Sensing & Communication Enhanced Control

• Sensing-and-Communication-Enhanced Control (SCEC) is a paradigm that integrates sensing, wireless communication, and control into a unified feedback loop to enhance system stability and real-time performance.
• It leverages joint optimization and co-design of sensor data fusion, finite-blocklength communication models, and advanced control policies (e.g., MPC) to manage latency and reliability constraints.
• Empirical studies show that balancing sensor count, bandwidth, and quantization is crucial for achieving URLLC-level performance in applications like industrial automation and autonomous systems.

Sensing-and-Communication-Enhanced Control (SCEC) refers to closed-loop control systems in which sensing, communication, and control are not treated as separable modules but are tightly coupled through joint modeling, performance constraints, and co-design of resource allocation and feedback policies. This paradigm has emerged to address the cross-domain challenges posed by wireless networked industrial automation, autonomous vehicular systems, and distributed robotics, where the timeliness, reliability, and informativeness of sensor data—and the quality of their transport—directly impact actuator behavior and closed-loop stability. SCEC leverages joint optimization and advanced models to guarantee system convergence and safety under stringent latency, bandwidth, and reliability constraints, moving beyond legacy designs that treat these domains in isolation (Meng et al., 2023, Meng et al., 2024).

1. Core System Architecture and Signal Flow

SCEC architectures are characterized by a closed information-control loop comprising three deeply interdependent components: sensing, wireless (typically finite-blocklength) communication, and an edge- or cloud-hosted control system (often Model Predictive Control, MPC). A generic signal flow is as follows:

• At each sampling instant, a set of $k_s$ sensors on the plant produce raw measurements $Y_i$, which are packetized and transmitted over an uplink queue (typically over a wireless channel with constrained $W$, $L$).
• The controller fuses received, potentially delayed/noisy sensor data (e.g., via maximum-likelihood or Kalman filtering), computes control commands, and relays them downstream through a downlink queue to actuators.
• This closed-loop cycle, with total loop delay $D_c$, repeats at the process time-scale (Meng et al., 2023).

Sensing, communication, and control are not independent:

• Increasing $k_s$ or sampling frequency improves state estimation but raises uplink arrival rate $\lambda_u$ and associated queuing, latency, and packet drop probability $\epsilon_u$.
• Communication bottlenecks (limited $W$, blocklength $L$, and effective capacity $C(\theta, W, \epsilon)$) propagate to the control layer as input staleness and uncertainty.
• Feedback policy design (e.g., delay-compensated MPC) must explicitly compensate for stochastic communication delay and loss, adjusting the controller gain $K$ according to the sensed uncertainty and real-time channel metrics (Meng et al., 2023, Meng et al., 2024).

2. Analytical Frameworks: Joint Models for Sensing, Communication, and Control

Sensing

For a physical state $X\in\mathbb{R}^n$, $k_s$ sensors with noise $N_i\sim\mathcal{N}(0,\sigma^2 I)$ yield $Y_i = X + N_i$. Max-likelihood fusion provides estimate error scaling as $E[\|\hat{X} - X\|^2] = \sigma^2 / k_s$ (Meng et al., 2023).

Communication

Links are modeled using effective capacity (EC) theory for both uplink and downlink:

• EC is a function of QoS exponent $\theta$, SNR, blocklength $L$, and bandwidth $W$, governing maximum stable arrival rates without queue overflow:

$$C(\theta,W,\epsilon)\approx\frac{1}{\theta}\ln[\mathrm{SNR}\,\beta](\frac{W\theta}{\ln2} - 1) - \frac{1}{\mathrm{SNR}\,\beta}$$

• Queuing delays $D_u,D_d$ and packet loss $\epsilon_u, \epsilon_d$ are connected to service rates and allocated bandwidth (Meng et al., 2024).

Control

Discrete-time linear/quasi-linear plant dynamics $X_{t+1} = \widetilde{A} X_t + \widetilde{B} U_t$, with delay- and loss-compensated feedback policy, where $U_t=K\hat{X}_t$ or involves batch MPC. Under packet drops (Bernoulli $\eta$), dynamics adapt to $$X_{t+1} = \widetilde{A} X_t + \eta\widetilde{B} Ku_t$$ (Meng et al., 2023).

Stability and Performance Trade-offs

SCEC systems enforce Lyapunov-based inequalities encapsulating the interaction:

$(1-\epsilon_c)F_1(X_t,K,\sigma) \leq F_2(X_t)$

where $F_1$ and $F_2$ quantify closed-loop contraction under packet loss and estimation error. In (Meng et al., 2024), this is further coupled with quantization bits $r$ and end-to-end delay $D_{c,\max}$, yielding an explicit inequality for performance-constrained stability.

3. Joint Optimization Formulations and Solution Methods

The SCEC design problem is generally posed as a nonconvex optimization that minimizes a finite-horizon cost function (typically of MPC/LQR form), with variables including control gain $K_t$, sensor count $k_s$, uplink/downlink bandwidths $W_u,W_d$, and packet-loss budgets $\epsilon_u,\epsilon_d$. Constraints capture:

• EC stability: $C_u \geq \lambda_u$, $C_d \geq \lambda_d$
• Latency budget: $D_{c,\max} \leq D_0$
• Plant stability: Lyapunov inequality as above
• Bandwidth resource: $W_u+W_d \leq W_0$
• Control dynamics under average packet loss: $$X_{t+1} = \widetilde{A} X_t + (1-\epsilon_c)\widetilde{B} K_t X_t$$ (Meng et al., 2023).

Representative solution techniques include:

• Interior-point nonlinear programming (e.g., IPOPT) for direct nonconvex programs (Meng et al., 2023)
• Evolutionary algorithms (differential evolution, genetic algorithms) for heuristic global optimization (Meng et al., 2024)
• Alternating optimization and convex relaxation for resource allocation with additional structure (e.g., in multi-agent or networked control settings) (Wang et al., 11 Aug 2025, Lei et al., 2024)

4. Empirical Insights and System-Level Trade-offs

Simulation and experimental studies across multiple SCEC works provide quantification of key resource allocations and trade zones. Empirically:

• Increasing $k_s$ improves convergence rate and reduces overshoot up to the point where uplink congestion sharply increases $\epsilon_u$, degrading stability (Meng et al., 2023).
• URLLC-level performance for factory control (latency $D_0 \sim 100 - 150$ ms, bandwidth $W_0 \sim 1$ MHz, SNR $\geq 20$ dB) is achievable, provided the sensor-communication-resource trade-off is balanced (Meng et al., 2023).
• Over-allocation of quantization bits $r$ (sensing resolution) without matching uplink bandwidth leads to increased delays and instability (Meng et al., 2024).

Key metrics include settling time, accumulated LQR cost, queuing delay distributions, and packet loss rates as a function of resource splits (Table 1 and Fig. 2/3 in (Meng et al., 2023, Meng et al., 2024)). The characteristic result is the existence of a "sweet spot"—e.g., $k_s \sim 10-12$, quantization $r \sim 6-8$ bits, latency $D_0 \sim 100$ ms—where system reliability and control speed are jointly optimized.

5. Design Guidelines and Practical Recommendations

SCEC research yields detailed design guidance:

• Sensing: Set $k_s$ moderate enough to reduce estimation error ($\sigma^2/k_s$) below a performance threshold, constrained by uplink EC capacity to avoid sharp packet-loss transitions (Meng et al., 2023).
• Communication: Allocate $W_u, W_d$ for $C_u, C_d$ exceeding arrival rates by a margin $\delta \sim 0.1$ for stochastic fade robustness. Balance total $W_0$ across devices for fairness.
• Latency: Do not over-constrain $D_0$ beyond the sum of average service times, as aggressive latency yields high $\epsilon_c$ with little gain in control performance.
• Stability: Enforce the Lyapunov inequality with a safety margin ($\mu \sim 0.05$) to tolerate model inaccuracy.
• Quantization: Bits per sample $r$ chosen to balance estimation and uplink load, typically $6-8$.
• Algorithmic: Global search heuristics are useful due to nonconvexities; EC-based models are preferred over simple fixed-delay approximations.

See summary design recommendations in Table 1:

Design Variable	Typical Value	Rationale
Sensors $k_s$	$10-12$	Sufficient estimation accuracy, avoid EC overload
Bandwidth $W_0$	$1$ MHz	URLLC compliance, per device
Latency $D_0$	$100-150$ ms	Minimizes total E2E delay, avoids excessive loss
Quantization bits $r$	$6-8$	Balance estimation error and channel load
SNR	$\geq 20$ dB	Supports EC constraints for reliable control

6. Extensions and Future Directions

Recent works have extended SCEC to distributed, multi-agent, and cognitive systems:

• Multi-loop scheduling (e.g., across $K$ robots) with shared network and compute resources on satellite/UAV-edge platforms (Lei et al., 2024).
• Bilevel formulations linking physical-layer ISAC resource optimization to high-level motion planning, with safety bounds derived from the Cramér–Rao bound and explicit occupancy inflation (Jin et al., 27 Oct 2025).
• Event- or goal-triggered scheduling, semantic communication, and reinforcement learning-based adaptation for resource-aware closed-loop control (Li et al., 6 May 2025, Pan et al., 22 Dec 2025, Liu et al., 2 Mar 2026).
• Integration with AI-networked fabrics (DS3C, orchestration layers in 6G) for vertical real-time industrial applications (Vukobratović et al., 2023).

A central insight is that cross-layer co-design, balancing sensing accuracy, communication resource, and control feedback gains, is essential. Over-optimization in any single domain (e.g., ultra-low latency, ultra-high sensing resolution) typically leads to resource bottlenecks and degraded overall closed-loop performance (Meng et al., 2023, Meng et al., 2024).

7. Standardization, Benchmarks, and Open Challenges

SCEC principles underpin emerging standards in URLLC (3GPP Rel. 16–19), 6G, and industrial wireless control. Benchmarks include sub-ms end-to-end latency, five-nines reliability, and resource-adaptive feedback control for mobile robots and AGVs (Meng et al., 2023, Vukobratović et al., 2023). Open challenges involve:

• Real-time distributed SCEC with heterogeneous edge/cloud/fog compute
• Scalable, learning-enabled adaptation under time-varying wireless channels
• Quantitative robustness to model inaccuracies and nonstationary workloads
• Seamless integration of semantic and goal-oriented communication within SCEC frameworks

Contemporary research converges on the necessity of explicitly modeling—then jointly optimizing—the intertwined effects of sensing, communication, and control to achieve provable, reliable, and efficient performance in demanding wireless-controlled industrial and autonomous systems (Meng et al., 2023, Meng et al., 2024, Jin et al., 27 Oct 2025).