betaFeedback

WorksComputer Science

Rydberg Atomic Quantum Receiver Empowered Hybrid Non-orthogonal Multiple Access

Abstract

In this paper, we propose a novel uplink Hybrid Non-Orthogonal Multiple Access (NOMA) architecture assisted by Rydberg Atomic Quantum Receivers (RAQRs) to improve energy efficiency by addressing the physical constraints of atomic sensing. We formulate a joint optimization problem encompassing phase measurements and multi-user power allocation to minimize the total transmit power. This problem is particularly challenging due to the non-linear amplitude-only observations of atomic vapor cells. To decomposes the resulting non-convexity, we develop a hierarchical three-layer decomposition optimization framework that systematically decouples the global problem. The inner layer employs a customized Expectation Maximization-Gerchberg-Saxton (EM-GS) algorithm to resolve phase ambiguity and estimate equivalent channel gains. The middle layer derives closed-form optimal power allocation solutions via Karush-Kuhn-Tucker (KKT) conditions. The outer layer utilizes linear Minimum Mean Square Error (MMSE) detection for high-reliability signal recovery. Simulation results demonstrate the superiority of the proposed framework over conventional RF-based Hybrid NOMA. Notably, the atomic architecture achieves approximately a 93% reduction in total power consumption compared to Orthogonal Multiple Access (OMA) baselines while maintaining strict backward compatibility, thereby laying a solid foundation for practical deployment in next-generation green communication networks.

Introduction

The main goals of sixth-generation (6G) wireless networks are to achieve: global coverage, full spectrum utilization, comprehensive application support, and robust security [1], [2]. Within this expanded scope, 6G must support the massive deployment of energy-constrained devices, ranging from low-power sensors to autonomous industrial terminals. However, accommodating such a sheer volume of connections poses a critical bottleneck due to the physical limitations of available bandwidth and the finite battery life of IoT devices. Therefore, a fundamental challenge in realizing this vision lies in designing highly efficient communication frameworks that can serve a massive number of users while rigorously optimizing scarce spectral and energy resources.

To address these constraints, innovations are needed in both access protocols and physical layer hardware. On the protocol front, the rigidity of conventional orthogonal access is increasingly ill-suited for the sporadic, massive nature of IoT traffic [2]. This has driven the shift toward Hybrid NOMA, which integrates non-orthogonal schemes with legacy systems to enhance spectral efficiency and flexibility [3]–[5]. In parallel, the uplink performance of energy-starved devices is often restricted by the thermal noise floor and sensitivity limits of classical radio-frequency (RF) receivers. To detect ultra-weak signals beyond these limits, Rydberg Atomic Quantum Receivers (RAQRs) have emerged as a revolutionary solution in quantum sensing [6]–[8]. Recent research has coalesced around these two pillars: the algorithmic evolution of multiple access flexibility [9]–[11] and the hardware advancement of receiver sensitivity [12]–[14].

Literature Review

Evolution Towards Flexible and Efficient Hybrid Multiple Access

To address the spectral efficiency limitations of OMA, power-domain NOMA was introduced. NOMA enables multiple users to share time-frequency resources via superposition coding and successive interference cancellation (SIC) [3], [15]. While identified as a key enabler for IMT-2030 [16], the rigid implementation of NOMA causes compatibility issues. Consequently, Hybrid NOMA has emerged as a critical evolution, designed to allow the dynamic coexistence of OMA and NOMA within a unified framework [4]. This architecture facilitates opportunistic access, where a weak user transmits in a time slot allocated to a strong user, provided the interference remains manageable thereby enabling adaptive multi-timeslot resource allocation [5], [9], [10].

This flexibility renders Hybrid NOMA highly effective for IoT scenarios characterized by sporadic traffic. Recent studies have optimized protocols to accommodate bursty uplink transmissions [11], focusing on joint resource allocation to maximize sum rates. Furthermore, the integration of Hybrid NOMA with simultaneous wireless information and power transfer (SWIPT) and backscatter communications is being actively explored. These approaches jointly optimize energy harvesting and data transmission, offering a pathway to support battery-free devices [17, 18, 19]. Collectively, these advancements position Hybrid NOMA as a leading candidate for Next-Generation Multiple Access (NGMA) in 6G, balancing high throughput with massive connectivity and energy sustainability.

RAQRs

While advanced multiple access schemes improve spectral utilization, uplink performance for energy-constrained devices remains bounded by receiver sensitivity. Conventional RF chains are inherently limited by thermal noise and finite linearity, creating a bottleneck for ultra-weak signal detection. Quantum sensing offers a paradigm shift to overcome these classical limitations [6, 7].

Among quantum platforms, RAQRs have shown significant potential for wireless communications [20, 21]. By leveraging the large electric dipole moments of Rydberg atoms and Electromagnetically Induced Transparency (EIT) [22], RAQRs convert incident RF signals into optical readouts with quantum-limited noise performance [12]. This mechanism provides distinct advantages, including sensitivity approaching the standard quantum limit, immunity to thermal noise, wideband frequency tunability, and high linearity [14, 23].

The development of RAQRs has progressed rapidly from theoretical physics to practical communication systems. Early experiments successfully demonstrated the reception of amplitude- and frequency-modulated signals using atomic vapors [24, 25, 26]. A significant milestone was the realization of the atomic superheterodyne receiver, which drastically improved sensitivity and validated the technology for precision metrology [27]. Recent comprehensive reviews provide detailed insights into the operational principles of these receivers and their expanding utility in electric field sensing [28, 29, 30].

On the theoretical front, significant progress has been made in bridging quantum physics with communication theory. Equivalent baseband channel models have been established for both single-input single-output (SISO) and multiple-input multiple-output (MIMO) configurations of RAQRs, allowing for the quantification of performance gains over conventional receivers [31, 32]. Crucially, specific detection and estimation algorithms have been developed to address the unique “amplitude-only” observation characteristic of certain RAQR implementations, which inherently results in phase ambiguity [31, 32]. Research has further accelerated into complex communication scenarios, investigating RAQR-specific precoding for MIMO systems [33, 34, 35], hardware-constrained channel estimation [36], and multi-band operation for spectrum aggregation [37, 38]. Additionally, applications in spatial signal processing, such as direction-of-arrival estimation, have highlighted the transition of RAQRs toward practical, ultra-sensitive communication hardware [39, 40].

Motivation and Contribution

Although significant research has advanced NOMA for next-generation networks and explored the high-sensitivity capabilities of RAQRs, their synergistic integration for communication systems remains a major gap. Specifically, leveraging RAQRs’ unique wideband and ultra-high-sensitivity to enhance NOMA’s flexibility and reliability in dynamic scenarios is unexplored [41]. Furthermore, their exceptional detection capability offers a novel physical-layer solution to address the persistent challenge of error propagation in SIC, promising reliable interference management at significantly lower power [8]. Moreover, while energy efficiency is a key NOMA objective, existing strategies focus on algorithmic power allocation. Phase-coherent RAQR arrays offer a novel method to significantly reduce transmit power, which is essential for energy-limited 6G IoT applications [42]. Thus, it is imperative to exploit a NOMA framework harnessing RAQR to simultaneously address flexibility, reliability, and energy efficiency.

To bridge this gap, this paper conceives a novel uplink Hybrid NOMA system assisted by an RAQR at the base station and develops a comprehensive optimization framework. To the best of the authors’ knowledge, the exploitation of RAQR-empowered Hybrid NOMA systems remains largely unexplored.

The main contributions of this paper are summarized as follows:

  • We first propose a RAQR-based Hybrid NOMA uplink network, motivated by the need to enhance system flexibility through dynamic access scheme adaptation and improve energy efficiency in power-constrained scenarios, explicitly incorporating the critical hardware constraint of amplitude-only observations. To evaluate the energy-efficient performance metric under this framework, we formulate the joint optimization problem that encompasses phase alignment, multi-user power allocation, and OMA compatibility requirements to minimize the total transmit power.

  • Next, a hierarchical three-layer solution strategy is proposed to address the non-convexity of the optimization problem while achieving reliable interference cancellation. This method systematically decomposes the intractable global problem into solvable sub-problems: (i) an inner layer for phase retrieval and channel estimation utilizing a customized EM-GS algorithm; (ii) a middle layer for resource allocation that derives closed-form power solutions via Karush-Kuhn-Tucker (KKT) conditions; and (iii) an outer layer for signal detection employing an online MMSE scheme.

  • We validate the superiority of the proposed architecture over conventional RF-based benchmarks through extensive simulations. The results demonstrate significant gains in both sum rate and energy efficiency across diverse channel conditions. This confirms the system’s potential in enabling sustainable, high-performance connectivity for next-generation energy-constrained IoT devices, while leveraging the high receiver sensitivity of RAQR arrays to achieve reliable weak signal detection at reduced transmit power.

The remainder of this paper is summarized as follows. Section II presents the system model. Section III details the proposed three-layer optimization framework. Simulation results are presented in § 5. Finally, the paper is concluded in Section VI.

Notations

For clarity, the following notation is adopted throughout this paper. Italic letters denote variables, bold letters denote vectors. R\mathbb{R} and C\mathbb{C} represent the real and complex number fields, respectively. The operators ∣⋅∣|\cdot| and ∥⋅∥\|\cdot\| denote the element-wise modulus and the Euclidean norm of a vector, respectively. R{⋅}\mathcal{R}\{\cdot\}, J{⋅}\mathcal{J}\{\cdot\}, and ∠(⋅)\angle(\cdot) denote the real part, imaginary part, and phase of a complex variable, respectively. The superscripts (⋅)T(\cdot)^{T} and (⋅)∗(\cdot)^{*} stand for the transpose and the optimal value under current conditions. Moreover, x^\hat{x} and x~\tilde{x} denote the estimated and transformed values of variable xx, while the superscript (⋅)opt(\cdot)^{\mathrm{opt}} refers to the optimally phase-aligned equivalent real-valued channel. Q(⋅)Q(\cdot) represents the Pulse Amplitude Modulation (PAM) demapping operation. Finally, CN(μ,Σ)\mathcal{CN}(\mu,\Sigma) is a circularly symmetric complex Gaussian distribution with mean μ\mu and covariance Σ\Sigma. ∘\circ represents the Hadamard (element-wise) product.

System model and problem statement

System Model

In this section, we consider an uplink network where two users, UkU_k (k∈{1,2}k \in\{1,2\}), are served by a single base station. Detailed physical principles of the RAQR can be found in [31]; here, we focus on the equivalent communication model. As illustrated in Figure 1, the base station replaces the conventional RF front-end with an MM-element RAQR. This array consists of MM vapor cells acting as atomic antennas, each filled with Rydberg atoms and coupled to a dedicated photodetector (PD). The optical readouts from these PDs are aggregated and processed digitally for signal detection.

A two-user Hybrid NOMA with RAQR

Figure 1. A two-user Hybrid NOMA with RAQR

Under this system architecture, the transmission protocol employed by the users is indicated in Figure 2. For comparison, Figure 2 (a) presents the conventional OMA (Time Division Multiple Access, TDMA) baseline, where resources are strictly orthogonal: time slot T1T_1 is dedicated exclusively to U1U_1, and T2T_2 to U2U_2, with each user transmitting at a baseline power POMAP^{\mathrm{OMA}}. While, the Hybrid NOMA framework adopted in this study, based on [10], is illustrated in Figure 2 (b). This scheme enables full-time resource sharing through dynamic role assignment. In time slot T1T_1, U1U_1 functions as the primary (strong) user, while U2U_2 is superimposed as the secondary (weak) user. Conversely, in time slot T2T_2, the roles are reversed, with U2U_2 becoming the primary user and U1U_1 the secondary user. Under this scheme, the transmit power for user kk in time slot tt is variable and denoted by Pk,tP_{k,t}.

A two-user Hybrid NOMA transmission scheme

Figure 2. A two-user Hybrid NOMA transmission scheme

Building upon the aforementioned network architecture and transmission protocol, let us define the complex baseband signal transmitted by user kk in time slot tt as sk,t=skejψks_{k,t}=s_k e^{j\psi_k}, where sks_k is the real-valued PAM information symbol and ψk\psi_k denotes the initial phase. The complex signal vector yt\mathbf{y}_t at the output of the Rydberg atomic quantum receiver (RAQR) is modeled as

yt=∑k=12hk,tsk,t+b+nt,(1)\mathbf{y}_t=\sum_{k=1}^{2}\mathbf{h}_{k,t}s_{k,t}+\mathbf{b}+\mathbf{n}_t, \tag*{(1)}

where hk,t∈CM×1\mathbf{h}_{k,t}\in\mathbb{C}^{M\times1} denotes the channel vector from user UkU_k to the MM-element atomic array, b∈CM×1\mathbf{b}\in\mathbb{C}^{M\times1} is the local-oscillator reference signal, and nt∼CN(0,σ2I)\mathbf{n}_t\sim\mathcal{CN}(0,\sigma^2\mathbf{I}) is the additive noise. Equation (1) represents the equivalent complex baseband model derived from the Rydberg electromagnetically induced transparency (EIT) principle. Under small-signal conditions, the down-converted optical field is a linear superposition of the users’ symbols {sk,t}\{s_{k,t}\} weighted by the channel vectors {hk,t}\{\mathbf{h}_{k,t}\}, plus the deterministic LO term b\mathbf{b}. The photodetector output thus follows the linear model in (1), analogous to a conventional RF front-end. In practical Rydberg receivers, the dominant microscopic noise mechanism is signal-dependent shot noise rather than purely additive Gaussian noise [43], but for analytical tractability we adopt the widely used circularly symmetric complex Gaussian approximation based on the central limit theorem, i.e., nt∼CN(0,σ2I)\mathbf{n}_t\sim\mathcal{CN}(0,\sigma^2\mathbf{I}), which enables the subsequent rate analysis and power optimization.

Unlike the traditional RF receivers that directly measure this complex signal yt\mathbf{y}_t, the RAQR in this work can only measure the amplitude of the received signal yt\mathbf{y}_t. Coherent detection of both amplitude and phase could alternatively be realized by employing a pair of lock-in amplifiers [25], [44]; investigating such receiver architectures for Hybrid NOMA is an interesting direction for future work.

zt=∣yt∣=∣∑k=12hk,tsk,t+b+nt∣∈RM×1,(2)\mathbf{z}_t=|\mathbf{y}_t|=\left|\sum_{k=1}^{2}\mathbf{h}_{k,t}s_{k,t}+\mathbf{b}+\mathbf{n}_t\right|\in\mathbb{R}^{M\times1}, \tag*{(2)}

where the amplitude-only observation ztz_t is directly linked to the Autler–Townes (AT) splitting interval of the underlying Rydberg EIT resonance, since the magnitude of the probe transmission at each atomic element is (after calibration) approximately proportional to the AT splitting induced by the total RF field ∑k=12hk,tsk,t+b+nt\sum_{k=1}^{2} h_{k,t}s_{k,t}+\mathbf{b}+\mathbf{n}_t. This non-linear amplitude relationship poses a significant challenge for signal detection. However, the signal vector ztz_t can be linearized by ensuring phase coherence between the received signal and the strong LO. Specifically, provided that the phase condition ∠(h1,ts1,t+h2,ts2,t)≈∠b\angle(h_{1,t}s_{1,t}+h_{2,t}s_{2,t})\approx\angle\mathbf{b} holds, the signal vector ztz_t is reformulated by the small-signal approximation, which is given by

zt≈∥b∥+h1,topts1,t+h2,topts2,t+n^t,(3)z_t \approx\lVert\mathbf{b}\rVert+h^{\mathrm{opt}}_{1,t}s_{1,t}+h^{\mathrm{opt}}_{2,t}s_{2,t}+\hat{n}_t, \tag*{(3)}

where hk,topth^{\mathrm{opt}}_{k,t} denotes the equivalent real-valued optical channel for user kk in time slot tt, and n^t\hat{n}_t represents the effective noise. It is worthwhile noted that achieving precise phase alignment transforms the EIT/AT-based non-linear observation in (2) into the linear model in (3), facilitating efficient user signal recovery and interference cancellation.

Problem Statement

Following the linearized observation model derived in (3), we define the effective channel gain Gk,t(ψ)G_{k,t}(\psi) as the squared Euclidean norm of the equivalent real-valued channel vector and express it as Gk,t(ψ)=∥hk,topt∥2=∑m=1M(hm,k,topt)2G_{k,t}(\psi)=\lVert h^{\mathrm{opt}}_{k,t}\rVert^2=\sum_{m=1}^{M}(h^{\mathrm{opt}}_{m,k,t})^2, where it quantifies the aggregate signal combining gain provided by the MM-element atomic MIMO array under a specific phase alignment ψ\psi. Since the base station prioritizes decoding the strong user U1_1 in the uplink, the presence of the weak user U2_2 would degrade the signal-to-interference-plus-noise ratio (SINR) of U1_1. To tackle this issue, we aim to prevent U2_2 from interfering with U1_1 in this time slot. To proceed, let us denote the achievable data rate of Uk_k at time slot Tt_t as Rk,tR_{k,t}. Hence, at the time slot T1_1, the two users’ data rates R1,1R_{1,1} and R2,1R_{2,1} are given, respectively, by

R1,1=log⁡2(1+P1,1⋅G1,1(ψ)σ2),R2,1=0.(4)R_{1,1}=\log_2\left(1+\frac{P_{1,1}\cdot G_{1,1}(\psi)}{\sigma^2}\right),\qquad R_{2,1}=0. \tag*{(4)}

In the time slot T2_2, NOMA transmission is employed to enhance system performance. By implementing SIC at the base station, the signal of the strong user U1_1 is decoded first. Thus, the achievable data rates R1,2R_{1,2} and R2,2R_{2,2} are expressed, respectively, as

R1,2=log⁡2(1+P1,2⋅G1,2(ψ)P2,2⋅G2,2(ψ)+σ2),(5)R_{1,2}=\log_2\left(1+\frac{P_{1,2}\cdot G_{1,2}(\psi)}{P_{2,2}\cdot G_{2,2}(\psi)+\sigma^2}\right), \tag*{(5)}
R2,2=log⁡2(1+P2,2⋅G2,2(ψ)σ2).(6)R_{2,2}=\log_2\left(1+\frac{P_{2,2}\cdot G_{2,2}(\psi)}{\sigma^2}\right). \tag*{(6)}

As indicated by (4) and (5), the legacy users achieve data rates identical to those of the OMA baseline within their respective allocated slots, thereby strictly guaranteeing backward compatibility. Consequently, the total achievable data rate for U1_1 is given by R1=R1,1+R1,2R_1=R_{1,1}+R_{1,2}, whereas U2_2 transmits exclusively in the second time slot, yielding R2=R2,2R_2=R_{2,2}. Furthermore, RR denotes the system target rate, which is determined by the specific quality-of-service (QoS) requirements.

On the basis of the aforementioned discussion, we formulate the optimization problem by minimizing the total power consumption of both users, subject to the achievable data rate requirement of each user as well as the Frobenius norm constraint of the composite effective channel phase at each time slot. Mathematically, this optimization problem can be expressed as follows:

min⁡Pk,t≥0,ψ∑k=12∑t=12Pk,t(7)\min_{P_{k,t}\geq0,\psi}\quad\sum_{k=1}^{2}\sum_{t=1}^{2}P_{k,t} \tag*{(7)}
s.t.R1≥R,(8)\mathrm{s.t.}\quad R_1\geq R, \tag*{(8)}
R2≥R,(9)R_2\geq R, \tag*{(9)}
∥ℑ{Hteq(ψ)}∥F2≤ϵ,∀t.(10)\left\lVert\Im\left\{\mathbf{H}^{\mathrm{eq}}_t(\psi)\right\}\right\rVert_F^2\leq\epsilon,\quad\forall t. \tag*{(10)}

where the effective channel gain, denoted as Gk,t(ψ)=∥ℜ{hk,t∘ejψ}∥2G_{k,t}(\psi)=\left\lVert\Re\left\{h_{k,t}\circ e^{j\psi}\right\}\right\rVert^2, explicitly depends on the phase vector ψ\psi. Also, the composite effective channel matrix is defined as Hteq(ψ)=[h1,tejψ1,h2,tejψ2]\mathbf{H}^{\mathrm{eq}}_t(\psi)=[h_{1,t}e^{j\psi_1},h_{2,t}e^{j\psi_2}]. In problem (6), the objective function (6a) is the total transmit power consumption; the constraints (6b) and (6b) guarantee that the achievable data rate of each user exceeds the target rate RR; the constraint (6d) guarantees phase alignment by restricting the energy of the residual imaginary component to be below a small tolerance ϵ>0\epsilon>0, thereby ensuring the validity of the real-valued approximation utilized in the linear observation model. The coupled variables of transmit power Pk,tP_{k,t} and phase alignment ϕ\phi leads to the non-convexity of problem (6). To tackle this non-convex issue, in the following, we propose a hierarchical three-layer decomposition framework to address (6), decoupling it into three tractable sub-problems, resolved in a sequential and iterative manner.

Proposed Three-Layer Decomposition Framework

To resolve the computational intractability caused by the non-convexity and strong variable coupling among optimization variablesidentified in the problem formulation, we propose a hierarchical three-layer decomposition framework, which is described as follows:

  • Inner Layer: Phase Alignment and Channel Estimation. Serving as the physical layer foundation, this module tackles the hardware-induced phase ambiguity. Specifically, for a given power allocation set, a tailored EM-GS algorithm is developed to jointly estimate the user symbols and optimize the receiver’s phase compensation vector ψ\psi. Consequently, the derived effective real-valued channel gain, Gk,tG_{k,t}, satisfies the linearization constraint in (3), thereby furnishing accurate channel state information (CSI) for the upper layers.

  • Middle Layer: Power Allocation. This layer serves as the core resource management component. By utilizing the phase-calibrated, real-valued effective channel gains provided by the Inner Layer, the originally non-convex power allocation problem is transformed into a convex formulation. We solve this efficiently using a Lagrangian multiplier method derived from the KKT conditions. This yields a closed-form solution that dynamically allocates transmit power Pk,tP_{k,t} to minimize total system consumption while strictly satisfying the target rate requirements RR.

Serving as the core for resource management, this layer leverages the phase-calibrated effective channel gains derived from the Inner Layer. Consequently, the originally non-convex power allocation problem is reformulated as a convex optimization problem. By analyzing the KKT conditions, a closed-form solution is derived to efficiently solve the problem. This solution dynamically allocates the transmit power Pk,tP_{k,t} to minimize total power consumption, while strictly satisfying the target rate requirements RR.

  • Outer Layer: Online Signal Detection. This layer is responsible for the real-time system operation after the Inner and Middle layers have optimized phase alignment and power allocation. It uses a simplified linear channel model and employs low-complexity linear MMSE detection to reliably recover multi-user signals, without resorting to complex non-linear decoding.

Inner-Layer: Phase Alignment and Channel Estimation

As delineated in the system model, the RAQR is constrained to acquire amplitude-only observations, denoted as ztz_t, thereby introducing inherent phase ambiguity. Consequently, the primary objective of the Inner Layer is to mitigate this hardware-induced limitation and furnish the precise, effective real-valued CSI necessitating for the subsequent power allocation layer. Specifically, given a fixed transmit power allocation set {Pk,t}\{P_{k,t}\}, a customized EM-GS algorithm is used to jointly estimate the user transmitted signals, sk,ts_{k,t}, and optimize the receiver’s phase compensation vector, ψ\psi.

Based on the observation model in (2), let st′≜[s1ejψ1,s2ejψ2]T\mathbf{s}'_t \triangleq[s_1e^{j\psi_1},s_2e^{j\psi_2}]^T denote the equivalent phase-shifted symbol vector, and At≜[h1,t,h2,t]\mathbf{A}_t \triangleq[\mathbf{h}_{1,t},\mathbf{h}_{2,t}] represent the aggregated channel matrix. Consequently, the observation equation can be recast in a compact vector form as follows:

zt=∣Atst′+b+nt∣.(11)\mathbf{z}_t = \left|\mathbf{A}_t\mathbf{s}'_t+\mathbf{b}+\mathbf{n}_t\right|. \tag*{(11)}

It is noteworthy that this formulation is mathematically isomorphic to the biased phase retrieval model established in [31]. Consequently, symbol detection in our setting can be regarded as a phase-retrieval problem, where the unknown symbol phases are treated as latent variables and the complex vector st′\mathbf{s}'_t is reconstructed solely from magnitude-based measurements.

To address the non-convexity inherent in this problem, the EM-GS algorithm introduces a set of latent variables defined as the unobserved phase vector of the received signal at the antenna array, denoted by θt=[θ1,t,…,θM,t]T∈RM\boldsymbol{\theta}_t=[\theta_{1,t},\ldots,\theta_{M,t}]^T\in\mathbb{R}^M. The underlying rationale is that, provided the received phases θt\boldsymbol{\theta}_t are known, the complex signal recovery task reduces to a computationally tractable linear Gaussian estimation problem. The algorithm proceeds by alternating between two steps: (1) the Expectation (E)-step, which computes the posterior statistics of the latent phase vector θt\boldsymbol{\theta}_t conditioned on the current parameter estimates; and (2) the Maximization (M)-step, which updates st′\mathbf{s}'_t (and by extension ψ\psi) to maximize the expected log-likelihood function. This iterative procedure effectively retrieves the missing phase information, thereby facilitating the linearization of the effective channel model.

E-step: Phase posterior expectation calculation

In the ii-th iteration, conditioned on the current symbol estimates s′(i)\mathbf{s}'^{(i)}, the conditional probability density function of the observation zm,tz_{m,t} is characterized by a Rician distribution:

p(zm,t∣s′(i))=2zm,tσ2exp⁡(−zm,t2+∣λm,t(i)∣2σ2)I0(2zm,t∣λm,t(i)∣σ2),(12)p(z_{m,t}|\mathbf{s}'^{(i)}) = \frac{2z_{m,t}}{\sigma^2}\exp\left(-\frac{z_{m,t}^2+|\lambda_{m,t}^{(i)}|^2}{\sigma^2}\right)I_0\left(\frac{2z_{m,t}|\lambda_{m,t}^{(i)}|}{\sigma^2}\right), \tag*{(12)}

where λm,t(i)=am,tTs′(i)+bm\lambda_{m,t}^{(i)}=\mathbf{a}_{m,t}^T\mathbf{s}'^{(i)}+b_m, am,tT\mathbf{a}_{m,t}^T is the mm-th row of A\mathbf{A} in time slot tt, I0(⋅)I_0(\cdot) is the modified Bessel function of the first kind of order zero.

Accordingly, the posterior distribution of the latent phase θm,t\theta_{m,t} follows a von Mises distribution, given by:

p(θm,t∣zm,t,s′(i))=exp⁡(κm,t(i)cos⁡(θm,t−ϑm,t(i)))2πI0(κm,t(i)),(13)p(\theta_{m,t}|z_{m,t},\mathbf{s}'^{(i)})= \frac{\exp\left(\kappa_{m,t}^{(i)}\cos(\theta_{m,t}-\vartheta_{m,t}^{(i)})\right)} {2\pi I_0(\kappa_{m,t}^{(i)})}, \tag*{(13)}

where ϑm,t(i)=∠λm,t(i)\vartheta_{m,t}^{(i)}=\angle\lambda_{m,t}^{(i)} represents the mean direction (or phase center) under the current estimate. The term κm,t(i)=2zm,t∣λm,t(i)∣σ2\kappa_{m,t}^{(i)}=\frac{2z_{m,t}|\lambda_{m,t}^{(i)}|}{\sigma^2} denotes the concentration parameter, which quantifies the reliability of the phase estimation, while λm,t(i)=am,tTs′(i)+bm\lambda_{m,t}^{(i)}=\mathbf{a}_{m,t}^T\mathbf{s}'^{(i)}+b_m corresponds to the estimated noiseless complex signal.

M-step: Update of Symbol Estimates

Subsequently, the symbol estimates are updated by maximizing the expected complete-data log-likelihood function. Solving this optimization problem yields the following closed-form expression:

s′(i+1)=(AtTAt)−1AtT(zt∘ejϑt(i)∘R(κt(i))−b),(14)\mathbf{s}'^{(i+1)}=(\mathbf{A}_t^T\mathbf{A}_t)^{-1}\mathbf{A}_t^T\left(\mathbf{z}_t\circ e^{j\boldsymbol{\vartheta}_t^{(i)}}\circ R(\boldsymbol{\kappa}_t^{(i)})-\mathbf{b}\right), \tag*{(14)}

where R(x)=I1(x)/I0(x)R(x)=I_1(x)/I_0(x) acts as a Signal-to-Noise Ratio (SNR)-dependent high-pass filter, and the operator ∘\circ denotes the element-wise Hadamard product.

Gradient derivation of phase updates

With the objective of directly optimizing the phase compensation vector ψ\psi to eliminate channel phase distortion, an explicit gradient descent-based update step is incorporated subsequent to each M-step of the EM-GS algorithm. To this end, the phase alignment cost function is formulated as the aggregate power of the imaginary components of the equivalent channel:

J(ψ)=∑m=1M∑k=12∣ℑ{hm,kejψk}∣2.(15)J(\psi)=\sum_{m=1}^{M}\sum_{k=1}^{2}\left|\Im\{h_{m,k}e^{j\psi_k}\}\right|^2. \tag*{(15)}

Minimizing the objective function in (11) inherently drives the equivalent channel coefficients hm,kejψkh_{m,k}e^{j\psi_k} towards the real domain, thereby effectively mitigating phase ambiguity. To facilitate the subsequent gradient derivation, we decompose the channel gain as hm,k=am,k+jbm,kh_{m,k}=a_{m,k}+jb_{m,k}. By applying Euler’s formula, the real and imaginary components of the phase-compensated channel can be explicitly defined as:

ℜ{hm,kejψk}=am,kcos⁡ψk−bm,ksin⁡ψk≜Rm,k,(16)\Re\{h_{m,k}e^{j\psi_k}\}=a_{m,k}\cos\psi_k-b_{m,k}\sin\psi_k\triangleq R_{m,k}, \tag*{(16)}
ℑ{hm,kejψk}=am,ksin⁡ψk+bm,kcos⁡ψk≜Im,k,(17)\Im\{h_{m,k}e^{j\psi_k}\}=a_{m,k}\sin\psi_k+b_{m,k}\cos\psi_k\triangleq I_{m,k}, \tag*{(17)}

where Rm,kR_{m,k} and Im,kI_{m,k} denote the real and imaginary terms, respectively.

Leveraging the definitions in (13), the gradient of the cost function can be derived via the chain rule. A critical observation is that the derivative of the imaginary component with respect to the phase parameter yields the real component, i.e., ∂Im,k∂ψk=Rm,k\frac{\partial I_{m,k}}{\partial\psi_k}=R_{m,k}. Consequently, the partial derivative of the quadratic term simplifies to:

∂(Im,k2)∂ψk=2Im,k⋅∂Im,k∂ψk=2⋅ℑ{hm,kejψk}⏟Im,k⋅ℜ{hm,kejψk}⏟Rm,k=2Im,kRm,k.\frac{\partial\left(I_{m,k}^{2}\right)}{\partial\psi_k}=2I_{m,k}\cdot\frac{\partial I_{m,k}}{\partial\psi_k}=2\cdot\underbrace{\Im\{h_{m,k}e^{j\psi_k}\}}_{I_{m,k}}\cdot\underbrace{\Re\{h_{m,k}e^{j\psi_k}\}}_{R_{m,k}}=2I_{m,k}R_{m,k}.

Aggregating the partial derivatives for each user k∈{1,2}k\in\{1,2\}, the gradient of the cost function J(ψ)J(\psi) can be compactly expressed in vector form as:

∇J(ψ)=[∂J∂ψ1,∂J∂ψ2]T=2[∑m=1Mℑ{hm,1ejψ1}ℜ{hm,1ejψ1}∑m=1Mℑ{hm,2ejψ2}ℜ{hm,2ejψ2}].(18)\nabla J(\psi)=\begin{bmatrix}\frac{\partial J}{\partial\psi_1},\frac{\partial J}{\partial\psi_2}\end{bmatrix}^{T} =2\begin{bmatrix}\sum_{m=1}^{M}\Im\{h_{m,1}e^{j\psi_1}\}\Re\{h_{m,1}e^{j\psi_1}\}\\[6pt]\sum_{m=1}^{M}\Im\{h_{m,2}e^{j\psi_2}\}\Re\{h_{m,2}e^{j\psi_2}\}\end{bmatrix}. \tag*{(18)}

The availability of this analytical gradient facilitates an efficient update using the gradient descent method. The iterative update rule is given by:

ψ(i+1)=ψ(i)−μ∇J(ψ(i)),(19)\psi^{(i+1)}=\psi^{(i)}-\mu\nabla J(\psi^{(i)}), \tag*{(19)}

where ii denotes the iteration index, and μ>0\mu>0 represents the step size (or learning rate), which governs the convergence speed.

In summary, the proposed inner-layer optimization framework achieves joint estimation of the user symbols sts_t and phase parameters ψ\psi by orchestrating the EM-GS algorithm with the gradient descent updates. The resulting optimal phase vector ψ∗\psi^* and the corresponding equivalent real channel gains, calculated as Gk,t=∥ℜ{hk,t∘ejψ∗}∥2G_{k,t}=\|\Re\{h_{k,t}\circ e^{j\psi^*}\}\|^2, are then fed as precise CSI into the subsequent outer-layer power allocation module.

Middle-Layer: Power Allocation

Once the inner layer converges, the system obtains the optimal phase compensation ψ∗\psi^* and the corresponding equivalent real channel gain Gk,t=∥ℜ{hk,t∘ejψ∗}∥2G_{k,t}=\|\Re\{h_{k,t}\circ e^{j\psi^*}\}\|^2. At this point, we need to determine whether the outer layer power allocation problem is convex. In the convexity proof, we will write the optimization problem as follows:

min⁡P1,1,P1,2,P2,2Ptotal=P1,1+P1,2+P2,2s.t.Rtarget−[log⁡2(1+P1,1G1,1σ2)+log⁡2(1+P1,2G1,2P2,2G2,2+σ2)]≤0,Rtarget−log⁡2(1+P2,2G2,2σ2)≤0,−P1,1≤0,−P1,2≤0,−P2,2≤0,P1,1−P1,max⁡≤0,P1,2−P1,max⁡≤0,P2,2−P2,max⁡≤0.(20)\begin{aligned} \min_{P_{1,1},P_{1,2},P_{2,2}}\quad&P_{\mathrm{total}}=P_{1,1}+P_{1,2}+P_{2,2}\\ \text{s.t.}\quad&R^{\mathrm{target}}-\left[\log_2\left(1+\frac{P_{1,1}G_{1,1}}{\sigma^2}\right)+\log_2\left(1+\frac{P_{1,2}G_{1,2}}{P_{2,2}G_{2,2}+\sigma^2}\right)\right]\leq0,\\ &R^{\mathrm{target}}-\log_2\left(1+\frac{P_{2,2}G_{2,2}}{\sigma^2}\right)\leq0,\\ &-P_{1,1}\leq0,\quad-P_{1,2}\leq0,\quad-P_{2,2}\leq0,\\ &P_{1,1}-P_{1,\max}\leq0,\quad P_{1,2}-P_{1,\max}\leq0,\\ &P_{2,2}-P_{2,\max}\leq0. \tag*{(20)} \end{aligned}

The feasibility of the optimization problem is subject to the non-negativity and maximum power limits specified in (16d) and (16e). These linear inequalities define a convex polyhedral set. Furthermore, since the objective function in (16a) and the constraints in (16d)–(16e) are affine, establishing the convexity of the problem reduces to analyzing the rate constraints.

Consider the constraint function for user 2 in (16c), denoted as:

fC2(P)=Rtarget−log⁡2(1+P2,2G2,2σ2).(21)f_{C2}(\mathbf{P})=R^{\mathrm{target}}-\log_2\left(1+\frac{P_{2,2}G_{2,2}}{\sigma^2}\right). \tag*{(21)}

Let g(P)=−log⁡2(1+aP)g(P)=-\log_2(1+aP) with a=G2,2/σ2>0a=G_{2,2}/\sigma^2>0. The convexity of fC2f_{C2} hinges on the convexity of g(P)g(P). By examining the second-order derivative:

g′′(P)=a2ln⁡(2)(1+aP)2>0,(22)g''(P)=\frac{a^2}{\ln(2)(1+aP)^2}>0, \tag*{(22)}

we confirm that g(P)g(P) is strictly convex for P≥0P\geq0, thereby proving the convexity of constraint (16c).

However, the constraint function fC1f_{C1} associated with (16b) is shown on top of the next page that expressed as: is non-convex due to the coupling of variables P1,2P_{1,2} and P2,2P_{2,2} within the term h2h_2. Consequently, the overall optimization problem is generally non-convex. To address this, we exploit the monotonicity of the objective function and the rate constraints. Since minimizing the total power requires the rate constraints to be active (i.e., satisfied with equality), the optimal solution must lie on the boundary of the feasible region. By converting the inequality constraint (16c) into an equality, we can directly derive the optimal power allocation for user 2, P2,2∗P_{2,2}^*, as:

log⁡2(1+P2,2G2,2σ2)=Rtarget⇒P2,2∗=σ2(2Rtarget−1)G2,2.(23)\log_2\left(1+\frac{P_{2,2}G_{2,2}}{\sigma^2}\right)=R^{\mathrm{target}}\Rightarrow P_{2,2}^*=\frac{\sigma^2(2^{R^{\mathrm{target}}}-1)}{G_{2,2}}. \tag*{(23)}

Substituting the optimal P2,2∗P_{2,2}^* into the original formulation reduces the problem to determining the optimal power distribution for User 1. The rate constraint fC1f_{C1} decomposes into a sum of two univariate logarithmic functions with respect to P1,1P_{1,1} and P1,2P_{1,2}. Since the sum of concave functions preserves concavity, the feasible region defined by the rate requirement is convex. Consequently, minimizing the affine objective function (total power) over this convex set constitutes a standard convex optimization problem.

Assuming the feasible region is non-empty (satisfying Slater’s condition), the KKT conditions provide necessary and sufficient criteria for optimality. The rate constraint (19) is rewritten as:

log⁡2(1+P1,1G1,1σ2)+log⁡2(1+P1,2G1,2I)≥Rtarget,\log_2\left(1+\frac{P_{1,1}G_{1,1}}{\sigma^2}\right)+\log_2\left(1+\frac{P_{1,2}G_{1,2}}{I}\right)\geq R^{\mathrm{target}},

where I=P2,2∗G2,2+σ2I=P_{2,2}^*G_{2,2}+\sigma^2 represents the constant aggregate noise-plus-interference term. The Lagrangian function is formulated as:

L(P1,1,P1,2,λ)=P1,1+P1,2−λ[log⁡2(1+P1,1G1,1σ2)+log⁡2(1+P1,2G1,2I)−Rtarget].(24)\mathcal{L}(P_{1,1},P_{1,2},\lambda)=P_{1,1}+P_{1,2}-\lambda\left[\log_2\left(1+\frac{P_{1,1}G_{1,1}}{\sigma^2}\right)+\log_2\left(1+\frac{P_{1,2}G_{1,2}}{I}\right)-R^{\mathrm{target}}\right]. \tag*{(24)}

Applying the stationarity condition of the KKT system, we set the partial derivatives of the Lagrangian with respect to P1,1P_{1,1} and P1,2P_{1,2} to zero, yielding the following closed-form expressions:

P1,1=λln⁡2−σ2G1,1,P1,2=λln⁡2−IG1,2.(25)P_{1,1}=\frac{\lambda}{\ln2}-\frac{\sigma^2}{G_{1,1}},\qquad P_{1,2}=\frac{\lambda}{\ln2}-\frac{I}{G_{1,2}}. \tag*{(25)}

Invoking the complementary slackness condition from the KKT framework, we observe that the optimal solution typically requires the rate constraint to be active, implying that the inequality holds with equality. Consequently, we have:

log⁡2(1+P1,1G1,1σ2)+log⁡2(1+P1,2G1,2I)=Rtarget.\log_2\left(1+\frac{P_{1,1}G_{1,1}}{\sigma^2}\right)+\log_2\left(1+\frac{P_{1,2}G_{1,2}}{I}\right)=R^{\mathrm{target}}.

Substituting the stationarity expressions from (22) into this equality allows us to solve for the optimal Lagrange multiplier

fC1(P)=Rtarget−[log⁡2(1+P1,1G1,1σ2)⏟h1(P1,1)+log⁡2(1+P1,2G1,2P2,2G2,2+σ2)⏟h2(P1,2,P2,2)](26)f_{C_1}(\mathbf{P})=R^{\mathrm{target}}-\left[\underbrace{\log_2\left(1+\frac{P_{1,1}G_{1,1}}{\sigma^2}\right)}_{h_1(P_{1,1})}+\underbrace{\log_2\left(1+\frac{P_{1,2}G_{1,2}}{P_{2,2}G_{2,2}+\sigma^2}\right)}_{h_2(P_{1,2},P_{2,2})}\right] \tag*{(26)}
λ∗=ln⁡2⋅2Rtarget/2⋅σ2IG1,1G1,2.(27)\lambda^*=\ln2\cdot2^{R^{\mathrm{target}}/2}\cdot\sqrt{\frac{\sigma^2 I}{G_{1,1}G_{1,2}}}. \tag*{(27)}

Recalling from the optimization of User 2 that I=P2,2∗G2,2+σ2=σ22RtargetI=P^*_{2,2}G_{2,2}+\sigma^2=\sigma^2 2^{R^{\mathrm{target}}}, we substitute this relationship and λ∗\lambda^* back into (22). This yields the final closed-form power allocation for User 1:

P1,1∗=σ2(2RtargetG1,1G1,2−1G1,1),(28)P^*_{1,1}=\sigma^2\left(\frac{2^{R^{\mathrm{target}}}}{\sqrt{G_{1,1}G_{1,2}}}-\frac{1}{G_{1,1}}\right), \tag*{(28)}
P1,2∗=σ2(2RtargetG1,1G1,2−2RtargetG1,2).(29)P^*_{1,2}=\sigma^2\left(\frac{2^{R^{\mathrm{target}}}}{\sqrt{G_{1,1}G_{1,2}}}-\frac{2^{R^{\mathrm{target}}}}{G_{1,2}}\right). \tag*{(29)}

Outer-Layer: Online Signal Detection

Upon the convergence of the alternating optimization between the inner and middle layers, the system establishes a steady-state operating point characterized by the optimal phase compensation vector ψ∗=[ψ1∗,ψ2∗]T\boldsymbol{\psi}^*=[\psi_1^*,\psi_2^*]^T and the optimal power allocation set {P1,1∗,P1,2∗,P2,2∗}\{P^*_{1,1},P^*_{1,2},P^*_{2,2}\}. With these parameters fixed, the objective of the third layer is to execute high-throughput, real-time detection of the transmitted data payload. In contrast to the iterative estimation required in the Inner Layer, this stage requires a low-complexity, non-iterative solution suitable for continuous data streams.

Phase Pre-compensation

1) Phase Pre-compensation: Utilizing the optimized phase vector ψ∗\boldsymbol{\psi}^*, phase compensation is applied to the complex baseband signal yty_t received in each time slot tt. This operation aligns the channel phase with that of the LO signal b\mathbf{b}, yielding the rotated signal:

y~t=yt∘e−jψ∗.(30)\widetilde{\mathbf{y}}_t=\mathbf{y}_t\circ e^{-j\boldsymbol{\psi}^*}. \tag*{(30)}

Signal Linearization and Detection

2) Signal Linearization and Detection: Following phase alignment, the signal structure simplifies to the form presented in (3). By subtracting the known amplitude of the LO signal, we establish an approximate linear real-valued observation model:

z^t=∣y~t∣−∣b∣≈Htoptsk,t+n~t,(31)\widehat{\mathbf{z}}_t=\lvert\widetilde{\mathbf{y}}_t\rvert-\lvert\mathbf{b}\rvert\approx\mathbf{H}^{\mathrm{opt}}_t\mathbf{s}_{k,t}+\widetilde{\mathbf{n}}_t, \tag*{(31)}

where Htopt=Re{Ht∘ejψ∗}∈RM×2\mathbf{H}^{\mathrm{opt}}_t=\mathop{\mathrm{Re}}\{\mathbf{H}_t\circ e^{j\boldsymbol{\psi}^*}\}\in\mathbb{R}^{M\times2} denotes the known optimal real-valued equivalent channel matrix.

Linear MMSE Detection

3) Linear MMSE Detection: Leveraging the linear real-valued observation model established in (26), standard linear estimation techniques becomes applicable. We employ a Linear MMSE detector, chosen for its robustness in interference-limited regimes. The LMMSE detector minimizes the mean square error between the transmitted symbols and their estimates, effectively balancing Gaussian noise suppression and the mitigation of residual inter-user interference inherent in the Hybrid NOMA protocol. The estimated symbol vector is given by:

\begin{bmatrix}\widehat{s}_{1,t}\\\widehat{s}_{2,t}\end{bmatrix}_{\mathrm{MMSE}}=\left((\mathbf{H}^{\mathrm{opt}}_t)^{T}\mathbf{H}^{\mathrm{opt}}_t+\sigma^2\mathbf{I}_2\right)^{-1}(\mathbf{H}^{\mathrm{opt}}_t)^{T}\widehat{\mathbf{z}}_t. \tag*{(32)}\end{bmatrix}

Hard Decision

4) Hard Decision: The final demodulation stage involves mapping the continuous soft estimates s^k,t\widehat{s}_{k,t} obtained from the MMSE detector to the nearest constellation points in the modulation scheme. This is achieved via a euclidean distance minimization, mathematically represented as:

s‾k,t=Q(s^k,t),(33)\overline{s}_{k,t}=\mathcal{Q}(\widehat{s}_{k,t}), \tag*{(33)}

where Q(⋅)\mathcal{Q}(\cdot) represents the PAM demapping operation.

Here is the complete process of the three-layer alternating optimization framework, where ε>0\varepsilon>0 is a very small value to denote the convergence threshold. This three-layer alternating optimization algorithm can be summarized in Algorithm 1.

Algorithm 1: Three-layer alternating optimization for atomic-receiver-assisted Hybrid NOMA

  1. Given: Initial power allocation P(0)={Pk,t(0)}\mathbf{P}^{(0)}=\{P_{k,t}^{(0)}\}, phase vector ψ(0)\boldsymbol{\psi}^{(0)}, tolerance δ>0\delta>0, maximum outer iterations Nmax⁡N_{\max}. a) Set n←0n\leftarrow0, Ptotal(0)←∑k,tPk,t(0)P_{\mathrm{total}}^{(0)}\leftarrow\sum_{k,t}P_{k,t}^{(0)}. b) Repeat: optimize ψ\boldsymbol{\psi} and P\mathbf{P} in an alternating manner. i) Inner-Layer: Phase Alignment and Channel Estimation

    Step 1: Fix P(n)\mathbf{P}^{(n)}, we find the optimal phase vector ψ(n+1)\boldsymbol{\psi}^{(n+1)} by using the EM-GS algorithm until convergence.

    Step 2: Compute equivalent channel gains: Gk,t(n+1)=∥Re{hk,t∘ejψ(n+1)}∥2G_{k,t}^{(n+1)}=\lVert\mathop{\mathrm{Re}}\{\mathbf{h}_{k,t}\circ e^{j\boldsymbol{\psi}^{(n+1)}}\}\rVert^2.

    ii) Middle-Layer: Power Allocation

    Step 1: Fix Gk,t(n+1)G_{k,t}^{(n+1)}, we find the optimal power allocation P(n+1)\mathbf{P}^{(n+1)} by solving the convex problem via KKT conditions.

    Step 2: Compute total power: Ptotal(n+1)=∑k,tPk,t(n+1)P_{\mathrm{total}}^{(n+1)}=\sum_{k,t}P_{k,t}^{(n+1)}.

    iii) Convergence Check

    Step 1: Compute relative change: Δ=∣Ptotal(n+1)−Ptotal(n)∣/Ptotal(n)\Delta=\lvert P_{\mathrm{total}}^{(n+1)}-P_{\mathrm{total}}^{(n)}\rvert/P_{\mathrm{total}}^{(n)}.

    Step 2: n←n+1n\leftarrow n+1.

    c) Until Δ<δ\Delta<\delta or n≥Nmax⁡n\geq N_{\max}.

  2. Outer-Layer: Online Signal Detection a) Phase Pre-compensation: Apply optimal phase ψ∗=ψ(n)\boldsymbol{\psi}^*=\boldsymbol{\psi}^{(n)} to received signal yt\mathbf{y}_t: y~t=yt∘e−jψ∗\widetilde{\mathbf{y}}_t=\mathbf{y}_t\circ e^{-j\boldsymbol{\psi}^*}. b) Amplitude Extraction: Obtain linear real observation: z^t=∣y~t∣−∣b∣\widehat{\mathbf{z}}_t=\lvert\widetilde{\mathbf{y}}_t\rvert-\lvert\mathbf{b}\rvert. c) Linear MMSE Detection: Compute symbol estimates: s^t=((Htopt)THtopt+σ2I2)−1(Htopt)Tz^t\widehat{\mathbf{s}}_t=((\mathbf{H}^{\mathrm{opt}}_t)^T\mathbf{H}^{\mathrm{opt}}_t+\sigma^2\mathbf{I}_2)^{-1}(\mathbf{H}^{\mathrm{opt}}_t)^T\widehat{\mathbf{z}}_t, where Htopt=Re{Ht∘ejψ∗}\mathbf{H}^{\mathrm{opt}}_t=\mathop{\mathrm{Re}}\{\mathbf{H}_t\circ e^{j\boldsymbol{\psi}^*}\}. d) Symbol Decision: Demodulate via PAM: s‾k,t=Q(s^k,t)\overline{s}_{k,t}=\mathcal{Q}(\widehat{s}_{k,t}), ∀k,t\forall k,t.

  1. Output:

  • Optimal phase: ψ∗=ψ(n)\psi^* = \psi^{(n)}

  • Optimal power: P∗=P(n)P^* = P^{(n)}

  • Equivalent channels: Gk,t∗G^*_{k,t}

  • Detected symbols: {sˉk,t}\{\bar{s}_{k,t}\} for all k,tk,t

Complexity and Convergence Analysis

To validate the practical feasibility and theoretical reliability of the proposed three-layer framework, we present a rigorous analysis of its computational complexity and convergence behavior.

Computational Complexity Analysis

  1. Computational Complexity Analysis: The computational burden is distributed across the three hierarchical layers. We analyze the complexity of each layer and the overall system as follows:

  • Inner Layer (Phase Alignment and Channel Estimation): The computational cost here is dominated by the Expectation-Step in the EM-GS algorithm and the gradient calculation for phase updates. Given the system configuration, these operations involve matrix–vector multiplications that scale linearly with the number of atomic antennas MM. Specifically, for a fixed number of users, the complexity per inner iteration is O(M)O(M). Assuming the inner layer requires IinI_{\mathrm{in}} iterations to converge, the cumulative complexity is O(Iin⋅M)O(I_{\mathrm{in}} \cdot M).

  • Middle Layer (Power Allocation): This layer leverages the derived analytical closed-form solutions (as seen in (20) and (24)). Since these computations involve only scalar arithmetic operations and elementary functions (logarithms/exponentials), the complexity is independent of the array size MM, i.e., O(1)O(1). Consequently, the computational overhead of the middle layer is negligible compared to the inner layer.

  • Outer Layer (Online Signal Detection): Once the parameters are optimized, the outer layer performs phase pre-compensation and linear MMSE detection. The pre-compensation involves element-wise vector operations with complexity O(M)O(M). The MMSE detection can be written as a multiplication between a 1×M1 \times M detection row vector and an M×MM \times M matrix, yielding a 1×M1 \times M output vector. Each of the MM output entries is a dot product of length MM, resulting in M⋅M=M2M \cdot M = M^2 multiplications and M⋅(M−1)=M2−MM \cdot(M - 1) = M^2 - M additions. Therefore, the overall online detection complexity is dominated by this operation and scales quadratically with the array size, i.e., O(M2)O(M^2).

  • Total Optimization Complexity: The joint optimization executes an alternating process between the Inner and Middle layers. Denoting the number of iterations between these two layers required for convergence as IloopI_{\mathrm{loop}}, the total worst-case computational complexity for the optimization phase is O(Iloop⋅Iin⋅M)O(I_{\mathrm{loop}} \cdot I_{\mathrm{in}} \cdot M), while the subsequent online detection stage incurs an additional per-symbol complexity of O(M2)O(M^2).

Comparison with Benchmarks: The proposed framework demonstrates a significant efficiency advantage. Its complexity scales linearly with the number of atomic antennas (O(M)O(M)). In stark contrast, standard Semi-Definite Programming (SDP) solvers typically exhibit a polynomial complexity of at least O(M3.5)O(M^{3.5}), while exhaustive search methods suffer from exponential scaling. This linear scaling characteristic renders the proposed approach highly suitable for large-scale RAQR systems.

Convergence Analysis

  1. Convergence Analysis: The optimization process within the framework operates effectively as a Block Coordinate Descent (BCD) method iterating between the Inner and Middle layers. We analyze the convergence properties as follows:

  • Inner Layer Convergence: The convergence is underpinned by the theoretical properties of the EM algorithm and gradient descent. The EM-GS component follows the monotonic likelihood ascent principle, guaranteeing that the expected complete-data log-likelihood is non-decreasing in each step. Simultaneously, the gradient descent for phase alignment minimizes the residual imaginary energy. With an appropriately chosen step size μ\mu, this generates a bounded, non-increasing sequence of the cost function values. By the Monotone Convergence Theorem, the inner layer is guaranteed to converge to a stationary point.

  • Global Convergence (Inner-Middle Interaction): The alternating optimization iteratively updates two blocks of variables: the phase/channel parameters (Inner Layer) and the power allocation (Middle Layer). The mechanism is as follows: (i) The Inner Layer maximizes the equivalent real channel gains Gˉk,t\bar{G}_{k,t} via optimal phase alignment; (ii) The Middle Layer, which is a convex problem, utilizes these improved gains to satisfy rate constraints with strictly lower (or equal) transmit power. Consequently, the total power objective function serves as a Lyapunov function, forming a monotonically non-increasing sequence {Ptotal(n)}\{P_{\mathrm{total}}^{(n)}\}. Since the transmit power is physically lower-bounded by zero, the sequence must converge.

  • Stability of Outer Layer: Although the Outer Layer (Detection) does not participate in the optimization loop, its performance relies on the convergence of the first two layers. The guaranteed convergence of the Inner and Middle layers ensures that the Outer Layer receives stable, optimal CSI and power parameters, thereby ensuring reliable symbol detection.

Simulation Results and Analysis

In this section, we present numerical results to evaluate the performance of the proposed RAQR-based Hybrid NOMA framework.

Convergence Analysis of the Algorithm

As illustrated in Figure 3, the phase error evolution presents a two-stage behavior: an initial temporary spike followed by a rapid decay. Notably, the phase error is artificially low (near zero) at the initial iteration (0-th iteration). This phenomenon occurs because the algorithm is typically initialized with default reference values before incorporating the actual noisy observations zm,tz_{m,t}. As the EM-GS algorithm executes its first few iterations, it begins to process the noisy received signals and actively explores the parameter space. This causes the phase estimates to deviate significantly from the naive initialization, resulting in the observed initial spike in the calculated error (peaking between 0.14 and 0.18). However, once the gradient search identifies the correct optimal descent direction, the phase error exhibits rapid decay, dropping to approximately 0.06 – 0.08 within the next 10 to 15 iterations, before achieving asymptotic stability below 0.02 after 35 iterations. This subsequent oscillation-free trajectory aligns with the monotonic convergence theory guaranteed by the EM framework, while the consistent behavior across varying channel initializations (Trials 1–3) confirms the algorithm’s robustness.

EM-GS: Phase Error Convergence

Figure 3. EM-GS: Phase Error Convergence.

The convergence of the inner-layer EM-GS algorithm is depicted in Figure 4. The residual imaginary energy decreases monotonically, stabilizing after approximately 10 iterations. These consistent trajectories demonstrate insensitivity to initial values, validating the theoretical prediction of effective phase alignment and low-latency interference suppression.

Cost function convergence behavior under different random initializations

Figure 4. Cost function convergence behavior under different random initializations.

Total Power Consumption Analysis

In Figure 5, we plot the total transmit power against the reference SNR (i.e., the inherent channel quality determined by path loss and noise variance) for target rates R∈{1.0,1.5,2.0}R \in\{1.0, 1.5, 2.0\} bits/s/Hz. As expected, the required power decreases with improving reference SNR and increases with higher RR. The proposed RAQR-Hybrid NOMA consistently outperforms the traditional benchmark, and the power saving gap widens as RR grows, highlighting its scalability for high-throughput demands. This advantage arises from the RAQR’s ability to coherently combine signals from multiple vapor cells, maximizing the effective channel gain and thus reducing

Total Power vs. SNR for Different Data Rates

Figure 5. Total Power vs. SNR for Different Data Rates (M=10M = 10).

transmit power. Although atomic receivers can ideally offer up to 20 dB sensitivity improvement, the observed gap in this figure is approximately 5 – 8 dB due to the limited number of cells (M=10M = 10) and multi-user interference. As shown later, scaling up MM unleashes the full potential of the atomic architecture.

Impact of Atomic Array Size (M)

Figure 6 investigates the impact of vapor cell count MM on total transmit power for M∈10,30,50,100M \in10, 30, 50, 100 across varying reference SNR levels. Increasing MM consistently reduces power consumption, with the most significant drop from M=10M = 10 to 30. Beyond that, further increases (e.g., 50 → 100) yield diminishing returns. This behavior stems not only from conventional diversity gain but also from enhanced spatial degrees of freedom that improve intra-cluster interference suppression in the multi-user NOMA setup. Although M=100M = 100 is currently impractical due to cost and complexity, this exploration reveals the asymptotic potential; a moderate array (M≈30M \approx30–50) can offer a favorable trade-off between energy efficiency, interference mitigation, and hardware feasibility.

Atomic Receiver Hybrid NOMA: Total Power vs. SNR for Different M Values

Figure 6. Atomic Receiver Hybrid NOMA: Total Power vs. SNR for Different MM Values (Target Rate R=1.5R = 1.5 bits/s/Hz).

A comprehensive bar chart comparison of the power consumption for the proposed scheme against the benchmarks (Traditional Receiver and OMA) is presented in Figure 7, spanning different SNR levels (0, 15, and 30 dB). A critical observation from these charts is that the power consumption of both Traditional-Receiver Hybrid NOMA and OMA baselines remains invariant with respect to MM. In these benchmarks, we intentionally fix a single conventional RF front-end and do not scale the number of RF antennas with MM; hence, they cannot exploit array or diversity gain without also incurring an MM-fold increase in hardware and circuit power. By contrast, MM in the proposed RAQR architecture parameterizes the number of low-cost atomic sensing elements behind one RF port, so increasing MM enhances the effective channel gain without replicating RF chains, and the blue bars therefore exhibit a pronounced reduction in transmit power as MM grows. Consequently, the performance advantage of the proposed scheme becomes increasingly pronounced with larger MM. Even at a relatively low SNR of 0 dB, scaling MM to 30 allows the proposed scheme to consume significantly less power than the benchmarks. This confirms that scaling the number of vapor cells is a highly effective strategy for overcoming adverse channel conditions and achieving superior energy efficiency compared to conventional approaches.

Power comparison of three transmission schemes for different $M$ values at SNR = 0 dB, 15 dB, and 30 dB

Figure 7. Power comparison of three transmission schemes for different MM values (R=1.5R = 1.5 bits/s/Hz).

To visualize the joint impact of SNR and MM on the total system power, Figure 8 provides a 3D surface plot. The curvature of the 3D surface visually reinforces the conclusions drawn from the 2D plots. The surface exhibits a steep gradient in the region where MM is small (10 to 30), particularly in low SNR regimes. This steep descent confirms that increasing the number of vapor cells is the most effective lever for reducing power consumption when signal quality is poor. As MM increases further, the surface flattens out, forming a “plateau” of high energy efficiency. This graphical representation succinctly summarizes the interplay between channel conditions and hardware resources, confirming that the proposed RAQR architecture can flexibly trade off array size for transmit power savings.

3D surface of total power versus SNR and $M$ for Hybrid NOMA with Atomic Receivers

Figure 8. 3D Surface: Total Power vs. SNR vs. MM for Hybrid NOMA with Atomic Receivers.

Detailed Power Allocation Analysis

To gain deeper insights into the internal mechanism driving the efficiency gains, we present a component-wise breakdown of power allocation in Figure 9. This figure dissects the power distributed to User 1 (across Slots 1 and 2) and User 2 (in Slot 2) for both the proposed and benchmark schemes.

Hybrid NOMA with Atomic Receivers

(a) Hybrid NOMA with Atomic Receivers

Hybrid NOMA with Traditional Receivers

(b) Hybrid NOMA with Traditional Receivers

Total Power Comparison

(c) Total Power Comparison

Figure 9. Detailed Power Allocation Analysis: Component-wise Breakdown.

As shown in subplots (a) and (b), the Traditional-Receiver scheme typically demands substantially higher transmit power, particularly for the interference-limited components, to satisfy the decoding constraints (SINR thresholds). In stark contrast, the RAQR scheme significantly reduces the power required across all signal components. Subplot (c) quantifies these savings, demonstrating a consistent and substantial reduction in total power consumption.

This reduction primarily reflects the coherent array gain provided by the MM-element atomic aperture: by performing precise phase compensation and combining the responses of many vapor cells, the receiver enhances the effective channel gain, thereby allowing the transmitter to meet the same SINR targets with much lower energy. In this sense, the observed power gain is analogous to classical multi-antenna array gain; exploiting deeper quantum limits is left for future work.

Furthermore, as depicted in Figure 10, a linear view of the allocation profile is offered to highlight structural differences. The

Hybrid NOMA with Atomic Receivers - Power Allocation

(a) Hybrid NOMA with Atomic Receivers - Power Allocation

Hybrid NOMA with Traditional Receivers - Power Allocation

(b) Hybrid NOMA with Traditional Receivers - Power Allocation

Figure 10. Power Allocation Comparison for Different Receivers (R=1.5R = 1.5 bits/s/Hz).

scale reveals that the RAQR achieves a more concentrated and balanced power distribution, characterized by smaller dynamic range gaps between different user components. Conversely, the traditional receiver exhibits a large dynamic range, implying that excessive power is wasted compensating for receiver-side limitations in specific time slots. This indicates that the RAQR improves not only the total power consumption but also the structural efficiency of resource allocation, leading to a more stable and balanced network operation.

Benchmarking against OMA and Robustness

Finally, we benchmark the proposed scheme against standard baselines and assess its robustness under varying channel statistics.

A comparison of the three transmission strategies across the SNR spectrum is shown in Figure 11, while Figure 12 provides a specific numerical snapshot at a target rate of R=1.5R = 1.5 bits/s/Hz.

Total Power vs. SNR for Three Transmission Schemes

Figure 11. Total Power vs. SNR for Three Transmission Schemes (M=10M = 10).

Total Power Comparison for Different Schemes

Figure 12. Total Power Comparison for Different Schemes (R=1.5R = 1.5 bits/s/Hz).

The results clearly establish the superiority of the RAQR-Hybrid NOMA scheme across all target rates. As shown in the bar chart, the RAQR achieves an approximate 93% power reduction compared to the OMA baseline. In comparison, the traditional receiver achieves only a 69% reduction. This significant margin underscores the transformative potential of integrating atomic sensing into NOMA networks.

Figure 13 evaluates the total transmit power versus target rate RR under symmetric (Case I) and asymmetric (Case II) channel conditions. The proposed RAQR-Hybrid NOMA consistently achieves the lowest power consumption, followed by traditional Hybrid NOMA while OMA requires the highest power. This underscores the robustness of the atomic architecture regardless of channel statistics.

Total Power versus Target Data Rate for Symmetric Average Signal Strength

(a) Case I: Symmetric Average Signal Strength

Total Power versus Target Data Rate for Asymmetric Average Signal Strength

(b) Case II: Asymmetric Average Signal Strength

Figure 13. Total Power Consumption vs. Target Data Rate under Different Channel Cases.

Beyond energy efficiency, the figure also reveals the spectral efficiency advantage of NOMA: for a given power budget, RAQR-Hybrid NOMA supports a higher RR than its counterparts, as evidenced by its slower power growth with RR. Moreover, when combined with the power-allocation profiles in Figure 10, the RAQR design enhances user fairness by yielding a more balanced power distribution across users. This is because the atomic receiver compensates for channel disparities through array gain and phase alignment, reducing the need for excessive power to weak users. These benefits hold under both symmetric and asymmetric channels, confirming that the proposed framework simultaneously improves energy efficiency, spectral utilization, and fairness.

Conclusion and Future Work

In this paper, we have proposed an innovative three-layer decomposition optimization framework for uplink Hybrid NOMA systems empowered by RAQR. By rigorously modeling the physical constraints of atomic sensing, specifically the non-linear amplitude-only observation, we successfully decoupled the intractable joint optimization problem into three mathematically tractable sub-modules. Specifically, the inner layer addresses fundamental phase ambiguity using a customized EM-GS algorithm to jointly estimate user symbols and phase compensation parameters. Subsequently, the middle layer derives a closed-form power allocation solution based on KKT conditions to guarantee optimal energy efficiency and finally, the outer layer employs linear MMSE detection for high-reliability signal recovery.

The results clearly demonstrates the superiority of the RAQR-Hybrid NOMA scheme across all target rates. The RAQR achieves an approximate 93% power reduction compared to the OMA baseline, whereas the traditional receiver achieves only a 69% reduction. This significant margin confirms the energy-efficiency synergy of the proposed architecture. However, investigating whether this high sensitivity can further mitigate NOMA-specific impairments, such as SIC error propagation or stringent power imbalance constraints, remains a challenging open issue for future research.

References

References

  1. [1]X. You, C.-X. Wang, J. Huang, X. Gao, Z. Zhang, M. Wang, Y. Huang, C. Zhang, Y. Jiang, J. Wang, et al., “Towards 6G Wireless Communication Networks: Vision, Enabling Technologies, and New Paradigm Shifts,” Science China Information Sciences, vol. 64, no. 1, pp. 1–74, 2021.
  2. [2]S. Sharma and K. Deka, “Toward Massive Type Communications in Ultra-dense Cellular IoT Networks: Current Issues and Machine Learning-assisted Solutions,” IEEE Communications Surveys & Tutorials, vol. 22, no. 1, pp. 426–471, 2021.
  3. [3]Y. Liu, Z. Qin, M. Elkashlan, Y. Gao, and L. Hanzo, “Evolution of NOMA Toward Next Generation Multiple Access (NGMA) for 6G,” IEEE Journal on Selected Areas in Communications, vol. 40, no. 4, pp. 1037–1071, 2022.DOI
  4. [4]Z. Ding, R. Schober, and H. V. Poor, “Next-Generation Multiple Access for IMT Towards 2030 and beyond,” IEEE Wireless Communications, vol. 31, no. 2, pp. 116–123, 2024.arxiv.org/abs/2404.04012
  5. [5]Y. Sun, Y. Li, and B. Vucetic, “Energy-Efficient Hybrid NOMA for Massive IoT Networks,” IEEE Transactions on Wireless Communications, vol. 24, no. 1, pp. 345–359, 2025.
  6. [6]T. Gong, A. Chandra, C. Yuen, Y. L. Guan, R. Dumke, C. M. S. See, M. Debbah, and L. Hanzo, “Rydberg Atomic Quantum Receivers for Classical Wireless Communication and Sensing,” IEEE Wireless Communications, 2025.DOI
  7. [7]C. L. Degen, F. Reinhard, and P. Cappellaro, “Quantum Sensing,” Reviews of modern physics, vol. 89, no. 3, p. 035002, 2017.DOI
  8. [8]Y. Chen, X. Guo, C. Yuan, Y. Zhao, Y. L. Guan, C. M. S. See, M. Debbah, and L. Hanzo, “Harnessing Rydberg Atomic Receivers: From Quantum Physics to Wireless Communications,” arXiv preprint arXiv:2501.11842, 2025.arxiv.org/abs/2501.11842
  9. [9]Z. Ding, P. Fan, and C. Tellambura, “Design and Analysis of Hybrid NOMA Systems with Imperfect CSI,” IEEE Transactions on Communications, vol. 72, no. 3, pp. 1568–1582, 2024.
  10. [10]Z. Ding, Y. Liu, and M. Elkashlan, “A Study on Grant-free Hybrid NOMA for URLLC in 6G,” IEEE Journal on Selected Areas in Communications, vol. 43, no. 2, pp. 445–458, 2025.
  11. [11]Z. Ding, M. Peng, and H. V. Poor, “Hybrid NOMA for IoT Networks with Bursty Traffic,” IEEE Internet of Things Journal, vol. 11, no. 4, pp. 5678–5692, 2024.
  12. [12]D. A. Anderson, R. E. Sapiro, and G. Raithel, “An Atomic Receiver for AM and FM Radio Communication,” IEEE Transactions on Antennas and Propagation, vol. 69, no. 5, pp. 2455–2462, 2021.
  13. [14]M. Cui, Q. Zeng, and K. Huang, “Quantum Atomic Receiver: Next Frontier of Wireless Communications,” IEEE Communications Magazine, 2025.arxiv.org/abs/2412.12485
  14. [15]M. Vaezi, Z. Ding, and H. V. Poor, Multiple access techniques for 5G wireless networks and beyond. Springer, 2019.
  15. [16]3GPP, “Evolved universal terrestrial radio access (E-UTRA); Physical channels and modulation (Release 15),” standard, 3rd Generation Partnership Project (3GPP), 2019.
  16. [17]C. She, R. Dong, Z. Gu, Z. Tang, T. Q. Quek, and Y. Li, “A Tutorial on NOMA-enhanced Backscatter Communications,” IEEE Wireless Communications, vol. 28, no. 4, pp. 98–104, 2021.
  17. [18]Z. Ding, Z. Yang, and H. Wang, “Backscatter Communication with Hybrid NOMA: Design and Performance Analysis,” IEEE Transactions on Green Communications and Networking, vol. 9, no. 1, pp. 245–259, 2025.
  18. [19]K. Huang, Z. Ding, and J. Zhang, “Energy Harvesting in Hybrid NOMA Networks: A Stochastic Geometry Approach,” IEEE Transactions on Wireless Communications, vol. 24, no. 3, pp. 1892–1906, 2025.
  19. [20]L. Gyongyosi and S. Imre, “A Survey on Quantum Computing Technology,” Computer Science Review, vol. 31, pp. 51–71, 2019.DOI
  20. [21]C. Wang and A. Rahman, “Quantum-enabled 6G Wireless Networks: Opportunities and Challenges,” IEEE Wireless Communications, vol. 29, no. 1, pp. 58–69, 2022.DOI
  21. [22]M. Saffman, T. G. Walker, and K. Mølmer, “Quantum Information with Rydberg Atoms,” Reviews of modern physics, vol. 82, no. 3, pp. 2313–2363, 2010.DOI
  22. [23]J. Liu, Y. Ma, and R. Tafazolli, “SA-MIMO: Scalable Quantum-Based Wireless Communications,” arXiv preprint arXiv:2504.19170, 2025.arxiv.org/abs/2504.19170
  23. [24]H. Fan, S. Kumar, J. Sedlacek, H. Kübler, S. Karimkashi, and J. P. Shaffer, “Atom Based RF Electric Field Sensing,” Journal of Physics B: Atomic, Molecular and Optical Physics, vol. 48, no. 20, p. 202001, 2015.
  24. [25]D. H. Meyer, K. C. Cox, F. K. Fatemi, and P. D. Kunz, “Digital Communication with Rydberg Atoms and Amplitude-modulated Microwave Fields,” Applied Physics Letters, vol. 112, no. 21, 2018.DOI
  25. [26]J. Yuan, T. Jin, L. Xiao, S. Jia, and L. Wang, “A Rydberg Atom-based Receiver with Amplitude Modulation Technique for the Fifth-generation Millimeter-wave Wireless Communication,” IEEE antennas and wireless propagation letters, vol. 22, no. 10, pp. 2580–2584, 2023.DOI
  26. [27]M. Jing, Y. Hu, J. Ma, H. Zhang, L. Zhang, L. Xiao, and S. Jia, “Atomic Superheterodyne Receiver Based on Microwave-dressed Rydberg Spectroscopy,” Nature Physics, vol. 16, no. 9, pp. 911–915, 2020.
  27. [28]M. T. Simons, A. B. Artusio-Glimpse, A. K. Robinson, N. Prajapati, and C. L. Holloway, “Rydberg Atom-based Sensors for Radio-frequency Electric Field Metrology, Sensing, and Communications,” Measurement: Sensors, vol. 18, p. 100273, 2021.DOI
  28. [29]C. T. Fancher, D. R. Scherer, M. C. S. John, and B. L. S. Marlow, “Rydberg Atom Electric Field Sensors for Communications and Sensing,” IEEE Transactions on Quantum Engineering, vol. 2, pp. 1–13, 2021.DOI
  29. [30]J. Yuan, W. Yang, M. Jing, H. Zhang, Y. Jiao, W. Li, L. Zhang, L. Xiao, and S. Jia, “Quantum Sensing of Microwave Electric Fields Based on Rydberg Atoms,” Reports on Progress in Physics, vol. 86, no. 10, p. 106001, 2023.DOI
  30. [31]M. Cui, Q. Zeng, and K. Huang, “Towards Atomic MIMO Receivers,” IEEE Journal on Selected Areas in Communications, 2025.arxiv.org/abs/2404.04864
  31. [32]T. Gong, J. Sun, C. Yuen, G. Hu, Y. Zhao, Y. L. Guan, C. M. S. See, M. Debbah, and L. Hanzo, “An Equivalent Baseband Signal Model for Rydberg Atomic Quantum Receiver Aided Wireless Communications and Sensing,” in 2025 International Conference on Quantum Communications, Networking, and Computing (QCNC), pp. 200–204, IEEE, 2025.DOI
  32. [33]Q. Peng, J. Liu, Q. Luo, Y. Ma, P. Xiao, M. Ekashlan, and G. K. Karagiannidis, “RIS-assisted Atomic MIMO Receiver,” arXiv preprint arXiv:2510.15763, 2025.arxiv.org/abs/2510.15763
  33. [34]T. Gong, C. Yuen, C. M. S. See, M. Debbah, and L. Hanzo, “Rydberg Atomic Quantum MIMO Receivers for the Multi-user Uplink,” arXiv preprint arXiv:2506.01355, 2025.arxiv.org/abs/2506.01355
  34. [35]M. Cui, Q. Zeng, and K. Huang, “MIMO Precoding for Rydberg Atomic Receivers Enabled Wireless Communications,” in ICC 2025 - IEEE International Conference on Communications, pp. 4792–4797, 2025.DOI
  35. [36]B. Xu, J. Zhang, Z. Chen, B. Cheng, Z. Liu, Y.-C. Wu, and B. Ai, “Channel Estimation for Rydberg Atomic Receivers,” IEEE Wireless Communications Letters, vol. 14, no. 9, pp. 2957–2961, 2025.arxiv.org/abs/2503.08985
  36. [37]M. Cui, Q. Zeng, Z. Wang, and K. Huang, “Rydberg Atomic Receivers for Multi-Band Communications and Sensing,” arXiv preprint arXiv:2505.24168, 2025.arxiv.org/abs/2505.24168
  37. [38]H. Wang, T. Gong, E. Björnson, and C. Yuen, “Multi-carrier rydberg atomic quantum receivers with enhanced bandwidth feature for communication and sensing,” arXiv preprint arXiv:2510.10473, 2025.arxiv.org/abs/2510.10473
  38. [39]T. Gong, C. Yuen, C. M. S. See, M. Debbah, and L. Hanzo, “Rydberg Atomic Quantum Receivers for Multi-Target DOA Estimation,” IEEE Transactions on Vehicular Technology, pp. 1–6, 2025.arxiv.org/abs/2501.02820
  39. [40]Y. Chen, C. Yuen, D. Arumugam, C. M. S. See, M. Debbah, and L. Hanzo, “Polarization-aware doa detection relying on a single rydberg atomic receiver,” arXiv preprint arXiv:2508.17179, 2025.arxiv.org/abs/2508.17179
  40. [41]D. H. Meyer, Z. A. Castillo, K. C. Cox, and P. D. Kunz, “Assessment of Rydberg Atoms for Wideband Electric Field Sensing,” Journal of Physics B: Atomic, Molecular and Optical Physics, vol. 53, no. 3, p. 034001, 2020.DOI
  41. [42]G. Santamaria-Botello, S. Verploeg, E. Bottomley, and Z. Popovic, “Comparison of Noise Temperature of Rydberg-atom and Electronic Microwave Receivers,” arXiv preprint arXiv:2209.00908, 2022.
  42. [43]J.-Y. Wang, X.-T. Fu, R.-R. Lu, J.-B. Wang, M. Lin, and J. Cheng, “Tight Capacity Bounds for Indoor Visible Light Communications with Signal-dependent Noise,” IEEE Transactions on Wireless Communications, vol. 20, no. 3, pp. 1700–1713, 2020.arxiv.org/abs/2011.00230
  43. [44]M. T. Simons, A. H. Haddab, J. A. Gordon, and C. L. Holloway, “A Rydberg Atom-based Mixer: Measuring the Phase of a Radio Frequency Wave,” Applied Physics Letters, vol. 114, no. 11, 2019.DOI

Paper details

vlnc:W_7h5vwd75
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Computer Science

Not recorded for this paper

View statement
Looking for related papers…
Contents