OFDM-Based Digital Semantic Communication with Importance Awareness

Chuanhong Liu, Caili Guo, Yang Yang, Wanli Ni, Tony Q. S. QuekIEEE Trans. Communications 2024

Abstract

Semantic communication (SemCom) has received considerable attention for its ability to reduce data transmission size while maintaining task performance. However, existing works mainly focus on analog SemCom with simple channel models, which may limit its practical application. To reduce this gap, we propose an orthogonal frequency division multiplexing (OFDM)-based SemCom system that is compatible with existing digital communication infrastructures. In the considered system, the extracted semantics is quantized by scalar quantizers, transformed into OFDM signal, and then transmitted over the frequency-selective channel. Moreover, we propose a semantic importance measurement method to build the relationship between target task and semantic features. Based on semantic importance, we formulate a sub-carrier and bit allocation problem to maximize communication performance. However, the optimization objective function cannot be accurately characterized using a mathematical expression due to the neural network-based semantic codec. Given the complex nature of the problem, we first propose a low-complexity sub-carrier allocation method that assigns sub-carriers with better channel conditions to more critical semantics. Then, we propose a deep reinforcement learning-based bit allocation algorithm with dynamic action space. Simulation results demonstrate that the proposed system achieves 9.7% and 28.7% performance gains compared to analog SemCom and conventional bit-based communication systems, respectively.

Introduction

Traditional communication systems focus on the accurate transmission of bits, often overlooking the underlying meaning or the specific requirements of the end task. Semantic Communication (SemCom) addresses this by prioritizing the transmission of "meaningful" information. However, most current SemCom research relies on analog joint source-channel coding (JSCC) and idealized additive white Gaussian noise (AWGN) channels. These models fail to account for the digital nature of modern infrastructure and the challenges of frequency-selective multipath fading.

This paper proposes an end-to-end digital SemCom framework that integrates with Orthogonal Frequency Division Multiplexing (OFDM). By quantifying the importance of different semantic features, the system intelligently allocates communication resources—bits and sub-carriers—to maximize task performance in realistic wireless environments.

System Architecture and Digital Pipeline

The proposed system bridges the gap between theoretical SemCom and practical digital pipelines. The architecture, illustrated in Figure 1, consists of a semantic transmitter and receiver. The transmitter employs a semantic encoder to extract features from input data. These features are then processed through scalar quantizers, which are more hardware-friendly than vector quantization methods. The resulting digital signals are mapped to OFDM sub-carriers for transmission over a wireless channel.

Overall architecture of the proposed OFDM-based digital semantic communication system, showing the integration of semantic codecs with digital OFDM components.

Semantic Importance Measurement

A key challenge in SemCom is that not all extracted features contribute equally to the final task. To address this, the authors developed a dual-module importance measurement framework:

  1. Semantics Task Relevance (STR): This module uses gradient-based analysis (similar to Grad-CAM) to quantify how much each feature influences the final task result (e.g., image classification).
  2. Inter-Semantics Relevance (ISR): This module employs cosine similarity to identify and penalize redundant features, ensuring that unique, high-impact information is prioritized.

As shown in Figure 2, these modules work together to generate a semantic importance vector that guides subsequent resource allocation.

The semantic importance measurement framework, detailing the STR and ISR modules used to prioritize feature transmission.

Resource Allocation via Deep Reinforcement Learning

Given the non-linear and non-differentiable nature of neural network-based codecs, traditional optimization methods are often insufficient for resource allocation. The proposed system utilizes two strategies:

  • Sub-carrier Allocation: A low-complexity greedy algorithm maps the most important semantic features to sub-carriers with the highest Signal-to-Noise Ratio (SNR).
  • Bit Allocation: A Dynamic Proximal Policy Optimization (DPPO) reinforcement learning agent is trained to distribute a bit budget across features. The agent uses a dynamic action space to ensure it stays within transmission constraints while maximizing task accuracy.

Experimental Results

The system was evaluated using the STL-10 image classification dataset over a SUI-5 multipath channel model. The results highlight several key advantages:

  • Superior Performance: The proposed digital SemCom (DSC) system achieved a 9.7% accuracy gain over analog SemCom (ASC) and a 28.7% gain over conventional separate source and channel coding (SSCC, using JPEG and LDPC) at 5 dB SNR.
  • Multipath Robustness: While analog SemCom performance degrades significantly as multipath complexity increases, the OFDM-based digital system remains stable by effectively combating inter-symbol interference.
  • Efficiency and Convergence: The DPPO algorithm demonstrated rapid convergence, reaching stable performance in fewer than 200 iterations, making it suitable for dynamic environments.

Conclusion

By combining semantic importance awareness with a digital OFDM-based pipeline, this work provides a practical and high-performance solution for next-generation communication. The integration of gradient-based importance measurement and DRL-driven resource allocation allows the system to outperform both traditional digital methods and modern analog SemCom approaches in realistic wireless scenarios.

Summary updated

Abstract

Semantic communication (SemCom) has received considerable attention for its ability to reduce data transmission size while maintaining task performance. However, existing works mainly focus on analog SemCom with simple channel models, which may limit its practical application. To reduce this gap, we propose an orthogonal frequency division multiplexing (OFDM)-based SemCom system that is compatible with existing digital communication infrastructures. In the considered system, the extracted semantics is quantized by scalar quantizers, transformed into OFDM signal, and then transmitted over the frequency-selective channel. Moreover, we propose a semantic importance measurement method to build the relationship between target task and semantic features. Based on semantic importance, we formulate a sub-carrier and bit allocation problem to maximize communication performance. However, the optimization objective function cannot be accurately characterized using a mathematical expression due to the neural network-based semantic codec. Given the complex nature of the problem, we first propose a low-complexity sub-carrier allocation method that assigns sub-carriers with better channel conditions to more critical semantics. Then, we propose a deep reinforcement learning-based bit allocation algorithm with dynamic action space. Simulation results demonstrate that the proposed system achieves 9.7% and 28.7% performance gains compared to analog SemCom and conventional bit-based communication systems, respectively.
BibTeX
@article{liu2024ofdm,
  author = {Chuanhong Liu and Caili Guo and Yang Yang and Wanli Ni and Tony Q. S. Quek},
  title = {{OFDM-Based Digital Semantic Communication with Importance Awareness}},
  journal = {IEEE Trans. Communications},
  year = {2024},
}
Figure