Cooperative RSMA: Relaying & CoMP Techniques

• Cooperative RSMA is a wireless transmission architecture that splits messages into common and private parts while adding a cooperative relaying or joint transmission phase.
• It employs a two-phase design to overcome the common-stream bottleneck, thereby enhancing system fairness and total achievable rates.
• Optimization methods such as WMMSE and SCA are key to addressing non-convex precoding and time-split challenges in cooperative RSMA setups.

Searching arXiv for recent and foundational papers on cooperative RSMA and closely related coordinated RSMA settings. arxiv_search: {"query":"Cooperative Rate-Splitting Multiple Access coordinated multipoint joint transmission user relaying RSMA", "max_results": 10} Cooperative Rate-Splitting Multiple Access (Cooperative RSMA, often abbreviated C-RSMA in user-relaying settings) denotes a class of wireless transmission architectures in which the common/private message structure of RSMA is combined with an explicit cooperative mechanism, most prominently user relaying of the common stream or coordinated multi-point joint transmission across multiple base stations. In its basic downlink form, each message is split into a common part and a private part, the common parts are jointly encoded into a common stream, and the private parts are separately encoded into private streams; receivers decode the common stream first, perform SIC, and then decode their private streams. Cooperative variants preserve this structure but add a second transmission path or a distributed transmitter, so that the common layer is no longer only a shared SIC layer but also a relayed or jointly transmitted resource (Mao et al., 2022, Xu et al., 2022, Mao et al., 2018, Elhattab et al., 2024).

1. Conceptual scope and taxonomy

Cooperative RSMA is best understood against the broader RSMA taxonomy. The survey literature identifies 1-layer RS, hierarchical RS, generalized RS, and RS-CMD as the principal downlink architectures. The basic 1-layer form uses one common stream decoded by all users plus user-specific private streams; generalized RS extends this by allowing streams intended for arbitrary user subsets; RS-CMD encodes split messages separately and is often associated with distributed settings (Mao et al., 2022).

Within that taxonomy, “cooperation” appears in more than one sense. The narrowest and most canonical meaning is user-assisted cooperative RS, also called cooperative rate-splitting (CRS), where one or more users decode the common stream and forward it to weaker users in a second phase (Mao et al., 2022). A second meaning is network-side cooperation, exemplified by Coordinated Multi-Point Joint Transmission (CoMP JT), where several base stations share user data and CSI and act as a distributed super-transmitter under per-BS power constraints (Mao et al., 2018). A broader and weaker usage treats RSMA as “cooperative” because all users decode a shared common layer or because the common layer coordinates multiple functions, but those cases do not constitute cooperative RSMA in the relaying sense (Dizdar et al., 2021, Xu et al., 2021).

A central technical reason cooperation aligns naturally with RSMA is that the common stream is already intended for multi-user decoding. In non-cooperative 1-layer RS, the total common-stream rate is constrained by the weakest common decoder. Cooperative RSMA attacks precisely that bottleneck by strengthening the common-stream path rather than trying to forward all private information. This design choice is repeatedly emphasized in user-relaying and multi-cell cooperative formulations (Xu et al., 2022, Elhattab et al., 2024).

The CoMP literature also positions RSMA as a bridge between SDMA and NOMA. In the fully cooperative CoMP JT setting, SDMA is obtained by allocating no power to common streams, while NOMA-type structures arise by activating specific decoding patterns; RSMA interpolates between “treat interference as noise” and “decode interference” and is therefore suited to a wider range of inter-user and inter-cell channel disparities (Mao et al., 2018).

2. Canonical signal model and decoding structure

The basic downlink 1-layer RS signal model is

$$\mathbf{x}=\mathbf{p}_{c}s_{c}+\sum_{k\in\mathcal{K}}\mathbf{p}_{k}s_{k},$$

where $s_c$ is the common stream, $s_k$ are private streams, and $\mathbf{p}_c,\mathbf{p}_k$ are precoders. At user $k$,

$$y_k=\mathbf{h}_{k}^{H}\mathbf{p}_{c}s_{c}+\mathbf{h}_{k}^{H}\mathbf{p}_{k}s_{k} +\sum_{j\in\mathcal{K},j\neq k}\mathbf{h}_{k}^{H}\mathbf{p}_{j}s_{j}+n_k.$$

The common-stream and private-stream rates are

$$R_{c,k}=\log_{2}\left(1+\frac{|\mathbf{h}_{k}^{H}\mathbf{p}_{c}|^{2}} {\sum_{j\in\mathcal{K}}|\mathbf{h}_{k}^{H}\mathbf{p}_{j}|^{2}+\sigma_{n,k}^{2}}\right),$$

$$R_{k}=\log_{2}\left(1+\frac{|\mathbf{h}_{k}^{H}\mathbf{p}_{k}|^{2}} {\sum_{j\in\mathcal{K},j\neq k}|\mathbf{h}_{k}^{H}\mathbf{p}_{j}|^{2}+\sigma_{n,k}^{2}}\right).$$

Since all users decode the common stream,

$$R_c=\min\{R_{c,1},\ldots,R_{c,K}\}, \qquad \sum_{k\in\mathcal{K}} C_k = R_c, \qquad R_{k,\mathrm{tot}}=C_k+R_k.$$

These relations are the basic analytical backbone of both cooperative user-relaying RSMA and coordinated multi-cell RSMA (Mao et al., 2022).

In cooperative user-relaying RSMA for multigroup multicast with finite blocklength, the first phase remains standard RSMA: $$\mathbf{x}^{[1]}=\mathbf{p}_c s_c+\sum_{m\in\mathcal{M}}\mathbf{p}_m s_m, \qquad \mathrm{tr}(\mathbf{P}\mathbf{P}^H)\le P_t.$$ Users decode the common stream first, then their group-private stream. The cooperative enhancement is a second phase in which users in a designated strong group relay only the decoded common stream: $s_c$0 For a user $s_c$1 in the other groups,

$s_c$2

The key rate aggregation is

$s_c$3

with users in the relaying group decoding the common stream from phase 1 only, while weaker groups combine direct and relayed common information (Xu et al., 2022).

In fully cooperative CoMP JT RSMA, the transmitter is distributed across $s_c$4 single-antenna BSs that jointly act as a super-BS. The aggregate transmit signal has the generalized-RS form

$s_c$5

where stream $s_c$6 is decoded by users in subset $s_c$7 and treated as noise by the others. SIC decodes higher-order streams before lower-order streams. Stream rates are limited by the weakest intended decoder, and common-rate allocation variables partition each stream’s rate across its intended users (Mao et al., 2018).

3. Principal cooperative architectures

The cooperative RSMA literature represented in the cited works divides naturally into user-relaying and network-coordinated forms.

Architecture	Cooperation mechanism	Representative paper
User-relaying C-RSMA	Strong users forward the decoded common stream in a second phase	(Xu et al., 2022)
CoMP JT RSMA	Multiple BSs jointly process data and beamform all RS streams	(Mao et al., 2018)
Multi-cell JT-CoMP C-RSMA	Joint BS transmission plus CCU relaying of the common stream in HD/FD	(Elhattab et al., 2024)

In the finite-blocklength cooperative RSMA model, a BS with $s_c$8 antennas serves $s_c$9 single-antenna users partitioned into multicast groups. Each group message $s_k$0 is split into $s_k$1 and $s_k$2, all common parts are merged into a super-common message, and the common stream is relayed only by users in $s_k$3, assumed to be cell-center users operating as half-duplex decode-and-forward relays under an NDF protocol (Xu et al., 2022). The paper explicitly states that non-cooperative RSMA is the special case $s_k$4, where the cooperative phase is turned off.

In the fully cooperative CoMP JT formulation, all BSs are connected through a central controller, all user messages are jointly processed, and each stream may be transmitted simultaneously from all BSs. Because each BS has one antenna, the system is equivalent to a distributed MISO BC with per-antenna or per-BS power constraints. Cooperation exists at both the data level and the beamforming level, while the receiver-side SIC structure remains the RSMA one (Mao et al., 2018).

The 2024 coordinated HD/FD C-RSMA model combines both forms of cooperation. All BSs in a multi-cell network use JT-CoMP to transmit the common and private streams to all users, while cell-center users relay the common stream to cell-edge users. In HD mode, the BS direct phase occupies a fraction $s_k$5 and the cooperative transmission phase occupies $s_k$6. In FD mode, CCUs receive the common stream from the BSs and simultaneously forward a delayed version to CEUs, at the cost of residual self-interference through a channel $s_k$7 modeled as $s_k$8 (Elhattab et al., 2024).

That multi-cell C-RSMA design is notable because all CCUs forward the same common stream. The paper contrasts this with cooperative NOMA, where different relayed streams coexist and create inter-user interference during the cooperative phase. In 1-layer C-RSMA, the relay phase is therefore structurally cleaner (Elhattab et al., 2024).

4. Optimization formulations and algorithmic machinery

Cooperative RSMA optimization is typically posed as weighted sum-rate or max-min fairness optimization over precoders, common-rate allocations, relay variables, and sometimes time partition variables. The resulting problems are non-convex because rates are logarithmic in SINRs, common-stream constraints involve minima across users, and cooperative variables couple numerators and denominators.

For CoMP JT generalized RSMA, the weighted sum-rate problem is

$s_k$9

subject to per-stream common-rate constraints, per-BS power constraints,

$\mathbf{p}_c,\mathbf{p}_k$0

QoS thresholds $\mathbf{p}_c,\mathbf{p}_k$1, and $\mathbf{p}_c,\mathbf{p}_k$2. The paper solves this by extending the WMMSE/AO framework: equalizers $\mathbf{p}_c,\mathbf{p}_k$3, MSE weights $\mathbf{p}_c,\mathbf{p}_k$4, and transformed common-rate variables $\mathbf{p}_c,\mathbf{p}_k$5 are introduced, and the $\mathbf{p}_c,\mathbf{p}_k$6-subproblem becomes a convex QCQP for fixed $\mathbf{p}_c,\mathbf{p}_k$7. The AO loop monotonically increases WSR and is guaranteed to converge because the objective is bounded above under per-BS power constraints (Mao et al., 2018).

For finite-blocklength cooperative RSMA, the objective is group-rate max-min fairness: $\mathbf{p}_c,\mathbf{p}_k$8 subject to common-rate feasibility and sum-power constraints. The finite-blocklength rate approximation is

$\mathbf{p}_c,\mathbf{p}_k$9

with

$k$0

The common phase/time split is optimized through $k$1. Because the FBL expressions are not amenable to the standard Shannon-rate WMMSE reformulation, the paper uses a 1D-SCA procedure: an outer one-dimensional search over $k$2 or $k$3, and an inner SCA loop that linearizes both the dispersion terms and the quadratic-over-linear SINR constraints (Xu et al., 2022).

For multi-cell JT-CoMP HD/FD C-RSMA, the objective is max-min user rate under common-rate, fronthaul, per-BS power, relay power, and, in HD, time-fraction constraints. The HD problem jointly optimizes BS beamforming $k$4, relay beamforming $k$5, common-rate allocation $k$6, and $k$7. The FD problem optimizes $k$8, $k$9, and $$y_k=\mathbf{h}_{k}^{H}\mathbf{p}_{c}s_{c}+\mathbf{h}_{k}^{H}\mathbf{p}_{k}s_{k} +\sum_{j\in\mathcal{K},j\neq k}\mathbf{h}_{k}^{H}\mathbf{p}_{j}s_{j}+n_k.$$0, while handling residual self-interference terms such as $$y_k=\mathbf{h}_{k}^{H}\mathbf{p}_{c}s_{c}+\mathbf{h}_{k}^{H}\mathbf{p}_{k}s_{k} +\sum_{j\in\mathcal{K},j\neq k}\mathbf{h}_{k}^{H}\mathbf{p}_{j}s_{j}+n_k.$$1 (Elhattab et al., 2024). Both are converted into SCA programs by introducing slack variables for fairness, SINR, and interference, then applying first-order Taylor lower bounds to quadratic-over-linear expressions. HD additionally uses exhaustive search over $$y_k=\mathbf{h}_{k}^{H}\mathbf{p}_{c}s_{c}+\mathbf{h}_{k}^{H}\mathbf{p}_{k}s_{k} +\sum_{j\in\mathcal{K},j\neq k}\mathbf{h}_{k}^{H}\mathbf{p}_{j}s_{j}+n_k.$$2 with step size $$y_k=\mathbf{h}_{k}^{H}\mathbf{p}_{c}s_{c}+\mathbf{h}_{k}^{H}\mathbf{p}_{k}s_{k} +\sum_{j\in\mathcal{K},j\neq k}\mathbf{h}_{k}^{H}\mathbf{p}_{j}s_{j}+n_k.$$3 (Elhattab et al., 2024).

A recurrent theme is that cooperative RSMA optimization remains close to mainstream RSMA optimization—common-rate allocation plus linear precoding plus SIC-aware constraints—but cooperation introduces either a second transmission phase or extra transmitter-side blocks. This suggests why WMMSE, SCA, and AO dominate the algorithmic literature.

5. Performance regimes and comparative evidence

The available evidence consistently reports that cooperative or coordinated RSMA improves robustness and fairness relative to conventional baselines, though the exact gain depends on channel disparity, duplex mode, and system loading.

In CoMP JT, RSMA achieves a larger achievable rate region than SDMA and NOMA across all tested inter-user and inter-cell disparity settings, and is consistently closer to DPC than either benchmark. In the two-user CoMP JT model, the paper summarizes the regime dependence as follows: SDMA is more suited to little inter-user disparity and large inter-cell disparity, NOMA to large inter-user disparity and little inter-cell disparity, while RSMA is suited to any deployment (Mao et al., 2018). In the three-user CoMP JT case, generalized RS and 1-layer RS show clear sum-rate improvements over MU-LP, SC-SIC, and SC-SIC per group, while full SC-SIC performs worst because it suffers a loss of multiplexing gain (Mao et al., 2018).

In finite-blocklength cooperative RSMA, the numerical results show

$$y_k=\mathbf{h}_{k}^{H}\mathbf{p}_{c}s_{c}+\mathbf{h}_{k}^{H}\mathbf{p}_{k}s_{k} +\sum_{j\in\mathcal{K},j\neq k}\mathbf{h}_{k}^{H}\mathbf{p}_{j}s_{j}+n_k.$$4

in most tested settings (Xu et al., 2022). The paper states that RSMA can achieve the same MMF rate as NOMA and SDMA with smaller blocklengths, and therefore lower latency, especially in cooperative transmission deployment (Xu et al., 2022). In the overloaded scenario with $$y_k=\mathbf{h}_{k}^{H}\mathbf{p}_{c}s_{c}+\mathbf{h}_{k}^{H}\mathbf{p}_{k}s_{k} +\sum_{j\in\mathcal{K},j\neq k}\mathbf{h}_{k}^{H}\mathbf{p}_{j}s_{j}+n_k.$$5 and blocklength $$y_k=\mathbf{h}_{k}^{H}\mathbf{p}_{c}s_{c}+\mathbf{h}_{k}^{H}\mathbf{p}_{k}s_{k} +\sum_{j\in\mathcal{K},j\neq k}\mathbf{h}_{k}^{H}\mathbf{p}_{j}s_{j}+n_k.$$6 bits, the relative gain of Fin C-RSMA over Fin SDMA increases from $$y_k=\mathbf{h}_{k}^{H}\mathbf{p}_{c}s_{c}+\mathbf{h}_{k}^{H}\mathbf{p}_{k}s_{k} +\sum_{j\in\mathcal{K},j\neq k}\mathbf{h}_{k}^{H}\mathbf{p}_{j}s_{j}+n_k.$$7 in the underloaded scenario to $$y_k=\mathbf{h}_{k}^{H}\mathbf{p}_{c}s_{c}+\mathbf{h}_{k}^{H}\mathbf{p}_{k}s_{k} +\sum_{j\in\mathcal{K},j\neq k}\mathbf{h}_{k}^{H}\mathbf{p}_{j}s_{j}+n_k.$$8 in the overloaded scenario, while the gain of Fin N-RSMA over Fin SDMA increases from $$y_k=\mathbf{h}_{k}^{H}\mathbf{p}_{c}s_{c}+\mathbf{h}_{k}^{H}\mathbf{p}_{k}s_{k} +\sum_{j\in\mathcal{K},j\neq k}\mathbf{h}_{k}^{H}\mathbf{p}_{j}s_{j}+n_k.$$9 to $$R_{c,k}=\log_{2}\left(1+\frac{|\mathbf{h}_{k}^{H}\mathbf{p}_{c}|^{2}} {\sum_{j\in\mathcal{K}}|\mathbf{h}_{k}^{H}\mathbf{p}_{j}|^{2}+\sigma_{n,k}^{2}}\right),$$0 (Xu et al., 2022). The same paper also reports that, in an overloaded non-cooperative multicast setting, RSMA improves performance by $$R_{c,k}=\log_{2}\left(1+\frac{|\mathbf{h}_{k}^{H}\mathbf{p}_{c}|^{2}} {\sum_{j\in\mathcal{K}}|\mathbf{h}_{k}^{H}\mathbf{p}_{j}|^{2}+\sigma_{n,k}^{2}}\right),$$1 times relative to SDMA at blocklength $$R_{c,k}=\log_{2}\left(1+\frac{|\mathbf{h}_{k}^{H}\mathbf{p}_{c}|^{2}} {\sum_{j\in\mathcal{K}}|\mathbf{h}_{k}^{H}\mathbf{p}_{j}|^{2}+\sigma_{n,k}^{2}}\right),$$2 (Xu et al., 2022).

In multi-cell coordinated HD/FD C-RSMA, the results show that at BS transmit power $$R_{c,k}=\log_{2}\left(1+\frac{|\mathbf{h}_{k}^{H}\mathbf{p}_{c}|^{2}} {\sum_{j\in\mathcal{K}}|\mathbf{h}_{k}^{H}\mathbf{p}_{j}|^{2}+\sigma_{n,k}^{2}}\right),$$3 dBm, the proposed FD C-RSMA achieves 25% over FD C-NOMA and the proposed HD C-RSMA achieves 19% over HD C-NOMA (Elhattab et al., 2024). FD is generally superior when residual self-interference is manageable, because it avoids the HD pre-log penalty; HD remains attractive when self-interference is severe or when relay simplicity is prioritized (Elhattab et al., 2024). The same study also reports that cooperative schemes outperform non-cooperative CoMP RSMA and CoMP NOMA in low-power regimes because relaying compensates for weak direct links to CEUs (Elhattab et al., 2024).

A coherent interpretation emerges across these papers. Cooperative RSMA gains are largest when the common-stream bottleneck is severe: overloaded operation, strong user heterogeneity, short-packet fairness constraints, or cell-edge weakness. This suggests that the common stream is most valuable not merely as a partial interference-decoding device, but as a robustness layer whose reliability can be selectively strengthened through cooperation.

6. Adjacent literatures and frequent misconceptions

A recurring source of confusion is the use of “cooperation” to describe any RSMA configuration with a common stream. That interpretation is too broad. Several important RSMA papers are highly informative for cooperative RSMA, yet do not study user cooperation, relaying, or multi-transmitter RS encoding in the strict sense.

The multi-carrier joint communications-and-jamming papers preserve the 1-layer RS logic in constrained multi-carrier settings with imperfect CSIT, pilot-subcarrier jamming, and per-subcarrier common-rate allocation. All intended users decode the common stream, perform SIC, and then decode their private streams, so the receivers “cooperate” only through a shared decoding layer; there is no relay or second hop (Dizdar et al., 2021, Dizdar et al., 2021). Those works are valuable because they show that the common/private architecture remains robust when additional functions and hard coexistence constraints are present, but they are not cooperative RSMA papers in the relay or CoMP sense.

The DFRC/ISAC papers likewise demonstrate that the common stream can serve as a shared functional resource. In DFRC, the RS common stream manages inter-user interference, manages communication-radar interference, and contributes to radar beampattern approximation; one paper explicitly states that the common stream fulfills this “triple function” (Xu et al., 2021). In RSMA-RadCom and multi-target ISAC, the same common/private decomposition is used to balance communication and sensing objectives, often outperforming SDMA in the communication-sensing trade-off (Xu et al., 2020, Chen et al., 2023). These works are adjacent rather than strictly cooperative.

The covert-communications and reliability-oriented splitter papers are similarly indirect. The covert RSMA work studies a stochastic control problem over beamforming, blocklength splitting, and common-rate allocation under a KL-divergence covertness constraint, but explicitly states that it is not cooperative RSMA in the relaying or CoMP sense (Hieu et al., 2022). The 2025 splitter paper proposes channel-dependent replication of vulnerable private symbols into the common stream in a multi-carrier downlink and notes that all users still decode the common stream first, yet there is no relay or user forwarding (Ali et al., 16 Apr 2025).

Finally, the uplink RSMA papers provide foundations that are methodologically relevant but architecturally non-cooperative. The 2016 Gaussian MAC paper gives a constructive algorithm for realizing any dominant-face rate tuple through virtual-user splitting and SIC order design (Mao et al., 2016), while the 2019 uplink RSMA paper optimizes splitting, uplink power allocation, and SIC order under proportional fairness (Yang et al., 2019). Both are useful for understanding message splitting and decoding order, but neither studies cooperation among nodes.

The most accurate boundary is therefore the following: cooperative RSMA refers primarily to architectures with explicit user relaying or network-side coordinated transmission; adjacent RSMA uses the same common/private machinery for shared decoding, multi-function coordination, or robustness, but without cooperative forwarding.

7. Limitations and research directions

The present cooperative RSMA literature remains relatively narrow in scope. The CoMP JT generalized RSMA study assumes perfect and instantaneous availability of both user data and CSI at the central controller, full BS cooperation, and no backhaul delay or CSI imperfection (Mao et al., 2018). The finite-blocklength user-relaying work assumes perfect CSI, perfect SIC, fixed relay group $$R_{c,k}=\log_{2}\left(1+\frac{|\mathbf{h}_{k}^{H}\mathbf{p}_{c}|^{2}} {\sum_{j\in\mathcal{K}}|\mathbf{h}_{k}^{H}\mathbf{p}_{j}|^{2}+\sigma_{n,k}^{2}}\right),$$4, half-duplex decode-and-forward relays, and a discretized one-dimensional search over the cooperative-phase blocklength (Xu et al., 2022). The multi-cell JT-CoMP C-RSMA paper assumes full CSI at a centralized controller, an underloaded regime, 1-layer RSMA only, and exhaustive search over the HD time fraction $$R_{c,k}=\log_{2}\left(1+\frac{|\mathbf{h}_{k}^{H}\mathbf{p}_{c}|^{2}} {\sum_{j\in\mathcal{K}}|\mathbf{h}_{k}^{H}\mathbf{p}_{j}|^{2}+\sigma_{n,k}^{2}}\right),$$5 (Elhattab et al., 2024).

The survey literature explicitly identifies several open directions. For cooperative user relaying, more attention is needed for CRS with imperfect CSIT, as existing CRS works largely assume perfect CSIT (Mao et al., 2022). The same survey also calls for joint relaying-user selection and precoding, rather than fixed or heuristic relay selection (Mao et al., 2022). On the network side, cooperative multi-cell RSMA naturally leads to questions on limited fronthaul, C-RAN/F-RAN realizations, cell-free massive MIMO, integrated terrestrial-satellite systems, SAGIN, UAV networks, V2X, security, and machine-learning-assisted optimization (Mao et al., 2022).

Application papers point to additional technical gaps. Multi-carrier RSMA with jamming highlights the absence of generalized or multi-layer RS across subcarriers and the use of carrier non-cooperative decoding instead of more general cross-subcarrier designs (Dizdar et al., 2021). The multi-cell cooperative paper suggests that practical scaling under residual self-interference, larger user populations, and distributed implementations remains open (Elhattab et al., 2024). The finite-blocklength work shows that infinite-blocklength designs are not sufficient surrogates for short-packet operation, which suggests that cooperative RSMA may need dedicated FBL-aware waveform and scheduling designs rather than simple adaptations of Shannon-rate precoders (Xu et al., 2022).

A plausible synthesis is that future cooperative RSMA will likely combine three themes already visible across the literature: a relayed or jointly transmitted common layer, heterogeneous decoding roles across users or nodes, and robust optimization under imperfect or asymmetric CSI. The existing evidence suggests that these ingredients are complementary rather than competing.