Economics of Semantic Communication System in Wireless Powered Internet of Things

Zi Qin Liew, Yanyu Cheng, Wei Yang Bryan Lim, Dusit Niyato, Chunyan Miao, Sumei Sun

Introduction

Traditional communication systems prioritize technical accuracy, aiming for bit-perfect transmission. However, the emerging paradigm of semantic communication focuses on the meaning of data, enabling significant efficiency gains by transmitting only necessary information. This is particularly relevant for energy-constrained Internet of Things (IoT) devices in Wireless Powered Communication Networks (WPCN), where a Hybrid Access Point (H-AP) provides energy for data transmission. This paper proposes an economic framework to manage resource allocation in such networks, treating wireless energy as a commodity auctioned to devices based on their semantic requirements.

System Model and Methodology

The proposed framework integrates a semantic communication system with a deep learning (DL)-based auction mechanism. The H-AP acts as an auctioneer, distributing energy to IoT devices that bid based on the perceived value of the information they can transmit.

System architecture of the wireless-powered semantic communication network.

Semantic Communication and Valuation

The system utilizes a DeepSC model based on the Transformer architecture for text encoding and decoding, trained on the Europarl dataset. To quantify the "value" of energy, devices evaluate potential transmission quality using two primary metrics:

  1. BLEU Score: Measures exact word-for-word matching between original and recovered text.
  2. Sentence Similarity: Uses a pre-trained BERT model to assess how well the original meaning is preserved, regardless of exact wording.

A device's valuation is a weighted sum of these metrics, allowing for diverse device priorities (e.g., medical precision vs. general sensor context). As shown in experimental results, increasing the output dimension ($D$) of the semantic encoder—which requires more energy—directly improves both BLEU and similarity scores, forming the basis for the devices' bidding strategies.

Impact of output dimension $D$ on semantic performance metrics (BLEU and similarity scores).

DL-Based Optimal Auction

To maximize revenue while ensuring fairness, the authors implement a DL-based auction. Unlike traditional Second-Price Auctions (SPA), this neural network approach is designed to satisfy Individual Rationality (IR) and Incentive Compatibility (IC), ensuring bidders remain truthful and profitable while the H-AP optimizes its returns.

Key Insights

  • Meaning over Bits: The valuation of energy is intrinsically tied to semantic quality. Higher energy availability allows for a larger output dimension, which significantly boosts the similarity score and BLEU metrics.
  • The Saturation Point: Experimental data shows that as wireless energy harvesting time increases, the average highest bid eventually saturates. This suggests a "diminishing return" on energy where additional power no longer significantly improves the semantic quality of the transmission.
  • Complexity Penalty: Increased text complexity, such as longer sentences or a higher volume of sentences, leads to lower bids. This occurs because devices must compress information more aggressively within their energy budgets, lowering the overall "value" of the resulting transmission.
  • Preference-Driven Bidding: The framework successfully captures device-specific needs, with bids reflecting whether a device prioritizes exact wording or general meaning.

Results and Discussion

The DL-based auction consistently outperformed the traditional SPA in terms of H-AP revenue across various preference settings. For example, when devices had a higher preference for semantic similarity, the average highest bid remained significantly higher than in scenarios prioritizing exact matching.

The system also demonstrated high sensitivity to physical constraints; bids dropped sharply as the distance between the H-AP and devices increased (particularly beyond 9 meters) due to the physics of path loss in wireless power transfer. Crucially, the auction maintained its game-theoretic properties (IC and IR), providing a stable economic environment for the network even as network parameters varied.

Conclusion

This work demonstrates that combining economic theory with semantic communication can effectively manage resources in wireless-powered IoT networks. By shifting the focus from bit-accuracy to meaning-preservation, the system achieves higher efficiency and revenue. Future research will explore extending this framework to multi-user interference scenarios and non-textual data types, such as image and video transmission, where semantic value metrics differ.

Summary updated

BibTeX
@inproceedings{liew2022economics,
  author = {Zi Qin Liew and Yanyu Cheng and Wei Yang Bryan Lim and Dusit Niyato and Chunyan Miao and Sumei Sun},
  title = {{Economics of Semantic Communication System in Wireless Powered Internet of Things}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2022},
  doi = {10.1109/icassp43922.2022.9746463},
}
Figure