An intelligent reflecting surface–assisted semantic communication architecture for 6G networks with hippopotamus–differential evolution optimized resource allocation

Abstract

A persistent obstacle in sixth-generation (6G) wireless systems is the simultaneous demand for extreme spectral efficiency and ultra-reliable low-latency delivery under constrained energy budgets. To address this obstacle, an integrated architecture is proposed in which intelligent reflecting surfaces (IRS) are coupled with a semantic communication engine, and the joint optimization of phase-shift configuration, semantic compression ratio, and transmit power is performed by a newly composed hippopotamus–differential evolution (HO–DE) metaheuristic. Rather than the transmission of raw bit streams, only task-relevant semantic features are extracted at the source and delivered over an IRS-shaped channel whose reflection coefficients are tuned to maximize an effective semantic signal-to-noise ratio. The semantic fidelity surrogate underlying this metric is calibrated and validated against a trained convolutional semantic codec on a concrete image-classification task, and the effective semantic signal-to-noise ratio is reported on a formally defined logit-decibel scale. The non-convex mixed-variable allocation problem is reformulated as a single-objective fitness landscape balancing fidelity, energy, and latency, and an adaptive switching mechanism transfers the search emphasis from exploration toward exploitation as the population matures. Across varying surface scale, user density, and channel severity, the proposed scheme reduced transmitted symbols per task by \(41.6\%\) , raised the effective semantic signal-to-noise ratio to 8.6 dB (a 3.8 dB gain), and lowered energy per task by \(27.4\%\) relative to the strongest benchmark, while attaining an aggregate fidelity of 0.93 and a terminal fitness of 3.28 with the lowest run-to-run deviation of 0.062. The ninety-fifth percentile latency fell to 3.0 ms, a \(25\%\) tail reduction. The reported gains are shown to be statistically significant against five benchmarks, including a success-history adaptive differential evolution variant and an alternating-optimization scheme with semidefinite relaxation, and to persist across a sensitivity sweep of the semantic mapping parameters. The contribution offers a deployable direction for semantic-aware, surface-assisted 6G access networks.