Pub Date : 2026-03-01Epub Date: 2026-01-02DOI: 10.1016/j.phycom.2026.102992
Zhongqiang Luo , Wenjie Wu , Xiang Dai , Qiang Han
Using unmanned aerial vehicles (UAVs) equipped with servers to assist multi-access edge computing (MEC) can provide computing support in areas with insufficient network coverage or hotspots. However, UAV-MEC systems under traditional cellular network support are susceptible to inter-cell interference and shadow fading, resulting in increased task processing delays and higher energy consumption. To address these challenges, this paper proposes a user-centric UAV-MEC architecture (UCUAV-MEC). This architecture integrates a user-centered transmission method, dynamically adjusting the UAV and access point (AP) to provide flexible computing and communication support for user equipment (UE). Additionally, dual connectivity (DC) technology is employed to enable parallel processing, alleviating resource competition and transmission interference. The delay and energy minimization problem is then formulated by jointly optimizing the UAV position, offloading decision, power allocation, and computing resource allocation within the UCUAV-MEC framework. To solve this problem, this paper proposes a multi-agent collaborative optimization scheme based on deep reinforcement learning (DRL) and convex optimization. Simulation results demonstrate that, compared to traditional UAV-MEC, the proposed optimization scheme based on UCUAV-MEC can reduce delay by up to 72.26% and energy consumption by 73.29%.
{"title":"Deep reinforcement learning-based computation offloading and resource allocation in user-centered UAV-MEC","authors":"Zhongqiang Luo , Wenjie Wu , Xiang Dai , Qiang Han","doi":"10.1016/j.phycom.2026.102992","DOIUrl":"10.1016/j.phycom.2026.102992","url":null,"abstract":"<div><div>Using unmanned aerial vehicles (UAVs) equipped with servers to assist multi-access edge computing (MEC) can provide computing support in areas with insufficient network coverage or hotspots. However, UAV-MEC systems under traditional cellular network support are susceptible to inter-cell interference and shadow fading, resulting in increased task processing delays and higher energy consumption. To address these challenges, this paper proposes a user-centric UAV-MEC architecture (UCUAV-MEC). This architecture integrates a user-centered transmission method, dynamically adjusting the UAV and access point (AP) to provide flexible computing and communication support for user equipment (UE). Additionally, dual connectivity (DC) technology is employed to enable parallel processing, alleviating resource competition and transmission interference. The delay and energy minimization problem is then formulated by jointly optimizing the UAV position, offloading decision, power allocation, and computing resource allocation within the UCUAV-MEC framework. To solve this problem, this paper proposes a multi-agent collaborative optimization scheme based on deep reinforcement learning (DRL) and convex optimization. Simulation results demonstrate that, compared to traditional UAV-MEC, the proposed optimization scheme based on UCUAV-MEC can reduce delay by up to 72.26% and energy consumption by 73.29%.</div></div>","PeriodicalId":48707,"journal":{"name":"Physical Communication","volume":"75 ","pages":"Article 102992"},"PeriodicalIF":2.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145980646","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-01-06DOI: 10.1016/j.phycom.2026.102993
A.O. Beena , S.M. Sameer
An investigation on the performance of a deep learning (DL)-based signal detection scheme over the traditional successive interference cancellation (SIC) method for a multi input - multi output (MIMO) based non-orthogonal multiple access (NOMA) downlink system for vehicular communication is presented in this paper. The proposed scheme is analyzed for a system with dynamic power allocation in a network experiencing Rayleigh fading channels. The DL-based MIMO-NOMA receiver detects signals from multiple users in a single-step process without explicitly estimating the channel state information (CSI). A long-short-term memory (LSTM) scheme is proposed for joint channel estimation and signal detection. Performance of the proposed DL-based detection scheme is compared with that of the traditional SIC based on least square (SIC-LS) and minimum mean square error (SIC-MMSE) estimations in terms of outage probability, sum rate, and symbol error rate (SER). Extensive simulations show that the proposed DL-based detection is more effective than conventional SIC techniques. SER studies revealed that there is 99.99 % detection accuracy at an SNR of 18 dB for the proposed DL-based detector. Also, at an SNR of 35 dB, the proposed DL-based detector with dynamic power allocation (DPA) achieves approximately 94 % lower outage probability for the weak user (Vehicle 2) and 83 % lower outage probability for the strong user (Vehicle 1) compared to the conventional SIC-based detectors. Furthermore, the SER performance of the DL-based scheme shows an improvement of up to 9 dB in SNR over the SIC-LS and SIC-MMSE schemes at a target SER of under various channel conditions. Additionally, the proposed DL-based detector achieves a higher Jain’s fairness index compared to conventional SIC-based detection, ensuring a more equitable assurance of quality of service (QoS) among users. Also, the DL-based technique shows better adaptability to variations in cyclic prefix (CP) length and the number of pilot symbols in the frame. It has been observed that even in situations of severe inter-symbol interference (ISI) or Doppler shift, the proposed DL-based method outperforms the LS and MMSE-based conventional SIC detectors. Further, it is shown that the computational complexity of the proposed scheme is much lower than the traditional schemes.
{"title":"Dynamic power allocation and a low-complexity deep learning based multi-user signal detection for NOMA-assisted vehicular communication","authors":"A.O. Beena , S.M. Sameer","doi":"10.1016/j.phycom.2026.102993","DOIUrl":"10.1016/j.phycom.2026.102993","url":null,"abstract":"<div><div>An investigation on the performance of a deep learning (DL)-based signal detection scheme over the traditional successive interference cancellation (SIC) method for a multi input - multi output (MIMO) based non-orthogonal multiple access (NOMA) downlink system for vehicular communication is presented in this paper. The proposed scheme is analyzed for a system with dynamic power allocation in a network experiencing Rayleigh fading channels. The DL-based MIMO-NOMA receiver detects signals from multiple users in a single-step process without explicitly estimating the channel state information (CSI). A long-short-term memory (LSTM) scheme is proposed for joint channel estimation and signal detection. Performance of the proposed DL-based detection scheme is compared with that of the traditional SIC based on least square (SIC-LS) and minimum mean square error (SIC-MMSE) estimations in terms of outage probability, sum rate, and symbol error rate (SER). Extensive simulations show that the proposed DL-based detection is more effective than conventional SIC techniques. SER studies revealed that there is 99.99 % detection accuracy at an SNR of 18 dB for the proposed DL-based detector. Also, at an SNR of 35 dB, the proposed DL-based detector with dynamic power allocation (DPA) achieves approximately 94 % lower outage probability for the weak user (Vehicle 2) and 83 % lower outage probability for the strong user (Vehicle 1) compared to the conventional SIC-based detectors. Furthermore, the SER performance of the DL-based scheme shows an improvement of up to 9 dB in SNR over the SIC-LS and SIC-MMSE schemes at a target SER of <span><math><msup><mn>10</mn><mrow><mo>−</mo><mn>4</mn></mrow></msup></math></span> under various channel conditions. Additionally, the proposed DL-based detector achieves a higher Jain’s fairness index compared to conventional SIC-based detection, ensuring a more equitable assurance of quality of service (QoS) among users. Also, the DL-based technique shows better adaptability to variations in cyclic prefix (CP) length and the number of pilot symbols in the frame. It has been observed that even in situations of severe inter-symbol interference (ISI) or Doppler shift, the proposed DL-based method outperforms the LS and MMSE-based conventional SIC detectors. Further, it is shown that the computational complexity of the proposed scheme is much lower than the traditional schemes.</div></div>","PeriodicalId":48707,"journal":{"name":"Physical Communication","volume":"75 ","pages":"Article 102993"},"PeriodicalIF":2.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145980713","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-17DOI: 10.1016/j.phycom.2026.103051
Long Cheng , Yuhao Kong
The performance of Wi-Fi CSI fingerprinting is often degraded by non-line-of-sight (NLOS) propagation, environmental dynamics, and device heterogeneity. To overcome these challenges, this work introduces CALGA-Net, an end-to-end coordinate regression framework designed for robust indoor localization. At its core, CALGA-Net employs a dual-path attention block that operates in parallel at the same spatial scale. Each residual unit combines large-kernel convolution with self-attention to integrate local priors and long-range dependencies. A lightweight fusion mechanism aggregates the two streams while preserving spatial consistency. The attention pathway retains directional semantics and strengthens non-local correlation modeling. A multi-scale atrous spatial pyramid pooling head aggregates context before predicting two-dimensional coordinates. The framework follows an offline training and online single-pass inference paradigm and requires no geometric ranging such as AoA or ToF or auxiliary sensors. We evaluate CALGA-Net on two public CSI localization datasets across three indoor environments: Lab, Meeting, and Hallway. CALGA-Net achieves mean localization errors of 0.088 m, 0.075 m, and 0.336 m in these environments, with standard deviations of 0.084 m, 0.122 m, and 0.406 m. These results support the effectiveness of combining large-kernel convolution with attention for CSI-based indoor localization.
{"title":"CALGA-Net: Convolution–attention with local–global aggregation for robust CSI-based indoor localization","authors":"Long Cheng , Yuhao Kong","doi":"10.1016/j.phycom.2026.103051","DOIUrl":"10.1016/j.phycom.2026.103051","url":null,"abstract":"<div><div>The performance of Wi-Fi CSI fingerprinting is often degraded by non-line-of-sight (NLOS) propagation, environmental dynamics, and device heterogeneity. To overcome these challenges, this work introduces CALGA-Net, an end-to-end coordinate regression framework designed for robust indoor localization. At its core, CALGA-Net employs a dual-path attention block that operates in parallel at the same spatial scale. Each residual unit combines large-kernel convolution with self-attention to integrate local priors and long-range dependencies. A lightweight fusion mechanism aggregates the two streams while preserving spatial consistency. The attention pathway retains directional semantics and strengthens non-local correlation modeling. A multi-scale atrous spatial pyramid pooling head aggregates context before predicting two-dimensional coordinates. The framework follows an offline training and online single-pass inference paradigm and requires no geometric ranging such as AoA or ToF or auxiliary sensors. We evaluate CALGA-Net on two public CSI localization datasets across three indoor environments: Lab, Meeting, and Hallway. CALGA-Net achieves mean localization errors of 0.088 m, 0.075 m, and 0.336 m in these environments, with standard deviations of 0.084 m, 0.122 m, and 0.406 m. These results support the effectiveness of combining large-kernel convolution with attention for CSI-based indoor localization.</div></div>","PeriodicalId":48707,"journal":{"name":"Physical Communication","volume":"75 ","pages":"Article 103051"},"PeriodicalIF":2.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147397583","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-06DOI: 10.1016/j.phycom.2026.103031
Xingcan Chen , Wendong Xiao
Owing to the advantages of no disturbance to the human body and low cost, the systems of human activity recognition (HAR) through WiFi channel state information (CSI) has been widely studied. A major problem that hinders the deployment of these systems in the real world is the position-dependence, i.e., the sensing performance of these systems will significantly degrade when the location or orientation of the human changes. To solve this problem, we propose a novel position-independent HAR system based on WiFi CSI physical information-guided deep learning models (WiPIHAR). Specifically, we first theoretically analyze the CSI physical information, i.e., the position-independent CSI. After that, we design an algorithm based on a single graph convolution layer (GAL) to enhance the features of the position-independent CSI and significantly reduce the feature dimension. Finally, we further analyze the enhanced position-independent CSI, and then design a multi-scale causal Transformer network (MSCTN) to automatically capture and fuse the local and global features that are helpful for position-independent HAR. Extensive experimental results on three datasets (OR, Widar3.0 and SC) demonstrate that WiPIHAR outperforms the existing state-of-the-art approaches. WiPIHAR achieves an average accuracy of 99.19% and 94.49% on the OR and Widar3.0 datasets respectively, and maintains high performance of 94.08% and 93.30% under cross-position and cross-orientation conditions on our self-collected dataset. Furthermore, WiPIHAR attains an average accuracy of 97.16% in the cross-dataset evaluation.
{"title":"WiFi-based position-independent activity sensing via physical information-guided feature fusion","authors":"Xingcan Chen , Wendong Xiao","doi":"10.1016/j.phycom.2026.103031","DOIUrl":"10.1016/j.phycom.2026.103031","url":null,"abstract":"<div><div>Owing to the advantages of no disturbance to the human body and low cost, the systems of human activity recognition (HAR) through WiFi channel state information (CSI) has been widely studied. A major problem that hinders the deployment of these systems in the real world is the position-dependence, i.e., the sensing performance of these systems will significantly degrade when the location or orientation of the human changes. To solve this problem, we propose a novel position-independent HAR system based on WiFi CSI physical information-guided deep learning models (WiPIHAR). Specifically, we first theoretically analyze the CSI physical information, i.e., the position-independent CSI. After that, we design an algorithm based on a single graph convolution layer (GAL) to enhance the features of the position-independent CSI and significantly reduce the feature dimension. Finally, we further analyze the enhanced position-independent CSI, and then design a multi-scale causal Transformer network (MSCTN) to automatically capture and fuse the local and global features that are helpful for position-independent HAR. Extensive experimental results on three datasets (OR, Widar3.0 and SC) demonstrate that WiPIHAR outperforms the existing state-of-the-art approaches. WiPIHAR achieves an average accuracy of 99.19% and 94.49% on the OR and Widar3.0 datasets respectively, and maintains high performance of 94.08% and 93.30% under cross-position and cross-orientation conditions on our self-collected dataset. Furthermore, WiPIHAR attains an average accuracy of 97.16% in the cross-dataset evaluation.</div></div>","PeriodicalId":48707,"journal":{"name":"Physical Communication","volume":"75 ","pages":"Article 103031"},"PeriodicalIF":2.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147397921","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-03DOI: 10.1016/j.phycom.2026.103035
Karthiga M , Deepa D
The coalescence of sensing and communication is one of the major requirements of 6 G wireless communication. Affine Frequency Division Multiplexing (AFDM) is propitious in the attainment of Integrated Sensing and Communication (ISAC). The signal processing technique in AFDM offers adaptable chirp parameters pivoting on channel conditions. The major concern in AFDM is the high Peak-to-Average-Power ratio (PAPR). The pliability of pre-chirp parameter expedites the PAPR reduction strategy in AFDM. In this paper, optimizing a single scalar pre-chirp parameter for PAPR reduction is considered a one-dimensional nonlinear nonconvex problem. The Projected Gradient Descent using Finite Difference method with Armijo Backtracking (PGA) minimizes the non-smooth, non-convex PAPR objective function. To further explore the global optimum, a multi-start strategy is adopted, which solves the optimization problem from several initial points and selects the best solution with the minimum PAPR. To facilitate robust AFDM-ISAC sensing parameter estimation, a bounded constraint is instilled on the pre-chirp parameter such that the superposition of complex exponentials at the receiver is negligible. Simulation results show that the proposed algorithm achieves a 51.75% reduction in PAPR without compromising BER. Additionally, the computational time and the distribution of the optimal pre-chirp parameter of the proposed algorithm are compared with those of other traditional algorithms.
{"title":"Multistart projected gradient descent optimization with Armijo backtracking-based pre-chirp tuning for PAPR reduction in AFDM","authors":"Karthiga M , Deepa D","doi":"10.1016/j.phycom.2026.103035","DOIUrl":"10.1016/j.phycom.2026.103035","url":null,"abstract":"<div><div>The coalescence of sensing and communication is one of the major requirements of 6 G wireless communication. Affine Frequency Division Multiplexing (AFDM) is propitious in the attainment of Integrated Sensing and Communication (ISAC). The signal processing technique in AFDM offers adaptable chirp parameters pivoting on channel conditions. The major concern in AFDM is the high Peak-to-Average-Power ratio (PAPR). The pliability of pre-chirp parameter expedites the PAPR reduction strategy in AFDM. In this paper, optimizing a single scalar pre-chirp parameter for PAPR reduction is considered a one-dimensional nonlinear nonconvex problem. The Projected Gradient Descent using Finite Difference method with Armijo Backtracking (PGA) minimizes the non-smooth, non-convex PAPR objective function. To further explore the global optimum, a multi-start strategy is adopted, which solves the optimization problem from several initial points and selects the best solution with the minimum PAPR. To facilitate robust AFDM-ISAC sensing parameter estimation, a bounded constraint is instilled on the pre-chirp parameter such that the superposition of complex exponentials at the receiver is negligible. Simulation results show that the proposed algorithm achieves a 51.75% reduction in PAPR without compromising BER. Additionally, the computational time and the distribution of the optimal pre-chirp parameter of the proposed algorithm are compared with those of other traditional algorithms.</div></div>","PeriodicalId":48707,"journal":{"name":"Physical Communication","volume":"75 ","pages":"Article 103035"},"PeriodicalIF":2.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147397924","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-01-31DOI: 10.1016/j.phycom.2026.103020
R. Anil Kumar , Sarala Patchala , R. Prakash Kumar , Ummiti Sreenivasulu , Shaik Fairooz
In modern wireless communication systems, Orthogonal Frequency Division Multiplexing (OFDM) is widely used in modulation technology. However, inter-subcarrier interference, along with nonlinear distortion, results in OFDM signals with a shorter cyclic prefix (CP) length and restricted pilot usages having lower performance in Channel Estimation (CE) or signal detection. Moreover, in conventional frequency or angle-domain based CE, performance degradation may occur due to natural sparsity in practical wireless communication channels. This study proposes an innovative Hybrid Polar and Frequency Domain Multi-Scale Depthwise Separable ResNet based Receiver (HPF-MSDSRRxr), whose aim is to overcome these limitations in recovering distorted OFDM signals due to non-linearities. Specifically, two new deep learning (DL) receiver architectures are formulated by combining the frequency domain and polar domain signal processing. To alleviate the performance degradation introduced by CE methods involving the angular domain, a Multi-Scale Depthwise Separable ResNet model called Polar Domain MSDSR (P-MSDSR) is first constructed. This model helps effectively exploit the channel sparseness property inherent to the polar domain. The results from the MSDSR model operating in the frequency domain, called F-MSDSR, which uses Demodulation Reference Signals (DMRS) along with accurate channel estimation and soft-bit detection, are then combined by seamlessly blending polar domain features. The introduced multi-scale idea, which efficiently detects both fine and coarse channel characteristics, enables dependable performance, including large nonlinear impairments and high error vector magnitude in the proposed architectural design. Simulation results verify that the HPF-MSDSRRxr scheme typically gains a constant Signal-to-Noise Ratio (SNR) margin of 2–3.5 dB with tolerable computing complexity over existing linear LMMSE and current DL baselines.
{"title":"A lightweight hybrid deep learning-based OFDM receiver for enhanced coverage and efficiency","authors":"R. Anil Kumar , Sarala Patchala , R. Prakash Kumar , Ummiti Sreenivasulu , Shaik Fairooz","doi":"10.1016/j.phycom.2026.103020","DOIUrl":"10.1016/j.phycom.2026.103020","url":null,"abstract":"<div><div>In modern wireless communication systems, Orthogonal Frequency Division Multiplexing (OFDM) is widely used in modulation technology. However, inter-subcarrier interference, along with nonlinear distortion, results in OFDM signals with a shorter cyclic prefix (CP) length and restricted pilot usages having lower performance in Channel Estimation (CE) or signal detection. Moreover, in conventional frequency or angle-domain based CE, performance degradation may occur due to natural sparsity in practical wireless communication channels. This study proposes an innovative Hybrid Polar and Frequency Domain Multi-Scale Depthwise Separable ResNet based Receiver (HPF-MSDSRRxr), whose aim is to overcome these limitations in recovering distorted OFDM signals due to non-linearities. Specifically, two new deep learning (DL) receiver architectures are formulated by combining the frequency domain and polar domain signal processing. To alleviate the performance degradation introduced by CE methods involving the angular domain, a Multi-Scale Depthwise Separable ResNet model called Polar Domain MSDSR (P-MSDSR) is first constructed. This model helps effectively exploit the channel sparseness property inherent to the polar domain. The results from the MSDSR model operating in the frequency domain, called F-MSDSR, which uses Demodulation Reference Signals (DMRS) along with accurate channel estimation and soft-bit detection, are then combined by seamlessly blending polar domain features. The introduced multi-scale idea, which efficiently detects both fine and coarse channel characteristics, enables dependable performance, including large nonlinear impairments and high error vector magnitude in the proposed architectural design. Simulation results verify that the HPF-MSDSRRxr scheme typically gains a constant Signal-to-Noise Ratio (SNR) margin of 2–3.5 dB with tolerable computing complexity over existing linear LMMSE and current DL baselines.</div></div>","PeriodicalId":48707,"journal":{"name":"Physical Communication","volume":"75 ","pages":"Article 103020"},"PeriodicalIF":2.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147397960","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-19DOI: 10.1016/j.phycom.2026.103048
Rashmi, Prashant Kumar
Device-to-Device (D2D) communication is vital in Fifth-Generation (5G) networks for reducing latency and offloading base stations, but its effectiveness is constrained by two persistent challenges: finding optimal multi-hop routes in dynamic conditions and preserving device battery life. Existing routing schemes typically optimize one at the expense of the other, leading to inefficient paths or premature device shutdown. This paper introduces a Distributional Dueling Q-learning (2-DQ) algorithm that decomposes the action-value (Q) function into state value and action advantage terms while explicitly enforcing a 30% minimum battery threshold. Extensive simulations show that 2-DQ delivers a 23% gain in route efficiency, a 19% improvement in adaptability under dense and heterogeneous network scenarios, and a 17% boost in energy optimization compared to standard D2D and single-dueling Q-learning approaches. Moreover, the algorithm consistently maintains device battery levels above operational thresholds in urban, rural, and industrial testbeds. These results position 2-DQ as a scalable and energy-aware framework for real-time D2D path selection in next-generation 5G deployments.
{"title":"Optimizing device-to-device path discovery in 5G networks using distributional dueling Q-learning with battery constraints","authors":"Rashmi, Prashant Kumar","doi":"10.1016/j.phycom.2026.103048","DOIUrl":"10.1016/j.phycom.2026.103048","url":null,"abstract":"<div><div>Device-to-Device (D2D) communication is vital in Fifth-Generation (5G) networks for reducing latency and offloading base stations, but its effectiveness is constrained by two persistent challenges: finding optimal multi-hop routes in dynamic conditions and preserving device battery life. Existing routing schemes typically optimize one at the expense of the other, leading to inefficient paths or premature device shutdown. This paper introduces a Distributional Dueling Q-learning (2-DQ) algorithm that decomposes the action-value (Q) function into state value and action advantage terms while explicitly enforcing a 30% minimum battery threshold. Extensive simulations show that 2-DQ delivers a 23% gain in route efficiency, a 19% improvement in adaptability under dense and heterogeneous network scenarios, and a 17% boost in energy optimization compared to standard D2D and single-dueling Q-learning approaches. Moreover, the algorithm consistently maintains device battery levels above operational thresholds in urban, rural, and industrial testbeds. These results position 2-DQ as a scalable and energy-aware framework for real-time D2D path selection in next-generation 5G deployments.</div></div>","PeriodicalId":48707,"journal":{"name":"Physical Communication","volume":"75 ","pages":"Article 103048"},"PeriodicalIF":2.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147397947","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-01-27DOI: 10.1016/j.phycom.2026.103019
Junhui Huang, Yan Guo, Hao Yuan, Xiaonan Cui, Xinliang Chen
The study of the target-attacker-defender (TAD) problem in three-dimensional (3D) space is of great significance to enhance the defense effectiveness in complex confrontation scenarios. However, the existing research has not fully considered the agent initiative, and the traditional reward function is difficult to motivate the defender to intercept actively; and the scenario simulation is mostly confined to the two-dimensional (2D) plane, while ignoring the differences in the distribution of the observed features, which leads to mixed information when the multi-agent algorithms deal with the non-independent and identically-distributed data. To solve these problems, this paper proposes the two-stream heterogeneous multi-agent proximal policy optimization algorithm (TSHMAPPO) algorithm. In the design of the reward mechanism, the reward relies on the combination of the previous moment’s state characteristics and the current state characteristics to provide effective reward signals for the active interception behavior of the defender, which solves the drawback of the traditional reward function that ignores the initiativ; in the observation processing, a two-stream feature extraction network is introduced into the MAPPO algorithm to realize adaptive characterization of non-independent and identically-distributed observation data, and to reduce the information confusions brought by the mixing of features. Experimental results show that the reward function design proposed in this paper improves the interception rate by 5.48% to 8.50% compared with the traditional method. The use of two-stream feature extraction network improves the defender’s interception rate by 1.47% to 10.73% in simulation experiments in different scenarios, while the TSHMAPPO algorithm has better interception performance than other comparison experiments.
{"title":"Dynamic resource allocation in air-ground defense via heterogeneous multi-agent tracking with cross-temporal state rewards","authors":"Junhui Huang, Yan Guo, Hao Yuan, Xiaonan Cui, Xinliang Chen","doi":"10.1016/j.phycom.2026.103019","DOIUrl":"10.1016/j.phycom.2026.103019","url":null,"abstract":"<div><div>The study of the target-attacker-defender (TAD) problem in three-dimensional (3D) space is of great significance to enhance the defense effectiveness in complex confrontation scenarios. However, the existing research has not fully considered the agent initiative, and the traditional reward function is difficult to motivate the defender to intercept actively; and the scenario simulation is mostly confined to the two-dimensional (2D) plane, while ignoring the differences in the distribution of the observed features, which leads to mixed information when the multi-agent algorithms deal with the non-independent and identically-distributed data. To solve these problems, this paper proposes the two-stream heterogeneous multi-agent proximal policy optimization algorithm (TSHMAPPO) algorithm. In the design of the reward mechanism, the reward relies on the combination of the previous moment’s state characteristics and the current state characteristics to provide effective reward signals for the active interception behavior of the defender, which solves the drawback of the traditional reward function that ignores the initiativ; in the observation processing, a two-stream feature extraction network is introduced into the MAPPO algorithm to realize adaptive characterization of non-independent and identically-distributed observation data, and to reduce the information confusions brought by the mixing of features. Experimental results show that the reward function design proposed in this paper improves the interception rate by 5.48% to 8.50% compared with the traditional method. The use of two-stream feature extraction network improves the defender’s interception rate by 1.47% to 10.73% in simulation experiments in different scenarios, while the TSHMAPPO algorithm has better interception performance than other comparison experiments.</div></div>","PeriodicalId":48707,"journal":{"name":"Physical Communication","volume":"75 ","pages":"Article 103019"},"PeriodicalIF":2.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147397951","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2025-12-10DOI: 10.1016/j.phycom.2025.102943
Haosu Zhang , Keyi Jing
With the continuous progress of society and the rapid development of information technology, the number of appeals on major government platforms is increasing rapidly, making traditional information processing methods difficult to efficiently handle and assign appeal information to relevant departments. To enhance government appeal handling efficiency, this paper proposes a semantic cognition architecture for government appeal classification, named SemGovNet-KG. This architecture constructs a domain-specific knowledge graph based on government information and designs a NeoBert-TextCNN based text classification model to achieve efficient semantic understanding and intelligent classification of government appeals. Addressing issues include semantic ambiguity, information silos, and weak knowledge associations in traditional government appeal processing, SemGovNet-KG integrates deep semantic modeling and knowledge graph reasoning mechanisms to build multi-level semantic representations and entity relationship networks. Experimental results show that SemGovNet-KG significantly improves classification accuracy, semantic interpretability, and system robustness on real government appeal datasets. This method provides a theoretical foundation and technical support for intelligent government services and efficient appeal response, with promising prospects for practical application.
{"title":"SemGovNet-KG: A semantic communication network for government affair appeals empowered by knowledge graph","authors":"Haosu Zhang , Keyi Jing","doi":"10.1016/j.phycom.2025.102943","DOIUrl":"10.1016/j.phycom.2025.102943","url":null,"abstract":"<div><div>With the continuous progress of society and the rapid development of information technology, the number of appeals on major government platforms is increasing rapidly, making traditional information processing methods difficult to efficiently handle and assign appeal information to relevant departments. To enhance government appeal handling efficiency, this paper proposes a semantic cognition architecture for government appeal classification, named SemGovNet-KG. This architecture constructs a domain-specific knowledge graph based on government information and designs a NeoBert-TextCNN based text classification model to achieve efficient semantic understanding and intelligent classification of government appeals. Addressing issues include semantic ambiguity, information silos, and weak knowledge associations in traditional government appeal processing, SemGovNet-KG integrates deep semantic modeling and knowledge graph reasoning mechanisms to build multi-level semantic representations and entity relationship networks. Experimental results show that SemGovNet-KG significantly improves classification accuracy, semantic interpretability, and system robustness on real government appeal datasets. This method provides a theoretical foundation and technical support for intelligent government services and efficient appeal response, with promising prospects for practical application.</div></div>","PeriodicalId":48707,"journal":{"name":"Physical Communication","volume":"75 ","pages":"Article 102943"},"PeriodicalIF":2.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147397592","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-03DOI: 10.1016/j.phycom.2026.103034
Jiai He, Tianxing Wang
The Orthogonal Time Frequency Space (OTFS) system has attracted significant research attention due to its superior performance in handling time-frequency doubly dispersive channels in high-mobility scenarios. As key reference signals for channel estimation, the design of pilots in the delay-Doppler (DD) domain crucially influences the final estimation accuracy. To address the issues of limited estimation accuracy and poor adaptability in existing pilot distribution schemes, this paper proposes a pilot distribution scheme based on a genetic algorithm. This scheme leverages reasonable channel prior information and efficiently searches for the optimal pilot positions and quantity in the DD domain through a genetic optimization process, aiming to minimize the channel estimation mean square error. In this process, a multi-objective fusion fitness function is designed and optimized jointly with the pilot count to balance channel estimation performance against pilot overhead; prior-fused initialization and selection operations are employed to narrow the search space and improve computational efficiency; an adaptive adjustment mechanism for crossover and mutation probabilities is introduced to balance global exploration and local exploitation; and elite retention coupled with local fine-tuning is applied to accelerate convergence and avoid premature stagnation. Simulation results demonstrate that the proposed scheme improves channel estimation performance compared to traditional pilot distribution methods, without significantly increasing computational time.
{"title":"Pilot design optimization of OTFS system based on genetic algorithm","authors":"Jiai He, Tianxing Wang","doi":"10.1016/j.phycom.2026.103034","DOIUrl":"10.1016/j.phycom.2026.103034","url":null,"abstract":"<div><div>The Orthogonal Time Frequency Space (OTFS) system has attracted significant research attention due to its superior performance in handling time-frequency doubly dispersive channels in high-mobility scenarios. As key reference signals for channel estimation, the design of pilots in the delay-Doppler (DD) domain crucially influences the final estimation accuracy. To address the issues of limited estimation accuracy and poor adaptability in existing pilot distribution schemes, this paper proposes a pilot distribution scheme based on a genetic algorithm. This scheme leverages reasonable channel prior information and efficiently searches for the optimal pilot positions and quantity in the DD domain through a genetic optimization process, aiming to minimize the channel estimation mean square error. In this process, a multi-objective fusion fitness function is designed and optimized jointly with the pilot count to balance channel estimation performance against pilot overhead; prior-fused initialization and selection operations are employed to narrow the search space and improve computational efficiency; an adaptive adjustment mechanism for crossover and mutation probabilities is introduced to balance global exploration and local exploitation; and elite retention coupled with local fine-tuning is applied to accelerate convergence and avoid premature stagnation. Simulation results demonstrate that the proposed scheme improves channel estimation performance compared to traditional pilot distribution methods, without significantly increasing computational time.</div></div>","PeriodicalId":48707,"journal":{"name":"Physical Communication","volume":"75 ","pages":"Article 103034"},"PeriodicalIF":2.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147397928","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号