When Semantic Communication Meets Queueing: Cross-Layer Latency and Task Fidelity Optimization Yalin E. Sagduyu and Tugba Erpek
arXiv:2605.05514v1 [cs.IT] 6 May 2026
Nexcepta, Gaithersburg, MD, USA
Abstract—Semantic communication (SemCom) with learned encoder-decoder architectures enables end-to-end learning of compact task-oriented representations optimized for the wireless channel, reducing channel resources needed to convey taskrelevant information and improving spectrum efficiency. This paper studies semantic image transmission over block Rayleigh fading with AWGN using a multi-task semantic autoencoder that jointly reconstructs images and predicts labels from the received waveform. The latent dimension (complex channel uses per source sample) serves as a cross-layer control variable governing semantic fidelity and channel resource usage. We characterize the resulting latency-task fidelity tradeoff: larger latent representations improve inference accuracy but increase service time, channel uses, and queueing delay. Building on this insight, we develop online semantic-rate controllers that adapt the latent dimension per update under a long-term semantic error constraint. A queue-aware drift-plus-penalty policy minimizes delay subject to an average semantic error cap, while a complementary age-aware policy minimizes time-average Age of Information (AoI). By adapting the semantic rate to congestion and fidelity requirements, the proposed framework improves spectrum utilization and enables timely semantic updates with significantly lower delay and AoI than fixed-rate baselines. Index Terms—Semantic communications, deep neural networks, latency, spectrum efficiency, cross-layer design.
I. I NTRODUCTION Wireless spectrum is scarce, particularly for emerging applications such as autonomous systems, edge intelligence, and distributed sensing. Conventional communication systems reliably deliver all source bits, even when many are irrelevant to the end task, leading to inefficient spectrum utilization when the objective is inference, detection, or situational awareness rather than exact signal reconstruction. Semantic communication (SemCom) addresses this inefficiency by transmitting compact task-relevant representations instead of raw data, reducing channel uses per update and improving spectrum efficiency and access to shared wireless resources. SemCom optimizes transmission for task-relevant meaning, rather than exact symbol recovery [1]–[3]. For highdimensional sources such as images, learned encoder-decoder models map each sample to a compact latent representation transmitted over the wireless channel and decoded into semantic outputs [4]–[7]. The latent dimension (complex channel uses per sample) defines the semantic rate: increasing it improves task fidelity and robustness but also increases transmission time and spectrum consumption, raising latency and limiting the number of semantic updates. This fidelity-service-
time coupling is critical in time-sensitive systems such as autonomous driving, edge video analytics, and UAV situational awareness, motivating joint optimization of communication, computation, and learning resources and highlighting the latent dimension as a key cross-layer control variable. From a network perspective, the latent dimension determines the spectrum resources consumed by each semantic update. Large latent representations waste channel uses and limit supported updates, while aggressive compression degrades semantic fidelity, making adaptive semantic-rate control essential for efficient spectrum utilization. Motivated by latency-critical sensing and situational awareness, we study semantic image transmission over block Rayleigh fading with AWGN. We develop an end-to-end semantic autoencoder that preserves semantic meaning at the receiver and characterize classification accuracy versus signal-to-noise ratio (SNR) and latent dimension, revealing saturation where additional channel uses yield diminishing gains but higher latency. Prior work highlights freshness and representation efficiency [8]–[12], but largely assumes fixed semantic rates, leaving online control under stochastic traffic and fidelity constraints underexplored. The central contribution of this work is a cross-layer control framework that treats the latent dimension as the action for each semantic update and adapts the semantic rate online under a long-term semantic fidelity requirement. We impose an average semantic error constraint and develop Lyapunov drift-plus-penalty controllers: a queue-aware policy that minimizes delay and a freshness-aware policy that minimizes time-average Age of Information (AoI). Both operate without retransmission and enforce the error cap using a virtual fidelity queue. Compared with prior SemCom designs that optimize distortion or task accuracy, the proposed framework exposes the accuracy-latency-freshness tradeoff induced by the latent dimension and enables near-optimal online operation under stochastic arrivals. Fixed-latent systems are inefficient: small latent dimensions underutilize semantic fidelity under favorable channels, while large latent dimensions incur unnecessary delay when fidelity pressure is low. The proposed cross-layer adaptation resolves this inefficiency by dynamically matching the semantic rate to queue state and fidelity debt. The remainder of the paper is organized as follows. Sec. II describes the SemCom system. Sec. III presents queue-aware semantic rate control under fidelity constraints. Sec. IV extends the formulation to AoI. Sec. V concludes the paper.
Latency-Driven Semantic Rate Adaptation Semantic Error subject to Semantic Error Constraints 𝑁 Select Latent Dimension 𝑁 Queue 𝑁 State Semantic Semantic Rayleigh Fading Image Encoder Decoder + AWGN Source (Transmitter) (Receiver) Channel
Fig. 1: SemCom system model.
II. S EMANTIC C OMMUNICATIONS
C. Reconstruction Decoder
The system model of SemCom, shown in Fig. 1, consists of a semantic encoder-decoder communicating over the wireless channel, with rate adaptation introduced in Secs. III-IV. A. Semantic Encoder Let x denote an input image with label l ∈ {0, . . . , L − 1}, and let x̃ denote its normalized version with pixel values in [0, 1]. The semantic encoder fenc maps x̃ to a complex baseband sequence s = fenc (x̃) ∈ CN ,
(1)
where N denotes the number of complex channel uses per image. Smaller N enforces stronger compression and semantic abstraction, while larger N provides additional degrees of freedom, improving semantic fidelity and robustness to fading. The encoder is implemented as a deep convolutional neural network (CNN) with residual learning and channelwise attention. An initial convolutional stem extracts lowlevel spatial features. This is followed by multiple stages of residual blocks that progressively increase channel depth while reducing spatial resolution through strided convolutions. Each residual block consists of two convolutional layers with batch normalization and ReLU activation, together with an identity shortcut that stabilizes gradient flow in deep networks. Within each residual unit, squeeze-and-excitation (SE) gating uses spatial pooling and a two-layer fully connected network with a reduction-and-expansion structure to generate adaptive channel weights, enabling the encoder to prioritize semantically informative features, especially when N is small. After the final convolutional stage, global average pooling produces a compact embedding vector e. Two fully connected layers map this embedding to a real-valued vector u ∈ R2N , which is reshaped into a complex baseband sequence s ∈ CN with elements sk = u2k−1 + j u2k ,
k = 1, . . . , N.
(2)
The resulting sequence is normalized per sample to satisfy a unit average power constraint, ensuring a fixed transmit SNR across latent dimensions. B. Channel Model and Receiver-Side Conditioning Transmission occurs over a block Rayleigh fading channel y = h s + n,
h ∼ CN (0, 1),
where a single fading coefficient h applies across the N channel uses of each image, n is additive complex Gaussian noise with variance σ 2 set by the target SNR under unit average transmit power. The receiver produces both a reconstructed image x̂ and semantic logits ẑ ∈ RL via learned decoders. To mitigate fading-induced variation under noncoherent operation, the received latent waveform y is normalized per sample, and a conditioning vector c is derived from its low-order statistics. No receiver channel state information is assumed; robustness follows from normalization and conditioning.
n ∼ CN (0, σ 2 IN ), (3)
The reconstruction decoder maps the received latent representation to an image estimate with the same spatial dimensions as the input (e.g., 32×32×3 for CIFAR-10). The normalized received latent sequence y is flattened, concatenated with a receiver-side conditioning vector c, and passed through two fully connected layers to produce an initial spatial feature map. This map is progressively upsampled using interpolation followed by residual convolutional refinement blocks with channel attention, enabling recovery of spatial structure while avoiding checkerboard artifacts. A final convolution with a bounded activation produces valid pixel values. When the latent dimension N is small, reconstruction is underdetermined and relies more on learned image priors, preserving global semantics but losing high-frequency detail. Increasing N improves Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM), reducing reconstruction ambiguity. D. Semantic Classifier Decoder In parallel, a semantic classifier infers labels from the received waveform. The receiver forms a multi-channel representation from the normalized received latent sequence y. A stack of 1D convolutions processes the sequence along the symbol dimension to extract waveform features. The resulting features are globally pooled, reweighted via channelwise gating, and augmented with the conditioning vector c to improve robustness to fading. The features are then passed through a fully connected network to produce class logits. Larger N increases the effective temporal context, improving robustness at the cost of more channel uses. The predicted label is given by ˆl = arg maxi ẑi . E. End-to-End Training The system is trained using a weighted composite loss that balances reconstruction fidelity and semantic accuracy. The loss consists of a reconstruction term combining mean-squared error (MSE) with a small ℓ1 component to capture global fidelity and edge sharpness, a structural term based on SSIM to preserve local contrast and texture, a classification term based on cross-entropy, and an auxiliary PSNR-based term that complements MSE by capturing reconstruction quality on a logarithmic scale. Training uses stochastic gradient optimization. For each mini-batch, images are encoded into N complex symbols, transmitted over the channel, and processed by both
the latent dimension N acts as the key control variable for semantic fidelity, improving accuracy at the cost of latency and motivating adaptive cross-layer semantic rate tuning.
0.9
0.8
III. Q UEUE -AWARE S EMANTIC R ATE C ONTROL This section develops a queue-aware SemCom controller that treats the latent dimension N as the per-update action. It minimizes delay under an average semantic error constraint via Lyapunov drift-plus-penalty control.
Accuracy
0.7
0.6
0.5 Latent dimension = 5 Latent dimension = 10 Latent dimension = 15 Latent dimension = 20
0.4
0.90
A. Queueing Model and Action Space Consider a single-server queue with exogenous Poisson arrivals {A(t)} at rate λ images per unit time, where A(t) denotes the number of arrivals in [0, t]. Let Q(t) denote the number of waiting images at time t and Qsys (t) denote the total number in the system (including the one in service):
0.85
Qsys (t) = Q(t) + 1{server busy}.
0.3 0
1
2
3
4
5 SNR
6
7
8
9
10
Fig. 2: Accuracy vs. SNR in SemCom.
We use Qsys (t) for all control decisions and delay analysis. Let {tk }k≥1 denote the service start times, and let k index the image entering service at time tk . At each service start time tk , the controller selects a latent dimension
0.80 Accuracy
(4)
0.75
0.70
Nk ∈ N ≜ {N1 , . . . , NM }, 0.65
SNR = 3dB SNR = 5dB SNR = 10dB
0.60 5
10
15
20
25 30 35 Latent Dimension
40
45
50
Fig. 3: Accuracy vs. latent dimension in SemCom.
where Nk is the number of complex channel uses allocated to the kth image and thereby determines its service time. Transmitting the kth image with latent dimension Nk consumes Nk channel uses and therefore incurs deterministic service time Sk = Nk .
decoders. The composite loss is backpropagated through the channel to update all parameters. Training uses the Adam optimizer with batch size 128 and is implemented on NVIDIA RTX PRO 6000 Blackwell GPUs. F. SemCom Performance and Latent-Dimension Tradeoff We use the CIFAR-10 dataset with L = 10 classes for training and evaluation. We first examine classification accuracy versus SNR for several latent dimensions. As shown in Fig. 2, accuracy increases with SNR for all N due to reduced channel distortion, with larger latent dimensions consistently achieving higher accuracy. The gap is largest at low SNR, reflecting implicit redundancy benefits, while saturation occurs at high SNR once semantic information is preserved. Fig. 3 shows accuracy versus latent dimension for several SNRs. Accuracy increases with N and then saturates, indicating diminishing gains per additional channel use. Larger N improves decoding fidelity but increases service time approximately linearly, leading to higher delay. Thus, small N yields low delay but reduced fidelity, whereas large N improves accuracy at higher delay. Fixed-latent designs are therefore inefficient: large N wastes resources under favorable conditions, while small N fails under deep fades or strict fidelity targets, suggesting operation near the saturation region rather than at maximum latent dimension. Reconstruction remains effective with PSNR 18.16 dB and SSIM 0.60. Overall,
(5)
(6)
This deterministic service-time model follows from time normalization and isolates the impact of the latent dimension on delay. The control process N (t) is piecewise constant, taking value Nk during service of the kth image. For each N ∈ N at a fixed operating SNR under Rayleigh fading, the semantic error probability is denoted by pe (N ) ∈ [0, 1], averaged over channel realizations and data samples, and typically decreasing with N . Realized semantic errors for the kth served image are modeled as Bernoulli indicators Ek ∈ {0, 1} with P(Ek = 1 | Nk = N ) = pe (N ).
(7)
Errors are conditionally independent across transmissions given the selected latent dimension. No retransmission is employed; each image is transmitted once without feedback. B. Minimum Delay Subject to Average Error Constraint We seek an online policy that stabilizes the queue while enforcing a long-term semantic fidelity constraint. Let Ek ∈ {0, 1} denote the semantic error indicator of the kth served image, and define the long-run average semantic error rate K
1 X E[Ek ]. Ē ≜ lim sup K→∞ K
(8)
k=1
The semantic fidelity requirement is Ē ≤ ε,
(9)
for a prescribed semantic error cap ε ∈ (0, 1). The latency objective is to minimize average delay. By Little’s law, under stability the long-term average delay satisfies Q̄sys , (10) λ so minimizing delay is equivalent to minimizing the average system backlog Q̄sys . Combining these, we consider the constrained stochastic optimization problem: W̄ =
min Q̄sys s.t. Ē ≤ ε, Nk ∈ N , sup E[Qsys (t)] < ∞, (11) t
where queue stability means that the expected queue length remains bounded. This formulation explicitly captures the semantic fidelity-latency tension. Selecting larger N reduces pe (N ) but increases the service time and thus queueing delay. C. Virtual Queue for the Semantic Fidelity Constraint To enforce Ē ≤ ε online, we introduce a virtual queue Z(k) ≥ 0 updated per served image using the realized indicator Ek . Define Z(k+1) = max {Z(k) + Ek − ε, 0} .
(12)
This is a standard deficit queue. If Ek = 1, the virtual backlog increases by 1 − ε; otherwise it decreases by ε (clipped at 0). Under mild ergodicity conditions, mean-rate stability of Z(k) ensures the time-average constraint: K
1 X E[Ek ] ≤ ε. K→∞ K
lim sup
(13)
k=1
Z(k) acts as an adaptive Lagrange multiplier, rising for aggressive (small N ) actions and falling for conservative ones. D. Lyapunov Drift-Plus-Penalty Control Law We derive a low-complexity online policy using Lyapunov optimization. Let the Lyapunov function be L(Qsys , Z) = 12 Q2sys + 21 Z 2 .
(14)
At each service decision epoch, we select N to approximately minimize an upper bound on the one-step conditional drift plus a service-time penalty. Intuitively, this reduces both backlogs Qsys and Z while penalizing large N that increases delay. A standard drift analysis yields a myopic rule of the form Nk ∈ arg min {Qk N + V Zk p̂e (N )} , N ∈N
(15)
where p̂e (N ) is an estimate of the semantic error probability P(Ek = 1 | Nk = N ), Qk ≜ Qsys (tk ) is the number of images in the system at the kth service-start epoch (including the image in service), Zk is the virtual backlog, and V ≥ 0 is a tunable weight trading delay against constraint enforcement. The term Qk N penalizes long service times under congestion, favoring smaller N , while Zk p̂e (N ) penalizes unreliable actions when the virtual queue is large, favoring larger N to reduce expected semantic error. Increasing V strengthens fidelity enforcement, typically lowering average error at the cost of higher delay.
E. Event-Driven Queue Dynamics and Delay Estimation Service is event-driven with deterministic service time Nk . Arrivals follow the exogenous process and departures occur at completion. Let Qsys (t) denote the number of images in the system (waiting plus the one in service). The time-average queue length is computed via the renewal-reward identity: Z 1 T Qsys (t) dt, (16) Q̄sys ≈ T 0 which is obtained by accumulating the area under Qsys (t) between events. The average delay follows from Little’s law, Q̄sys , λ provided the system is stable and the horizon is long. W̄ ≈
(17)
F. Minimum Delay Subject to Semantic Fidelity Constraint The control law (15) yields a family of policies indexed by V . For fixed arrival rate λ and error cap ε, one can evaluate the achieved pair (W̄ (V ), Ē(V )) and identify the minimum delay among those satisfying the semantic fidelity requirement: W̄ ⋆ (λ, ε) = min W̄ (V ) V ≥0
s.t.
Ē(V ) ≤ ε.
(18)
This traces the accuracy-latency frontier. Small V favors low latency but may violate the error constraint, while large V prioritizes semantic fidelity at higher delay. The controller adapts the semantic rate based on the estimated semantic error, using larger N only when needed to meet the fidelity constraint and smaller N otherwise to reduce delay. The mechanism is a coupled primal-dual system linking SemCom and queueing dynamics. SemCom provides N and pe (N ), the queue provides the state Qk , and the virtual queue Zk enforces the error constraint. The lightweight policy (15) adapts in real time to congestion and fidelity debt. Since N jointly controls latency and semantic fidelity, drift-plus-penalty control minimizes delay under the error bound. G. Delay Performance Under Semantic Fidelity Constraints We evaluate the queue-aware semantic rate controller under varying traffic loads and fidelity requirements. Fig. 4 shows average delay versus arrival rate λ for the adaptive latentdimension policy and fixed-N baselines under different error caps ε. The results highlight the fidelity-latency coupling. Delay increases slowly at light load and rises sharply as λ approaches service capacity, consistent with single-server behavior where delay scales as (1 − ρ)−1 with utilization ρ = λE[N ], where E[N ] is the average latent dimension selected by the policy. The sharp upturn indicates the onset of instability as service time becomes insufficient for the arrival load. Tightening the fidelity cap from ε = 0.3 to 0.25 and 0.2 shifts delay curves upward and shrinks the stability region, since stricter fidelity requires more frequent use of larger N , increasing mean service time. The figure therefore quantifies the cost of enforcing semantic fidelity in SemCom.
Minimum Delay subject to error < ε
250
ε = 0.3, Dynamic N ε = 0.25, Dynamic N ε = 0.2, Dynamic N ε = 0.25, 0.3, N = 10 ε = 0.25, 0.3, N = 15 ε = 0.2, 0.25, 0.3, N = 20
225 200 175 150
in the same cross-layer coupling. Larger N improves semantic fidelity but increases service time and may worsen freshness. B. Optimization Problem and Virtual Semantic Fidelity Queue The AoI-aware semantic control problem is ¯ min ∆
125
s.t.
Ē ≤ ε, Nk ∈ N , {Qsys (t)} is stable. (21)
100 75 50 25 0 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10 0.11 0.12 0.13 0.14
Arrival Rate
Fig. 4: Delay vs. arrival rate.
The dynamic policy achieves lower delay than fixed-N baselines by adapting N using the backlog and virtual fidelity queue, selecting smaller N when backlog is low and increasing N only to maintain Ē ≤ ε. Fixed-N schemes show clear feasibility limits: N = 10 supports only ε = 0.3, N = 15 meets ε = 0.25 but struggles at ε = 0.2, and only N = 20 satisfies ε ∈ {0.2, 0.25, 0.3} at much higher delay. Thus, no fixed semantic rate achieves both low delay and strict fidelity. Near the stability boundary, fixed-N schemes diverge earlier, whereas the adaptive policy compresses more aggressively under backlog, extending the feasible arrival region while maintaining the constraint. The results confirm that the latent dimension N is the key cross-layer control variable, with adaptive control approaching the optimal fidelity-latency balance, while fixed-N designs either violate fidelity or increase delay. IV. AO I-AWARE S EMANTIC R ATE C ONTROL We develop an AoI-aware controller that adapts the semantic rate (latent dimension) to minimize time-average AoI under a semantic fidelity constraint, directly targeting freshness. A. AoI System Model Consider the same single-server SemCom system with instantaneous AoI ∆(t) defined as the time since the most recently successfully delivered image was generated. The queue operates in FIFO order, with each update represented by its generation time. Let gk and dk denote the generation and departure times of the kth update, and Ek ∈ {0, 1} the semantic error indicator (1 failure, 0 success). The AoI resets only upon successful decoding: ( dk − gk , Ek = 0, ∆(d+ (19) k)= ∆(d− Ek = 1, k ), where ∆(d− k ) denotes the AoI just before completion of the kth service. Between events, AoI increases linearly with unit d ∆(t) = 1. The objective is the time-average AoI slope, i.e., dt Z T ¯ ≜ lim sup 1 ∆ ∆(t) dt. (20) T →∞ T 0 As before, selecting latent dimension N ∈ N determines both the service time and the semantic error probability pe (N )
Compared with the delay formulation in Section II, the key difference is that the controller must account for the age penalty incurred by long service times. We reuse the same virtual queue to enforce the semantic error constraint. At each service completion k, Z(k+1) = max{Z(k) + Ek − ε, 0} .
(22)
As discussed earlier, stability of Z(k) implies satisfaction of the long-term semantic fidelity requirement. The key distinction from the delay-focused design lies in the per-decision cost used in the drift-plus-penalty step. C. AoI-Driven Drift-Plus-Penalty Policy Using standard AoI sample-path arguments, the incremental AoI cost incurred by serving a semantic update of duration N when the current age is ∆ is 1 (23) ∆N + N 2 , 2 which captures the trapezoidal area accumulated during service. Following the same Lyapunov optimization principle used earlier, but replacing the backlog penalty with the AoI cost, the online semantic-rate decision becomes 1 Nk ∈ arg min ∆k N + N 2 + V Zk p̂e (N ) . (24) N ∈N 2 Here, ∆k ≜ ∆(t− k ) denotes the AoI at the start of the kth service. This rule differs from the queue-driven policy in Section II. The term ∆k N penalizes long service when information is already stale, while the quadratic term 12 N 2 captures the intrinsic AoI curvature associated with deterministic service. The third term enforces the semantic fidelity constraint through the virtual queue. Unlike the delay-optimal policy, decisions depend on instantaneous age ∆k rather than backlog Qk , leading to smaller N when information is stale. Since AoI resets only upon semantic success, fidelity and freshness are tightly coupled. When semantic debt grows, the virtual queue Zk shifts the policy in (24) toward larger N to reduce p̂e (N ) while maintaining Ē ≤ ε. D. Selection of Minimum Feasible AoI For each arrival rate λ, we sweep the control weight V to obtain the AoI-semantic fidelity tradeoff and select ¯ ⋆ (λ, ε) = min ∆(V ¯ ) : Ē(V ) ≤ ε , ∆ (25) V ≥0
the minimum AoI over operating points satisfying Ē ≤ ε. This procedure parallels the delay analysis but targets freshness rather than backlog. The AoI-aware semantic controller provides a complementary operating regime to the
Minimum AoI subject to error < ε
250
ε = 0.2, Dynamic N ε = 0.25, Dynamic N ε = 0.3, Dynamic N ε = 0.25, 0.3, N = 10 ε = 0.25, 0.3, N = 15 ε = 0.2, 0.25, 0.3, N = 20
225 200 175 150 125 100 75 50 25 0
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10 0.11 0.12 0.13 0.14
Arrival rate
Fig. 5: AoI vs. arrival rate.
queue-aware design of Section II. While the earlier policy is backlog-driven and prioritizes congestion mitigation, the present mechanism is freshness-driven and prioritizes timely semantic updates through the latent-dimension adaptation.
V. C ONCLUSION We considered semantic image communication over Rayleigh fading channels with an end-to-end learned encoderdecoder and identified the latent dimension as the key semantic-rate control variable coupling fidelity and service time. We showed that accuracy improves with latent dimension but saturates, while channel use (and latency) grows linearly, making fixed-rate designs inefficient. To address this, we developed semantic-fidelity-constrained rate controllers that adapt the latent dimension per update. The queue-aware policy minimizes delay under an average semantic error cap via a virtual fidelity queue and drift-plus-penalty selection, outperforming fixed-rate baselines and expanding the feasible load region. The AoI-aware policy targets time-average AoI under the same constraint and improves freshness across arrival rates. Overall, adaptive semantic-rate control provides a unified framework for balancing latency, fidelity, and freshness under dynamic traffic and channel conditions. R EFERENCES
E. AoI Performance Under Semantic Fidelity Constraints We next evaluate the AoI-aware semantic rate controller across varying traffic intensities. Fig. 5 shows the minimum achievable time-average AoI versus arrival rate λ under error caps ε ∈ {0.2, 0.25, 0.3} for both the dynamic latentdimension policy and fixed-N baselines. The results exhibit the characteristic U-shaped AoI versus load. At very small λ, the system is update-starved and AoI is dominated by inter-arrival time; as λ increases, AoI decreases with more frequent updates. Beyond a critical load, service becomes the bottleneck and queueing dominates, causing AoI to rise sharply and marking the freshness stability boundary. The semantic fidelity constraint strongly impacts the adaptive policy. Tightening the cap from ε = 0.3 to 0.25 and 0.2 shifts the AoI curves upward and leftward because stricter fidelity forces more frequent selection of larger N , increasing service time and degrading freshness. The figure thus quantifies the intrinsic cost of fidelity in AoI-driven SemCom. The dynamic latent-dimension policy consistently achieves the lowest AoI over the feasible region. At moderate load, it opportunistically uses smaller N when the virtual fidelity queue is small, reducing service time while maintaining Ē ≤ ε. Near heavy traffic, it increases N only as needed to prevent fidelity violations, operating near the optimal freshness-fidelity balance. Fixed-N baselines show clear feasibility limits. With N = 10, only ε ∈ {0.25, 0.3} is feasible; N = 15 meets ε = 0.25 and 0.3 over a wider range but struggles at ε = 0.2; only N = 20 satisfies all targets {0.2, 0.25, 0.3}, at the cost of substantially higher AoI. Thus, no fixed semantic rate achieves both low AoI and strict fidelity. The adaptive policy enlarges the AoI-stable region, as fixed-rate curves diverge earlier, while the dynamic controller maintains finite AoI over a broader λ range by reducing the rate as age grows. Overall, AoI-aware semantic rate control enables lower AoI while automatically navigating the fidelity-freshness tradeoff.
[1] D. Gündüz, Z. Qin, I. E. Aguerri, H. S. Dhillon, Z. Yang, A. Yener, K. K. Wong, and C. S. Chae, “Beyond transmitting bits: Context, semantics, and task-oriented communications,” IEEE Journal on Selected Areas in Communications, vol. 41, no. 1, pp. 5–41, 2023. [2] E. Uysal, O. Kaya, A. Ephremides, J. Gross, M. Codreanu, P. Popovski, M. Assaad, G. Liva, A. Munari, T. Soleymani, B. Soret, and H. Johansson, “Semantic communications in networked systems,” IEEE Network, vol. 36, no. 4, pp. 233–240, 2022. [3] Y. E. Sagduyu, T. Erpek, A. Yener, and S. Ulukus, “Will 6G be semantic communications? opportunities and challenges from task-oriented and secure communications to integrated sensing,” IEEE Network, vol. 38, no. 6, pp. 72–80, 2024. [4] E. Bourtsoulatze, D. B. Kurka, and D. Gündüz, “Deep joint sourcechannel coding for wireless image transmission,” IEEE Transactions on Cognitive Communications and Networking, vol. 5, no. 3, pp. 567–579, 2019. [5] M. Jankowski, D. Gündüz, and K. Mikolajczyk, “Wireless image retrieval at the edge,” IEEE Journal on Selected Areas in Communications, vol. 39, no. 1, pp. 89–100, 2021. [6] E. Erdemir, T. Tung, P. L. Dragotti, and D. Gündüz, “Generative joint source-channel coding for semantic image transmission,” IEEE Journal on Selected Areas in Communications, vol. 41, no. 8, pp. 2645–2657, 2023. [7] Z. Lyu, G. Zhu, J. Xu, B. Ai, and S. Cui, “Semantic communications for image recovery and classification via deep joint source and channel coding,” IEEE Transactions on Wireless Communications, vol. 23, no. 8, pp. 8388–8404, 2024. [8] Y. E. Sagduyu, T. Erpek, A. Yener, and S. Ulukus, “Age of information in deep learning-driven task-oriented communications,” in IEEE INFOCOM AoI Workshop, 2023. [9] Y. E. Sagduyu, T. Erpek, A. Yener, and S. Ulukus, “Low-latency taskoriented communications with multi-round, multi-task deep learning,” in ACM MobiCom Workshop on Machine Learning for NextG Networks, 2024. [10] N. Singh, C. K. Thomas, W. Saad, and E. C. Strinati, “On the computing and communication tradeoff in reasoning-based multi-user semantic communications,” in IEEE Wireless Communications and Networking Conference (WCNC), 2025. [11] Z. Yang, M. Chen, Z. Zhang, and C. Huang, “Energy efficient semantic communication over wireless networks with rate splitting,” IEEE Journal on Selected Areas in Communications, vol. 41, no. 5, pp. 1484–1495, 2023. [12] Y. E. Sagduyu, T. Erpek, A. Yener, and S. Ulukus, “Joint sensing and task-oriented communications with image and wireless data modalities for dynamic spectrum access,” in IEEE International Symposium on Dynamic Spectrum Access Networks (DySPAN), 2024.