CD-Kennedy Receiver: Principles & Applications

• The CD-Kennedy receiver is a quantum optical system that displaces one hypothesis toward vacuum and then uses photon counting for binary state discrimination.
• It utilizes both static coherent displacement and conditionally-dynamic strategies to optimize detection performance and improve error rates against Gaussian receivers.
• Experimental implementations and theoretical extensions show its adaptability in telecom platforms, hybrid receiver designs, and enhanced phase-reference management.

The CD-Kennedy receiver is a family of Kennedy-type quantum optical receivers in which an incoming optical state is first displaced in phase space and then discriminated by photon counting. In the literature represented here, the label is used in two closely related senses. In one, it emphasizes the canonical Kennedy front end of coherent displacement plus detection; in another, especially for multistage free-space and CV-QKD receivers, it denotes a conditionally-dynamic Kennedy architecture in which the displacement is updated from channel estimates or prior measurement outcomes. Across these usages, the common mechanism is the same: one candidate hypothesis is displaced toward vacuum, and the resulting click or photon-number statistics implement a non-Gaussian measurement that can outperform Gaussian receivers such as homodyne detection in appropriate regimes (Bina et al., 2014, Izumi et al., 2017, Yuan et al., 24 Sep 2025).

1. Terminology and defining architecture

The Kennedy receiver was introduced for binary phase-shift keyed coherent states, and its basic structure remains the template for CD-Kennedy variants. The receiver applies a displacement operator $\hat D(\beta)$ so that one hypothesis becomes vacuum or vacuum-like, and then measures the displaced field with an on-off detector or, in generalized forms, a photon-number-resolving detector. Operationally, the displacement is implemented by interfering the signal with a strong local oscillator on a highly transmissive beam splitter, after which the receiver maps “no click” and “click” to the two hypotheses (Bina et al., 2014, Shcherbatenko et al., 2019).

Usage of “CD”	Defining feature	Representative source
Coherent displacement + detection	Static displacement followed by on-off or PNR detection	(Izumi et al., 2017, Bina et al., 2014)
Conditionally-dynamic	Displacement chosen from estimated transmittance and previous-stage decisions	(Yuan et al., 24 Sep 2025)

For coherent-state BPSK, the canonical alphabet is $\{|\alpha\rangle,|-\alpha\rangle\}$. The ideal Kennedy choice nulls one state exactly:

$$|-\alpha\rangle \xrightarrow{\hat D(\alpha)} |0\rangle,\qquad |\alpha\rangle \xrightarrow{\hat D(\alpha)} |2\alpha\rangle.$$

The detector then realizes the on-off POVM

$$\Pi_0 = |0\rangle\langle 0|,\qquad \Pi_1 = \sum_{n>0}|n\rangle\langle n|,$$

with “off” assigned to the nulled hypothesis and “on” to the other (Bina et al., 2014). This is a simple non-Gaussian, non-projective measurement, and it is the reference point for generalized, optimized, and conditional-displacement extensions (Izumi et al., 2017, Notarnicola et al., 2022).

2. Binary coherent-state discrimination and generalized Kennedy reception

The underlying discrimination task is binary state discrimination for nonorthogonal coherent states. For BPSK coherent states with equal priors and mean photon number $N=|\alpha|^2$, the Helstrom minimum error probability is

$$P_{\text{Hel}}=\frac12\left(1-\sqrt{1-e^{-4N}}\right),$$

whereas the ideal Kennedy receiver yields

$P_e=\frac12 e^{-4N}.$

Thus the Kennedy receiver is quasi-optimal but not optimal; in the high-energy regime its error remains approximately twice the Helstrom bound (Bina et al., 2014, Shcherbatenko et al., 2019).

A generalized Kennedy receiver replaces exact nulling by an arbitrary displacement. In one parameterization, the displaced hypotheses are written as $|\beta\rangle$ and $|2\alpha+\beta\rangle$, with conditional no-click and click probabilities determined by Poisson vacuum probabilities after displacement (Bhadani et al., 2020). This formulation makes clear that the original Kennedy design is only one operating point in a broader family of displacement receivers.

The generalized formulation also exposes a distinction between single-shot error minimization and capacity maximization. For a receiver with symbol prior $p$ and displacement $\{|\alpha\rangle,|-\alpha\rangle\}$0, the average error probability and the induced mutual information define different optimization problems. The capacity attained by the Kennedy receiver is

$\{|\alpha\rangle,|-\alpha\rangle\}$1

and the displacement that maximizes $\{|\alpha\rangle,|-\alpha\rangle\}$2 is different from the displacement that minimizes one-shot discrimination error (Bhadani et al., 2020). This result is significant because it shows that a Kennedy-type receiver may need to be “programmed” differently for communication rate than for hypothesis testing.

At the limit of complexity, the Dolinar receiver adds real-time feedback to the Kennedy principle and attains the Helstrom bound for binary coherent states. CD-Kennedy architectures occupy the intermediate regime: more flexible than static exact nulling, but simpler than full continuous-time Dolinar feedback (Shcherbatenko et al., 2019, Notarnicola et al., 2022).

3. Conditional, optimized, and hybrid variants

A major line of development replaces fixed on-off decision logic with optimized displacement and thresholding. For a generalized Kennedy receiver with thermal noise, dark counts, finite efficiency, and imperfect visibility, maximum a posteriori detection is provably equivalent to threshold detection on the observed photon number. This yields the framework of optimally displaced threshold detection (ODTD), in which one jointly optimizes the displacement $\{|\alpha\rangle,|-\alpha\rangle\}$3 and the photon-count threshold $\{|\alpha\rangle,|-\alpha\rangle\}$4 to minimize error (Yuan et al., 2020).

Within that model, the detector statistics are written in terms of Laguerre-polynomial expressions for $\{|\alpha\rangle,|-\alpha\rangle\}$5, incorporating detector efficiency $\{|\alpha\rangle,|-\alpha\rangle\}$6, thermal photons $\{|\alpha\rangle,|-\alpha\rangle\}$7, dark counts $\{|\alpha\rangle,|-\alpha\rangle\}$8, and interference visibility $\{|\alpha\rangle,|-\alpha\rangle\}$9 (Yuan et al., 2020). Two asymptotic connections are especially important. First, as the signal power grows, the optimum displacement tends to $$|-\alpha\rangle \xrightarrow{\hat D(\alpha)} |0\rangle,\qquad |\alpha\rangle \xrightarrow{\hat D(\alpha)} |2\alpha\rangle.$$0, so ODTD degenerates to a Kennedy receiver with threshold detection. Second, as the displacement becomes very large, the thresholded Kennedy receiver approaches one-port homodyne detection (Yuan et al., 2020). These limits place optimized Kennedy reception on a continuum between exact nulling and quadrature measurement.

A distinct conditional-displacement construction is the hybrid near-optimum binary receiver based on a first homodyne-like stage and a second displacement-counting stage. The input is split by a beam splitter of transmissivity $$|-\alpha\rangle \xrightarrow{\hat D(\alpha)} |0\rangle,\qquad |\alpha\rangle \xrightarrow{\hat D(\alpha)} |2\alpha\rangle.$$1. A homodyne-like measurement using PNR detectors produces a difference photocurrent $$|-\alpha\rangle \xrightarrow{\hat D(\alpha)} |0\rangle,\qquad |\alpha\rangle \xrightarrow{\hat D(\alpha)} |2\alpha\rangle.$$2, and the sign of $$|-\alpha\rangle \xrightarrow{\hat D(\alpha)} |0\rangle,\qquad |\alpha\rangle \xrightarrow{\hat D(\alpha)} |2\alpha\rangle.$$3 determines the sign of the nulling displacement applied to the remaining signal before direct detection. This is a genuine CD-Kennedy mechanism: the displacement is conditioned on prior measurement information (Notarnicola et al., 2022).

In the ideal homodyne-limit version of that hybrid receiver, the optimized error probability beats both the Kennedy receiver and the standard-quantum-limit benchmark. At large $$|-\alpha\rangle \xrightarrow{\hat D(\alpha)} |0\rangle,\qquad |\alpha\rangle \xrightarrow{\hat D(\alpha)} |2\alpha\rangle.$$4, the ratio to the Kennedy error saturates at

$$|-\alpha\rangle \xrightarrow{\hat D(\alpha)} |0\rangle,\qquad |\alpha\rangle \xrightarrow{\hat D(\alpha)} |2\alpha\rangle.$$5

so the hybrid receiver has about $$|-\alpha\rangle \xrightarrow{\hat D(\alpha)} |0\rangle,\qquad |\alpha\rangle \xrightarrow{\hat D(\alpha)} |2\alpha\rangle.$$6–$$|-\alpha\rangle \xrightarrow{\hat D(\alpha)} |0\rangle,\qquad |\alpha\rangle \xrightarrow{\hat D(\alpha)} |2\alpha\rangle.$$7 lower error than Kennedy in that regime (Notarnicola et al., 2022). With finite PNR resolution, non-unit efficiency, dark counts, and visibility reduction, the hybrid scheme continues to outperform both the standard Kennedy receiver and a pure displacement-PNR receiver (Notarnicola et al., 2022).

4. Phase reference, noise, and practical limitations

Because Kennedy reception is intrinsically phase-sensitive, phase-reference management is a central issue. A Kennedy-like BPSK receiver can be analyzed as a displaced mixture of opposite-phase coherent states, producing photon-number statistics

$$|-\alpha\rangle \xrightarrow{\hat D(\alpha)} |0\rangle,\qquad |\alpha\rangle \xrightarrow{\hat D(\alpha)} |2\alpha\rangle.$$8

These statistics carry information not only about the transmitted symbol but also about the relative phase $$|-\alpha\rangle \xrightarrow{\hat D(\alpha)} |0\rangle,\qquad |\alpha\rangle \xrightarrow{\hat D(\alpha)} |2\alpha\rangle.$$9 between signal and local oscillator (Bina et al., 2014).

A key result is that the same data used for shot-by-shot discrimination can also be processed, via Bayesian inference, to estimate $$\Pi_0 = |0\rangle\langle 0|,\qquad \Pi_1 = \sum_{n>0}|n\rangle\langle n|,$$0 in real time. For PNR detection, the likelihood is

$$\Pi_0 = |0\rangle\langle 0|,\qquad \Pi_1 = \sum_{n>0}|n\rangle\langle n|,$$1

and the associated estimator asymptotically achieves variance $$\Pi_0 = |0\rangle\langle 0|,\qquad \Pi_1 = \sum_{n>0}|n\rangle\langle n|,$$2, where $$\Pi_0 = |0\rangle\langle 0|,\qquad \Pi_1 = \sum_{n>0}|n\rangle\langle n|,$$3 is the classical Fisher information. PNR detection yields higher Fisher information than on-off detection, and the method remains robust under uniform phase noise modeled by bracket states (Bina et al., 2014). This establishes a dual-use role for Kennedy-type receivers: discrimination and phase-reference monitoring within the same optical front end.

The principal limitation is that Kennedy nulling is highly vulnerable to phase diffusion. For BPSK channels with Gaussian phase diffusion of width $$\Pi_0 = |0\rangle\langle 0|,\qquad \Pi_1 = \sum_{n>0}|n\rangle\langle n|,$$4, the Kennedy error probability degrades rapidly, whereas homodyne detection is robust for any channel energy and reaches the Helstrom bound in the limit of large noise (1305.4201). The same work shows that homodyne beats the Kennedy receiver as the signal energy increases, with an energy-dependent threshold in $$\Pi_0 = |0\rangle\langle 0|,\qquad \Pi_1 = \sum_{n>0}|n\rangle\langle n|,$$5 above which homodyne is superior (1305.4201). This directly counters the common misconception that displacement-based reception is uniformly preferable to homodyne: its advantage is conditional on phase stability.

Under realistic device imperfections, the main penalties arise from non-unit visibility, dark counts, and finite extinction. In an all-fiber Kennedy receiver model, a practical imperfect-error expression is

$$\Pi_0 = |0\rangle\langle 0|,\qquad \Pi_1 = \sum_{n>0}|n\rangle\langle n|,$$6

where $$\Pi_0 = |0\rangle\langle 0|,\qquad \Pi_1 = \sum_{n>0}|n\rangle\langle n|,$$7 is the dark-count contribution per symbol bin and $$\Pi_0 = |0\rangle\langle 0|,\qquad \Pi_1 = \sum_{n>0}|n\rangle\langle n|,$$8 is the interference extinction ratio (Shcherbatenko et al., 2019). In optimized-threshold analyses, thermal noise is even more detrimental than dark counts, while interference visibility strongly controls the range in which Kennedy-type receivers can beat homodyne (Yuan et al., 2020).

5. Experimental platforms and benchmarked performance

Experimentally, Kennedy reception has been realized in both free-space-style interferometric settings and all-fiber telecom platforms. A notable fiber implementation at $$\Pi_0 = |0\rangle\langle 0|,\qquad \Pi_1 = \sum_{n>0}|n\rangle\langle n|,$$9 assembled the receiver entirely from standard fiber-optic elements, transmitted signal and local oscillator through different fibers, and used a superconducting nanowire single-photon detector. The system achieved extinction $N=|\alpha|^2$0, detector efficiency $N=|\alpha|^2$1, and dark count rate $N=|\alpha|^2$2 (Shcherbatenko et al., 2019).

That experiment reported sub-SQL discrimination for $N=|\alpha|^2$3, and “the lowest error rate, normalized to SQL, was obtained for the signal intensity 1.3 photons and is equal to 0.4 SQL” (Shcherbatenko et al., 2019). The significance of this result is architectural as much as metrological: it shows that the basic Kennedy front end—high-contrast displacement plus single-photon detection—can be implemented in a telecom-compatible platform suitable for further extension to multichannel or feedforward designs.

Earlier Kennedy-like experiments also demonstrated that real-time phase monitoring can be embedded directly into receiver operation without interrupting data transmission, using either on-off or PNR detection and Bayesian post-processing (Bina et al., 2014). Together with the fiber implementation, these results establish a practical progression: static Kennedy nulling, improved phase-reference handling, and eventual integration of more advanced conditional or multistage control.

The experimental record also clarifies the role of benchmark choice. Depending on the paper, the comparison baseline is a homodyne or heterodyne SQL-type Gaussian receiver. Kennedy-type receivers can beat these Gaussian benchmarks in low-photon-number regimes, but they do not reach the Helstrom bound without additional adaptivity or more elaborate front-end transformations (Shcherbatenko et al., 2019, Notarnicola et al., 2022).

6. Extensions beyond coherent-state BPSK

The Kennedy principle has been extended well beyond its original binary coherent-state setting. In single-rail photonic qubits, the displacement photon counter studied for the states

$N=|\alpha|^2$4

is explicitly identified as the single-rail analogue of a CD-Kennedy receiver (Izumi et al., 2017). Its ideal POVM is

$N=|\alpha|^2$5

restricted to the $N=|\alpha|^2$6 subspace. Quantum detector tomography reconstructed the actual POVM in a truncated four-dimensional Fock space and found fidelities above $N=|\alpha|^2$7 with the ideal Kennedy POVM in the relevant two-dimensional subspace (Izumi et al., 2017). For this task, the minimum Kennedy error is

$N=|\alpha|^2$8

which is below the best Gaussian-plus-homodyne benchmark

$N=|\alpha|^2$9

The result shows that non-Gaussian displacement-plus-counting can outperform any Gaussian preprocessing followed by homodyne even when the signal alphabet lies in the vacuum–single-photon sector (Izumi et al., 2017).

For displaced-squeezed BPSK, the “Inverse-squeezing Kennedy” receiver generalizes the Kennedy architecture by inserting an inverse-squeezing operator after the conventional nulling displacement. With S-BPSK states $$P_{\text{Hel}}=\frac12\left(1-\sqrt{1-e^{-4N}}\right),$$0 and optimal squeezing fraction, the front end $$P_{\text{Hel}}=\frac12\left(1-\sqrt{1-e^{-4N}}\right),$$1 maps the alphabet to $$P_{\text{Hel}}=\frac12\left(1-\sqrt{1-e^{-4N}}\right),$$2 with effective energy

$$P_{\text{Hel}}=\frac12\left(1-\sqrt{1-e^{-4N}}\right),$$3

Its ideal error probability becomes

$$P_{\text{Hel}}=\frac12\left(1-\sqrt{1-e^{-4N}}\right),$$4

remaining within a factor of $$P_{\text{Hel}}=\frac12\left(1-\sqrt{1-e^{-4N}}\right),$$5 of the S-BPSK Helstrom bound across the full energy range and giving an error below $$P_{\text{Hel}}=\frac12\left(1-\sqrt{1-e^{-4N}}\right),$$6 near $$P_{\text{Hel}}=\frac12\left(1-\sqrt{1-e^{-4N}}\right),$$7 (Bai et al., 27 Jan 2026). This suggests a broader interpretation of the Kennedy paradigm: suitably chosen Gaussian front-end operations can convert transmitter-side resources such as squeezing into an effective displacement gain before photon counting.

The multistage QPSK receiver proposed for CV-QKD in turbulent channels uses “CD-Kennedy” in the explicitly conditionally-dynamic sense. A received QPSK coherent state is split into $$P_{\text{Hel}}=\frac12\left(1-\sqrt{1-e^{-4N}}\right),$$8 equal-intensity branches, each branch is displaced by $$P_{\text{Hel}}=\frac12\left(1-\sqrt{1-e^{-4N}}\right),$$9 according to the estimated transmittance $P_e=\frac12 e^{-4N}.$0 and prior-stage MAP decisions, and each displaced branch is measured by a photon-number-resolving detector with Poisson statistics determined by

$P_e=\frac12 e^{-4N}.$1

(Yuan et al., 24 Sep 2025). Three receiver types are defined: Type-I with $P_e=\frac12 e^{-4N}.$2 at every stage, Type-II with a globally optimized common displacement ratio $P_e=\frac12 e^{-4N}.$3, and Type-III with an optimized first-stage ratio $P_e=\frac12 e^{-4N}.$4 followed by full cancellation (Yuan et al., 24 Sep 2025). Numerical results show that all three outperform a classical coherent receiver in both error probability and secret key rate under turbulent channels; Type-II tolerates worse channel conditions in raw error performance, whereas Type-III offers a particularly effective complexity-performance trade-off for SKR (Yuan et al., 24 Sep 2025).

Taken together, these extensions show that the CD-Kennedy receiver is no longer a single narrowly defined device. It is a design family organized around one durable principle—displace one hypothesis toward vacuum and use non-Gaussian counting statistics to decide—which has been adapted to coherent-state communications, single-rail qubits, squeezed-state discrimination, hybrid receivers, and multistage QPSK detection under turbulence (Izumi et al., 2017, Notarnicola et al., 2022, Yuan et al., 24 Sep 2025, Bai et al., 27 Jan 2026).