UniCast: User-Specific Communication Models

• UniCast is user-specific communication characterized by streams intended for a single destination, employing techniques such as beamforming and resource optimization.
• It is applied in diverse settings including mobile ad hoc networks, publish–subscribe systems, and multi-antenna cellular downlinks to enhance efficiency.
• Research on UniCast emphasizes optimizing coding strategies, interference management, and energy efficiency to improve overall network performance.

Searching arXiv for recent and relevant papers on unicast and joint unicast–multicast transmission. UniCast denotes user-specific or destination-specific communication rather than common delivery to all users or to a multicast group. In the cited literature, the term appears in several distinct technical settings: one-to-one communication in mobile ad hoc networks, direct flow-to-user delivery in publish–subscribe systems, point-to-point optical circuit switching, user-specific downlink streams in multi-antenna cellular systems, source–destination communication over relay networks, and message-specific coding problems such as multiple unicast and index coding. Across these settings, unicast is best understood as a service model whose efficiency depends on topology, CSI, interference structure, coding strategy, and coexistence with multicast or broadcast traffic (Debnath et al., 2010, Mao et al., 2018).

1. Canonical meaning and formal models

In the networking literature represented here, unicast is naturally appropriate for one-to-one communication, but it is also used more broadly for any stream intended for exactly one receiver or one user terminal (Debnath et al., 2010). In a multi-antenna downlink, this appears as $K$ user-specific messages $W_1,\ldots,W_K$ superposed with one common multicast message $W_0$, with transmitted signal

$$\mathbf{x}=\mathbf{p}_0 s_0+\sum_{k\in\mathcal{K}}\mathbf{p}_k s_k,$$

where $s_k$ is the stream intended only for user $k$ (Mao et al., 2018). In coordinated multi-cell LDM systems, the same idea appears as one common multicast or broadcast layer plus one dedicated unicast stream per user, with each user decoding the common layer first and then its own unicast stream (Chen et al., 2017).

At the network-information-theoretic level, unicast is formalized as source–destination communication. A composite unicast relay network contains one source, one destination, and $N$ relays, with channel law indexed by $\theta=(\theta_r,\theta_d)$; the source does not know the realized $\theta$, the destination knows the full channel realization, and the relays know partial CSI $\theta_r$ (Behboodi et al., 2012). In multiple-unicast network coding, the directed acyclic graph $W_1,\ldots,W_K$0 supports $W_1,\ldots,W_K$1 sessions $W_1,\ldots,W_K$2, each with its own source $W_1,\ldots,W_K$3 and sink $W_1,\ldots,W_K$4, and the central question is whether routing alone can achieve the whole rate region or whether network coding is required (Meng et al., 2014).

A further abstraction appears in publish–subscribe systems. There, unicast is represented explicitly by a transport decision matrix $W_1,\ldots,W_K$5, where $W_1,\ldots,W_K$6 means flow $W_1,\ldots,W_K$7 is sent to user $W_1,\ldots,W_K$8 using unicast, alongside multicast assignment matrices $W_1,\ldots,W_K$9 and $W_0$0 satisfying $W_0$1 (0901.2687). This formulation makes unicast not merely a default transport, but an optimization variable.

2. Multi-antenna downlink unicast as beamforming and power control

In contemporary cellular and massive-MIMO treatments, unicast is typically the user-specific layer in a jointly optimized downlink. In cooperative multi-cell transmission with limited backhaul, each of $W_0$2 users has exactly one dedicated unicast request, and each unicast message $W_0$3 is associated with its own beamformer $W_0$4 and adaptive BS-cluster indicators $W_0$5 (Chen et al., 2017). After decoding the multicast layer, user $W_0$6 sees unicast SINR

$W_0$7

so unicast interference is managed by beamforming, clustering, and power allocation rather than by decoding other users’ private streams (Chen et al., 2017). The corresponding optimization maximizes a weighted sum of multicast and unicast rates under per-BS power and backhaul constraints, and the paper develops both a branch-and-bound global algorithm and a lower-complexity convex–concave procedure (Chen et al., 2017).

In single-cell massive MIMO under MRT, unicast is studied through closed-form achievable spectral efficiency. For user $W_0$8,

$W_0$9

with

$$\mathbf{x}=\mathbf{p}_0 s_0+\sum_{k\in\mathcal{K}}\mathbf{p}_k s_k,$$0

where $$\mathbf{x}=\mathbf{p}_0 s_0+\sum_{k\in\mathcal{K}}\mathbf{p}_k s_k,$$1 is the unicast downlink power, $$\mathbf{x}=\mathbf{p}_0 s_0+\sum_{k\in\mathcal{K}}\mathbf{p}_k s_k,$$2 is the channel-estimation quality term, and $$\mathbf{x}=\mathbf{p}_0 s_0+\sum_{k\in\mathcal{K}}\mathbf{p}_k s_k,$$3 is the shared power budget (Sadeghi et al., 2021). The unicast objective is a weighted sum spectral efficiency, and the optimal downlink allocation takes the water-filling-like form

$$\mathbf{x}=\mathbf{p}_0 s_0+\sum_{k\in\mathcal{K}}\mathbf{p}_k s_k,$$4

with $$\mathbf{x}=\mathbf{p}_0 s_0+\sum_{k\in\mathcal{K}}\mathbf{p}_k s_k,$$5 and $$\mathbf{x}=\mathbf{p}_0 s_0+\sum_{k\in\mathcal{K}}\mathbf{p}_k s_k,$$6 (Sadeghi et al., 2021). The same work proves that the Pareto region of the joint unicast–multicast problem is convex, hence the system should serve the unicast and multicast UTs at the same time-frequency resource (Sadeghi et al., 2021).

In multicell LDM-based broadcast–unicast transmission, broadcast is decoded first and canceled before unicast decoding, so the unicast SINR matches the TDM unicast SINR while avoiding a time-sharing penalty. The optimization minimizes total transmit power under user-specific unicast-rate constraints and a common broadcast-rate constraint, with robust treatment of imperfect CSI and imperfect channel coding via SCA, SDR-based lower bounds, and a distributed dual-decomposition method for the unicast beamformers (Zhao et al., 2019).

3. Non-orthogonal coexistence with multicast and broadcast

A major contemporary theme is that unicast need not be isolated from common traffic. In non-orthogonal unicast–multicast transmission, each receiver already needs one layer of SIC to remove the multicast stream before decoding its own unicast stream. Rate-splitting exploits that same receiver architecture by splitting each unicast message $$\mathbf{x}=\mathbf{p}_0 s_0+\sum_{k\in\mathcal{K}}\mathbf{p}_k s_k,$$7 into a common part $$\mathbf{x}=\mathbf{p}_0 s_0+\sum_{k\in\mathcal{K}}\mathbf{p}_k s_k,$$8 and a private part $$\mathbf{x}=\mathbf{p}_0 s_0+\sum_{k\in\mathcal{K}}\mathbf{p}_k s_k,$$9, and encoding the multicast message together with all unicast common parts into a super-common stream $s_k$0 decoded by all users (Mao et al., 2018). The total unicast rate of user $s_k$1 becomes

$s_k$2

while the multicast QoS is enforced through $s_k$3 (Mao et al., 2018). The precoders are designed by maximizing the weighted sum rate of unicast messages subject to multicast QoS and sum-power constraints, and the nonconvex WSR problem is solved through an equivalent WMMSE reformulation and alternating optimization (Mao et al., 2018).

Conceptually, RS softly bridges MU-LP/SDMA and NOMA/SC–SIC. MU-LP decodes no unicast interference and treats all of it as noise; SC–SIC decodes all or most of a strong interferer’s stream; RS decodes only a common part of multi-user interference and leaves the rest in private streams treated as noise (Mao et al., 2018). In the reported two-user simulations, RS always achieves a rate region equal to or larger than MU-LP and SC–SIC, and when $s_k$4 it reduces to MU-LP behavior (Mao et al., 2018). Because one SIC layer is already required to separate multicast from unicast, the paper’s main practical claim is that RS provides rate and QoS enhancements without increasing receiver complexity compared with MU-LP in joint unicast–multicast transmission (Mao et al., 2018).

The same architectural point was experimentally validated in a two-user MISO SDR platform for NOUM. RSMA-based NOUM uses the common stream for both multicast and parts of the users’ unicast messages, whereas MULP-based NOUM uses the common stream only for multicast (Lyu et al., 2024). Under measured MCS-limited throughputs, RSMA-based NOUM provides higher unicast throughput than MULP-based NOUM whenever the multicast data rate is below $s_k$5 Mbps in Case 1, $s_k$6 Mbps in Case 2, and $s_k$7 Mbps in Cases 3 and 4 (Lyu et al., 2024). The same study reports that the optimal power-split parameter $s_k$8 decreases from $s_k$9 to $k$0 as channel correlation increases, with the common stream’s share of maximum sum throughput rising from $k$1 to $k$2, and it notes that values of $k$3 close to but not equal to $k$4 can cause common-stream decoding failure and SIC error propagation (Lyu et al., 2024).

4. Network-layer scaling, dissemination, and transport selection

At the network layer, unicast can be structurally inefficient when the delivery objective is one-to-many or many-to-many rather than one-to-one. In MANETs, the paper comparing DSR and BCAST emphasizes that if there are $k$5 senders and $k$6 receivers, a unicast approach requires $k$7 logical connections, and each sender–receiver pair may require route discovery, maintenance, repair, and MAC-layer support (Debnath et al., 2010). Under identical ns-2 scenarios with 50 mobile nodes, pause time $k$8, and CBR traffic at 2 packets/s, DSR degrades sharply with sender scaling: with five senders, BCAST achieves a PDR of $k$9 while DSR achieves $N$0; DSR’s normalized routing load is $N$1 packets versus $N$2 for BCAST; and throughput is $N$3 kbps for DSR versus $N$4 kbps for BCAST (Debnath et al., 2010). The same study gives the concrete example that five senders and 30 receivers require $N$5 unicast connections under DSR (Debnath et al., 2010). This directly counters the common misconception that unicast is a scalable default for group dissemination.

In enterprise publish–subscribe systems, the transport choice is cast as an optimization between multicast overdelivery and unicast duplication. The hybrid formulation introduces $N$6 so that requests are satisfied by $N$7, with total cost

$N$8

combining multicast receive cost, multicast send cost, and unicast cost (0901.2687). The paper shows that the hybrid channelization problem is NP-hard, proposes a two-step heuristic framework, and reports that the K-means channelization method is the strongest multicast-only baseline while greedy user or greedy flow heuristics are generally best for moving traffic to unicast, depending on the workload (0901.2687). The main empirical conclusion is that hybridization can reduce total host and network resource consumption and, in some settings, beat both multicast-only and unicast-only dissemination (0901.2687). Here unicast is not rejected; it is used selectively where exact delivery is worth the extra sender/network overhead.

In optical compute-cluster fabrics, unicast is the native strength of OCS. The paper on Shufflecast states that OCS are only capable of providing point-to-point unicast circuit connections, and that OCS-based cores are agnostic to data-rate, have negligible or zero power consumption, need no O/E/O conversion in the core, and can achieve close to non-blocking network performance for point-to-point communication (Das et al., 2021). Shufflecast is presented as a multicast complement to any unicast-capable OCS-based network core, using passive optical splitters. Its multicast fabric scales as

$N$9

supports all ToRs within at most $\theta=(\theta_r,\theta_d)$0 hops, and consumes $\theta=(\theta_r,\theta_d)$1 transceiver ports per ToR (Das et al., 2021). The broader implication is that a strong unicast fabric does not automatically provide efficient one-to-many service; multicast may need an additional architecture specifically designed not to sacrifice the optical properties that make OCS attractive for unicast.

5. Information-theoretic and coding-theoretic treatments

In relay networks, unicast appears as a single end-to-end message flow whose best relaying mode depends on local channel conditions. The Selective Coding Strategy paper studies composite unicast networks where the source does not know the channel realization, the destination knows the full realization, and relays know partial CSI $\theta=(\theta_r,\theta_d)$2 (Behboodi et al., 2012). Each relay defines a DF decision region $\theta=(\theta_r,\theta_d)$3 and dynamically chooses between decode-and-forward and compress-and-forward; the paper generalizes noisy network coding to mixed DF/CF relay sets and allows DF relays to exploit CF relays via offset coding (Behboodi et al., 2012). In the Gaussian two-relay example, no single non-selective strategy is uniformly best over fading realizations, whereas SCS significantly improves the asymptotic error probability and becomes close to the cutset bound (Behboodi et al., 2012). This shows that unicast reliability in relay networks can benefit from state-dependent coding-mode selection rather than fixed relay roles.

In network coding, the fundamental question is whether multiple unicast requires coding gains at all. The paper on routing-optimal networks introduces information-distributive networks, defined by cumulative cut-set sequences, distributive cut-set sequences, and extendable path-set sequences, and proves that if a network is information-distributive, then it is routing-optimal, namely

$\theta=(\theta_r,\theta_d)$4

(Meng et al., 2014). The proof constructs routing flows directly from the conditional mutual informations assigned to ordered cut-set edges in a network-coding solution. This sharply limits a frequent overgeneralization: coding gain is not universal across multiple-unicast networks; there exist topological classes where routing alone achieves the whole rate region (Meng et al., 2014).

A related algebraic treatment appears in unicast-uniprior index coding. There, each receiver demands a unique subset of messages and knows another unique subset of messages, and the task is to compute the minrank of the corresponding fitting matrix over $\theta=(\theta_r,\theta_d)$5 (Ambadi, 2019). The paper establishes structural properties special to unicast-uniprior instances, including that a fitting matrix cannot have more than two identical rows and that minimally dependent row sets correspond to unicycles in the side-information graph (Ambadi, 2019). The proposed algorithm constructs $\theta=(\theta_r,\theta_d)$6 demand tables, treats each column as a single-unicast uniprior problem, computes each column minrank, and then minimizes the sum over all valid tables. In the reported examples, this reduces instances that would require $\theta=(\theta_r,\theta_d)$7 or $\theta=(\theta_r,\theta_d)$8 brute-force fitting-matrix checks to $\theta=(\theta_r,\theta_d)$9 and $\theta$0 structured subproblems, respectively (Ambadi, 2019).

6. Specialized directions: cooperative beamforming, ISAC, and energy-efficient joint designs

In ad hoc wireless networks under a free-space line-of-sight model, unicast can be accelerated by distributed beamforming. The beamforming paper shows that on a $\theta$1 grid, with each node’s transmit power restricted so a single node can only reach its nearest horizontal or vertical neighbors, a relay-rectangle construction achieves unicast in

$\theta$2

rounds and $\theta$3 energy, and for $\theta$4 there is a matching lower bound $\theta$5 (Janson et al., 2014). For random node placements with transmission range $\theta$6 and $\theta$7, the same problem is solved in $\theta$8 rounds with $\theta$9 energy (Janson et al., 2014). This is a specialized physical-layer result, but it broadens the encyclopedia picture: unicast delay can sometimes be reduced far below ordinary multi-hop behavior by exploiting coherent cooperation.

In integrated radar–communication, unicast is optimized subject to both multicast QoS and sensing constraints. The NOMA-aided joint radar and multicast–unicast framework uses a multicast beamformer $\theta_r$0 and a unicast beamformer $\theta_r$1, with transmitted signal

$\theta_r$2

The C-user decodes multicast first and then, after SIC, achieves unicast rate

$\theta_r$3

while the actual multicast rate is limited by the weaker of the multicast decoding rates at the C-user and R-user (Mu et al., 2022). The design maximizes unicast rate under a multicast-rate constraint and a radar beampattern mismatch constraint, and the resulting nonconvex problem is solved by a penalty-based iterative algorithm (Mu et al., 2022). The system-level lesson is that unicast can remain the design objective even when the same waveform must simultaneously satisfy common-service and sensing requirements.

A different specialized direction is energy-efficient joint unicast and multicast beamforming with multi-antenna user terminals. In that model, each user terminal can receive one group-specific common multicast stream and up to $\theta_r$4 private unicast streams, where $\theta_r$5 is the number of receive antennas (Tervo et al., 2017). Private and common streams are decoded simultaneously and independently, with all interference treated as noise, and the optimization maximizes total delivered throughput divided by total consumed power under per-BS power constraints and minimum multicast-rate requirements (Tervo et al., 2017). The paper’s main practical observation is that multicast-only transmission can underuse stronger users’ receive spatial dimensions, whereas adding private streams exploits those dimensions and improves network energy efficiency (Tervo et al., 2017).

Taken together, these works show that UniCast is not a single protocol family but a recurring architectural primitive. In some settings it is the natural one-to-one service mode; in others it is a beamformed layer coexisting with multicast or broadcast; elsewhere it is an optimization variable, a coding objective, or a component inside radar–communication integration. The literature repeatedly converges on the same technical conclusion: the performance of unicast depends less on the label itself than on how precisely the system represents destination specificity, manages shared resources, and exploits or avoids interaction with common traffic.