Real-time shaping of entangled photons by classical

control and feedback

arXiv:1902.06653v2 [quant-ph] 27 Feb 2020

Ohad Lib, Giora Hasson, Yaron Bromberg∗

Racah Institute of Physics, The Hebrew University of Jerusalem, Jerusalem, 91904 Israel

∗
To whom correspondence should be addressed; E-mail: [email protected].

Abstract
Quantum technologies hold great promise for revolutionizing photonic applications
such as cryptography. Yet their implementation in real-world scenarios is held back, mostly
due to sensitivity of quantum light to scattering. Recent developments in optimizing the
shape of single photons introduce new ways to control quantum light. Nevertheless, shap-
ing single photons in real-time remains a challenge due to the weak associated signals,
which are too noisy for optimization processes. Here, we overcome this challenge and con-
trol scattering of entangled photons by shaping the classical laser beam that stimulates their
creation. We discover that since the classical beam and the entangled photons follow the
same path, the strong classical signal can be used for optimizing the weak quantum signal.
We show that this approach can increase the length of free-space turbulent quantum links
by up to two orders of magnitude, opening the door for employing wavefront shaping for
quantum communications.

Introduction
Nearly a century after quantum mechanics revolutionized the way we understand nature, quan-
tum resources such as superposition and entanglement are beginning to enter and transform
technology [1, 2]. Many implementations of quantum technologies such as quantum communi-
cation [3–5] and quantum imaging [6–8], are based on photonic platforms that encode quantum
bits (qubits) using single photons. One of the main challenges in such applications is the low
flux of single photons that can be sent per communication channel or image pixel, resulting in
extremely low capacities of these systems. A promising approach for boosting the capacity of
quantum systems is to encode multi-level quantum bits (coined qudits) using a single photon in
a d-dimensional Hilbert space [9–11]. To this end, photonic qudits have been implemented in

1
the temporal [12], spectral [13] and spatial [14,15] domains. The spatial domain is in particular
attractive, since using spatial light modulators (SLMs) it is possible to arbitrary rotate qudits in
the d-dimensional space, simply by shaping the spatial distribution of photons [16]. However,
this also implies that the information carried by spatial qudits is extremely sensitive to scattering
and aberrations, acting as random rotations on the qudits. For example, scattering of entangled
photons that encode spatial qudits, scrambles their unique quantum correlations, resulting in
a random grainy spatial correlation pattern, coined two-photon speckle [17, 18]. Scattering of
photonic qudits is therefore a limiting factor in implementing photonic quantum technologies
in real-world applications, such as ground-satellite quantum communication [3–5] or quantum
imaging of biological samples [19]. In the past few years, extensive research was devoted for
protecting the information carried by spatial qudits, by encoding them in spatial modes that are
immune to scattering and aberrations [20–23]. Nevertheless, for significant scattering all spatial
modes will eventually suffer from scrambling of the information carried by the photons [24,25].
A promising approach for cancelling scattering of classical light is wavefront shaping. Over
a decade after the pioneering work of Vellekoop and Mosk [26], who focused classical light
through scattering media using an SLM, a remarkable set of tools for controlling light in random
media has been developed [27–29]. It is therefore appealing to adopt these tools to the quantum
regime. Over the past few years several important developments towards spatial control of
quantum light have been reported [30–34]. Nevertheless, for practical applications such as
quantum technologies, real-time optimization and feedback must be performed, which is in
particular challenging due to the inherently weak quantum signals. In most demonstrations to
date, photon-pairs generated by spontaneous parametric down conversion (SPDC) were sent to
an SLM that directly shaped their spatial distribution before hitting a scattering sample, and the
optimization was performed in advance using an auxiliary classical laser at the same wavelength
as the entangled photons [30–33]. In a few other demonstrations, the SLM modulated the
bright laser beam that stimulates the SPDC process (coined pump beam), and feedback was
provided by the inherently weak quantum signal [34,35]. Thus, although the significant progress
made, all demonstrations to date do not overcome the fundamental challenge of providing fast
feedback in real-time, and thus cannot be implemented in real-world applications.
In this work, we demonstrate a novel method for compensating the scattering of entangled
photons which allows, for the first time, real-time optimization of quantum correlations (Fig.1a).
Our method consists of two main components: shaping and feedback. First, instead of shaping
the entangled photons directly, we use the well known technique of pump shaping [34–36] to
control the correlations of the entangled photons, thus avoiding additional loss to the quantum
link. Second and most importantly, we utilize for the first time the intensity of the classical pump
beam for the feedback on the entangled photons. Since we use the classical pump beam both
for control and feedback, the well-established toolbox of classical wavefront shaping can be
trivially extended to the quantum regime, opening the door for implementing wavefront shaping
in quantum technologies. We explain this striking result by showing, both theoretically and
experimentally, that when the pump beam scatters by the same random sample as the entangled
photons, the spatial distribution of its intensity is identical to the spatial correlations of the

2
entangled photons. Hence, any manipulation of the classical pump intensity has an identical
effect on the entangled photons correlations. Since wavefront shaping of a classical bright
beam is much faster and more efficient than shaping weak fluxes of entangled photons, we
were able, for the first time, to demonstrate real-time wavefront correction of entangled photon
scattered by a dynamically moving diffuser. Finally, we show using numerical simulations that
our method allows significant improvement of free-space quantum links by compensating the
effect of turbulence on the entangled photons.

Results
To explain why scattering of two entangled photons corresponds to scattering of a single pump
photon at half the wavelength, we write the quantum state of the two photons (coined signal
and idler photons), in terms of their transverse wavevector components qs and qi ,
|ψi = dqs dqi ψ(qs , qi )a† (qs )a† (qi )|0i. Here a† (q) is the creation operator of a photon with
R

a transverse momentum q, |0i is the vacuum state, and we assume the signal and idler pho-
tons have the same frequency and polarization. The two-photon amplitude ψ(qs , qi ) can be
expressed in terms of the angular spectrum of the pump beam v(q) [36],

ψ(qs , qi ) = v (qs + qi ) Φ (qs − qi ) , (1)

L 2
where Φ(q) ∝ Sinc( 4k q ) is the phase matching function of the SPDC crystal, L is the crystal
length and k is the pump wavenumber inside the crystal. The number of inseparable modes in
the superposition of the two-photon state, quantified by the Schmidt number K, is proportional
to the ratio between the width of the phase matching function Φ(q) (determined by the crystal
length L), and the width of the pump angular spectrum function v(q) (determined by the width
of the pump beam). In the so called thin-crystal regime, K  1, the two-photon state can
be approximated by ψ(qs , qi ) ∝ v (qs + qi ) [36], yielding an Einstein-Podolsky-Rosen (EPR)
entangled state [37]. In this regime, we can precisely control the two-photon amplitude, by tai-
loring the angular spectrum of the pump beam or equivalently, by controlling its spatial profile,
W (ρ), at the input plane of the crystal.
We start by considering a thin diffuser placed right afterR the crystal, that can be modeled by
a linear transformation on the creation operator a† (q) → dρa† (ρ)Ad (ρ) exp(iρ · q), where
Ad (ρ) is the amplitude transfer function of the thin diffuser and ρ is the transverse spatial
coordinate. We note that if the diffuser is not located right after the crystal, it can always be
re-imaged on the crystal plane, as in conjugate-plane adaptive optics [38]. Experimentally, the
two-photon quantum state is measured using two single photon detectors placed at the far-field
of the crystal. The rate of coincidence events, i.e. detection of two photons simultaneously, one
photon with transverse wavevector qs and the other with transverse wavevector qi , is given by
C(qs , qi ) = |h0|a(qi )a(qs )|ψi|2 , yielding (see Supplementary) [39],
Z 2
dρW (ρ)A2d (ρ) exp(−iρ
 
C(qs , qi ) ∝ 
 · (qs + qi )) (2)

3
For random diffusers, the spatial distribution of the coincidence pattern C(qs , qi ) exhibits a
random two-photon speckle pattern [17,18]. In the absence of loss, the right hand side of Eq. (2)
corresponds to the far-field intensity profile of the pump beam in the direction q = qi + qs (see
Supplementary). Hence, by measuring the intensity profile of the pump beam, we get the spatial
distribution of the two-photon state. Since this is true for any pump profile W (ρ), using an SLM
we can optimize the pump profile to get a focused pump spot at the far-field of the diffuser, and
the quantum correlations between the photons will become spatially localized, simultaneously.
In case the diffuser is lossy, the pump intensity and the coincidence pattern may not be identical.
Nevertheless, optimizing the pump profile will still localize the two-photon spatial correlations
(See Supplementary). The fact that pump-optimization simultaneously localizes the two-photon
correlations is a remarkable feature of spatially entangled photons. It offers orders of magnitude
faster feedback than would have been possible with the inherently weak signal provided by the
coincidence rate, allowing us to extend wavefront shaping to the quantum domain, by applying
classical wavefront shaping to the bright pump beam.
The quantum wavefront shaping experimental setup is depicted in Fig.1a. A 2mm long
PPKTP crystal is pumped by a λ = 404nm continuous-wave laser, which is shaped by a phase-
only SLM imaged on the input facet of the crystal. The SPDC process in the crystal generates a
continuous flux of entangled photon-pairs, with a Schmidt number of K ≈ 680 (Fig.1b,c, Sup-
plementary). After the crystal, both the pump beam and the entangled photons are scattered by a
thin diffuser located at the image plane of the crystal, creating a fully-developed speckle pattern
with no ballistic component. The pump intensity and the two-photon coincidence patterns are
measured at the far-field of the crystal, after separating the pump and entangled photons using a
dichroic mirror. The coincidence patterns at the far-field are measured using two single photon
detectors, where one detector is scanning while the other is always kept stationary.

4
Figure 1: Experimental setup. (a) Spatially entangled photons are created by pumping a
nonlinear crystal (PPKTP) with a λ = 404nm continues-wave laser. Both the pump beam and
the entangled photons pass through a thin diffuser, which is imaged on the crystal and SLM
planes (by lenses L1 − L4), and measured at the far-field (lens L5). An optimization method is
employed for compensating the scattering of the pump beam using the SLM. (b) The far-field
single counts, in the absence of the diffuser, is much wider than the coincidence distribution (c),
indicating high spatial entanglement.

Figures 2a and 2b depict the measured pump intensity and two-photon coincidence rate, at
the far-field of the diffuser. Even though the wavelength of the entangled photons is twice the
wavelength of the pump photons, the two signals exhibit strikingly similar patterns, as predicted
by Eq. (2). The wavelength difference comes into play only in the scaling of the two patterns,
as the two-photon pattern is stretched by a factor of two compared to the pump pattern (see
Supplementary). Since the pump and two-photon speckle patterns are remarkably similar, we
can apply wavefront shaping optimization to the classical pump beam, and the quantum two-
photon correlations will be optimized simultaneously. Specifically, we use the partitioning
optimization algorithm to enhance the intensity of the pump beam at an arbitrary point at the far-
field of the diffuser [29] (Fig.2d). With the exact same phase mask applied to the pump beam,
the two-photon coincidence pattern is measured, showing a clear enhancement and localization
of the two-photon correlations at the target area (Fig.2e). Interestingly, the single photon counts
(Fig.2c,f) are not affected by the scattering nor by the optimization, due to the multimode nature
of SPDC light in the high Schmidt number regime [18].

5
Figure 2: Quantum wavefront shaping. The pump beam and the entangled photons pass
through the same diffuser, forming similar speckle patterns in the intensity (a) and coincidence
(b) pictures, respectively. By performing classical wavefront shaping on the bright pump beam,
a single speckle grain is enhanced (d), yielding a simultaneous enhancement and localization of
the quantum correlations at the corresponding location (e). The single photon counts (c,f) are
not affected by either the scattering or the optimization.

One of the main challenges in adopting wavefront shaping to real-world applications is

scattering by a varying medium, for example in communication through turbulent atmosphere
[4,19,24], as it requires fast modulation rates and sufficient signal-to-noise ratios (SNR) at short
integration times. The need for fast modulation rates is common for both classical and quan-
tum wavefront shaping. This can be solved by recent breakthroughs providing high modulation
rates, ranging from a few tens of kHz using digital mirror devices (DMDs) [40, 41] to 350kHz
using 1D MEMS [42]. The need for high SNR at short integration times, is in particular crit-
ical for quantum light, since quantum signals are too weak for real-time optimization [34, 43].
This is usually solved by performing the wavefront optimization on an auxiliary bright laser,
that is carefully co-aligned with the entangled photons so that it undergoes the exact same scat-
tering [30–33]. While this approach enables fast optimization, it does not allow simultaneous
transmission of the entangled photons since the auxiliary laser must have the same wavelength
and polarization as the entangled photons and thus cannot be filtered out. For this reason, dy-
namical shaping, although crucial to many quantum technologies, has not been achieved so far
for entangled photons. Since in our method the optimization is done entirely on the classical
bright pump beam, the optimization rates can in principle be as fast as record-high classical
wavefront shaping, making quantum wavefront shaping applicable for real-time applications.
To demonstrate this feature, we emulate dynamical scattering by placing a diffuser on a moving

6
stage, producing a time-dependent speckle pattern. First, we use a diffuser with a relatively
long correlation time, for which the coincidence signal can be measured during the optimiza-
tion (Fig.3a). When the optimization is turned on, both the pump (blue curve) and coincidence
(red curve) signals are enhanced simultaneously, even though the diffuser is constantly moving.
Although the diffuser has a relatively long correlation time, when the same optimization algo-
rithm is used with the coincidence rate for the feedback, no enhancement is observed due to the
poor SNR of the coincidence signal (black curve). In Fig.3b, we increase the speed of the mov-
ing diffuser, yielding a much shorter correlation time, which limits the number of coincidence
measurements that can be performed. Nevertheless, the pump beam intensity is still enhanced
by the optimization process (blue curve), causing the coincidence signal to increase as well
(red points), highlighting the strength of our use of the pump beam as feedback for real-time
optimization.

Figure 3: Real-time shaping. A diffuser is placed on a moving stage, creating a time dependent
speckle pattern at the far-field. By using the intensity of the pump beam as feedback, real-
time optimization is obtained for both the pump beam (blue) and the entangled pairs (red).
When the optimization is turned off, degradation of both signals occurs due to the movement
of the diffuser. In (a), the diffuser speed is relatively slow, so that the correspondence between
the pump beam and the coincidence signal could be observed. Even in this case, real-time
optimization is not possible when the coincidence signal is used for the feedback (black), due to
its inherently low signal-to-noise ratio. In (b), we significantly increase the speed of the diffuser,
showing the strength of our method of using the pump beam for real-time optimization. Here,
the optimization speed is limited by the SLM response time and the feedback electronics (100
ms), yet orders of magnitude faster optimization rates can be achieved using deformable mirror
devices and fast electronics.

Next, we turn to discuss the expected classical and quantum enhancements of the optimiza-
(2)
tion process. Let βp(1) and βDC be the pump intensity and entangled photons coincidence counts
at the chosen target area, respectively, normalized by the total signal. We distinguish between

7
the effects of absorption and scattering on β. In the case of absorption, the relation between
(2) 2
the pump and coincidence signals is quadratic βDC = βp(1) , as we experimentally confirm us-
ing a variable attenuator (Fig.4a,b). This relation results from the fact that for a coincidence
event, both photons must be transmitted through the absorbing media. However, for a purely
scattering sample, the situation can be quite different. Although in the weak pumping regime
the coincidence signal is always linear with the intensity of the pump beam [44], this relation
becomes more complicated when considering the signals at a target area smaller than the whole
emission area, where a quadratic dependency might be expected (Supplementary). To illus-
(2)
trate this, we measure the two-photon signal βDC and the corresponding pump signal βp(1) at
different stages of the optimization process (Fig.4a). Remarkably, we get a clear linear depen-
(2)
dency, βDC = βp(1) (Fig.4b). The linear rather than quadratic dependence is a unique feature
of entangled photons in the high Schmidt number regime K  1, which are scattered by a
thin diffuser. To elucidate this, we numerically simulated the pump and two-photon speckle
patterns formed at the far-field. The simulation was performed for two-photon states with dif-
ferent Schmidt numbers (corresponding to different crystal lengths), using the double-Gaussian
approximation [45]. In Fig.4c, the calculated correlation coefficient between the two patterns
is plotted versus the Schmidt number, K. At K = 1, there is almost no correlation between
the patterns, as expected from two different wavelengths. However, as the Schmidt number in-
creases, the patterns become increasingly correlated, yielding almost perfect correlation in our
experimental conditions at K ≈ 680. Indeed, the experimental correlation coefficient between
the two patterns presented in Fig.2a,b is 0.83. This result completes the picture given by Eq. (2)
and Fig.2a,b, and explains the correspondence between the pump beam and entangled photon
(2)
pairs in the high Schmidt number regime, that yields βDC = βp(1) . We can now easily derive the
(2)
expected enhancement ηDC of our optimization method, defined by the ratio of the coincidence
rate at the target area after optimization and the average coincidence rate at the target area be-
fore optimization. As a consequence of the linear correspondence we established between the
(2)
pump and two-photon signals, the coincidence enhancement ηDC must be exactly equal to the
classical enhancement of the pump beam [29],
(2) π
ηDC = ηp(1) = (N − 1) + 1, (3)
4
where N is the number of degrees of freedom used in the optimization. We therefore conclude
that in our method, the efficiency of the quantum optimization is identical to the efficiency of
classical wavefront shaping.

8
 (2)
Figure 4: Quantum vs. classical efficiency. (a) The two-photon coincidence rate βDC , and
the intensity of the pump beam βp(1) , both measured at the same target area and normalized
by the total signal, are measured in two different configurations. In the first configuration, a
linear polarizer is rotated in the optical path of both beams, to induce a variable absorption
(2)
loss, yielding a clear quadratic relation between βDC and βp(1) (b, green points). In the second
(2)
configuration, the polarizer is replaced by a diffuser, and (βp(1) ,βDC ) is registered during the
(2)
optimization process (b, black points). Remarkably, βDC and βp(1) exhibit a linear dependency
(black curve) instead of the classically expected quadratic one (green curve). (c) The correlation
coefficient between the speckle pattern of the pump beam and the coincidence pattern is pre-
sented as a function of Schmidt number. The correlation coefficient was calculated numerically
under the double Gaussian approximation. In our experiment, K ≈ 680 (marked with a dashed
line), yielding a high correlation between the pump and coincidence patterns which explains the
linear dependency in (b).

Thus far, we considered a transparent thin diffuser placed at the image plane of the non-
linear crystal. Since in this configuration the signal and idler photons pass the diffuser at the

9
same location, it induces the same phase on both photons. The two-photon amplitude therefore
accumulates the same phase as the pump amplitude, and the far-field speckle patterns of the
pump beam and the entangled photons are identical. In fact, the signal and idler photons do
not have to pass through the same location in the diffuser, as long as they accumulate the same
phase. It is therefore enough to require that the photons pass the diffuser through the same co-
herence area, defined by a typical scale d over which the phases induced by the diffuser become
uncorrelated [46]. Pump feedback can therefore work even when the diffuser is not thin nor
perfectly imaged to the crystal, as long as the thickness of the diffuser or its distance from the
image plane is smaller than d2 /λ (see Supplementary). The efficiency of the optimization will
also depend on the position of the stationary detector at the far-field, since when the thickness
of the diffuser is larger than its coherence length d, it exhibits a finite ’memory effect’, i.e. a
finite range of angles over which the speckle patterns remain correlated [47]. The finite memory
effect limits the range of target areas that can be optimized simultaneously.
In Fig.5a,b, we demonstrate the use of pump shaping to compensate volume scattering of
entangled photons by two thin diffusers separated by z ≈ 3mm, creating an effective diffuser
with thickness z such that d  z < d2 /λ, where d ≈ 100µm. This double diffuser config-
uration exhibits a narrow memory effect range (see Supplementary). Indeed, optimizing the
pump beam makes the coincidence pattern localized (Fig.5b). When displacing the stationary
idler detector, the coincidence rate at the target area decreases due to the finite memory effect
associated with the double diffuser configuration.
The d2 /λ limitation on the thickness of the diffuser restricts the range of practical appli-
cations that can benefit from pump optimization. Nevertheless, for free-space quantum links
through turbulent atmosphere, pump optimization can dramatically improve the efficiency of
the link, since the transverse coherence length of the atmosphere d is many orders of magnitude
larger than the optical wavelength λ (Fig.5c). We utilize Kolmogorov’s phase screens model for
turbulence, to numerically simulate the efficiency of our pump optimization method for free-
space optical links (see Supplementary). Figure 5d compares the coincidence rate at the target
area as a function of the length of the optical link, with and without employing pump optimiza-
tion. The simulation is performed for three different turbulence strengths, ranging from weak
to moderate turbulence, quantified by the structure constant of the refractive index fluctuations
Cn2 [48]. A two orders of magnitude improvement in the accessible link length is achieved for
all turbulent conditions. Remarkably, the improvement grows with the turbulence strength (see
Supplementary). To estimate theoretically the maximal expected link length, it is useful to in-
troduce the Rytov variance σR2 , which quantifies the strength of scintillation in the optical link,
where σR2 > 1 is considered moderate to strong scintillation [49]. Using the reasoning presented
above for a volume diffuser, we found that our method will allow efficient optimization for links
with moderate scintillation, of up to σR2 ≈ 2.5 (see Supplementary).

10
Figure 5: Double diffuser configuration and atmospheric links. Two thin diffusers are placed at
a distance of 3mm from each other to emulate a volume diffuser with a finite memory effect (a).
By optimizing the pump beam, the correlations between the entangled photons at a target area
on the optical axis (xi = 0) are enhanced. Due to the finite memory effect, when the stationary
idler detector is displaced from the optical axis (|xi | > 0), the enhancement degrades (b). To
demonstrate the applicability of pump optimization and feedback to free space optical links
through turbulent atmosphere, we simulate ground to ground links with different lengths and
turbulence strengths (c). The coincidence rate at the receiver end for non-optimized link (dots),
decays after 10 km for weak turbulence (brown) and 100m for moderate turbulence (green).
Using pump optimization (solid curves), the link length increases by two orders of magnitude,
ranging from 100km for weak turbulence to 10 km for moderate turbulence. Remarkably, the
increase in the link length grows with the turbulence strength (See Supplementary).

Conclusions
Our approach for shaping entangled photons, by creating them with the correct spatial correla-
tions to compensate for the scattering, has several unique advantages over directly shaping them
after their creation. First, since it is based on shaping the classical pump beam without interact-
ing with the entangled photons themselves, it does not introduce any loss to the quantum light.
Second, since the classical pump beam is used for the optimization feedback, the optimization
can be as fast as shaping of bright classical light. Moreover, since we could easily separate the
pump photons from the entangled photons, we were able, for the first time, to cancel scattering
of entangled-photons from a dynamically moving diffuser, without interfering their continuous
transmission. This is an important step towards implementing quantum wavefront shaping in
real-life scenarios, since there is no down-time for the transmission of the photons while the

11
optimization process is performed.
Since the pump optimization works best for scattering layers with thicknesses that are
2
smaller than dλ , implementation of our method in practical applications requires that the trans-
verse coherence length of the scattering layer d will be much larger than the optical wavelength
λ. As quantum communications through turbulent atmosphere falls exactly in this realm, we be-
lieve the method we developed in this work can play an important role in extending the available
communication distances in scenarios where atmospheric turbulence limits the communication
links [4, 24, 35]. Finally, we note that pump shaping can be used to create a two-photon source
with tailored correlations and coherence properties, by shaping both the amplitude and phase
of the pump beam [50, 51], opening the door for exploiting the capacity of high dimensional
qudits implemented in the spatial domain.

Methods
The experimental setup is presented in Fig.1a. A 2mm long type-0 PPKTP crystal is pumped
by a 50mW , λ = 404nm continuous-wave laser. The wavefront of the pump beam is shaped
by a phase-only SLM, imaged on the crystal by two lenses with focal lengths L1 = 200mm
and L2 = 100mm respectively. Without shaping, the pump profile at the crystal plane is
approximately Gaussian with a waist of 0.7mm. Both the pump beam and the entangled photons
are then imaged onto a 0.25◦ thin diffuser by two lenses with focal lengths L3 = 100mm and
L4 = 50mm. The pump beam and the entangled photons are separated using a dichroic mirror
and measured at the far-field by an CMOS camera and 100um multimode fibers coupled to
single photon detectors, respectively. The far-field measurements are obtained after passing
through a L5 = 150mm lens. For the coincidence measurements, 10nm interference filters
around 808nm are used. In the experiment presented in Fig.5a,b, the thin diffuser is replaced
with a double diffuser configuration, consisting of two thin diffusers (0.16◦ ,0.25◦ ) separated by
3mm.

Data availability
The data that support the findings of this study are available from the corresponding author upon
reasonable request.

References
[1] Dowling, J. P. & Milburn, G. J. Quantum technology: the second quantum revolution
(2003).

[2] O’brien, J. L., Furusawa, A. & Vučković, J. Photonic quantum technologies (2009).

12
 [3] Lo, H.-K., Curty, M. & Tamaki, K. Secure quantum key distribution. Nature Photonics 8,
595 (2014).

[4] Liao, S.-K. et al. Satellite-to-ground quantum key distribution. Nature 549, 43 (2017).

[5] Takenaka, H. et al. Satellite-to-ground quantum-limited communication using a 50-kg-

class microsatellite. Nature photonics 11, 502 (2017).

[6] Brida, G., Genovese, M. & Berchera, I. R. Experimental realization of sub-shot-noise

quantum imaging. Nature Photonics 4, 227 (2010).

[7] Lemos, G. B. et al. Quantum imaging with undetected photons. Nature 512, 409 (2014).

[8] Tenne, R. et al. Super-resolution enhancement by quantum image scanning microscopy.

Nature Photonics 13, 116 (2019).

[9] Bechmann-Pasquinucci, H. & Tittel, W. Quantum cryptography using larger alphabets.

Physical Review A 61, 062308 (2000).

[10] Cerf, N. J., Bourennane, M., Karlsson, A. & Gisin, N. Security of quantum key distribution
using d-level systems. Physical Review Letters 88, 127902 (2002).

[11] Erhard, M., Fickler, R., Krenn, M. & Zeilinger, A. Twisted photons: new quantum per-
spectives in high dimensions. Light: Science & Applications 7, 17146 (2018).

[12] Islam, N. T., Lim, C. C. W., Cahall, C., Kim, J. & Gauthier, D. J. Provably secure and
high-rate quantum key distribution with time-bin qudits. Science advances 3, e1701491
(2017).

[13] Lukens, J. M. & Lougovski, P. Frequency-encoded photonic qubits for scalable quantum
information processing. Optica 4, 8–16 (2017).

[14] OSullivan-Hale, M. N., Khan, I. A., Boyd, R. W. & Howell, J. C. Pixel entanglement:
experimental realization of optically entangled d= 3 and d= 6 qudits. Physical review
letters 94, 220501 (2005).

[15] Walborn, S., Lemelle, D., Almeida, M. & Ribeiro, P. S. Quantum key distribution with
higher-order alphabets using spatially encoded qudits. Physical review letters 96, 090501
(2006).

[16] Kagalwala, K. H., Giuseppe, G., Abouraddy, A. F. & Saleh, B. E. Single-photon three-
qubit quantum logic using spatial light modulators. Nature Communications 8, 739 (2017).

[17] Beenakker, C., Venderbos, J. & Van Exter, M. Two-photon speckle as a probe of multi-
dimensional entanglement. Physical review letters 102, 193601 (2009).

13
[18] Peeters, W., Moerman, J. & Van Exter, M. Observation of two-photon speckle patterns.
Physical review letters 104, 173601 (2010).

[19] Jin, J. et al. Genuine time-bin-encoded quantum key distribution over a turbulent depolar-
izing free-space channel. arXiv preprint arXiv:1903.06954 (2019).

[20] Vallone, G. et al. Free-space quantum key distribution by rotation-invariant twisted pho-
tons. Physical review letters 113, 060503 (2014).

[21] Mirhosseini, M. et al. High-dimensional quantum cryptography with twisted light. New
Journal of Physics 17, 033033 (2015).

[22] Krenn, M. et al. Twisted light transmission over 143 km. Proceedings of the National
Academy of Sciences 113, 13648–13653 (2016).

[23] Sit, A. et al. High-dimensional intracity quantum cryptography with structured photons.
Optica 4, 1006–1010 (2017).

[24] Paterson, C. Atmospheric turbulence and orbital angular momentum of single photons for
optical communication. Physical review letters 94, 153901 (2005).

[25] Tyler, G. A. & Boyd, R. W. Influence of atmospheric turbulence on the propagation of

quantum states of light carrying orbital angular momentum. Optics letters 34, 142–144
(2009).

[26] Vellekoop, I. M. & Mosk, A. Focusing coherent light through opaque strongly scattering
media. Optics letters 32, 2309–2311 (2007).

[27] Mosk, A. P., Lagendijk, A., Lerosey, G. & Fink, M. Controlling waves in space and time
for imaging and focusing in complex media. Nature photonics 6, 283 (2012).

[28] Horstmeyer, R., Ruan, H. & Yang, C. Guidestar-assisted wavefront-shaping methods for
focusing light into biological tissue. Nature photonics 9, 563 (2015).

[29] Vellekoop, I. M. Feedback-based wavefront shaping. Optics express 23, 12189–12206

(2015).

[30] Defienne, H. et al. Nonclassical light manipulation in a multiple-scattering medium. Op-

tics letters 39, 6090–6093 (2014).

[31] Defienne, H., Barbieri, M., Walmsley, I. A., Smith, B. J. & Gigan, S. Two-photon quantum
walk in a multimode fiber. Science advances 2, e1501054 (2016).

[32] Wolterink, T. A. et al. Programmable two-photon quantum interference in 10 3 channels

in opaque scattering media. Physical Review A 93, 053817 (2016).

14
[33] Defienne, H., Reichert, M. & Fleischer, J. W. Adaptive quantum optics with spatially
entangled photon pairs. Phys. Rev. Lett. 121, 233601 (2018).

[34] Peng, Y., Qiao, Y., Xiang, T. & Chen, X. Manipulation of the spontaneous parametric
down-conversion process in space and frequency domains via wavefront shaping. Optics
letters 43, 3985–3988 (2018).

[35] Pugh, C. J., Kolenderski, P., Scarcella, C., Tosi, A. & Jennewein, T. Towards correcting
atmospheric beam wander via pump beam control in a down conversion process. Optics
express 24, 20947–20955 (2016).

[36] Monken, C. H., Ribeiro, P. S. & Pádua, S. Transfer of angular spectrum and image forma-
tion in spontaneous parametric down-conversion. Physical Review A 57, 3123 (1998).

[37] Howell, J. C., Bennink, R. S., Bentley, S. J. & Boyd, R. Realization of the einstein-
podolsky-rosen paradox using momentum-and position-entangled photons from sponta-
neous parametric down conversion. Physical Review Letters 92, 210403 (2004).

[38] Beckers, J. M. Increasing the size of the isoplanatic patch with multiconjugate adap-
tive optics. In European Southern Observatory Conference and Workshop Proceedings,
vol. 30, 693 (1988).

[39] Walborn, S. P., Monken, C., Pádua, S. & Ribeiro, P. S. Spatial correlations in parametric
down-conversion. Physics Reports 495, 87–139 (2010).

[40] Conkey, D. B., Caravaca-Aguirre, A. M. & Piestun, R. High-speed scattering medium

characterization with application to focusing light through turbid media. Optics express
20, 1733–1740 (2012).

[41] Blochet, B., Bourdieu, L. & Gigan, S. Focusing light through dynamical samples using
fast continuous wavefront optimization. Optics letters 42, 4994–4997 (2017).

[42] Tzang, O. et al. Wavefront shaping in complex media with a 350 khz modulator via a
1d-to-2d transform. Nature Photonics 1–6 (2019).

[43] Minozzi, M., Bonora, S., Sergienko, A., Vallone, G. & Villoresi, P. Optimization of two-
photon wave function in parametric down conversion by adaptive optics control of the
pump radiation. Optics letters 38, 489–491 (2013).

[44] Gerry, C., Knight, P. & Knight, P. L. Introductory quantum optics (Cambridge university
press, 2005).

[45] Law, C. & Eberly, J. Analysis and interpretation of high transverse entanglement in optical
parametric down conversion. Physical review letters 92, 127903 (2004).

15
[46] Gatti, A., Magatti, D. & Ferri, F. Three-dimensional coherence of light speckles: theory.
Physical Review A 78, 063806 (2008).
[47] Freund, I., Rosenbluh, M. & Feng, S. Memory effects in propagation of optical waves
through disordered media. Physical review letters 61, 2328 (1988).
[48] Goodman, J. W. Statistical optics (John Wiley & Sons, 2015).
[49] Gbur, G. Partially coherent beam propagation in atmospheric turbulence. JOSA A 31,
2038–2045 (2014).
[50] Defienne, H. & Gigan, S. Spatially-entangled photon-pairs generation using partial spa-
tially coherent pump beam. arXiv preprint arXiv:1812.02046 (2018).
[51] Zhang, W., Fickler, R., Giese, E., Chen, L. & Boyd, R. W. The influence of pump coher-
ence on the generation of position-momentum entanglement in down-conversion. arXiv
preprint arXiv:1812.09532 (2018).
[52] Rickenstorff, C., Rodrigo, J. A. & Alieva, T. Programmable simulator for beam propaga-
tion in turbulent atmosphere. Optics express 24, 10000–10012 (2016).
[53] Fried, D. L. Statistics of a geometric representation of wavefront distortion. JoSA 55,
1427–1435 (1965).
[54] Rodenburg, B. et al. Simulating thick atmospheric turbulence in the lab with application
to orbital angular momentum communication. New Journal of Physics 16, 033020 (2014).
[55] Schmidt, J. D. Numerical simulation of optical wave propagation with examples in MAT-
LAB (2010).
[56] Lane, R., Glindemann, A., Dainty, J. et al. Simulation of a kolmogorov phase screen.
Waves in random media 2, 209–224 (1992).
[57] Andrews, L. C. Atmospheric optics. SPIE Field Guides (2004).
[58] Van Exter, M., Aiello, A., Oemrawsingh, S., Nienhuis, G. & Woerdman, J. Effect of
spatial filtering on the schmidt decomposition of entangled photons. Physical Review A
74, 012309 (2006).

Acknowledgments
The authors thank Hugo Defienne for useful discussions. Funding: This work is supported by
the Zuckerman STEM Leadership Program, the Israel Science Foundation (grant No. 1268/16)
and the Israeli ministry of science. Authors contributions: O.L. and Y.B. designed the experi-
ment. O.L. and G.H. built the experimental setup and performed measurements. O.L. analyzed
and interpreted the data. All authors contributed to the manuscript.

16
Supplementary materials
Correspondence between the pump beam and the entangled
photon pairs
As discussed in the main text, in the high spatial entanglement regime (K  1) there is a
remarkable correspondence between the pump intensity distribution and the entangled photons
coincidence pattern at the far-field. In this section, we demonstrate this correspondence in a sim-
pler setting in which the diffuser in Fig.1a is replaced by the phase mask φ(x, y) = sign(y) π2 ,
for the wavelength of the entangled photons. Since the pump wavelength is half the wavelength
of the entangled photons, the phase mask acts as a 2π-step and does not affect its wavefront
(Fig.S1 insets). For the entangled pair the result is less obvious, since each individual photon
will be affected by the induced π-step (Fig.S1a). However, when considering the pair of entan-
gled photons, due to the high spatial entanglement, both photons hit the phase mask at the same
location, thus accumulating twice the single photon phase. Therefore, although diffracting each
individual photon, the two-photon wavefront is not affected by the π-step and the coincidence
pattern does not change (Fig.S1b,c). In the next section, this surprising correspondence is ex-
tended to a general thin diffuser, under moderate dispersion, allowing our unique use of the
pump beam for the optimization feedback.

Figure S1: Pump beam and entangled photons correspondence. The diffuser in the experi-
mental setup was replaced by an SLM, applying a π-step for the wavelength of the entangled
photons (a). When no phase mask is applied (b), both the pump beam (inset) and coincidence
pattern exhibit a peak with a width determined by the angular spectrum of the pump beam.
When applying the π-step (c), both the pump beam and surprisingly the coincidence pattern
remain unchanged.

Scattering of entangled pairs

In the thin crystal regime, the quantum state of the entangled photons is given by
|ψi = dqs dqs v(qs + qi )a† (qs )a† (qi )|0i, where v(q) is the angular spectrum of the pump
R

17
beam, a† (q) is the creation operator of a photon with a transverse momentum q and |0i is
the vacuum state. For simplicity, we assume here that the photons are created with the same
frequency (corresponding to degenerate SPDC with narrow-band interference filters), and the
same polarization (corresponding to type-0 SPDC process). The effect of a thin diffuser placed
right afterRthe crystal, can be modeled by a unitary transformation on the creation operator
a† (q) → dρa† (ρ)Ad (ρ) exp(iρ · q), where Ad (ρ) is the amplitude transfer function of the
thin diffuser and ρ is the transverse spatial coordinate. By substituting the above relation into
the quantum state we get that:
Z
|ψi = dρW (ρ)Ad (ρ)2 a† (ρ)2 |0i (S1)

where W (ρ) is the spatial profile of the pump beam at the crystal plane. This representation of
the quantum state stresses that, in the thin crystal regime, each pair of photons are created at
the same spatial location. Substituting the quantum state into the expression for the coincidence
pattern, C(qs , qi ) = |h0|a(qi )a(qs )|ψi|2 , yields Eq. (2):
Z 2
dρW (ρ)A2d (ρ) exp(−iρ
 
C(qs , qi ) ∝ 
 · (qs + qi )) (S2)

Neglecting loss and material dispersion, we can model the diffuser as a random phase mask
of the form Ad (ρ) = exp(iφ(ρ)). In this case, the scattering of the two entangled photons is
determined by an effective diffuser, given by A2d (ρ) = exp(i2φ(ρ)), which corresponds to the
phase accumulated by the pump beam. Thus, the created two-photon speckle pattern is identical
to that of the classical pump beam, up to a scaling factor of two due to the different wavelength.
In the derivation, we assumed that the crystal is thin enough so that the phase matching
condition can be neglected. To quantitatively determine the limitations on the crystals thickness,
we consider a thin diffuser with a scattering angle θ ∼ λ/d, where d is the transverse coherence
length of the diffuser, i.e. the length scale over which the phase imposed by the diffuser is
correlated [46]. For our method to work, we need to fulfill two conditions. First, both signal and
idler photons must accumulate the same phase when passing through the diffuser. Therefore, we
require that the spatial correlation length dc between the entangled photons is smaller √ than the
transverse coherence length of the diffuser d, i.e. dc < d. Since for SPDC light dc ∝ λL [39],
we get that this condition is met for L < d2 /λ, where L is the thickness of the nonlinear crystal.
Second, the angular spectrum of the shaped pump beam needs to be transferred to the quantum
state of the entangled photons. This implies that the angular spread of the SPDC φSP DC has to
be wider than the angular spread of the pump beam. Since the pump beam is shaped in order
to compensate the scattering induced by the diffuser, the width of its angular spectrum, given
in terms of angle, is θ. We q therefore require that θ < φSP DC . Since the angular spread of
SPDC is given by φSP DC ∝ λ/L [39], we conclude that Lθ < d, or equivalently L < d2 /λ.
Interestingly, this is the same requirement we obtained by the first condition.

18
Lossy diffuser
In this section, we consider the performance of our pump shaping optimization method in the
presence of loss. We model the lossy diffuser as an amplitude transfer function of the form
A(ρ) = t(ρ) exp(iφ(ρ)), where φ(ρ) and 0 ≤ t(ρ) ≤ 1 are the phase and amplitude modula-
tions imposed by the diffuser, respectively. In this case, the coincidence pattern takes the form
(Eq. (2))
Z 2
2
 
C(qs , qi ) ∝ 
 dρW (ρ)t (ρ) exp(i2φ(ρ)) exp(−iρ(qs + qi )) (S3)

Similarly, for the classical pump beam, the intensity profile at the far-field is given by
Z 2
 
I(q) = 
 dρW (ρ)t(ρ) exp(i2φ(ρ)) exp(−iρq) (S4)

Inspection of the above equations shows that even in the presence of significant loss (t2 (ρ) < 1),
since by definition both t(ρ) and t2 (ρ) are real and positive functions, the optimized phase
correction for the pump beam, exp(−i2φ(ρ)), will simultaneously optimize the phase of the
two-photon amplitude. Remarkably, this means that in the presence of loss, the far-field speckle
patterns of the pump beam and the entangled photons may not be the same, yet pump optimiza-
tion will still optimize the coincidence signal, as demonstrated in Fig.S2.

19
Figure S2: Numerical simulation of a lossy diffuser. In the presence of loss, the speckle pattern
of the pump beam (a) and the entangled photons (b) are not identical. Nevertheless, optimizing
the phase of the pump beam (c) simultaneously optimize the coincidence signal as well (d).
In this case, the effect of loss was simulated by imposing random transmission coefficients,
uniformly distributed between 0 and 1 (see text).

One might wonder what will be the performance of phase-only optimization in the presence
of significant amplitude modulations. In the common case of imperfect optimization, which can
result for example from a limited number of degrees of freedom used in the optimization, the
enhancement achieved by phase-only correction is lower only by a factor of π/4 compared to
phase and amplitude correction [29]. In the other limit, where all phases are perfectly corrected,
the residual amplitude modulations will be the limiting factor for the efficiency. However, in
contrast to phase aberrations, amplitude modulations by themselves cannot severely degrade the

20
signal, as all fields interfere constructively at the far-field target area (Fig.S2). To quantify the
upper bound for the efficiency set by the residual amplitude modulations, we assume the random
transmission coefficients ti are distributed with a given mean µ and standard deviation σ. In this
case, the fraction of transmitted signal at the target area is given by h( ni=1 ti )2 i/hn ni=1 t2i i,
P P

where n is the number of random transmission coefficients. For large n, this upper bound for
the efficiency approaches (1 + σ 2 /µ2 )−1 .
To demonstrate the performance of our pump shaping method in the presence of loss, we
numerically simulate a segmented lossy diffuser, where each segment is assigned with random
phase taken from a uniform distribution between 0 to 2π, φ ∼ unif (0, 2π) and a random
amplitude taken from a uniform distribution t ∼ unif (1−s, 1), where 0 ≤ s ≤ 1 is a parameter
controlling the amount of loss in the simulated diffuser. When considering scattering due to
amplitude modulations, the average transmission value is not by itself important, but rather the
strength of the modulations around the mean. Therefore, aside from its simplicity, we chose
our model such that it demonstrates strong amplitude fluctuations for s = 1 (transmission
coefficients are randomly distributed between zero and one) and no loss nor fluctuations for
s = 0 (the transmission is one).
From Fig.S3a, one can see that by optimizing the phase of the pump beam, the coincidence
counts of the entangled photons are enhanced as well, for both perfect and imperfect phase-
only corrections, even in the presence of significant loss, t ∼ unif (0, 1). The case of imperfect
phase only correction was simulated by limiting the amount of degrees of freedom used in the
optimization. In Fig.S3b, we consider the enhancement η achieved by optimizing the pump
beam, as a function of the loss strength. In the case of perfect phase-only correction, the resid-
ual amplitude modulations cause the enhancement to decrease relatively to the lossless case.
The decrease in the enhancement sets an upper bound for the efficiency of the optimization in
the perfect phase correction case, which depends on the statistics of the random transmission
coefficients. For our model with s=1, the maximal efficiency that could be achieved is bounded
by ∼ 0.5. In the case of imperfect phase-only optimization, since significant uncorrected phase
aberrations are still present, the effect of the amplitude modulations is less pronounced, yielding
almost constant enhancement. Therefore, we can conclude that in the case of a lossy diffuser,
one can still use phase-only pump shaping to compensate the scattering of the entangled pho-
tons, which will significantly enhance the signal at the target area.

21
 (2)
Figure S3: Lossy diffuser. (a) The normalized coincidence signal at the target area, βDC , as a
function of the loss strength parameter s chosen in the simulation. For s = 1, the transmission
coefficients follow t ∼ unif (0, 1). (b) The enhancement achieved by the pump optimiza-
tion method as a function of the loss strength, for perfect and imperfect phase-only correction.
Imperfect correction was simulated by limiting the number of degrees of freedom used in the
correction. In all numerical simulations, as in the experiments, the feedback for the optimization
is always based on the intensity of the pump beam at the far-field target area.

Optimization efficiency
In this section, we give an intuitive explanation for the surprising linear correspondence between
the classical pump beam and the entangled photons (Fig.4). When an entangled pair passes
through the diffuser, both of the photons are scattered, each having a certain probability to
arrive at the target area, which is much smaller than the total emission area. Following this
consideration, since for a coincidence event both photons must arrive at the target area, one
might naively expect a quadratic relation between the efficiency of the coincidence signal and
that of a laser beam with the same optical properties. However, this is not true when considering
spatially entangled photons in a high dimensional Hilbert space. In this case, the full quantum
state, consisting of a large superposition of entangled pairs, must be taken into account. To
explain the effect of this large superposition, we define A, B as the events where the signal and
idler photons arrive at the target area, respectively. Therefore, requiring that for a coincidence
event both photons arrive at the target area, the efficiency of the coincidence signal is given
by, P (A ∩ B) = P (B)P (A|B) , where ∩ is the logical ’and’, P (B) is the ratio between
the rates of the event B with and without the diffuser, and A|B is the conditional event of A
happening, given B. Since the quantum state consists of a large number of spatial modes, the
contrast of the single photon speckle pattern approaches zero in the high entanglement regime
(Fig.2c,f). Hence, the rate of idler photons arriving at the target position does not change due
to the scattering, yielding P (B) = 1. The second term, P (A|B), is determined by the spatial

22
distribution to which the signal photon collapses after passing through the diffuser, given that
the idler photon was measured at the target area. However, according to Eq. (2) and Fig.2a,b,
this spatial distribution corresponds to that of the classical pump beam, yielding the surprising
linear correspondence between the pump beam intensity and the entangled photons correlations.

Volume diffusers
As described in the previous sections, our method relies on the fact that when the signal and
idler photons pass through the diffuser at the same location, their joint quantum state accumu-
lates twice the single photon phase, which is thus equal to the phase accumulated by the pump
beam. In fact, the photons do not have to pass the diffuser at exactly the same location to ac-
cumulate the same phase. It is enough to demand that the distance between them is smaller
than the coherence length of the diffuser d. Thus, for a diffuser with a scattering angle θ, the
performance of our pump shaping method will depend on the dimensionless parameter d/zθ,
where z is thickness of the diffuser. One might note that we use the scattering angle θ in the
above expression and not the angular spread of the entangled photons, φSP DC . This is because
even if θ  φSP DC for a given target area at the far-field, only photons emitted at a range of
angles θ will be scrambled by the diffuser and contribute to the signal at the target point. As for
classical light, the range of far-field target positions for which the optimization works will be
limited by the memory effect [47]. Therefore, within the memory effect, the feedback provided
by the pump beam will work up to a thickness of z ∝ d/θ ∼ d2 /λ. To show this, we performed
a numerical simulation of volume diffusers, emulated by two thin diffusers located at variable
distance from each other (Fig.S4). Indeed, for all simulated diffuser strengths, the same scaling
is observed, showing that the optimization works up to the relevant d2 /λ scale.

23
Figure S4: Volume scattering. The fraction of signal at the target area, which is our defini-
tion for the efficiency of the optimization, is simulated for a double diffuser configuration as
a function of the distance between the diffusers. It is clearly seen that the efficiency of the
optimization depends on the typical length scale zrd = πd2 /λ.

The same considerations could be applied in the case of a non-imaged diffuser as well,
which can be viewed as a simplified case of volume scattering in which propagation occurs
prior to scattering. Therefore, by identifying the thickness z from the previous analysis to be the
distance of the diffuser from the image plane of the crystal, we get that the same dimensionless
parameter d2 /λz will govern the efficiency of the optimization. Thus, optimizing the pump
beam could be useful in cases of volume scattering and non-imaged diffusers as well, within
the memory effect, as long as the propagation distance is smaller than the typical propagation
length d2 /λ determined by the diffuser properties.

Diffuser characterization
In this section, we measure the memory effect for both diffusers used in the experiments. The
diffusers were placed on a rotating stage, allowing us to measure the far-field speckle pattern
as a function of the diffusers angle (Fig.S5 inset). The memory effect of the double diffuser
configuration, consisting of two thin diffusers separated by 3mm, is significantly narrower than
the one obtained with a single thin diffuser (Fig.S5). The transverse coherence length of the
double diffuser, d ∼ λ/θ, is approximately 100µm, satisfying d  z.

24
Figure S5: The memory effect of the thin diffuser (blue) and the double diffuser configuration
emulating a volume diffuser (orange) was measured by rotating the diffuser and record the
far-field angle dependent speckle pattern (inset). The correlation between the speckle patterns
decrease significantly faster with angle for the double diffuser configuration.

Propagation through turbulent atmosphere

In this section, we explain in detail the simulation presented in Fig.5d. First, we begin by a
brief review on light scattering in the atmosphere. The refractive index of air, including the
effect of dispersion, is given by n(P, T, λ) = 1 + 77.6(1 + 7.52 ∗ (10)−3 /λ2 ) PT ∗ (10)−6 ,
where λ is the wavelength in microns, P is the atmospheric pressure in millibars and T is
the temperature in Kelvins [48]. The refractive index variations in the atmosphere of earth
are dominated by temperature fluctuations, resulting from turbulent air flow. These refractive
index fluctuations can be described by their power spectral density (PSD) function, which,
according to Kolmogorov’s theory of turbulence, is given by Φn (k) = 0.033Cn2 k −11/3 , where
k = (kx , ky , kz ) is the spatial frequencies vector and Cn2 is the structure constant of the refractive
index fluctuations, which quantifies the local turbulence strength [48]. To obtain a more accurate
model, two corrections are often added to the PSD function, yielding the von-Karman spectrum

Φn (kx , ky , kz ) = 0.033Cn2 (kx2 + ky2 + kz2 + ko2 )−11/6 exp(−(kx2 + ky2 + kz2 )/km
2
) (S5)

25
 where ko = 2π/lo , lo is the so-called outer scale of the turbulence, which is usually in the
1-100m range, km = 5.32/li and li is the inner turbulence scale, which has a typical value
of a few millimeters [48]. To simulate the propagation of light through turbulent atmosphere,
the von-Karman PSD function is transferred into a two dimensional form, representing a phase
screen rather than the three dimensional refractive index fluctuations [52]
−5/3
Φφ (kx , ky ) = 0.49r0 (kx2 + ky2 + ko2 )−11/6 exp (−(kx2 + ky2 )/km
2
) (S6)

where r0 = (0.4229(2π/λ)2 zCn2 )−3/5 is the atmosphere coherence width, sometimes called
the Fried parameter [52, 53], and z is the atmospheric link length. For simplicity we assumed
in the expression for r0 a constant Cn2 profile. To simulate propagation through thick atmo-
sphere, we utilize a well-known two phase screen model [52, 54]. The screens are positioned in
an equally spaced configuration, and the Fried parameter of each screen is chosen such that
both the Fried parameter and scintillation effects of the entire link are correctly accounted
for [52]. Indeed, we found the effect of simulating more than two phase screens to be neg-
ligible. We simulate the random phase screens from the von-Karman PSD function using the
standard inverse-Fourier transform method, with additional ten subharmonics to account for the
low spatial frequency components [55, 56]. We use typical values of lo = 10m and li = 5mm
for the outer and inner turbulence scales. In all simulations, the pump is taken to be a Gaus-
sian beam with a one-meter waist, chosen to ensure significant scattering by the atmosphere.
We note that in all simulations the dispersion of the atmosphere was taken into account, via the
above expression for n(P, T, λ). Nevertheless, we find that the effect of dispersion is negligible.

26
 (2)
Figure S6: Thick atmosphere scaling. The efficiency βDC is plotted as a function of the link
length for different values of r0 (inset). The degradation in efficiency is determined by the
r0
typical length scale zra = πρ20 /λ, where ρ0 = 2.1 .

As in the case of a volume diffuser discussed earlier, the efficiency of our method will
depend on the ratio z/zra , where zra = (πρ20 )/λ, λ is the wavelength of the pump beam and
ρ0 = r0 /2.1 is the atmosphere coherence radius. The coherence radius ρ0 is the atmospheric
equivalent of the diffusers transverse coherence length d discussed in the previous section. To
(2)
show this, we simulated atmospheric links with different values of r0 , and calculated βDC as a
function of the link length (Fig.S6). We can therefore derive a condition for the applicability of
pump shaping to free space atmospheric links, by demanding that z < 7zra , where the numerical
prefactor was taken for a 1/e decay. It is convenient to rewrite this condition using the Rytov
variance, σR2 = 1.23(2π/λ)7/6 Cn2 z 11/6 , which quantifies the scintillation strength in the optical
link, yielding σR2 < 2.5. As σR2 defines the transition between weak to strong turbulence [49],
we conclude that our pump shaping method could be used in a wide range of free space optical
links, including both ground to ground and ground to satellite links with moderate turbulent
conditions. In Fig.5d in the main text, three ground to ground links with different Cn2 values
were simulated using the method described before. In this case, Cn2 was kept constant while the
link length is increased, yielding stronger scattering (smaller ro ) and scintillation. A two orders
of magnitude improvement in the link length is observed.
The enhancement of the link length due to pump optimization can be calculated based on the
analysis presented above. We define zo to be the link length for which the efficiency of the quan-
tum signal drops to half, when pump optimization is employed. Similarly, zno is the equivalent

27
length in the non-optimized case. Since the efficiency of the pump optimized quantum signal
depends on (πρ20 )/λz, we get that zo satisfies 0.5 ∼ (πρ20 )/(λzo ). For the non-optimized case,
the efficiency depends on (r0 /w)2 , where w is the waist of the Gaussian beam [57]. Therefore,
zno satisfies 0.5 ∼ (r0 /w)2 . Combining these two results and substituting the expressions for r0
and ρ0 , we get that the enhancement of the link length remarkably increases with the turbulence
strength Cn2 , and is given by
zo w5/3
∝ 15/11 (Cn2 )5/11 (S7)
zno λ
We demonstrate this result in Fig.S7, where we simulate zo /zno as a function of (Cn2 )5/11 ,
showing a clear linear relation.

Figure S7: The ratio between the optimized and non-optimized link lengths, defined by the
length for which the efficiency drops to half, is presented as a function of (Cn2 )5/11 . Dots are
obtained from a numerical simulation, and the curve is a linear fit given by the equation in the
inset. The clear linear relation indicates that the improvement offered by our method remarkably
increases as the turbulence becomes stronger.

Finally, we note again that in order to ensure a significant degradation by the turbulent
atmosphere, we used the extreme case of a Gaussian beam with a large, one-meter waist. To
show that our approach provides a significant advantage in more common cases as well, we
repeated the simulation in Fig.5d with a waist of 15cm. The results are presented in Fig.S8,
showing a clear advantage provided by our approach in this case as well.

28
Figure S8: The simulation presented in Fig.5d is repeated using a Gaussian beam with an 15cm
waist at the transmitter plane. Our optimization method demonstrates significant improvement
in the link length in this case as well.

Measurement of the Schmidt number

In order to estimate the Schmidt number K in our experimental apparatus, we use the double-
Gaussian approximation for the phase matching condition, yielding a two-photon distribution
of the form [45],
(qs + qi )2
!
 
2 2
ψ(qs , qi ) ∝ exp − exp −b (q s − q i ) (S8)
σ2
L
where b2 = 4k , k is the pump wavenumber inside the crystal, and σ is the waist of the Gaussian
pump beam.
Under the double-Gaussian approximation, the Schmidt number K could be computed ex-
actly, and is given by [45],
2
1 1

K= + bσ (S9)
4 bσ
Thus, in the thin crystal or weak focusing regimes, where bσ  1, one could approximate the
1
Schmidt number as K = (2bσ) 2 . This expression for the Schmidt number can be understood

intuitively by noticing its relation to the number of transverse spatial modes in the system [58].
Therefore, the Schmidt number could be estimated experimentally by using the width of the
far-field single and coincidence counts distributions as follows. The coincidence pattern at
2
the far-field is given by C(qs , qi = 0) ∝ |exp(−q2s /σ 2 ) exp(−b2 q2s )| , where we assumed for
simplicity that the stationary detector is placed such that qi = 0. Since in the thin crystal regime
2
σ  1/b, we can approximate C(qs , qi = 0) ∝ |exp(−q2s /σ 2 )| , and thus measure σ from the

29
width of the coincidence pattern. The single photon counts distribution, which is obtained by
scanning one of the detectors and recording the photon rate, is given by I(q) = |a(q)|ψi|2 ∝
2
dq0 |exp (−(q + q0 )2 /σ 2 ) exp (−b2 (q − q0 )2 )| . Since σ  1/b, we can approximate q0 ≈
R
2
−q, yielding I(q) ∝ |exp (−b2 (2q)2 )| . Thus, b can be measured from the single counts
distribution width. Combining these two measurements, we obtained a Schmidt number K =
680 ± 60 in our experiment, indicating high spatial entanglement.

30