RSMA: Rate-Splitting Multiple Access

• RSMA is a physical layer multiple access method that splits each user’s message into common and private parts to manage interference and improve fairness.
• It utilizes linear precoding and single-layer SIC, enabling robust performance under varied channel conditions and imperfect CSIT.
• RSMA achieves significant gains in throughput, energy efficiency, and latency reduction, outperforming traditional SDMA and NOMA in diverse network scenarios.

Rate-Splitting Multiple Access (RSMA) is a physical layer multiple access paradigm for downlink multi-antenna wireless networks that strategically enables simultaneous transmission of common and private messages to multiple users. RSMA’s key innovation is the application of “rate splitting,” in which each user’s message is divided into a private part, decoded only by its target user, and a common part, jointly decoded by all users. The transmission is superposed, linearly precoded, and decoded using single-layer Successive Interference Cancellation (SIC) at each receiver. This structure generalizes Space Division Multiple Access (SDMA, which treats multi-user interference as noise) and Non-Orthogonal Multiple Access (NOMA, which decodes all interference in an ordered fashion) as special cases. RSMA explicitly addresses the challenges of spectral efficiency, QoS fairness, energy efficiency, and robustness to imperfect Channel State Information at the Transmitter (CSIT), supporting a flexible trade-off between interference decoding and treating interference as noise across a broad range of network loading scenarios and user deployments.

1. Transceiver Architecture and Signal Model

RSMA operates over the MISO Broadcast Channel (BC), where a multi-antenna transmitter serves $K$ single-antenna users. For each user $k$, the message $W_k$ is split into $W_{c,k}$ (common, for all) and $W_{p,k}$ (private). All $K$ common parts are merged and jointly encoded into a common stream $s_c$, while each private part is encoded into its own stream $s_k$. Denoting $\mathbf{s} = [s_c, s_1, \ldots, s_K]^T$, the transmit signal is given by

$$\mathbf{x} = \mathbf{p}_c s_c + \sum_{k=1}^K \mathbf{p}_k s_k, \;\; \mathbb{E}[\mathbf{s}\mathbf{s}^H] = \mathbf{I}$$

where $\mathbf{p}_c, \{\mathbf{p}_k\}_{k=1}^K$ are linear precoding vectors.

The received signal at user $k$:

$$y_k = \mathbf{h}_k^H \mathbf{x} + n_k, \quad n_k\sim\mathcal{CN}(0,\sigma_k^2)$$

Each user first decodes the common stream (treating all privates as noise), strips it off, then decodes its private stream (treating residual interference as noise). The SINRs are:

$$\gamma_{c,k} = \frac{|\mathbf{h}_k^H\mathbf{p}_c|^2}{\sum_{j=1}^K |\mathbf{h}_k^H\mathbf{p}_j|^2 + \sigma_k^2}$$

$$\gamma_{p,k} = \frac{|\mathbf{h}_k^H\mathbf{p}_k|^2}{\sum_{j\neq k}|\mathbf{h}_k^H\mathbf{p}_j|^2 + \sigma_k^2}$$

The maximum decodable common-stream rate is $R_c = \min_k \log_2(1+\gamma_{c,k})$, which is split as $R_c = \sum_k C_k$ among users. Each user’s total rate is $R_k^{\text{tot}} = C_k + \log_2(1+\gamma_{p,k})$ (Dizdar et al., 2021).

2. Physical Layer Optimization and Power Allocation

Robust precoder design is central for RSMA, given practical constraints of imperfect CSIT (e.g., feedback delays, user mobility). For user mobility, the temporal correlation in CSI is modeled as $$\mathbf{h}_k[m] = \epsilon\hat{\mathbf{h}}_k[m-\Delta] + \sqrt{1-\epsilon^2}\,\mathbf{e}_k[m]$$, where $\epsilon$ is the time-correlation coefficient. In this setting, precoders can be chosen to maximize a lower bound on the ergodic sum-rate, yielding closed-form power-allocation:

$$\mathbf{p}_c = \sqrt{P(1-t)}\,\mathbf{f}_c, \quad \mathbf{p}_k = \sqrt{\frac{P\,t}{K}}\,\mathbf{f}_k$$

The scalar $t\in[0,1]$ allocates power between the common ($1-t$) and private ($t$) streams. The ergodic sum-rate lower bound is tractably maximized in $t$, yielding

$$t_{\rm opt} = \frac{\rho\,(K-1)}{\rho(\omega+K) - K}$$

with $\omega,\rho$ explicit functions of the power, antenna/user count, and $\epsilon$ (Dizdar et al., 2021, Dizdar et al., 2021). In the perfect CSIT case, $t_{\rm opt}\to1$, reducing RSMA to SDMA; for highly outdated CSI, RSMA pushes more power into the common stream.

Optimization of RSMA transmission under these constraints is typically formulated as (weighted) sum-rate or max-min fairness problems subject to transmit power and per-user minimum rate constraints. The resulting non-convex programs are solved by alternating optimization (AO) strategies such as Weighted MMSE (WMMSE) or successive convex approximation (SCA), guaranteeing monotonic convergence (Mao et al., 2017, Dizdar et al., 2021).

3. Reliability, Latency, and Energy Efficiency in RSMA

RSMA supports both enhanced Mobile Broadband (eMBB) and Ultra-Reliable Low-Latency Communication (URLLC):

• URLLC: For short packets, finite-blocklength information theory is essential, as Shannon capacity overestimates achievable rates. The normal approximation yields

$$R(\gamma)\approx \log_2(1+\gamma) - \sqrt{V(\gamma)/N}\,Q^{-1}(\xi)$$

where $V(\gamma)$ is the channel dispersion. RSMA achieves a given sum-rate with orders-of-magnitude shorter blocklengths (as low as 100 bits vs. 2,500 bits for SDMA) under channel alignment, yielding up to 20$\times$ latency reduction at fixed reliability (Dizdar et al., 2021).

• eMBB: In high-mobility, high-throughput scenarios, RSMA's common/private split is robust to CSIT aging, with non-saturating sum-rate growth at high SNR and only mild degradation with user speed. Throughput gains of 20–50% are observed versus SDMA/NOMA in large-scale scenarios (e.g., 32 antennas, 8 users, 100–200 km/h) (Dizdar et al., 2021, Dizdar et al., 2021).
• Energy Efficiency: RSMA generally outperforms SDMA and NOMA in energy efficiency (EE), exhibiting 20–30% higher EE under mixed channel conditions and circuit power budgets (Mao et al., 2018).

Implementation complexity remains comparable to NOMA (one layer of SIC per receiver) and only slightly higher than SDMA. All SCA-based schemes converge rapidly with per-iteration costs similar across paradigms (Mao et al., 2018, Mao et al., 2017).

4. Generalizations: Overloading, Intelligence, and Special Cases

RSMA’s flexibility extends to advanced settings:

• Overloaded Networks: In $K > N_t$ regimes (more users than antennas), RSMA overcomes the DoF collapse of SDMA, retaining non-saturating rates and fairness even in massive connectivity or IoT (5–8 dB SNR gains in max–min rate at high SNR over SDMA) (Dizdar et al., 2023).
• Physical Layer Enhancements: New splitter designs replicate information from low-SNR subcarriers (in multi-carrier RSMA) into the common stream, yielding dramatic reductions in packet failure and delay (40–83% delay reduction, 3–10$\times$ BER gains at low SNR, 20% sum-rate uplift over plain RSMA) (Ali et al., 16 Apr 2025).
• Integration with Beyond-5G Technologies: RSMA forms synergistic designs with intelligent reflecting surfaces (IRSs) to boost spectral efficiency and robustness. IRS-enabled RSMA can provide >3 bpcu uplift in common-stream rate, greatly enhanced resilience to SIC and CSI errors, and robustness for UAV, LEO satellite, and mmWave systems (Sena et al., 2022).
• Hierarchical Multi-Cluster and Joint Transmission: For dense heterogeneous deployments (C-RAN, CoMP), RSMA supports multi-layer splitting (multiple common streams associated with user clusters), with efficient stream and clustering selection procedures to maintain practical complexity and outpace single-stream RSMA and conventional SDMA/NOMA (Yu et al., 2019, Mao et al., 2018). In multi-point joint transmission, RSMA achieves significant weighted sum-rate improvements across all channel disparity regimes, generalizing SDMA (all-private split) and NOMA (all-common split) (Mao et al., 2018).
• Integrated Sensing and Communication (ISAC): RSMA in dual-functional systems (e.g., joint radar and communications, or multi-user and multi-target sensing) enables joint max-min fairness in communications while optimizing the Cramér–Rao bound of target parameter estimation, sustaining strictly superior communication–sensing tradeoff regions even with many targets (Chen et al., 2023, Xu et al., 2020). The common stream enables simultaneous multiuser interference management and radar beampattern shaping.

5. Algorithmic and Design Guidelines

Key algorithmic and system-design insights for RSMA implementations:

• Precoder Design: Under imperfect CSIT, ZF is preferred for private beams, while the common beam can be chosen as a random or dominant-eigenvector direction. IRS/MA positioning can be alternated with beamformer and rate-split updates for additional performance gains (Dizdar et al., 2021, Zhang et al., 2024).
• Power Splitting: A single scalar parameter $t$ governs the common/private split, determinable in closed form from large-scale network statistics and Doppler/external fading estimations (Dizdar et al., 2021, Dizdar et al., 2021).
• Blocklength and Split: For URLLC, splitting blocklength between streams and tuning the number of low-SNR subcarriers to replica in the common stream leads to optimal reliability/latency tradeoffs (Dizdar et al., 2021, Ali et al., 16 Apr 2025).
• Complexity: Resource allocation, clustering for multi-stream RSMA, and power/rate-splitting can all be executed with polynomial or near-real-time complexity (SCA, WMMSE, power iteration), applicable to large-scale systems. Receiver processing is single-layer SIC.
• Adaptability: RSMA bridges the full “decode-all” and “treat-as-noise” extremes, automatically morphing to SDMA or NOMA when these become optimal (e.g., fully orthogonal or severely aligned/gain-disparate channels, respectively) (Mao et al., 2017).

6. Comparative Performance and Limitations

Extensive numerical and analytical results consistently show that RSMA:

• Strictly outperforms SDMA and NOMA in both underloaded and overloaded regimes, with rate, fairness, and energy efficiency gains ranging from 20–50% depending on SNR, channel orthogonality, and CSIT fidelity (Dizdar et al., 2021, Mao et al., 2018, Dizdar et al., 2023).
• Remains robust and non-saturating under realistic mobility and feedback delays: sum-rate and throughput scale logarithmically in SNR, even with aged CSIT and large user numbers (Dizdar et al., 2021).
• Enables new trade-off frontiers in ISAC (joint multi-user communication and multi-target sensing), outperforming SDMA in both CRB and communication rate (Chen et al., 2023).
• Is especially advantageous in scenarios with non-orthogonal user channels, high mobility, strong or uncertain inter-user interference, and multi-mission (communication+sensing) requirements.
• Adds only one layer of SIC complexity per receiver, and generally does not require user ordering/grouping at the transmitter.

However, RSMA systems are subject to:

• Increased complexity at the signal splitting stage, especially when supporting multiple common streams in cluster-based RSMA designs, though efficient hierarchical procedures mitigate this (Yu et al., 2019).
• SNR penalty and potential sum-rate tradeoff if excessive redundancy is introduced to maximize reliability in the channel-dependent splitting approaches (Ali et al., 16 Apr 2025).
• Need for accurate large-scale channel statistics to optimally tune the power/rate split parameter in highly dynamic environments.

7. Practical Impact and Prospects

RSMA is positioned as a foundational physical layer for 6G and beyond, functioning as a robust, scalable, and low-complexity generalization of SDMA and NOMA that can be dynamically adapted to a diversity of deployment scenarios. Its unified treatment of interference management, support for extreme heterogeneity in requirements (URLLC, mMTC, eMBB), and synergy with emerging technologies (IRSs, movable antennas, and ISAC) recommend its adoption for future wireless systems (Dizdar et al., 2021, Sena et al., 2022, Zhang et al., 2024, Chen et al., 2023).

The evolutionary path for RSMA includes machine learning-based splitting and clustering, multi-cell and multi-layer extensions, and adaptability to coded modulation and cross-layer optimization. These research directions aim to further improve latency, reliability, and energy efficiency while reducing control overhead and scaling to massive connectivity scenarios.