Bridging Neural Networks and Wireless Systems with MIMO-OFDM Semantic Communications PubID: pubid: 0000–0000/00$00.00 © 2021 IEEE

Hanju Yoo, Dongha Choi, Yonghwi Kim, Yoontae Kim, Songkuk Kim, Chan-Byoung Chae, Robert W. Heath, Jr · 2025

Read Bridging Neural Networks and Wireless Systems with MIMO-OFDM Semantic Communications PubID: pubid: 0000–0000/00$00.00 © 2021 IEEE by Hanju Yoo, Dongha Choi, Yonghwi Kim, Yoontae Kim, Songkuk Kim, Chan-Byoung Chae, Robert W. Heath, Jr (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-16726.txt.

Introduction

Semantic communication sends meaning instead of raw bits, promising leaner wireless traffic. Hanju Yoo and co-authors test such a system built on MIMO antennas and OFDM signaling, the workhorses of modern networks. Lab results often shine in simulation, then sag on real hardware, and the team digs into why.

Two culprits stand out: power amplifiers bending signals out of shape, and swings in peak-to-average power ratio. The channel's frequency selectivity turns out to drive much of the lost ground. With fixes aimed at those weak points, the system nearly matches its ideal scores. The 2025 article lays out design lessons for taking semantic links out of the lab.

Catalog record

CREATORHanju Yoo, Dongha Choi, Yonghwi Kim, Yoontae Kim, Songkuk Kim, Chan-Byoung Chae, Robert W. Heath, Jr
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE31 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-16726

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Bridging Neural Networks and Wireless Systems with MIMO-OFDM Semantic Communications PubID: pubid: 0000–0000/00$00.00 © 2021 IEEE Hanju Yoo, Dongha Choi, Yonghwi Kim, Yoontae Kim, Songkuk Kim, Chan-Byoung Chae, Robert W. Heath, Jr Abstract Semantic communications aim to enhance transmission efficiency by jointly optimizing source coding, channel coding, and modulation. While prior research has demonstrated promising performance in simulations, real-world implementations often face significant challenges, including noise variability and nonlinear distortions, leading to performance gaps. This article investigates these challenges in a multiple-input multiple-output (MIMO) and orthogonal frequency division multiplexing (OFDM)-based semantic communication system, focusing on the practical impacts of power amplifier (PA) nonlinearity and peak-to-average power ratio (PAPR) variations. Our analysis identifies frequency selectivity of the actual channel as a critical factor in performance degradation and demonstrates that targeted mitigation strategies can enable semantic systems to approach theoretical performance. By addressing key limitations in existing designs, we provide actionable insights for advancing semantic communications in practical wireless environments. This work establishes a foundation for bridging the gap between theoretical models and real-world deployment, highlighting essential considerations for system design and optimization. I Introduction As next-generation wireless networks push toward more efficient and intelligent data delivery, conventional digital communication paradigms—based on separate source coding, channel coding, and modulation—begin to show their limitations. While classic information-theoretic results [ 1 ] support this modular structure, real-world conditions often deviate from the ideal assumptions of large block lengths, stationary channels, and additive white Gaussian noise (AWGN) channel. As a result, practical systems struggle to fully realize the theoretical gains promised by such layered approaches [ 2 ] . Semantic communications have emerged as a promising alternative, integrating source and channel coding within a single, end-to-end learning framework [ 3 , 4 ] . Rather than focusing solely on symbol-level accuracy, semantic systems directly map application-relevant information such as key visual features [ 5 , 6 ] or essential text components [ 4 , 7 ] into symbols. This approach has demonstrated notable improvements over traditional methods in simulations, achieving better performance across metrics ranging from mean squared error to perceptual similarity [ 5 , 6 , 4 , 7 ] . However, their end-to-end design, which bypasses traditional bit-based processing and directly operates on symbols, inherently increases their susceptibility to channel noise and variability. This raises a critical question: Can the theoretical and simulated gains of semantic communications be reliably achieved in real-world wireless scenarios? In this paper, we provide new insights into how realistic communication components and settings limit semantic communications, and suggest approaches for overcoming these limitations. To understand the feasibility, we anchor the discussion around the lessons learned from implementing a MIMO-OFDM prototype. We address challenges commonly overlooked in simulations, such as power amplifier (PA) nonlinearity and varying noise levels across subcarriers and data streams. Our measurements reveal how these factors degrade performance and explain previously observed gaps between simulated and experimental results, offering insights into the feasibility of semantic communications in practical systems and providing key considerations for system design. To the best of our knowledge, this is the first work to validate the performance of semantic communications in MIMO-OFDM environments. Our contributions are as follows: • We examine challenges encountered in real wireless environments, such as error variations across subcarriers and data streams, to identify the causes of performance deviations between simulations and prototype results. • We demonstrate that semantic communications can achieve near-theoretical performance in real systems when channel variations are properly addressed. • We validate that peak-to-average-power ratio (PAPR) mitigation techniques, which typically degrade performance in statistical channel-based simulations without PA nonlinearity, can enhance system performance in practical systems within the nonlinear power region. • We make our source code and hardware setup publicly available 1 1 1 https://github.com/kmsiapps/semantic-mimo-ofdm to provide a foundation for developing semantic communications systems in real-world conditions. Fig. 1: Basic block diagram comparing conventional digital communications and semantic communications system models. Conventional systems rely on separate source coding, channel coding, and modulation stages, whereas semantic communications adopt an end-to-end approach. This approach directly maps source data to wireless symbols, bypassing intermediate bitstreams, and reconstructs the data from noisy symbols. While it enables goal-oriented training, it is more sensitive to channel characteristics. II An Overview of Semantic Communications II-A System Architecture In this section, we introduce the typical architecture of semantic communications systems and compare it to conventional communications systems. Fig. 1 illustrates the basic building blocks of conventional and semantic communications system models. In traditional communication systems, transforming input data into in-phase and quadrature-phase (I/Q) symbols involves multiple stages, including source coding (e.g., JPEG), channel coding (e.g., LDPC), and modulation (e.g., QAM). The output of the source and channel coding blocks is in the form of bits, which are then converted into complex values during the modulation stage. In contrast, semantic communications systems bypass this bit-based processing by directly mapping input data (e.g., images) to wireless symbols. This method eliminates the need for separate source coding, channel coding, and modulation stages. By jointly optimizing these components with an end-to-end training approach, semantic communications systems enhance efficiency and performance. They consist of an encoder that reduces the dimensionality of the input and a decoder that generates the desired output from the encoded data, typically constructed using deep neural networks. The primary objective of the encoder neural networks in semantic communications systems is to produce complex symbols suitable for wireless transmission. To achieve this, the encoder processes the input data to generate features significantly smaller than the original input, effectively compressing the data. These real-valued features are then paired to form complex-valued symbols. A channel layer is incorporated to mimic the physical wireless channel. As the channel layer should be differentiable for training, it typically consists of simple statistical channel models such as AWGN or Rayleigh fading channels. The impaired symbols are then unpacked into corrupted, real-valued features and passed to the decoder. The decoder’s role is to effectively achieve predefined goals using the corrupted features. These goals can vary, ranging from accurate reconstruction (e.g., image or text reconstruction) to analysis tasks (e.g., object detection or image classification) based on the input data. The extent of achievement is measured using a loss function, which could be a perceptual metric, mean squared error (for reconstruction tasks), or classification accuracy, depending on the system’s configuration. As the encoder and decoder jointly learn to achieve these goals despite channel-corrupted symbols, the encoder is trained to generate noise-robust wireless symbols, while the decoder is optimized to efficiently realize the objectives from the encoder-produced symbols. This end-to-end training process of the semantic communications system, as demonstrated in prior work [ 5 , 7 , 6 , 8 ] , has shown improvements in transmission efficiency compared to traditional systems. Transforming features into wireless symbols, however, requires conscientious handling, mainly when those neural network-generated symbols are transmitted over actual wireless channels. We investigate the procedures of symbol transmission and identify the associated challenges in the following sections. Fig. 2: System architecture of a semantic communications system with a MIMO-OFDM prototype setup. Red boxes indicate practical issues arising from the radio hardware and communication channels. II-B Prior Work There have been few implementations of wireless semantic communications prototypes. The first effort to

Read the complete text