a broad range of sampling rates, including rates close to and somewhat below fNf_N, despite the incomplete representation of the continuous spatial field. This robustness results from the structured nature of the inverse problem together with the regularization used in the reconstruction. As expected, however, reducing the sampling rate increases the sensitivity to noise and to the sampling grid, and for the lowest sampling rate considered, 0.5fN0.5f_N, the reconstruction becomes unstable.

Figure 4. Reconstruction of the temporal refractive-index profile from incompletely sampled output fields. The spatial sampling rate is varied relative to the reference Nyquist rate fNf_N associated with the half-power bandwidth of the reflected signal. Measurement noise is included in all cases. Accurate reconstruction is maintained over a broad range of sampling rates, while at sufficiently low sampling rates the increased ill-conditioning and noise sensitivity lead to deterioration of the recovered profile. In the legend, ‘Sim’ stands for the ground truth, while the next items labeled ’Rec’ correspond to different levels of reconstruction incompleteness associated with the sampling rates 10fN10f_N, 5fN5f_N, fNf_N, 0.9fN0.9f_N, 0.8fN0.8f_N, and 0.5fN0.5f_N

Scope and limitations

The present analysis considers finite, nonperiodic temporal modulations in a lossless and nondispersive medium, with additive measurement noise applied to the output fields. The transfer-matrix formulation remains formally applicable to finite periodic modulations, including regimes exhibiting parametric gain; however, strong amplification may substantially increase the dynamic range and ill-conditioning of the forward operator, thereby increasing the sensitivity of the reconstruction to noise. Noise already present in the incident field or introduced during the modulation process may also undergo parametric amplification. Material loss and dispersion

introduce additional challenges: loss reduces the available signal-to-noise ratio, whereas dispersion prevents propagation between temporal interfaces from being represented by simple spatial translations of the incident waveform. In principle, the framework may be generalized to a prescribed causal dispersive model, such as a Drude or Lorentz medium, by augmenting the electromagnetic state with the corresponding material variables and formulating the peeling procedure in this enlarged state space. Such extensions are expected to introduce additional conditioning and identifiability challenges and are left for future work.

Conclusions

In this paper, we have established a robust and direct methodology for the reconstruction of temporal refractive index profiles from field measurements. By exploiting the unique FIR nature of the spatial impulse response, we have shown that the layer-peeling concept can be successfully adapted to provide direct, step-by-step inversion of temporal layers. The transition from a general piecewise model to Gouipillaud medium allows the use of zz-transform formalisms, enabling the recovery of continuous modulations through simple optimization methods. As technological capabilities in high-speed material modulation continue to evolve, the proposed temporal layer-peeling method offers a critical analytical tool for the characterization and diagnostic evaluation of devices with time varying properties. Moreover, the derived methods and conditions for admissibility can be readily used for the design of time-varying media. Extensions of the present framework to strongly amplifying, lossy, and dispersive temporal media constitute natural directions for future work.

Spatial and Temporal Transfer Matrices

In this section we present realizability conditions and the condition for conservation of momentum for temporal layers with equal traveling distance. This supplementary material will use a slightly different expression for the transfer matrix from the main text (up to a constant). This is to have a better vision of the (mathematical) duality between spatial-domain and time-domain transfer matrices.

We consider a spatial layer as a transition between two materials, followed by propagation distance Δx\Delta x (see Figure 5-left). The transfer matrix of a single spatial layer reads:

Fields before and after a single spatial and temporal layer

Figure 5. Fields before and after a single spatial (left) and temporal (right) layer

\begin{bmatrix} \widetilde{E}_0^+(t,x) \\ \widetilde{E}_0^-(t,x) \end{bmatrix} = \widetilde{Q}_{01} \begin{bmatrix} \widetilde{E}_1^+(t,x+\Delta x) \\ \widetilde{E}_1^-(t,x+\Delta x) \end{bmatrix}. \tag*{(20)}\end{bmatrix}

Here we adopt the notation of transfer from left to right, as in [39]. We consider a temporal layer as an instantaneous transition between two materials, followed by propagation time Δt\Delta t (see Figure 5-right). The transfer matrix of a single temporal layer reads:

\begin{bmatrix} \widehat{E}_1^+(t+\Delta t,x) \\ \widehat{E}_1^-(t+\Delta t,x) \end{bmatrix} = \widehat{Q}_{01} \begin{bmatrix} \widehat{E}_0^+(t,x) \\ \widehat{E}_0^-(t,x) \end{bmatrix}. \tag*{(21)}\end{bmatrix}

Henceforth, the tilde ( ~\widetilde{\ }) denotes spatial-domain quantities, and the hat ( ^\widehat{\ }) will denote time-domain quantities. The waves are reflected by the interface (spatial/temporal), and are then allowed to propagate for some distance/time. The transfer matrix in space would be:

Q~01=12n0[(n0+n1)ejk1Δx1(n0−n1)e−jk1Δx1(n0−n1)ejk1Δx1(n0+n1)e−jk1Δx1].(22)\widetilde{Q}_{01}=\frac{1}{2n_0} \begin{bmatrix} (n_0+n_1)e^{jk_1\Delta x_1} & (n_0-n_1)e^{-jk_1\Delta x_1} \\ (n_0-n_1)e^{jk_1\Delta x_1} & (n_0+n_1)e^{-jk_1\Delta x_1} \end{bmatrix}. \tag*{(22)}

and in time:

Q^01=n02n12[(n0+n1)e−jω1Δt1(n0−n1)e−jω1Δt1(n0−n1)ejω1Δt1(n0+n1)ejω1Δt1].(23)\widehat{Q}_{01}=\frac{n_0}{2n_1^2} \begin{bmatrix} (n_0+n_1)e^{-j\omega_1\Delta t_1} & (n_0-n_1)e^{-j\omega_1\Delta t_1} \\ (n_0-n_1)e^{j\omega_1\Delta t_1} & (n_0+n_1)e^{j\omega_1\Delta t_1} \end{bmatrix}. \tag*{(23)}

In the following, we will assume that there are multiple spatial layers, for which: Δti=Δxi/ci=Δt\Delta t_i=\Delta x_i/c_i=\Delta t, i.e., the traveling time in each layer is constant. In an analogous fashion, we will assume that for multiple temporal layers, Δxi=Δti⋅ci=Δx\Delta x_i=\Delta t_i\cdot c_i=\Delta x, i.e., the traveling distance in each temporal layer is equal. Thus, we denote:

z~=e2jωΔt,z^=e2jkΔx(24)\widetilde{z}=e^{2j\omega\Delta t},\qquad\widehat{z}=e^{2jk\Delta x} \tag*{(24)}

Using this notation, the transfer matrix from any layer ii to layer i+1i+1 becomes:

Q~i,i+1=z~1/22ni[(ni+ni+1)(ni−ni+1)z~−1(ni−ni+1)(ni+ni+1)z~−1](25)\widetilde{Q}_{i,i+1}=\frac{\widetilde{z}^{1/2}}{2n_i} \begin{bmatrix} (n_i+n_{i+1}) & (n_i-n_{i+1})\widetilde{z}^{-1} \\ (n_i-n_{i+1}) & (n_i+n_{i+1})\widetilde{z}^{-1} \end{bmatrix} \tag*{(25)}
Q^i,i+1=z^1/2ni2ni+12[(ni+ni+1)z^−1(ni−ni+1)z^−1(ni−ni+1)(ni+ni+1)].(26)\widehat{Q}_{i,i+1}=\frac{\widehat{z}^{1/2}n_i}{2n_{i+1}^2} \begin{bmatrix} (n_i+n_{i+1})\widehat{z}^{-1} & (n_i-n_{i+1})\widehat{z}^{-1} \\ (n_i-n_{i+1}) & (n_i+n_{i+1}) \end{bmatrix}. \tag*{(26)}

The overall transfer matrix of NN interfaces (N+1N+1 layers) in space and time would be:

Q~=∏i=0N−1Q~i,i+1,Q^=∏i=0N−1Q^i,i+1,(27)\widetilde{Q}=\prod_{i=0}^{N-1}\widetilde{Q}_{i,i+1},\qquad\widehat{Q}=\prod_{i=0}^{N-1}\widehat{Q}_{i,i+1}, \tag*{(27)}

so that:

\begin{bmatrix} \widetilde{E}_0^+(t,x) \\ \widetilde{E}_0^-(t,x) \end{bmatrix} = \widetilde{Q} \begin{bmatrix} \widetilde{E}_N^+(t,x+\sum\Delta x_i) \\ \widetilde{E}_N^-(t,x+\sum\Delta x_i) \end{bmatrix}, \tag*{(28)}\end{bmatrix}
\begin{bmatrix} \widehat{E}_N^+(t+\sum\Delta t_i,x) \\ \widehat{E}_N^-(t+\sum\Delta t_i,x) \end{bmatrix} = \widehat{Q} \begin{bmatrix} \widehat{E}_0^+(t,x) \\ \widehat{E}_0^-(t,x) \end{bmatrix}. \tag*{(29)}\end{bmatrix}

The matrices can be written as:

Q~=z~N/2νS2N∏i=0N−1[(ni+ni+1)(ni−ni+1)z~−1(ni−ni+1)(ni+ni+1)z~−1](30)\widetilde{Q}=\frac{\widetilde{z}^{N/2}\nu_S}{2^N}\prod_{i=0}^{N-1} \begin{bmatrix} (n_i+n_{i+1}) & (n_i-n_{i+1})\widetilde{z}^{-1} \\ (n_i-n_{i+1}) & (n_i+n_{i+1})\widetilde{z}^{-1} \end{bmatrix} \tag*{(30)}
Q^=z^N/2νT2N∏i=0N−1[(ni+ni+1)z^−1(ni−ni+1)z^−1(ni−ni+1)(ni+ni+1)](31)\widehat{Q}=\frac{\widehat{z}^{N/2}\nu_T}{2^N}\prod_{i=0}^{N-1} \begin{bmatrix} (n_i+n_{i+1})\widehat{z}^{-1} & (n_i-n_{i+1})\widehat{z}^{-1} \\ (n_i-n_{i+1}) & (n_i+n_{i+1}) \end{bmatrix} \tag*{(31)}

where the multiplication is performed so that Q~=Q~0,1⋅…⋅Q~N−1,N\widetilde{Q}=\widetilde{Q}_{0,1}\cdot\ldots\cdot\widetilde{Q}_{N-1,N} and Q^=Q^N−1,N⋅…⋅Q^0,1\widehat{Q}=\widehat{Q}_{N-1,N}\cdot\ldots\cdot\widehat{Q}_{0,1}, and:

νS=∏i=0N−11ni,νT=∏i=0N−1nini+12.(32)\nu_S=\prod_{i=0}^{N-1}\frac{1}{n_i},\qquad\nu_T=\prod_{i=0}^{N-1}\frac{n_i}{n_{i+1}^2}. \tag*{(32)}

Realizability condition for equal travel distance layers

For the dual discretizations, we choose Δx=cΔt\Delta x=c\Delta t, so that z^(ω/c)=z~(ω)\widehat{z}(\omega/c)=\widetilde{z}(\omega). From Eqs. (A6) and (A7) we see that:

Q^i,i+1(ω/c)=(nini+1)2JQ~i,i+1T(ω)J(33)\hat{Q}_{i,i+1}(\omega/c)=\left(\frac{n_i}{n_{i+1}}\right)^2J\tilde{Q}^{T}_{i,i+1}(\omega)J \tag*{(33)}

where we denoted Q^(ω/c)=Q^(e2jωcΔx)\hat{Q}(\omega/c)=\hat{Q}(e^{2j\frac{\omega}{c}\Delta x}) and Q~(ω)=Q~(e2jωΔt)\tilde{Q}(\omega)=\tilde{Q}(e^{2j\omega\Delta t}). Also, c=c0/n0c=c_0/n_0, c0c_0 being the speed of light in vacuum, and:

J=[0110](34)J=\begin{bmatrix}0&1\\1&0\end{bmatrix} \tag*{(34)}

Due to the fact that JJ=IJJ=I, II being the 2x22\mathrm{x}2 identity matrix, we get:

Q^=(n0nN)2JQ~TJ.(35)\hat{Q}=\left(\frac{n_0}{n_N}\right)^2J\tilde{Q}^{T}J. \tag*{(35)}

Let’s denote the entries of Q~\tilde{Q} and Q^\hat{Q} as per Eq. (8):

Q~=(n0nN)z~N/2[A~(z~)B~(z~)C~(z~)D~(z~)],(36)\tilde{Q}=\left(\frac{n_0}{n_N}\right)\tilde{z}^{N/2} \begin{bmatrix} \tilde{A}(\tilde{z})&\tilde{B}(\tilde{z})\\ \tilde{C}(\tilde{z})&\tilde{D}(\tilde{z}) \end{bmatrix}, \tag*{(36)}
Q^=(n0nN)z^N/2[A^(z^)B^(z^)C^(z^)D^(z^)].(37)\hat{Q}=\left(\frac{n_0}{n_N}\right)\hat{z}^{N/2} \begin{bmatrix} \hat{A}(\hat{z})&\hat{B}(\hat{z})\\ \hat{C}(\hat{z})&\hat{D}(\hat{z}) \end{bmatrix}. \tag*{(37)}

Based on Eq. (35):

D^(ω/c)=(n0nN)2A~(ω)(38)\hat{D}(\omega/c)=\left(\frac{n_0}{n_N}\right)^2\tilde{A}(\omega) \tag*{(38)}

From Eqs. (A9) and (B4), the reflection and transmission coefficients for the spatial and temporal NN layers are defined [39]:

Γ~(ω)=C~(z~=e2jωΔt)A~(z~=e2jωΔt),T~(ω)=1γ~A~(z~=e2jωΔt),(39)\tilde{\Gamma}(\omega)=\frac{\tilde{C}(\tilde{z}=e^{2j\omega\Delta t})}{\tilde{A}(\tilde{z}=e^{2j\omega\Delta t})},\qquad \tilde{T}(\omega)=\frac{1}{\tilde{\gamma}\tilde{A}(\tilde{z}=e^{2j\omega\Delta t})}, \tag*{(39)}

and:

T^(k)=γ^A^(z^=e2jkΔx),Γ^(k)=γ^C^(z^=e2jkΔx),(40)\hat{T}(k)=\hat{\gamma}\hat{A}(\hat{z}=e^{2jk\Delta x}),\qquad \hat{\Gamma}(k)=\hat{\gamma}\hat{C}(\hat{z}=e^{2jk\Delta x}), \tag*{(40)}

where for convenience we have denoted: γ~=(n0nN)ejωNΔt\tilde{\gamma}=\left(\frac{n_0}{n_N}\right)e^{j\omega N\Delta t} and γ^=(n0nN)ejkNΔx\hat{\gamma}=\left(\frac{n_0}{n_N}\right)e^{jkN\Delta x}. Realizability condition of the spatial layers is that the zeros of A~\tilde{A} are inside the unit-circle, which guarantees the system is stable [39]. This also indicates that the spatial reflection coefficients (Fresnel coefficients) in each interface are smaller than 1 in absolute value (positive refraction values). An alternative definition is that A~\tilde{A} is a minimum-phase system. Using Eqs. (38), (39) and (40), we obtain the dual realizability condition for temporal layers, which is that D^\hat{D} is a minimum-phase system. Equivalently, we may notice that D^(z^)\hat{D}(\hat{z}) is the reverse polynomial of A^(z^)\hat{A}(\hat{z}). This means that the vector representing the coefficients of A^(z^)\hat{A}(\hat{z}): [aN,…,a0][a_N,\ldots,a_0] is a flipped version of the vector representing the coefficients of D^(z^)\hat{D}(\hat{z}): [dN,…,d0][d_N,\ldots,d_0]. This indicates that an equivalent condition would be that A^(z^)\hat{A}(\hat{z}) is maximum-phase. Lastly, momentum conservation implies that ∀k\forall k [40]:

nNn0∣A^(k)∣2−nNn0∣C^(k)∣2=1(41)\frac{n_N}{n_0}|\hat{A}(k)|^2-\frac{n_N}{n_0}|\hat{C}(k)|^2=1 \tag*{(41)}

References

  1. [1]F. Morgenthaler, Velocity modulation of electromagnetic waves, IRE Trans. Micro. Theory Tech. 6, 167 (1958).DOI
  2. [2]E. Cassedy and A. Oliner, Dispersion relations in time-space periodic media: Part i—stable interactions, Proceedings of the IEEE 51, 1342 (1963).DOI
  3. [3]L. Felsen and G. Whitman, Wave propagation in time-varying media, IEEE Trans. Ant. Prop. 18, 242 (1970).DOI
  4. [4]R. Fante, Transmission of electromagnetic waves into time-varying media, IEEE Trans. Ant. Prop. 19, 417 (1971).DOI
  5. [5]Y. Yu, H. Hu, Q. Yang, L. Zou, D. Liu, H. C. Zhang, and Y. Luo, Fundamentals and applications of time-varying media: A review (2026), arXiv:2604.15629 [physics.optics].arxiv.org/abs/2604.15629
  6. [6]Y. Hadad, D. L. Sounas, and A. Alu, Space-time gradient metasurfaces, Phys. Rev. B 92, 100304(R) (2015).DOI
  7. [7]Y. Hadad, J. C. Soric, and A. Alu, Breaking temporal symmetries for emission and absorption, Proceedings of the National Academy of Sciences 113, 3471 (2016).DOI
  8. [8]E. Lustig, Y. Sharabi, and M. Segev, Topological aspects of photonic time crystals, Optica 5, 1390 (2018).
  9. [9]A. M. Shaltout, V. M. Shalaev, and M. L. Brongersma, Spatiotemporal light control with active metasurfaces, Science 364, eaat3100 (2019), https://www.science.org/doi/pdf/10.1126/science.aat3100.DOI
  10. [10]M. G. Ozlu, V. Mkhitaryan, C. B. Fruhling, A. Boltasseva, and V. M. Shalaev, Floquet quantum engineering of polaritonic amplification in dispersive photonic time crystals, Phys. Rev. Res. 7, 023214 (2025).arxiv.org/abs/2408.00552
  11. [11]J. F. Sustaeta-Osuna, T. F. Allard, F. J. García-Vidal, and P. A. Huidobro, Near-field gain and far-field control via a plasmonic time crystal slab, Phys. Rev. Lett. 136, 136903 (2026).
  12. [12]H. Lira, Z. Yu, S. Fan, and M. Lipson, Electrically driven nonreciprocity induced by interband photonic transition on a silicon chip, Phys. Rev. Lett. 109, 033901 (2012).pubmed.ncbi.nlm.nih.gov/22861851
  13. [13]N. A. Estep, D. L. Sounas, J. Soric, and A. Alù, Magnetic-free non-reciprocity and isolation based on parametrically modulated coupled-resonator loops, Nature Physics 10, 923 (2014).DOI
  14. [14]S. Taravati, N. Chamanara, and C. Caloz, Nonreciprocal electromagnetic scattering from a periodically space-time modulated slab and application to a quasisonic isolator, Phys. Rev. B 96, 165144 (2017).DOI
  15. [15]W. Jaffray, S. Stengel, A. Boltasseva, V. M. Shalaev, M. A. Vincenti, D. de Ceglia, M. Scalora, C. Rizza, and M. Ferrera, All-optical polarization control in time-varying low-index films via plasma symmetry breaking, Nature Photonics 20, 628 (2026).DOI
  16. [16]Y. Li, S. Vezzoli, T. Klee, J. J. Broughton, R. Tirole, A. C. Harwood, H. Allegre, S. A. R. Horsley, R. Sapienza, J. B. Pendry, and J. W. G. Tisch, Ultrafast switching of optical properties in a doped semiconductor by intense femtosecond laser pulses, Light: Science & Applications 15, 336 (2026).DOI
  17. [17]N. Reiskarimian and H. Krishnaswamy, Magnetic-free non-reciprocity based on staggered commutation, Nature Communications 7, 11217 (2016).DOI
  18. [18]E. Lustig, O. Segal, S. Saha, E. Bordo, S. N. Chowdhury, Y. Sharabi, A. Fleischer, A. Boltseva, O. Cohen, V. M. Shalaev, and M. Segev, Time-refraction optics with single cycle modulation, Nanophotonics 12, 2221 (2023), https://onlinelibrary.wiley.com/doi/pdf/10.1515/nanoph-2023-0126.DOI
  19. [19]R. Tirole, S. Vezzoli, E. Galiffi, I. Roberston, D. Maurice, B. Tilmann, S. A. Maier, J. B. Pendry, and R. Sapienza, Double-slit time diffraction at optical frequencies, Nature Physics 19, 999 (2023).DOI
  20. [20]A. Bruckstein and T. Kailath, Inverse scattering for discrete transmission-line models, Siam Review - SIAM REV 29 (1987).DOI
  21. [21]D. Colton and A. Kirsch, A simple method for solving inverse scattering problems in the resonance region, Inverse Problems 12, 383 (1996).DOI
  22. [22]P. M. van den Berg and R. E. Kleinman, A contrast source inversion method, Inverse Problems 13, 1607 (1997).DOI
  23. [23]M. Burger, S. Osher, and E. Yablonovitch, Inverse problem techniques for the design of photonic crystals, IEICE Transactions on Electronics E87C, 258 (2004).
  24. [24]S. Molesky, Z. Lin, A. Y. Piggott, W. Jin, J. Vucković, and A. W. Rodriguez, Inverse design in nanophotonics, Nature Photonics 12, 659 (2018).
  25. [25]D. Ramaccia, A. Alù, A. Toscano, and F. Bilotti, Temporal multilayer structures for designing higher-order transfer functions using time-varying metamaterials, Applied Physics Letters 118, 101901 (2021).DOI
  26. [26]G. Castaldi, V. Pacheco-Peña, M. Moccia, N. Engheta, and V. Galdi, Exploiting space-time duality in the synthesis of impedance transformers via temporal metamaterials, Nanophotonics 10, 3687 (2021), https://onlinelibrary.wiley.com/doi/pdf/10.1515/nanoph-2021-0231.DOI
  27. [27]O. Silbiger and Y. Hadad, Optimization-free filter and matched-filter design through spatial and temporal soft switching of the dielectric constant, Physical Review Applied 19, 014047 (2023).DOI
  28. [28]E. Galiffi, G. Xu, S. Yin, H. Moussa, Y. Ra’di, and A. Alù, Broadband coherent wave control through photonic collisions at time interfaces, Nature Physics 19, 1703 (2023).arxiv.org/abs/2212.02647
  29. [29]G. P. Agrawal, Space–time duality in optics: Its origin and applications, Photonics 12, 10.3390/photonics12060611 (2025).DOI
  30. [30]J. Xu, W. Mai, and D. H. Werner, Generalized temporal transfer matrix method: A systematic approach to solving electromagnetic wave scattering in temporally stratified structures, Nanophotonics 11, 1309 (2022).DOI
  31. [31]Y. Xiao, D. N. Maywar, and G. P. Agrawal, Reflection and transmission of electromagnetic waves at a temporal boundary, Opt. Lett. 39, 574 (2014).DOI
  32. [32]A. Shlivinski and Y. Hadad, Beyond the bode-fano bound: Wideband impedance matching for short pulses using temporal switching of transmission-line parameters, Phys. Rev. Lett. 121, 204301 (2018).DOI
  33. [33]J. L. Valdez-García and P. Halevi, Parametric resonances in a photonic time crystal with periodic square modulation of its permittivity ε(t), Phys. Rev. A 109, 063517 (2024).
  34. [34]V. Pacheco-Peña, M. Fink, and N. Engheta, Temporal chirp, temporal lensing, and temporal routing via space-time interfaces, Physical Review B 111, L100306 (2025).
  35. [35]S. Taravati, A. A. Kishk, and G. V. Eleftheriades, Finite-difference time-domain simulation of wave transmission through space-time-varying media, arXiv preprint arXiv:2409.19923 (2024).arxiv.org/abs/2409.19923
  36. [36]O. Silbiger and Y. Hadad, Inverse design of time-varying transmission-line filters for broadband spatial signal processing, Phys. Rev. B 111, 054307 (2025).DOI
  37. [37]G. Castaldi, M. Coppolaro, M. Moccia, C. Rizza, N. Engheta, and V. Galdi, Analytic inverse design of temporal metamaterials via space-time duality, Physical Review Letters 136, 206901 (2026).DOI
  38. [38]P. L. Goupillaud, An approach to inverse filtering of near-surface layer effects from seismic records, Geophysics 26, 754 (1961), https://pubs.geoscienceworld.org/seg/geophysics/article-pdf/26/6/754/3152780/754.pdf.DOI
  39. [39]S. J. Orfanidis, Electromagnetic waves and antennas (El autor, 2016).
  40. [40]S. Yin, E. Galiffi, G. Xu, and A. Alù, Scattering at temporal interfaces: An overview from an antennas and propagation engineering perspective, IEEE Antennas and Propagation Magazine 65, 21 (2023).DOI

Paper details

vlnc:W_qhpkxv4p
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents