STRMAC Protocols in Multi-Domain Networks

• STRMAC is a framework of space-time adaptive protocols that enhance resource sharing via dynamic scheduling, routing, and coding in diverse communication systems.
• It employs state-aware agent selection, time-reversal processing, and space-time rate-splitting to achieve improvements such as higher throughput and reduced latency.
• Empirical results show gains of 4.4–23.8% accuracy in LLM collaboration and 2–3× throughput in physical networks, underscoring its practical efficiency.

STRMAC refers to several distinct but converging concepts and protocols in high-performance communication, collaboration, and networking, unified by the use of spatial or temporal dynamics for medium access, coordination, or routing. Representative STRMAC frameworks and protocols span multi-agent collaboration, wireless network-on-chip (WNoC) design, multi-hop underwater acoustic networking, and satellite communications. This article surveys the principal STRMAC definitions and instantiations, organizing them by domain, method, and technical foundation.

1. Definitions and Domains of STRMAC

STRMAC encompasses protocols and frameworks that (i) perform explicit space-time or state-aware medium access or agent scheduling, and (ii) leverage spatial/temporal focusing, routing, or coding to improve throughput, efficiency, or robustness. Notable instances include:

• State-aware routing for multi-agent LLM collaboration—STRMAC as a routing framework that enables efficient sequential agent selection based on global state and agent knowledge (Wang et al., 4 Nov 2025).
• Time-reversal (TR) MAC for wireless and underwater networks—STRMAC (sometimes written "Spatial Time-Reversal MAC") protocols leveraging space-time focusing to enable parallel transmissions with collision minimization, both intra-chip (WNoC) (Bandara et al., 14 Aug 2024) and in underwater acoustic environments (Zhao et al., 2019).
• Space-Time Rate-Splitting Multiple Access (ST-RSMA) for satellite multi-user systems—integrating space-time coding into the MAC and access layer to maximize diversity and fairness (Seong et al., 20 Oct 2025).

A commonality is the dynamic adaptation to spatial, temporal, or state information to optimize resource sharing or agent selection in distributed settings.

2. STRMAC in Multi-Agent LLM Collaboration

The STRMAC framework for multi-agent systems addresses the agent scheduling problem in compositional LLM-based collaborations. Unlike static pipelines or graph-based strategies, STRMAC integrates the following mechanisms (Wang et al., 4 Nov 2025):

• State Encoding ($S_t$): The evolving problem-solving state (original query and agent interaction history) is encoded by a lightweight router encoder $E_R$, generating a state embedding $z_t$.
• Agent Expertise Embedding ($e_i$): Each agent’s private context is encoded via a fixed large LLM-based encoder $E_L$ to obtain an expertise embedding $e_i$.
• Dynamic Routing Policy: At decision step $t$, the router scores all agents via $\langle z_t, e_i \rangle$ and selects the most suitable agent for that state. The routing policy is defined as:

$$\pi(i \mid S_t) = \mathrm{softmax}_j \langle z_t, e_j \rangle, \quad i_t = \arg\max_i \langle z_t, e_i \rangle$$

• Self-Evolving Data Generation: The system employs solution-aware pruning and router-guided expansion to drastically reduce the number of annotated paths required for supervised training, achieving up to 90.1% reduction in data generation overhead.

Quantitative results: STRMAC increases performance relative to random or static chains with typical absolute gains of +4.4–23.8% accuracy and radical data-efficiency improvements (Wang et al., 4 Nov 2025). The design is inherently sequential, selecting only one agent per step, with training performed using a contrastive cross-entropy loss over the cosine similarities between state and agent embeddings.

3. Time-Reversal STRMAC Protocols in Physical Networks

3.1 Underwater Acoustic Networks

STRMAC (Spatiotemporal Reversal-based MAC) utilizes active time-reversal (TR) physical-layer processing to exploit multipath focusing, enabling near-interference-free concurrent communications in multi-hop underwater acoustic networks (Zhao et al., 2019). The key protocol elements include:

• Active TR handshake: Nodes exchange probe (P-R/Pro) packets to estimate the channel impulse response (CIR), then transmit TR-Data packet waveforms obtained by convolving data with the normalized, time-reversed CIR.
• Collision avoidance: Link cross-correlation is measured; if normalized cross-correlation $|\eta_{AB,AJ}|$ exceeds a threshold $\eta_\mathrm{th}$, transmission is deferred.
• Scheduling and ARQ: Packet-scheduling leverages probe overhearing and collision/backoff procedures, avoiding conventional carrier-sense (ineffective underwater).

Performance: Simulations yield 2–3× higher throughput, 30–50% lower end-to-end delay, and data drop rates below 8% (vs. >40% for CSMA/CA) under high contention (Zhao et al., 2019).

3.2 Wireless Network-on-Chip (WNoC)

Within WNoC, STRMAC utilizes TR precoding to carve out multiple spatially separated “beams” per frequency channel. The protocol's primary steps are (Bandara et al., 14 Aug 2024):

• Offline CIR characterization.
• TR filtered preamble transmission for channel reservation.
• TR-based data and ACK exchange using a three-phase handshake.
• Support for $N_\mathrm{PT}$ parallel transmissions (with cross-correlation constraints) in a single frequency slot.

Results: STRMAC in WNoC achieves aggregate throughput up to 120 Gbps (with $N_\mathrm{PT}=4$ parallel streams), 20–40% lower median latency at observable loads, and <5% collision probability up to saturation, all with single-channel spectral efficiency.

4. STRMAC as Space-Time Rate-Splitting Multiple Access

The ST-RSMA (Space-Time Rate-Splitting Multiple Access) framework is designed for multibeam LEO satellite networks, integrating space-time coding into the RSMA paradigm (Seong et al., 20 Oct 2025):

• System architecture: Each user’s message is decomposed into a common part (encoded via orthogonal Alamouti code for diversity) and a private part (user-specific linear precoding).
• Decoding order: Users first decode common streams jointly, perform SIC, then decode their private streams.
• Resource allocation: Power and beamforming vectors are optimized via a WMMSE-based alternating optimization to maximize minimum user rates under imperfect CSI.

Empirical findings: ST-RSMA demonstrates up to 44% improvement in minimum user rate compared to conventional WMMSE-RSMA under high channel uncertainty or user loading, with full diversity gain across all users (Seong et al., 20 Oct 2025).

5. Comparison of STRMAC Protocols Across Domains

STRMAC Variant	Domain/Context	Key Mechanism	Design Outcome	Source
Multi-agent LLM	Multi-agent inference	State-aware router	Adaptive, interpretable agent task-split	(Wang et al., 4 Nov 2025)
TR-MAC (WNoC)	On-chip wireless	TR spatial beams	Parallel low-latency link contention	(Bandara et al., 14 Aug 2024)
TR-MAC (UWAN)	Acoustic networking	Probe handshake, TR	Collision-avoidance, throughput scaling	(Zhao et al., 2019)
ST-RSMA	LEO satellite MIMO	Space-time RSMA	Full diversity, min-rate maximization	(Seong et al., 20 Oct 2025)

Across these diverse domains, STRMAC protocols share methodological foundations in adaptive or focusing-based access (state, space, or time), and all quantitatively outperform static baselines in key performance metrics, including throughput, reliability, delay, fairness, or data efficiency.

6. Technical and Practical Considerations

STRMAC implementations subject to physical-layer time-reversal (TR) require:

• Explicit channel state measurement and estimation, supporting TR filter computation and energy focusing.
• Careful design of thresholding and scheduling to avoid interference in high-density topologies.
• For agent-based STRMAC, persistent embedding quality (agent context, router generalization), and mechanisms for relabeling or retraining as agent capabilities or task distributions evolve.

System constraints (e.g., channel coherence time in underwater networks, or static CIR in WNoC) determine the feasibility of TR-based STRMAC. In LLM-based multi-agent settings, scalability is achieved via path-pruned data augmentation and supervised learning, reducing the data demands exponentially.

7. Open Issues and Prospective Directions

Current limitations of STRMAC frameworks include:

• Sequentiality in agent routing: STRMAC for LLM-based agents currently accommodates only single-agent execution per step; generalization to parallel (fan-in/fan-out) dialogue remains an open area (Wang et al., 4 Nov 2025).
• Physical implementation trade-offs: TR signal processing hinges on the hardware’s ability to generate and sense arbitrary TR waveforms and to perform rapid, coherent estimation of CIRs (Zhao et al., 2019, Bandara et al., 14 Aug 2024).
• Robustness to evolved contexts: In routing-based STRMAC, static agent embeddings may require periodic refresh; richer routing mechanisms (beyond cosine similarity) and RL-based dynamic adaptation are possible future developments (Wang et al., 4 Nov 2025).
• Optimization complexity: WMMSE and alternating minimization are numerically efficient for current problem scales, but further work is required for ultra-large or ultra-dense deployments (e.g., 1,000+ core WNoC, dense LEO satellite constellations) (Seong et al., 20 Oct 2025, Bandara et al., 14 Aug 2024).

STRMAC protocols collectively highlight a paradigm of leveraging space, time, and state information for principled, scalable, and robust resource coordination across heterogenous collaborative and communication environments.