---
title: Phase-Conjugate Circuits Overview
url: https://www.emergentmind.com/topics/phase-conjugate-circuits
type: topic
---

# Phase-Conjugate Circuits Overview

Searching arXiv for the cited papers and closely related work on phase-conjugate circuits.
arXiv search query: "phase conjugation nonlinear active metamaterials 1008.1999"
Phase-conjugate circuits are electromagnetic or optical circuit configurations that generate a field proportional to the complex conjugate of an incident field, commonly written as $E_{\text{out}}(x,y)=E_{\text{in}}^*(x,y)$. In this operation, the returned or transmitted wavefront retraces aberrated propagation paths, reverses phase distortions, and can realize time-reversal, retrodirectivity, or negative refraction depending on geometry and frequency conversion. The term encompasses several distinct implementations: varactor-loaded split-ring resonator arrays driven by parametric pumping, Michelson interferometric beam-combining networks terminated by a phase-conjugate mirror, passive diffractive processors trained to approximate optical phase conjugation, and four-wave-mixing resonators in rubidium vapor that alternate between probe and conjugate colors [1008.1999], [1311.6703], [2311.04473], [2406.11071].

## 1. Fundamental operation and circuit concept

The core operation of a phase-conjugate circuit is the inversion of the optical or electromagnetic phase while preserving the amplitude structure sufficiently for the output to retrace the original propagation. In the diffractive formulation, the target operation is stated directly as
$$
E_{\text{in}}(x,y)\Rightarrow E_{\text{out}}(x,y)=E_{\text{in}}^*(x,y),
$$
so that the conjugate wavefront undoes aberrations on a subsequent pass through the same distortion [2311.04473]. In the four-wave-mixing formulation, the boundary condition of a phase-conjugate mirror is written as
$$
E_{\text{out}}(x,y,\omega)=R_{PC}(\omega)\,E_{\text{in}}^*(x,y,\omega),
$$
which differs from the ordinary cavity condition $E_{\text{out}}=rE_{\text{in}}$ because the field is phase-reversed rather than simply reflected [2406.11071].

In RF metamaterial implementations, the same principle is realized through parametric frequency conversion rather than a conventional mirror. A split-ring resonator of inductance $L$ and capacitance $C$ has resonance $\Omega_0=1/\sqrt{LC}$, and varactor modulation at $2\omega_0$ produces a conjugate component at $\omega_2=2\omega_0-\omega_1$ when the element is illuminated at $\omega_1\approx\omega_0$. The reradiated component at $\omega_2$ has reversed phase in the transverse coordinate, which yields retrodirectivity or negative refraction depending on the observation geometry [1008.1999].

Across these platforms, phase conjugation is therefore not a single hardware topology but a functional class of circuits. Exact or nearly exact conjugation appears in nonlinear mixing regimes such as degenerate or non-degenerate four-wave mixing, whereas passive diffractive processors approximate the same operator over a trained distribution of input fields. This distinction is central to interpreting reported performance and scope [1311.6703], [2311.04473].

## 2. Nonlinear active metamaterial elements and RF phase-conjugate arrays

A discrete RF phase-conjugate circuit can be built from nonlinear active metamaterial elements based on varactor-loaded split-ring resonators. In the reported configuration, each element is a copper SRR on FR-4 designed to resonate at $\Omega_0\approx2\pi\cdot950\,\text{MHz}$ when loaded with two anti-series varactor diodes. A strong RF pump at $2\omega_0$ is applied directly to the SRR gap through a small bias-tee or high-impedance feed, and isolation resistors $R\approx10\,\text{k}\Omega$ prevent pump leakage into neighboring elements or measurement equipment. No additional mixers, PLLs, or active semiconductors are required; the nonlinearity of the varactors plus the strong pump generates the conjugate response [1008.1999].

The lumped-element model treats each SRR as a parametrically driven dipolar oscillator:
$$
\frac{d^2p}{dt^2}+\gamma\frac{dp}{dt}+\Omega_0^2[1+h_0\sin(2\omega_0 t)]p
=
\Omega_0^2 a^3 E_{\text{inc}}(t).
$$
With the ansatz
$$
p(t)=a_1 e^{-i\omega_1 t}+b_2 e^{+i\omega_2 t}+\text{c.c.},\qquad \omega_2=2\omega_0-\omega_1,
$$
the coefficient $b_2$ describes the phase-conjugate component. For the $j$th element at $y=y_j$,
$$
a_1^{(j)} = \frac{\Omega_0^2-\omega_2^2+2i\gamma\omega_2}{D(\omega_1)}\,e^{ik_y y_j},
\qquad
b_2^{(j)} = \frac{i h_0 \Omega_0^2}{2D(\omega_1)}\,e^{ik_y y_j},
$$
with
$$
D(\omega_1)=
[\Omega_0^2-\omega_1^2-2i\gamma\omega_1]
[\Omega_0^2-\omega_2^2+2i\gamma\omega_2]
-\frac{h_0^2\Omega_0^4}{4}.
$$
The term proportional to $b_2$ reradiates at $\omega_2$ with reversed phase in $y$, and the array therefore behaves as a time-reversal surface [1008.1999].

When $N$ such SRRs are placed in a plane $z=0$, pumped in phase, and spaced by $\Delta y<\lambda/2$, the array acts as a discrete thin slab of time-reversal medium. Theory predicts $\theta_{\text{out}}=-\theta_{\text{inc}}$ because the transverse wavevector changes sign. The reported three-element experiment used spacing $\lambda/4$, total array width $\lambda/2$, and a slab thickness $t\approx0.12\lambda$. A monopole source at $950\,\text{MHz}$ placed at $z=-\lambda$ generated a phase-conjugate beam at $900\,\text{MHz}$ that refracted to the “wrong” side of the normal, in agreement with a simple $r^{-1}$ dipole-pattern model of three discrete radiators [1008.1999].

The single-element spectrum test used $f_{\text{pump}}=1850\,\text{MHz}$, $f_{\text{src}}=950\,\text{MHz}$, and observed $f_{pc}=900\,\text{MHz}=f_{\text{pump}}-f_{\text{src}}$. The maximum source power at $950\,\text{MHz}$ was $-25.2\,\text{dBm}$ and the maximum phase-conjugate power at $900\,\text{MHz}$ was $-49.6\,\text{dBm}$, corresponding to conversion approximately $24.4\,\text{dB}$ down. In a two-element interference test, excitation phases of $0^\circ$ and $-90^\circ$ at $950\,\text{MHz}$ shifted the fringe in one direction, while the phase-conjugate field at $900\,\text{MHz}$ shifted the fringe to the opposite side, presented as direct proof of phase conjugation [1008.1999].

The same report identifies the main constraints of this circuit family. Operational bandwidth is set by the resonance linewidth $\gamma$ and the parametric stability condition $h_0\omega_0<2\gamma$; in practice, approximately $\pm10\,\text{MHz}$ around $950\,\text{MHz}$ was used without noticeable drop in phase-conjugate amplitude. Excessive $h_0$ leads to self-oscillation, and higher-frequency scaling requires ultrafast modulators such as optical varactors or carrier-driven resonances [1008.1999].

## 3. Michelson phase-conjugating networks for coherent fiber-array recombination

A second major class of phase-conjugate circuits is the Michelson phase-conjugating configuration for coherent beam combining in large fiber-amplifier arrays. In this layout, a master oscillator produces a short transform-limited pulse that is stretched and chirped, then divided by a binary beamsplitter tree into $N_f=2^{N_{ex}}$ identical sub-pulses. Each sub-pulse is injected into one of $N_f$ single-mode fiber amplifiers of length $L_{mn}$, each accumulating an unknown phase-piston $\Delta\phi_{mn}$. The amplified outputs propagate into a common pupil and illuminate a degenerate four-wave-mixing phase-conjugate mirror, where counter-propagating pumps $E_1$ and $E_2$ record the instantaneous speckled field in a $\chi^{(3)}$ hologram. The mirror returns a conjugated replica $E_b$ proportional to the complex conjugate of the forward field $E_f$, and the same beamsplitter tree recombines the backward pulse into a smooth collimated output. A time-dependent half-wave Pockels cell imposes a $\pi$ phase shift on the returning beam to decouple it from the master oscillator [1311.6703].

The forward field exiting the array is written as
$$
\bar{E}_f(z,t,\mathbf r)=
e^{-i\omega t+i k_z z+i\theta(\tau)} f(\tau)
\sum_{mn} E^o_{mn}
\exp\!\left[-\frac{(\mathbf r-\mathbf r_{mn})^2}{2D^2}+i\Delta\phi_{mn}\right],
$$
with $\tau=t-z/v_g$. Inside a thin DFWM mirror the total field $E^{PC}=E_1+E_2+E_f+E_b$ obeys
$$
i\frac{\partial E^{PC}}{\partial z}
+\frac{1}{2k}\nabla_\perp^2 E^{PC}
+\frac{\gamma_{PC}}{2}|E^{PC}|^2E^{PC}=0,
\qquad
\gamma_{PC}=k\cdot\frac{3\chi^{(3)}}{8\pi}.
$$
In the undepleted-pump, small-signal regime,
$$
\frac{\partial E_b}{\partial z}
=
i\frac{\gamma_{PC}}{2}E_1E_2 f^2(\tau) E_f^*,
$$
so the returning field contains $\Delta\phi_{mn}\rightarrow-\Delta\phi_{mn}$. After double passage through the same fiber, the net phase shift becomes $\Delta\phi+(-\Delta\phi)=0$, and the circuit therefore performs automatic compensation of slow thermal or stress-induced path-length errors without active multi-channel phase control [1311.6703].

The mirror performance is quantified by the instantaneous reflectivity
$$
R_{PCM}(\tau)=
\frac{|E_b(0,\tau)|^2}{|E_f(0,\tau)|^2}
=
f^4(\tau)L_{PCM}^2\gamma_{PC}^2|E_1E_2|,
$$
and by the fidelity
$$
K_{PCM}=
\frac{\left|\iint E_f E_b^*\,dx\,dy\right|^2}
{\iint |E_f|^2\,dx\,dy\;\iint |E_b|^2\,dx\,dy}
\le 1.
$$
The design parameters listed for this architecture include $N_f=2^{12\ldots15}\approx4\,\text{k}\ldots32\,\text{k}$ elements, core diameter $\lesssim50\,\mu\text{m}$, mode diameter $2D\approx100\,\mu\text{m}$, pitch $p\approx D$, fiber length $L_f=1\ldots10\,\text{m}$, per-fiber small-signal gain $G\approx2\ldots10$, and output energy $\lesssim2$–$3\,\text{mJ}$ at repetition rate $1$–$10\,\text{kHz}$. The DFWM slice is described as a $\chi^{(3)}$ medium such as BaTiO$_3$ or photorefractive glass with thickness $L_{PCM}\sim0.1\ldots1\,\text{mm}$, undepleted-pump reflectivity $R_{PCM}\lesssim0.2$, pump pulse energies $E_1,E_2\sim0.1$–$0.5\,\text{J}$, and fidelity $K_{PCM}\gtrsim0.9$ [1311.6703].

A notable limitation is that phase-piston cancellation does not remove all pulse distortions. Backward amplification in each Yb-doped fiber follows an incoherent Frantz–Nodvik model, and under deep saturation the Gaussian envelope is reported to be strongly distorted, with its chirp position shifted by more than $1\,\text{ns}$, whereas the $sech$ envelope remains more symmetric, shifts only a few hundred ps, and preserves a nearly linear chirp across the main lobe. The far-field interference figure of merit therefore becomes limited by residual chirp mismatch rather than by the compensated static phase pistons. In the large-$N_f$ limit, the global visibility is reported to be approximately set by the two-beam residual chirp mismatch, $FOM_{\text{array}}\approx FOM_{12}$ [1311.6703].

## 4. Passive diffractive phase-conjugate processors

A more recent circuit realization replaces nonlinear mixing with a passive, spatially engineered diffractive volume optimized by deep learning to approximate optical phase conjugation. Under scalar-wave theory, propagation between planes is modeled with an angular-spectrum transfer function
$$
U(x,y;z+\Delta z)=
\mathcal{F}^{-1}
\left\{
\mathcal{F}[U(x,y;z)]\cdot H(f_x,f_y;\Delta z)
\right\},
$$
where
$$
H(f_x,f_y;\Delta z)=
\exp\!\left[i\,2\pi\,\Delta z\sqrt{\frac{1}{\lambda^2}-f_x^2-f_y^2}\right],
$$
or, under the Fresnel approximation,
$$
H(f_x,f_y;\Delta z)\approx
\exp[-i\pi\lambda\Delta z(f_x^2+f_y^2)].
$$
Each thin layer is a phase mask,
$$
t_n(x,y)=\exp[i\phi_n(x,y)],
\qquad
\phi_n(x,y)=\frac{2\pi(n-n_{\text{air}})h_n(x,y)}{\lambda}.
$$
The composite transmissive operator is
$$
U_{\text{out}}=P_{\Delta z}\,t_K\cdots P_{\Delta z}\,t_2\,P_{\Delta z}\,t_1\,U_{\text{in}}.
$$
A reflective phase-conjugate mirror variant is realized by placing a flat mirror after the diffractive stack so that the field double-passes the same layers [2311.04473].

The numerical transmissive processor used $K=8$ layers, inter-layer spacing $\Delta z=12\lambda$, an aperture approximately $106\times106\,\text{mm}^2$ discretized into $200\times200$ pixels with lateral pitch approximately $\lambda/2$, refractive index $n\approx1.7$ with $k\approx0$, and maximum learnable thickness $h_{\max}\approx1.07\,\text{mm}$ at $\lambda=0.75\,\text{mm}$. The experimental terahertz proof-of-concept used $K=3$ layers, $\Delta z=20\lambda$ between successive layers, total axial span approximately $80\lambda$, input and output apertures of $28\times28$ pixels over approximately $30\times30\,\text{mm}^2$, and layer features $120\times120$ fabricated in dielectric material with $n\approx1.7$ and $k\approx0.017$ using an Objet30 Pro 3D printer [2311.04473].

Training employed $200\,000$ pairs of complex fields $\{E_{\text{in}},E_{\text{target}}=E_{\text{in}}^*\}$, with phase profiles generated from random combinations of Zernike polynomials and training phase range $\Delta\phi_{\text{train}}\in[0.2\pi,1.0\pi]$. Optimization used back-propagation and Adam with learning-rate decay, typically for $100$ epochs on an NVIDIA RTX 3090 GPU, reported as approximately $20\,\text{h}$ for $K=8$. The loss function was a weighted sum of amplitude and phase mean-square error with an optional diffraction-efficiency penalty,
$$
\mathcal{L}
=
\alpha\,MSE[|U_K|,|E^*|]
+
\beta\,MSE[\arg U_K,\arg E^*]
-
\gamma\,\eta_{\text{out}},
$$
where
$$
\eta_{\text{out}}=
\frac{\iint|U_K|^2\,dxdy}{\iint|E_{\text{in}}|^2\,dxdy}.
$$
This formulation makes explicit that the diffractive device is trained to approximate conjugation over a distribution of inputs rather than deriving the operation from a nonlinear medium [2311.04473].

The reported performance is correspondingly statistical. In blind testing of the $K=8$ design with $\Delta\phi_{\text{test}}=\pi$, $10\,000$ unseen two-Zernike fields yielded phase MAE $=1.38\pm0.12\%$ and amplitude MAE $=8.9\pm1.9\%$; single-Zernike tests over $28$ modes gave phase MAE $=1.71\pm0.14\%$ and amplitude MAE $=13.1\pm2.3\%$; three-Zernike tests gave phase MAE $=1.09\pm0.06\%$ and amplitude MAE $=8.4\pm1.3\%$. Over $\Delta\phi_{\text{test}}\in[0.2\pi,1.99\pi]$, phase MAE remained below $2\%$ for $\Delta\phi\le\pi$ and rose to approximately $6.8\%$ at $1.99\pi$. In the terahertz experiment with $K=3$ and $\Delta\phi=0.5\pi$, an aberrated spherical wave passed through the diffractive OPC processor, then through the identical aberrator, and recovered a tight focal spot of $4\,\text{mm}$ measured by a single-pixel THz detector; without OPC, the output was a scattered pattern [2311.04473].

This circuit family occupies a distinct position relative to conventional analog OPC and digital OPC. The comparison given in the source states that photorefractive or four-wave-mixing crystals have intrinsically low reflectivity, listed as less than $1\%$, while digital OPC is bulky and slow on the millisecond scale because it requires an interferometer, SLM, camera, and digital computation. The diffractive processor is described instead as passive, real-time, compact, monolithic, polarization-insensitive with isotropic dielectrics, and scalable across the spectrum by preserving the proportionality of lateral pitch and layer spacing to wavelength [2311.04473].

## 5. Phase-conjugate resonators, four-wave mixing, and color-switching oscillation

Phase-conjugate circuitry can also define a resonator whose boundary conditions are created by the phase-conjugating element itself. In the reported optical parametric oscillator, a phase-conjugate resonator is formed from non-degenerate four-wave mixing in hot $^{85}$Rb vapor. Two counter-propagating pump beams at frequency $\omega_p$ are derived from a single CW Ti:Sapphire laser of approximately $500\,\text{mW}$ and $1\,\text{mm}$ diameter, detuned by approximately $+800\,\text{MHz}$ to the blue of the $5S_{1/2}\rightarrow5P_{1/2}$ transition at $795\,\text{nm}$. The pumps traverse a $12\,\text{mm}$ vapor cell at approximately $100^\circ\text{C}$ and atomic density approximately $10^{13}\,\text{cm}^{-3}$. A weak probe $A_s$ detuned by approximately $-3\,\text{GHz}$ from $\omega_p$ crosses the pumps at approximately $0.5^\circ$, is amplified, and generates a phase-conjugate beam $A_c$ at frequency $\omega_c=\omega_p+3\,\text{GHz}$ satisfying $2k_p\approx k_s+k_c$. The output port is fed back to the input port through a single-mode polarization-maintaining fiber loop of length from several meters to $105\,\text{m}$, so that the PCM output closes the resonator [2406.11071].

In the slowly varying envelope approximation, the intracell probe and conjugate satisfy
$$
\frac{dA_s}{dz}=i\kappa A_p^2 A_c^* e^{i\Delta k z},
\qquad
\frac{dA_c}{dz}=i\kappa A_p^2 A_s^* e^{i\Delta k z},
$$
with $\Delta k=2k_p-k_s-k_c$. Under perfect phase matching and constant pumps,
$$
A_c(L)\simeq i\sinh(GL)A_s^*(0),
\qquad
A_s(L)\simeq \cosh(GL)A_s(0),
$$
where $G=\kappa|A_p|^2$. For a single pass of length $L_{\text{cell}}=12\,\text{mm}$, a typical small-signal gain of approximately $10$ is reported. The phase-conjugate boundary condition is then
$$
E_{\text{out}}(x,y,\omega)=R_{PC}(\omega)E_{\text{in}}^*(x,y,\omega),
$$
with $R_{PC}\approx1$ on resonance. Because a mode must pass once as a probe and once as a conjugate to reproduce itself, the resonator requires two successive passes before closure, and the longitudinal mode spacing is
$$
\delta f=\frac{c}{2(n_fL_f+L_{\text{free}})}.
$$
This is the basis of the statement that no empty-cavity modes exist without the gain of the phase-conjugate mirror [2406.11071].

Oscillation occurs when round-trip gain compensates loss:
$$
|\Gamma|^2=\left|e^{gL_{\text{cell}}-\alpha L_{\text{total}}}\right|^2=1,
$$
so the threshold condition is $gL_{\text{cell}}=\alpha L_{\text{total}}$. The source gives the corresponding pump intensity as
$$
I_p\approx\frac{500\,\text{mW}}{\pi(0.5\,\text{mm})^2}\approx6\times10^4\,\text{W/m}^2
$$
for vapor at $100^\circ\text{C}$. Passive mode-locking is attributed to the combination of narrow gain bandwidth and mode coupling. The natural linewidth is approximately $10\,\text{MHz}$, much smaller than the inverse round-trip time, given as approximately $2\,\text{MHz}$ for $L_f=105\,\text{m}$, so many longitudinal modes fit under the gain envelope. Two-beam coupling mixes adjacent probe modes within approximately $6\,\text{MHz}$ of each other, and a discrete-time model,
$$
A_s^{n+1}(t)=G\,A_c^{n*}(t-T)e^{-\alpha/2},
\qquad
A_c^{n+1}(t)=G\,A_s^{n*}(t-T)e^{-\alpha/2},
$$
produces alternation between probe and conjugate colors on successive passes. In the time domain, this appears as square-like pulses alternating color every $T$ [2406.11071].

The reported measured parameters include fiber loop lengths from $3\,\text{m}$ to $105\,\text{m}$, repetition rate from approximately $1\,\text{MHz}$ at $105\,\text{m}$ to greater than $50\,\text{MHz}$ at $3\,\text{m}$, probe-conjugate frequency separation $\omega_c-\omega_s\approx2\pi\times6\,\text{GHz}$, pulse widths of $0.7$–$1.0\,T$, and approximately $0.5\,\mu\text{s}$ pulse width for $105\,\text{m}$ fiber. The small-signal gain bandwidth is approximately $20\,\text{MHz}$ at $3\,\text{dB}$, and the conversion efficiency to OPO output is approximately $6\%$, stated as $30\,\text{mW}$ out at threshold pump power. The resonator is also reported to remove thermal or acoustic instabilities on a MHz or slower timescale because any slow phase drift $\phi$ is canceled after two passes as $\phi-\phi=0$ [2406.11071].

## 6. Comparative scope, applications, and recurrent misconceptions

The applications explicitly identified for phase-conjugate circuits span thin-slab negative-refraction RF and imaging lenses, retrodirective arrays for communications and radar, passive subwavelength imaging via time-reversal focusing, wave-front correction in complex propagation environments, coherent chirped-pulse beam combining, turbidity suppression, aberration correction, and quantum-network or squeezed-light platforms [1008.1999], [1311.6703], [2311.04473], [2406.11071]. The common mechanism across these otherwise different systems is the use of phase reversal to retrace propagation and suppress accumulated distortions, but the implementation-level trade-offs differ substantially.

One recurrent misconception is that phase conjugation is equivalent to ordinary reflection. The rubidium resonator source states the distinction directly: a conventional Fabry–Pérot condition takes the form $E_{\text{out}}=rE_{\text{in}}$, whereas a phase-conjugate boundary returns $E_{\text{in}}^*$ and may require two successive passes before the mode closes on itself [2406.11071]. A second misconception is that phase-conjugate compensation removes every form of distortion. The Michelson fiber-array analysis shows that piston errors $\Delta\phi_{mn}$ are canceled, but gain saturation can still deform the chirped envelope and reduce interference visibility, particularly for Gaussian input pulses [1311.6703]. A third misconception is that all phase-conjugate devices are necessarily nonlinear in operation. The diffractive processor is passive in deployment and approximates the conjugation operator through a trained stack of phase masks rather than through in situ nonlinear mixing [2311.04473].

The limitations listed in the sources are likewise implementation-specific. In varactor-loaded SRR arrays, conversion is limited by varactor $Q$ and coupling, bandwidth is set by resonator linewidth and the condition $h_0\omega_0<2\gamma$, and excessive modulation depth causes parametric instability [1008.1999]. In the Michelson-PCM architecture, dispersion matching and nonlinear phase management are required, with the total $B$-integral to be kept below $3$–$5\,\text{rad}$ to avoid soliton breakup [1311.6703]. In diffractive OPC, output diffraction efficiency and phase accuracy are coupled through the training objective; the source reports $\eta_{\text{out}}\approx14\%\rightarrow24\%$ for $K=4,\ldots,10$ with phase MAE decreasing from approximately $2.37\%$ to approximately $1.55\%$, while an efficiency penalty can raise $\eta_{\text{out}}$ to $85$–$92\%$ at the cost of modest MAE increase [2311.04473]. In the rubidium resonator, oscillation depends on balancing 4WM gain against fiber coupling losses, etalon loss, splice loss, and atomic absorption, while the available gain bandwidth and mode coupling determine the passive mode-locking behavior [2406.11071].

Taken together, these reports suggest a technically useful classification of phase-conjugate circuits into three operational regimes. One regime is nonlinear frequency-converting circuitry, exemplified by SRR arrays and 4WM resonators. A second is self-adjusting interferometric circuitry, exemplified by the Michelson fiber network terminated by a DFWM mirror. A third is passive operator-approximation circuitry, exemplified by the diffractive wavefront processor. This suggests that “phase-conjugate circuit” is best understood not as a single device category but as a family of architectures whose unifying feature is engineered complex-conjugation of a propagated field under specific bandwidth, fidelity, and stability constraints [1008.1999], [1311.6703], [2311.04473], [2406.11071].

Source: https://www.emergentmind.com/topics/phase-conjugate-circuits