Hybrid AI Framework for Real‑Time Signal Denoising and Error

Correction in 5G/6G Wireless Communication Systems

Purushotham Endla

, Rajkumar Mandal

, R. Purushothaman

, Nimmagadda Padmaja

Tandra Nagarjuna

and Syed Zahidur Rashid

Department of Physics, School of Sciences and Humanities, SR University, Warangal 506371, Telangana, India

Faculty of Information Technology and Engineering, Gopal Narayan Singh University, Bihar, India

Department of Electronics and Communication Engineering, J.J. College of Engineering and Technology, Tiruchirappalli,

Tamil Nadu, India

Department of ECE, School of Engineering, Mohan Babu University (erstwhile Sree Vidyanikethan Engineering College),

Tirupati, Andhra Pradesh, India

Department of Computer Science and Engineering, MLR Institute of Technology, Hyderabad, Telangana, India

Department of Electronic and Telecommunication Engineering, International Islamic University Chittagong, Chittagong,

Bangladesh

Keywords: Signal Denoising, Error Correction, Wireless Communication, Deep Learning, 6G Networks.

Abstract: In 5G and future 6G networks, it is important to guarantee strong and reliable signal transmission in the

presence of noise, interference, and data degradation. A new trainable hybrid AI framework with deep

learning, denoising diffusion models, and attention-based autoencoders is suggested to conduct the real-time

noise removal and error correction in the dynamic wireless channels. Unlike its predecessors, which are either

simulation-oriented, hardware-based, or application-specific, our technique is validated on the real-time

communication testbeds and learns multi-level AI layers to combat the physical corruption, semantic

discrepancy, and packet-level mistakes. Combining positional accuracy and adaptive signal enhancement, this

design does not only decrease the bit error rate (BER) and signal-to-noise ratio (SNR) which, but it also makes

the wireless communication reliable for various channel scenarios. Experimental results show significant

improvements over existing models, indicating that the proposed method is highly promising for URLLC in

future networks.

1 INTRODUCTION

With the ongoing deployment of 5G and the future

6G wireless communication systems, there is a

significant surge in the requirement of ultra-reliable,

low-latency, large-throughput services. Latest

applications based on exchange of error free signals

abound as in self-driving cars or remote surgery

dipping into virtual immersion to real time industry

automation. Nevertheless, due to the higher

complexity and crowding of communication

channels, wireless communication systems become

more vulnerable to noises, interferences and

degradations. However, the traditional signal

processing techniques cannot efficiently manage the

dynamic, asymmetric ed of future wireless systems.

New frontiers have opened however, after recent

improvements in artificial intelligence. Artificial

intelligence models are capable of extracting complex

patterns from noisy data and reproducing the original

signal with a high degree of fidelity. But although a

solution, few of the related researches directly solve

the end-to-end optimization, but limited to POSR or

channel estimation and with a single-layer denoising

and error correction, and are not complete solutions

since they are dedicated to modulations without

holistic real-timely processing.

This work presents an end-to-end AI-based

architecture based on denoising diffusion models,

attention mechanisms based on transformers, and

deep learning classifiers to improve the quality of

signals throughout the transmission-reception path.

By incorporating intelligences at both physical and

500

Endla, P., Mandal, R., Purushothaman, R., Padmaja, N., Nagarjuna, T. and Rashid, S. Z.

Hybrid AI Framework for Real-Time Signal Denoising and Error Correction in 5G/6G Wireless Communication Systems.

DOI: 10.5220/0013868200004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 1, pages

500-507

ISBN: 978-989-758-777-1

semantic layers, this framework not only recovers

(attenuated version of) corrupted signals, but also

predicts and rectifies transmission insufficiencies, all

in the context of arbitrary channel states. Our

methodology goes beyond simulation, incorporating

hardware-based verification for practicability. The

objective of this work is to lay a strong foundation of

smart, resilient and intelligent wireless infrastructure

capable of addressing the future needs and

requirements of 5G and 6G wireless networks.

2 PROBLEM STATEMENT

The advent of 5G and the future 6G wireless

communications systems can potentially

revolutionize the high data transmission rate, large-

scale devices connection, and ultra-reliable low-

latency communication (URLLC). Nevertheless,

there is only so much improvement that is practically

possible since wireless channels in which such

interfering, noise, fading and distortion of the signal

are encountered is a ubiquitous, unsolvable problem

and is constantly becoming more complex.

Conventional signal processing techniques, such as

linear filtering and error correction codes, are unable

to adapt well to the nonlinear, non-stationary and

rapidly changing environment in today's

communication settings. However, they are usually

validated based on static scenarios, they do not adapt

in real-time to the environmental challenge with the

changes due to environmental variability, mobility,

and presence of heterogeneous types of devices.

Furthermore, the AI has shown effectiveness able

to handle some subparts of the wireless pipeline

separately, such as modulation recognition or the

channel estimation, however a strong necessity of

intelligent system for the denoising and the error cor-

rection based on the unified mechanism is still open

to develop in a real-time performance. These recent

AIbased solutions also tend to be applicable only to

standalone cases, and rely strongly on simulation data

without being thoroughly validated on real hardware,

possibly limiting their applicability to large-scale

real-world scenarios. This gap motivates a holistic

plan in which the state-of-the-art AI solutions

combined with well-designed system level to ensure

stable signal integrity, iterable error mitigation and

flexibility towards different wireless communication

scenarios.

3 LITERATURE SURVEY

Wireless Communications and Artificial Intelligence

The intersection of AI and wireless communications

has been gaining increasing momentum with the

development of 5G and emerging 6G systems. There

are a few attempts where researchers have attempted

various AI methods for signal extraction and errors

suppression especially under noisy and dynamic

situations. Adaptive AI algorithms for signal

denoising have been studied in Mao (2024), however

the application setting was more general and was not

focused on real-time telecom scenarios. For example,

Roy and Islam (2025) developed channel estimation

and signal processing models for MIMO systems

applying advanced machine learning but mainly

carried out their research through simulations without

hardware validation. Zhang and Li (2025) also

proposed AI-assisted detection for MIMO systems,

but their work was not on the end-to-end signal

recovery.

Recently, there are growing interests in

employing diffusion models in physical layer

communications. For example, based on denoising

diffusion process, Neshaastegaran and Jian (2025)

built CoDiPhy as a general framework and Wu et al.

(2023) and Letafati et al. (2023) independently

pointed out the interest of such models for channel

denoising. These references have attempted to solve

this problem; however, they have concentrated

primarily on the diffusion aspect and have not

provided an overall integration of error correction

circuitry for such an error correcting mapper. Wu et

al. (2023) further expanded this analysis to semantic

communication tasks, with an emergent layered

notion of the integration of AI, but still with denoising

as a side issue.

In a more general view, Chae (2025) published a

survey on AI-based communication systems for the

6G, addressing possible challenges and future

directions only rather than providing specific

algorithmic designs. Zhao, Shen, and Wang (2025)

investigated generative AI models for improving the

security at the physical layer, where the AI algorithm

can provide the defense to signal quality as well, since

it diminishes adversary disturbances. Other research,

(Zhang and Li 2024) applied attention based

denoising networks into modulation classification,

but this was more or less processed as preprocessing

steps, instead of core functionality.

Wireless sensing has also been optimized

through artificial intelligence. proposed a data

augmentation approach with generative model to

improve the robustness wireless signal sensing, but it

Hybrid AI Framework for Real-Time Signal Denoising and Error Correction in 5G/6G Wireless Communication Systems

501

was not tailored for the denoising and error-correction

pipeline. Zhang and Li (2025) made several efforts,

e.g., AI in ABC and wireless positioning (2025a,

2025b), demonstrating broad applications of AI in

future networks, but none proposed an integrated

solution of signal fidelity. In addition, Zhang and Li

(2025c) repetition work of positioning were

interesting but didn’t make its own contribution in

the denoising and error correction field.

Some of the recent works focussed on

architectural integration. The work in (Zhang and Li,

2025d) also employed a neural network and parsed

deep learning layers of Autoencoder in a hybrid

model for better preserving the noise but their

test/timeliness e?orts are pure theoretical. More on

the hardware side, Zhang and Li (2024a) studied AI

for passive electronic filters, aiming at building a

platform for real-time embedded systems, excluding

however full communication stack support. Other

works such as Zhang and Li (2024b) which

concentrated on the tensor signal modeling, was

limited in its range of applications in noise-perturbed

situations. Differently, AI-based architecture was

proposed in Zhang and Li (2025e) in the context of

noise-robustness communication receivers, but it did

not include cross-layer optimization.

Cross-domain AI methods were also shown

effective in Zhang and Li (2024c) in geophysical

denoise via masked autoencoders, but except

promising results, the whole architecture was not

especially designed to address wireless transmission

constraints. Also, Zhang and Li (2024d) considered

AI on the generative wireless sensing, in a way of

enhancing signal interpretability, which however did

not have perfect error correction. A few recent works

from Zhang and Li also have more entries (2025f)

that discuss AI in wp, however, the focus was always

on spatial accuracy without an emphasis on temporal

signal recovery.

Overall, while existing work does provide strong

foundations on which AI can be built for individual

components of the wireless pipeline, there is a lack of

a real-time, hardware-validated, AI-based system for

jointly learning signal denoising and error

correction. Towards filling this gap, in this paper, we

propose a hybrid framework to exploit the merits of

diverse AI models for achieving robust performance

in noisy, low latency and multi-user NGWs.

4 METHODOLOGY

The approach involves the design and assembly of a

layered AI architecture, engineered to perform on-

the-fly signal denoising and error mitigation for the

forthcoming era of wireless communications. The

system design is developed to combat diverse channel

noise and errors in dynamic environments like urban

canyons, high-speed mobile sites, and dense IoT

deployments.

Figure 1: Workflow of the Hybrid AI Framework for Signal

Denoising and Error Correction.

The process begins with the capture of raw wireless

signals with software-defined radios (SDRs) in

diverse controlled settings and realistic scenarios in

the wild, representing a wide range of signal

corruptions such as Gaussian noise, Rayleigh fading,

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

502

impulsive interference, and burst packet errors.

Figure 1 show the Workflow of the Hybrid AI

Framework for Signal Denoising and Error

Correction. These signals are processed through a

series of pre-processing stages, which normalize,

window and segment the signal to ensure temporal

and spatial uniformity across instances. The kernel of

the framework is a hybrid deep learning architecture

composed of denoising diffusion probabilistic

models (DDPMs), attention-guided autoencoders and

recurrent error correction networks.

Table 1 show the

Dataset Specifications for Model Training and

Evaluation.

Table 1: Dataset Specifications for Model Training and Evaluation.

Dataset Type

Source

Type

SNR

Range

(

)

Modulation

Formats

Channel

Types

Size

(Samples)

Synthetic

Noise Data

MATLAB

+ AWGN

-5 to 20

BPSK, QPSK,

16QAM

AWGN,

lei

100,000

Real-World

Captures

SDR

(USRP)

0 to 25

QPSK,

64QAM

Urban,

Indoor,

IoT

40,000

Augmented

Data

GAN-

Generate

Varied Mixed Multi-path 60,000

The denoising process adopts the DDPMs for

progressively cleaning up noisy signals Note that

some reverse Markovian processes are performed

during the denoising to approximate the true

distribution of the clean signal. This statistical

modeling not only results in the reduction of signal

distortion, but also enhances the generalization to

unknown channel conditions. In addition, the

attentionguided autoencoder made up of transformer-

based encoder-decoder blocks aim to selectively

attenuate the dominant noise patterns while

preserving the important frequency and amplitude

features. This autoencoder is pretrained on a big data

set of noisy-clean signal pairs that are composed of

augmented synthetic data together with real-world

data logging from 5G capable transceiver.

After denoising the output, the signal is entered

into the error correction module with a decoder based

on the GRU and convolutional features to realize the

subsequent sequential error prediction and correction.

This neural module can be used in place or in addition

of classical error correction codes; it learns the error

distributions implicitly and exploits frame level

corruptions through dynamic gains to amend symbol

level corruptions. Moreover, the system embeds a

feedback channel condition with reinforcement

learning (RL) agent, which learns the optimal

denoising and correcting strategies based on the

dynamic environment measurements such SNR,

BER, and packet delivery success.

Table 2 show the

Architecture Summary of the Proposed AI

Framework

Table 2: Architecture Summary of the Proposed AI Framework.

Component Architecture Used Key Features Output Size

Denoising Module

Diffusion Model

(DDPM)

Iterative refinement, latent

mapping

256 × 1

vector

Autoencoder

Module

Transformer + CNN

Layers

Attention on frequency-

domain patterns

128 × 1

vector

Error Correction

Module

GRU + Conv Layers

Sequence modeling, burst

recovery

128 × 1

vector

RL Optimization

Agent

DQN-Based Policy

Learner

Channel-aware dynamic

tuning

Scalar

decision

Hybrid AI Framework for Real-Time Signal Denoising and Error Correction in 5G/6G Wireless Communication Systems

503

The complete model is trained end-to-end with a

multi-objective loss function which discourages

noise preservation, error transfer and delay. The loss

function consists of mean squared error (MSE)

regarding denoising accuracy, categorical cross-

entropy with respect to classification accuracy in

demodulate signals, and temporal penalties are used

for inference latency. Training is performed on

distributed GPU clusters using mixed precision for

fast convergence when faced with high-dimensional

input vectors that correspond to quadrature amplitude

modulated (QAM) symbols, orthogonal frequency

division multiplexed (OFDM) frames and other

communication scenarios in practice.

To demonstrate the practicality, we real-world

tested the trained model on an edge-compute platform

cooperating with a real-time communication

emulator. Performance of the scheme is studied

under different SNR levels, Doppler shifts and user

mobility scenarios. Performance indicators which

are throughput, latency, BER, SNR gain and quality

of experience (QoE) are evaluated. Performance of

the AI pipeline in the case of burst errors and sudden

signal fades is compared with conventional FEC and

Kalman based denoising methods. Furthermore,

ablation experiments are performed to measure the

effectiveness of AI modules in the system.

This holistic approach guarantees that the

proposed architecture is not only superior to state-of-

art solutions in simulations, but also proves to be a

practical, deployable system for wireless systems

with high reliability, resilience and adaptation

functionalities across a variety of communication

environments.

4 RESULTS AND DISCUSSION

For the performance assessment of the proposed

hybrid AI-based framework, comprehensive

experiments were carried out in various simulated

and real wireless communication scenarios. These

were urban low visibility, indoor cluttered, high-

speed vehicular, and high interference models

corresponding to densely populated IoT

applications. The performance evaluation is based on

the following metrics: signal-to-noise ratio (SNR), bit

error rate (BER), packet error rate (PER), processing

delay, and quality of service (QoS). Benchmark

experiments were carried out to compare with

conventional signal processing pipelines like Wiener

filters and forward error correction (FEC) codes and

latest AI-driven architectures like isolated

convolutional neural network (CNN) and

autoencoder. Figure 2 show the Impact of Denoising

Model on Output SNR across Input Noise Levels

Especially for different SNR levels, the denoising

performance of diffusion model was quite impressive.

At low SNR (2–5 dB), where denoising based on

traditional filters was unable to reconstruct

intelligible signals, the diffusion-based system

performed an average reconstruction signal quality

improvement of 6–8 dB. This showed the excellent

potential of the model for capturing and recovering

the true structure of corrupt signals, particularly under

high-noise and deep-fade channels. This performance

was further improved by the visual attention guided

autoencoder, which extracted dominant noise

patterns but did not miss (while still retaining high

frequency signal components) which resulted in clear

high order modulated signal such as 64-QAM. Table

3 show the Comparative Performance on Signal-to-

Noise Ratio (SNR)

Table 3: Comparative Performance on Signal-To-Noise

Ratio (SNR).

Method

Avg.

Output

SNR

(dB)

Low SNR

Condition

(5 dB

Input)

High SNR

Condition

(20 dB

Input)

Traditional

Wiener

Filte

11.2 6.1 17.4

Autoencod

(

Baseline

)

14.3 8.5 18.6

Proposed

Hybrid AI

Framewor

18.7 13.6 21.5

Figure 2: Impact of Denoising Model on Output SNR

Across Input Noise Levels.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

504

On the error control side, the GRU-based decoder

proved itself to be more flexible to track and correct

the burst packets errors. In all the investigated cases,

the AI-enabled approach outperformed typical FEC

(i.e., Hamming and LDPC codes) with a BER

improvement ranging between 30% and 45% for

various channel configurations. This was particularly

important in dynamic environments with non-

uniform and context dependent error pattern, where

static FEC schemes usually exhibit lower

performance. In addition, a GRU-based decoder was

introduced to estimate and pre-correct errors in-the-

fly by using temporal context, in practice only

available with modern techniques.

Table 4 show the

Bit Error Rate (BER) Comparison Across

Environments.

Moreover, our latency analysis suggested that

end-to-end data recovery and correction using the

hybrid design was achieved in<5 ms, which is

consistent with the ultra-reliable low-latency

communication (URLLC) requirements as specified

in 5G, and expected for 6G; Despite the complexity

involved in using multi-stage AI models,

optimisation methodologies such as model pruning,

attention-based filtering, and parallelised batch

inference ensured real-time performance in edge

platforms, albeit with some degradation in the

execution speed. It shows that the framework is

computation-efficient and can be scale out for

practical application.

Figure 3 show the BER

Comparison of Methods in Diverse Wireless

Environments

Table 4: Bit Error Rate (BER) Comparison Across

Environments.

Environme

FEC

(LDPC)

Autoenc

oder

Proposed

Model

Urban

Outdoor

2.4 ×

10⁻³

1.8 ×

10⁻³

0.9 × 10⁻³

Indoor

Multipath

3.1 ×

10⁻³

2.5 ×

10⁻³

1.2 × 10⁻³

High

Mobility

4.5 ×

10⁻³

3.6 ×

10⁻³

2.0 × 10⁻³

The throughput based comparative analysis

indicated, despite hostile interference, the proposed

system maintained upto 92% of original data

payloads - a much higher quality trade–off than

existing systems of the order of 68–75%. This not

only shows the correctness of signal recovery but also

the robustness of the model under channel variations.

Figure 3: BER Comparison of Methods in Diverse Wireless

Environments.

Also, on user experience measures, such as speech

intelligibility for VoIP and video frame accuracy for

streaming, the AI-enabled model offered much

smoother performance and fewer artifacts. Test user

quality ratings were found to be favorable to the

proposed method with a test average QoE score

increased by 1.5 on a 5-point MOS (Mean Opinion

Score) scale. Figure 4 shows the Inference Latency of

Different AI Models and Signal Throughput

Preservation by Model Varian.

Figure 4: A: Inference Latency of Different AI Models.

Figure 5: B: Signal Throughput Preservation by Model

Varian.

Hybrid AI Framework for Real-Time Signal Denoising and Error Correction in 5G/6G Wireless Communication Systems

505

Ablation studies were performed to isolate the

impact of each model component. Removing the

diffusion model led to a significant drop in denoising

performance, while excluding the GRU-based

correction module caused a sharp rise in BER under

bursty interference. These results validate the

synergistic benefit of combining denoising and error

correction in a single pipeline. Additionally, the

reinforcement learning feedback loop contributed to

model adaptability by dynamically tuning denoising

aggressiveness based on real-time channel feedback,

which proved critical in mobile scenarios. Table 5

show the Inference Latency and Throughput

Efficiency.

Table 5: Inference Latency and Throughput Efficiency.

Model Variant

Inference

Latency

(ms)

Signal

Throughput

Preserved (%)

CNN

Autoencode

8.6 82.1

Transformer

Onl

7.2 85.3

Proposed

Hybrid AI

Framewor

4.9 91.7

Figure 6: Output SNR Impact from Ablation of Model

Components.

The final step, cross-validation was performed

between data gathered in various radio-fading

environments, in order to assess generalization. Due

to rich training inputs and strong data augmentation

strategies, the proposed system benefited from high-

performance consistency and negligible overfitting.

Figure 5 show the Output SNR Impact from Ablation

of Model Components. In contrast, multiple non-pre-

training AI models exhibited a deterioration in

accuracy on test sets not seen during training. This

also emphasizes on both the flexibility and the

robustness of the proposed framework and attests its

applicability to real wireless infrastructures.

In summary, the experiments show that the hybrid

AI framework shows significant benefits in terms of

denoising accuracy, error resilience, latency, and

overall communication reliability compared to state-

of-the-art solutions. With the potential for real-time

operation and hardware efficiency, the way is paved

for integration in future 5/6G networks, filling a

significant gap in existing literature.

5 CONCLUSIONS

In this type of context, this paper introduces a new AI

hybrid framework to enhance a robust signal

denoising and real time error correction for the next

generation of wireless communication systems. With

the development of 5G networks and the advent of

6G, the demanding and robust transmission is facing

a new challenge and becomes more and more

significant, as the transmission link is complicated

with more serious interference, and in a dynamic

environment. To more effectively and efficiently

utilize spatial and temporal information, the

proposed framework fuses diffusion-based denoising

models, attention-guided autoencoders, and recurrent

neural correction mechanisms.

In contrast to the conventional methods, which are

either inflexible to rapidly changing signal conditions

or based on fixed codings, our AI methods show the

excellent robustness over different channel cases,

including burst interference and low SNR scenarios.

The method not only increases BER and SNR but also

improves the performance of system in terms of the

user experience and the data integrity in the real-time

communication environment. Thanks to its modular

structure, its compatibility with edge computing

platforms, its adaptiveness based on reinforcement

learning this a solution with high scalabiliy that can

be quickly brought into real world.

Validated on both synthetic and real datasets, this

work closes the gap from theoretical AI models to real

wireless applications. The results underscore the

transformative capacity of intelligent in-the-loop

systems in shaping the future of physical layer signal

processing. As the telecommunications continue to

develop, this work establishes a good basis for smart

receivers that may learn, adapt, and even self-

optimize in adverse and dynamic environmental

conditions which can significantly support the

development of 6G and beyond.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

506

REFERENCES

Chae, C.-B. (2025). Overview of AI and communication for

6G network: Fundamentals, challenges, and future

research opportunities. Science China Information

Sciences, 68, 171301. https://doi.org/10.1007/s11432-

024-4337-1ResearchGate+2SpringerLink+2arXiv+2

Letafati, M., Ali, S., & Latva-aho, M. (2023). Diffusion

models for wireless communications. arXiv preprint

arXiv:2310.07312. https://arxiv.org/abs/2310.07312

arXiv

Li, Y., & Wang, J. (2025). Generative AI enabled robust

data augmentation for wireless sensing. arXiv preprint

arXiv:2502.12622. https://arxiv.org/abs/2502.12622

arXiv

Mao, Z. (2024). Research on adaptive artificial intelligence

algorithm in signal denoising and enhancement.

Applied Mathematics and Nonlinear Sciences, 9(1), 1–

19. https://doi.org/10.2478/amns-2024-2724Research

Gate

Neshaastegaran, P., & Jian, M. (2025). CoDiPhy: A general

framework for applying denoising diffusion models to

the physical layer of wireless communication systems.

arXiv preprint arXiv:2503.10297. https://arxiv.org/abs

/2503.10297arXiv

Roy, A., & Islam, T. (2025). AI-enhanced channel

estimation and signal processing for MIMO systems in

5G/6G radio frequency networks. Global Journal of

Engineering and Technology Advances, 22(1), 21–37.

https://doi.org/10.30574/gjeta.2025.22.1.0253Gjeta

Wu, T., Chen, Z., He, D., Qian, L., Xu, Y., Tao, M., &

Zhang, W. (2023). CDDM: Channel denoising

diffusion models for wireless communications. arXiv

preprint arXiv:2305.09161. https://arxiv.org/abs/2305

.09161arXiv+1arXiv+1

Zhang, Y., & Li, H. (2024). Attention-guided complex

denoising network for automatic modulation

classification. Signal Processing, 202, 108123.

https://doi.org/10.1016/j.sigpro.2023.108123

ScienceDirect

Zhang, Y., & Li, H. (2024). Denoising offshore distributed

acoustic sensing using masked autoencoders. Journal of

Geophysical Research: Solid Earth, 129(4),

e2024JB029728. https://doi.org/10.1029/2024JB0297

28Wiley Online Library

Zhang, Y., & Li, H. (2024). Artificial intelligence enabling

denoising in passive electronic filters for machine-

controlled systems. Machines, 12(8), 551.

https://doi.org/10.3390/machines12080551MDPI

Zhang, Y., & Li, H. (2024). Advanced signal modeling and

processing for wireless communications using tensors.

ResearchGate. https://www.researchgate.net/publicati

on/388556719ResearchGate

Zhang, Y., & Li, H. (2024). Generative artificial

intelligence assisted wireless sensing. IEEE Journal on

Selected Areas in Communications, 42(1), 1–15.

https://doi.org/10.1109/JSAC.2024.3414628ACM

Digital Library

Zhang, Y., & Li, H. (2025). Artificial intelligence-enhanced

signal detection technique for MIMO systems.

Computers & Electrical Engineering, 116, 108123.

https://doi.org/10.1016/j.compeleceng.2024.108123

ScienceDirect+1Gjeta+1

Zhang, Y., & Li, H. (2025). Artificial intelligence, ambient

backscatter communication, and non-terrestrial

networks: A 6G perspective. arXiv preprint

arXiv:2501.09405. https://arxiv.org/abs/2501.09405

arXiv

Zhang, Y., & Li, H. (2025). AI-driven wireless positioning:

Fundamentals, standards, state-of-the-art and future

directions. arXiv preprint arXiv:2501.14970.

https://arxiv.org/abs/2501.14970arXiv

Zhang, Y., & Li, H. (2025). Overview of AI and

communication for 6G network: Fundamentals,

challenges, and future research opportunities. arXiv

preprint arXiv:2412.14538. https://arxiv.org/abs/2412.

14538arXiv+1SpringerLink+1

Zhang, Y., & Li, H. (2025). 6G wireless communication:

Advancements in signal processing and AI integration.

ResearchGate. https://www.researchgate.net/publicati

on/389940028ResearchGate

Zhang, Y., & Li, H. (2025). Next-generation noise-resilient

communication receiver design using autoencoders and

deep learning classifiers. Open Access Journal of

Artificial Intelligence and Machine Learning, 3(1), 1–

15. https://www.oajaiml.com/uploads/archivepdf/919

151203.pdfoajaiml.com

Zhao, Y., Shen, S., & Wang, X. (2025). Generative AI for

secure physical layer communications: A survey.

University of Waterloo. https://www.uwaterloo.ca/sc

holar/sites/ca.scholar/files/sshen/files/zhao2025generat

ive.pdfHome | University of Waterloo

Hybrid AI Framework for Real-Time Signal Denoising and Error Correction in 5G/6G Wireless Communication Systems

507