A key challenge in multi-agent collaborative tasks is reducing uncertainty about teammates to enhance cooperative performance. Explicit communication methods can reduce uncertainty about teammates, but the associated high communication costs limit their practicality. Alternatively, implicit consensus learning can promote cooperation without incurring communication costs. However, its performance declines significantly when local observations are severely limited. This paper introduces a novel multi-agent learning framework that combines the strengths of these methods. In our framework, agents generate a consensus about the group based on their local observations and then use both the consensus and local observations to produce messages. Since the consensus provides a certain level of global guidance, communication can be disabled when not essential, thereby reducing overhead. Meanwhile, communication can provide supplementary information to the consensus when necessary. Experimental results demonstrate that our algorithm significantly reduces inter-agent communication overhead while ensuring efficient collaboration.

Generalizing multi-agent reinforcement learning (MARL) to accommodate variations in problem configurations remains a critical challenge in real-world applications, where even subtle differences in task setups can cause pre-trained policies to fail. To address this, we propose Context-Aware Identity Generation (CAID), a novel framework to enhance MARL performance under the Contextual MARL (CMARL) setting. CAID dynamically generates unique agent identities through the agent identity decoder built on a causal Transformer architecture. These identities provide contextualized representations that align corresponding agents across similar problem variants, facilitating policy reuse and improving sample efficiency. Furthermore, the action regulator in CAID incorporates these agent identities into the action-value space, enabling seamless adaptation to varying contexts. Extensive experiments on CMARL benchmarks demonstrate that CAID significantly outperforms existing approaches by enhancing both sample efficiency and generalization across diverse context variants.

Parameter sharing, as an important technique in multi-agent systems, can effectively solve the scalability issue in large-scale agent problems. However, the effectiveness of parameter sharing largely depends on the environment setting. When agents have different identities or tasks, naive parameter sharing makes it difficult to generate sufficiently differentiated strategies for agents. Inspired by research pertaining to the brain in biology, we propose a novel parameter sharing method. It maps each type of agent to different regions within a shared network based on their identity, resulting in distinct subnetworks. Therefore, our method can increase the diversity of strategies among different agents without introducing additional training parameters. Through experiments conducted in multiple environments, our method has shown better performance than other parameter sharing methods.

Extremely large antenna array (ELAA) is able to significantly improve the spectral efficiency, so it is regarded as one of the most important technologies for the next-generation networks. However, larger antenna aperture and higher frequency make the Rayleigh distances dramatically increased, so in sixth-generation (6G) networks, more and more communication scenarios will take place in the near-field region. Different from traditional far-field communication, electromagnetic waves transmitted in the near-field communication is widely considered to be spherical rather than planar, so techniques designed for far-field scenarios are no longer applicable. In this paper, we study the near-field communication scenario in which the base station is equipped with a cylindrical antenna array (CLA). Specifically, by exploiting the geometrical relationship between CLA elements and the near-field user, codebook design for beamforming in near-field CLA systems is studied for the first time. We first derive the beamforming gain in the elevation angle domain, the azimuth angle domain and the distance domain respectively. Then a 3-D near-field CLA codebook is proposed to make beamforming more effectively in near-field CLA systems. We obtain the sampling method in each domain by controlling the correlation between different codewords. Simulation results prove the effectiveness of the proposed codebook.

Asynchronous action coordination presents a pervasive challenge in Multi-Agent Systems (MAS), which can be represented as a Stackelberg game (SG). However, the scalability of existing Multi-Agent Reinforcement Learning (MARL) methods based on SG is severely restricted by network architectures or environmental settings. To address this issue, we propose the Stackelberg Decision Transformer (STEER). It efficiently manages decision-making processes by incorporating the hierarchical decision structure of SG, the modeling capability of autoregressive sequence models, and the exploratory learning methodology of MARL. Our approach exhibits broad applicability across diverse task types and environmental configurations in MAS. Experimental results demonstrate both the convergence of our method towards Stackelberg equilibrium strategies and its superiority over strong baselines in complex scenarios.

Almost all multi-agent reinforcement learning algorithms without communication follow the principle of centralized training with decentralized execution. During the centralized training, agents can be guided by the same signals, such as the global state. However, agents lack the shared signal and choose actions given local observations during execution. Inspired by viewpoint invariance and contrastive learning, we propose consensus learning for cooperative multi-agent reinforcement learning in this study. Although based on local observations, different agents can infer the same consensus in discrete spaces without communication. We feed the inferred one-hot consensus to the network of agents as an explicit input in a decentralized way, thereby fostering their cooperative spirit. With minor model modifications, our suggested framework can be extended to a variety of multi-agent reinforcement learning algorithms. Moreover, we carry out these variants on some fully cooperative tasks and get convincing results.

Recently, some challenging tasks in multi-agent systems have been solved by some hierarchical reinforcement learning methods. Inspired by the intra-level and inter-level coordination in the human nervous system, we propose a novel value decomposition framework HAVEN based on hierarchical reinforcement learning for fully cooperative multi-agent problems. To address the instability arising from the concurrent optimization of policies between various levels and agents, we introduce the dual coordination mechanism of inter-level and inter-agent strategies by designing reward functions in a two-level hierarchy. HAVEN does not require domain knowledge and pre-training, and can be applied to any value decomposition variant. Our method achieves desirable results on different decentralized partially observable Markov decision process domains and outperforms other popular multi-agent hierarchical reinforcement learning algorithms.

In multi-agent reinforcement learning (MARL), self-interested agents attempt to establish equilibrium and achieve coordination depending on game structure. However, existing MARL approaches are mostly bound by the simultaneous actions of all agents in the Markov game (MG) framework, and few works consider the formation of equilibrium strategies via asynchronous action coordination. In view of the advantages of Stackelberg equilibrium (SE) over Nash equilibrium, we construct a spatio-temporal sequential decision-making structure derived from the MG and propose an N-level policy model based on a conditional hypernetwork shared by all agents. This approach allows for asymmetric training with symmetric execution, with each agent responding optimally conditioned on the decisions made by superior agents. Agents can learn heterogeneous SE policies while still maintaining parameter sharing, which leads to reduced cost for learning and storage and enhanced scalability as the number of agents increases. Experiments demonstrate that our method effectively converges to the SE policies in repeated matrix game scenarios, and performs admirably in immensely complex settings including cooperative tasks and mixed tasks.

Spatial information is essential in various fields. How to explicitly model according to the spatial location of agents is also very important for the multi-agent problem, especially when the number of agents is changing and the scale is enormous. Inspired by the point cloud task in computer vision, we propose a spatial information extraction structure for multi-agent reinforcement learning in this paper. Agents can effectively share the neighborhood and global information through a spatially encoder-decoder structure. Our method follows the centralized training with decentralized execution (CTDE) paradigm. In addition, our structure can be applied to various existing mainstream reinforcement learning algorithms with minor modifications and can deal with the problem with a variable number of agents. The experiments in several multi-agent scenarios show that the existing methods can get convincing results by adding our spatially explicit architecture.

Value decomposition methods have gained popularity in the field of cooperative multi-agent reinforcement learning. However, almost all existing methods follow the principle of Individual Global Max (IGM) or its variants, which limits their problem-solving capabilities. To address this, we propose a dual self-awareness value decomposition framework, inspired by the notion of dual self-awareness in psychology, that entirely rejects the IGM premise. Each agent consists of an ego policy for action selection and an alter ego value function to solve the credit assignment problem. The value function factorization can ignore the IGM assumption by utilizing an explicit search procedure. On the basis of the above, we also suggest a novel anti-ego exploration mechanism to avoid the algorithm becoming stuck in a local optimum. As the first fully IGM-free value decomposition method, our proposed framework achieves desirable performance in various cooperative tasks.

Unsupervised dimension reduction and clustering are frequently used as two separate steps to conduct clustering tasks in subspace. However, the two-step clustering methods may not necessarily reflect the cluster structure in the subspace. In addition, the existing subspace clustering methods do not consider the relationship between the low-dimensional representation and local structure in the input space. To address the above issues, we propose a robust discriminant subspace (RDS) clustering model with adaptive local structure embedding. Specifically, unlike the existing methods which incorporate dimension reduction and clustering via regularizer, thereby introducing extra parameters, RDS first integrates them into a unified matrix factorization (MF) model through theoretical proof. Furthermore, a similarity graph is constructed to learn the local structure. A constraint is imposed on the graph to guarantee that it has the same connected components with low-dimensional representation. In this spirit, the similarity graph serves as a tradeoff that adaptively balances the learning process between the low-dimensional space and the original space. Finally, RDS adopts the$\ell _{2,1}$-norm to measure the residual error, which enhances the robustness to noise. Using the property of the$\ell _{2,1}$-norm, RDS can be optimized efficiently without introducing more penalty terms. Experimental results on real-world benchmark datasets show that RDS can provide more interpretable clustering results and also outperform other state-of-the-art alternatives.

Recently, model-based agents have achieved better performance than model-free ones using the same computational budget and training time in single-agent environments. However, due to the complexity of multi-agent systems, it is tough to learn the model of the environment. The significant compounding error may hinder the learning process when model-based methods are applied to multi-agent tasks. This paper proposes an implicit model-based multi-agent reinforcement learning method based on value decomposition methods. Under this method, agents can interact with the learned virtual environment and evaluate the current state value according to imagined future states in the latent space, making agents have the foresight. Our approach can be applied to any multi-agent value decomposition method. The experimental results show that our method improves the sample efficiency in different partially observable Markov decision process domains.

We propose a learned factor graph based on convolutional neural network (CNN) and Bi-directional Long Short Term Memory (BiLSTM) to realize signal detection under the scenario of no channel model. It can solve the inevitable over-reliance on channel state information (CSI) of model-based signal detection methods and avoid the shortcomings of large training scale of general data-driven methods by using relatively small training samples. The proposed method uses a network of CNN-BiLSTM structure with strong learning capabilities to determine the statistical relationship of the channel model which is what traditional model-based methods rely on. Based on above, the parameter estimation (Gaussian mixture model considering Akaike information criterion) and non-parametric estimation (adaptive kernel density) are adopted to learn a factor node together. The simulations show that, the proposed method can guarantee the accuracy of signal detection and robustness to the training of imperfect CSI.

In order to reduce the consumption cost for successive interference cancellation in non-orthogonal multiple access(NOMA), we propose a resource allocation scheme that involves both orthogonal multiple access and NOMA technologies. The scheme uses deep learning to choose the appropriate access according to the communication environment. Moreover, the scheme jointly allocates subcarrier and power resources for users by utilizing a deep Q network and a multi-agent deep deterministic policy gradient network. Meanwhile, an adaptive mechanism combining online learning and offline learning is introduced into allocation scheme to flexibly adapt to the communication environment. Results show that the proposed scheme can achieve better system performance in sum-rate. In order to better cope with changes in the environment and make the resource allocation strategy more robust, we propose a novel resource allocation algorithm combining transfer learning and deep reinforcement learning. The algorithm can effectively improve the model convergence speed when changing the communication environment. Furthermore, the algorithm allows us to transfer the subcarrier allocation network and the power allocation network simultaneously or separately depending on the environment.

Abstract The accurate and robust automatic segmentation of cardiac structures in magnetic resonance imaging (MRI) is significant in calculating cardiac clinical functional indices, and diagnosing heart diseases. Most U‐Net based methods use pooling, transposed convolution, and skip connection operations to integrate the multiscale features for improved segmentation in cardiac MRI. However, this architecture lacks adequate semantic connection between the channel and spatial information, and robustness in segmenting objects with significant shape variations. In this paper, a new multiscale feature attentive U‐Net for cardiac MRI structural segmentation method is proposed. An attention mechanism is adopted after concatenating the multi‐level features to aggregate different scale features and determine on which features to focus. Cascade and parallel dilated convolution is also employed in the decoder blocks and skip connection is employed to enhance the ability of sensing receptive fields for multiscale context information. Furthermore, deep supervision approach with a loss function that combines the dice and cross‐entropy losses to reduce overfitting and ensure better prediction is introduced. The proposed method was evaluated on three public cardiac datasets. The experimental results indicate that the method achieved competitive segmentation performance with the three datasets, which verifies the robustness and generalisability of the proposed network. In comparison with conventional U‐Net methods, the model leverages attention mechanism and dilated convolution block, which increases the semantic connection between the channel and the spatial information, and improves the robustness of the right ventricle segmentation performance. From the view of the Dice scores and segmentation results, the multiscale feature attentive U‐Net method is one of effective methods in segmenting cardiac MRI structures.

In the real world, many tasks require multiple agents to cooperate with each other under the condition of local observations. To solve such problems, many multi-agent reinforcement learning methods based on Centralized Training with Decentralized Execution have been proposed. One representative class of work is value decomposition, which decomposes the global joint Q-value Qjtinto individual Q-values Qato guide individuals' behaviors, e.g. VDN (Value-Decomposition Networks) and QMIX. However, these baselines often ignore the randomness in the situation. We propose MMD-MIX, a method that combines distributional reinforcement learning and value decomposition to alleviate the above weaknesses. Besides, to improve data sampling efficiency, we were inspired by REM (Random Ensemble Mixture) which is a robust RL algorithm to explicitly introduce randomness into the MMD-MIX. The experiments demonstrate that MMD-MIX outperforms prior baselines in the StarCraft Multi-Agent Challenge (SMAC) environment.

Abstract This paper proposes a data‐driven method of mmWave beam selection in multi‐cell systems to achieve a near‐optimal fast beam allocation with low complexity. In particular, an online learning algorithm based on support vector machine (SVM) equipped with the radial basis function kernel, namely SVM‐based online beam selection (SBOS) algorithm is proposed. The proposed algorithm starts with an adaptive beam selection process for certain traffic pattern that uses an SVM learning model to adaptively refine the beam selection strategy. Specifically, SVM‐based model labels the feedback (the average information rate) from the cellular system, then learns from samples, and makes the scheme space smaller by maximising samples' minimum distances to all labelled samples in the sample space constrained by newly learned boundaries. Then, according to the aggregated data about the traffic patterns and the performance of corresponding beam selection strategy, SBOS algorithm exploits beam selection schemes recorded in the database or explores new schemes for unknown situations, respectively, and how to tune the hyperparameters for the SBOS algorithm is discussed. Furthermore, the extensive simulation results show that the proposed algorithm achieves a better performance versus upper confidence bound and Random methods.

Millimeter-wave (mmWave) bands are expected to be an important choice for future vehicular communication to support Gbps links for reliable data transfer in high-rate applications. The recent online learning technologies addressed the problem of fast beam tracking by exploiting user location information and mining received data in mmWave vehicular systems to adapt to the vehicle's environmental situation. However, the fairness and efficiency over mmWave beams are difficult to maintain on the move, especially for high-density traffic, since the number of available beams is quite limited by hardware and cost for current antenna arrays. Fortunately, the social structure of preferences between the neighboring smart cars and their passengers can be leveraged to improve the beam coverage efficiency by performing the broadcast transmission via a single beam. In this article, we propose a double-layer online learning algorithm, namely, context- and social-aware machine learning (CSML), that is based on the context and social preference information of vehicles and passengers, to realize fast beam access with broadcast coverage in mmWave communication systems. Based on the multiarmed bandit model, CSML embodies the selection of appropriate beams in the first layer and steers the broadcast angle along these beams in the second layer by aggregating the received data. Furthermore, CSML needs to adjust the timing of exploration and exploitation based on the social information, i.e., the probability of vehicles meeting with each other that have the same preference. Finally, we perform an extensive evaluation using realistic traffic patterns and show that CSML increases the efficiency of mmWave base stations by using social data and can achieve near-optimal system performance.

In this article, we study the edge caching problem by considering the heterogeneous context with unknown users' preferences. The cache provider (CP) can personalize the users' storage based on available data to maximize the overall cache hit rate, accounting for the dynamic natures of both mobile edge cache scenarios and the users' preferences. Toward this end, we introduce an online Bayesian clustering caching algorithm for the CP to autonomously learn the users' interactive cache hit data in a collaborative way while maintaining sustainable scalability. Specifically, a Bayesian generative framework called the Dirichlet multinomial mixture (DMM) model is used to describe the uncertainty about the latent number of users' clusters, each of which consists of the users with the same preference. Then, a dynamic clustering policy is proposed to obtain both the underlying mapping of users to clusters and the preferences of each cluster by using a collapsed Gibbs sampling algorithm. Subsequently, cache decisions are made according to the generated mappings by extending the traditional cache bandit algorithm to a new bandit mechanism with clusters of arms, capable of expediting the learning process between the exploitation and exploration. We theoretically characterize the value of dynamic Bayesian clustering for the long-term edge caching scenario with respect to the regret incurred by the noncluster schemes. Finally, using a real-world data set, our numerical results show that the proposed scheme outperforms the caching algorithms without clustering in the uncertain network scenario.

With the certainty of the high-speed railway(HSR) route, high-speed train(HST) is always driving periodically, and it is quite meaningful to assist HSR wireless signal detection through historical big data. One key challenge in this detection is that the HSR wireless channel is varying when the HST drives to different places, thus the data under various channel environment needs to be analyzed separately. In this paper, we propose a deep learning algorithm to detect the multiple input multiple output(MIMO) signal for HSR scenarios, and the entire algorithm framework is divided into two phases: offline training phase and online detection phase. At the offline training phase, we first analyze the data of HSR at each location, and explore a division scheme to further divide each scene into multiple smaller regions so that data in each divided region can share the same network. Then, the deep neural network(DNN) is constructed and trained for each divided region. At the online detection phase, the HST locates the current region according to the location information achieved by GPS, and selects the corresponding DNN model to detect the signal in real time. In addition, this DNN structure combines channel estimation and signal detection. Thus, the HSR detection system can detect the MIMO signal directly without the step of channel estimation. Finally, the simulation results show that the deep learning detection algorithm has better accuracy than those traditional detection algorithms, such as the least square(LS) algorithm and the minimum mean-square error(MMSE) algorithm.

In this paper, a renewable energy powered millimeter wave (mmW) backhaul network is studied. In the considered model, the wireless operator must request renewable energy from multiple renewable power suppliers (RPSs) to serve the end mobile users using the mmW backhaul. The unit price of renewable energy depends on the RPS's production capacity/lead time for the corresponding backhaul node. A lead time-dependent pricing scheme is proposed thus enabling the operator to manage the traffic latency over the backhaul and co-ordinate independent RPSs' decisions on the renewable energy storage levels with uncertain wireless traffic demand. Toward this end, the problem is formulated as a Stackelberg game between the operator and multiple RPSs. In this game, the operator first specifies a pricing scheme for RPSs and each RPS should then make its own decision in stocking the renewable energy. Then, efficient distributed algorithms are proposed to find the operator's optimal pricing scheme, and the RPSs' Pareto equilibrium storage strategies, respectively. Our results provide useful insights for understanding the tradeoff between the benefit of energy savings and the cost of quality-of-service (QoS) reducing for the operator. Also, simulation results show how renewable energy production capacities affect the revenue of individual RPSs in decentralized renewable-powered backhaul systems. The results also show that the proposed scheme can enable the operator to achieve more profit compared to a centralized solution.

In this paper, a green wireless communication system in which base stations are powered by renewable energy sources is considered. This system consists of a capacity-constrained renewable power supplier (RPS) and a base station (BS) that faces a predictable random connection demand from mobile user equipments (UEs). In this model, the BS, which is powered via a combination of a renewable power source and the conventional electric grid, seeks to specify the renewable power inventory policy, i.e., the power storage level. On the other hand, the RPS must strategically choose the energy amount that is supplied to the BS. An M/M/1 make-to-stock queuing model is proposed to investigate the decentralized decisions when the two parties optimize their individual costs in a noncooperative manner. The problem is formulated as a noncooperative game whose Nash equilibrium (NE) strategies are characterized to identify the causes of inefficiency in the decentralized operation. A set of simple linear contracts are introduced to coordinate the system so as to achieve an optimal system performance. The proposed approach is then extended to a setting with one monopolistic RPS and N BSs that are privately informed of their optimal energy inventory levels. In this scenario, we show that the widely used proportional allocation mechanism is no longer socially optimal. To make the BSs truthfully report their energy demand, an incentive compatible (IC) mechanism is proposed for our model. Simulation results show that using the green energy can present significant traditional energy savings for the BS when the connection demand is not heavy. Moreover, the proposed scheme provides valuable energy cost savings by allowing the BSs to smartly use a combination of renewable and traditional energy, even when the BS has a heavy traffic of connections. Also, the results show that performance of the proposed IC mechanism will be close to the social optimal when the green energy production capacity increases.

Relay stations are introduced into cellular systems to extend the coverage of the cell, improve the throughput and outage probability of the system. However, present resource-allocation algorithms for traditional cellular system are not directly applicable to the system with relay stations. In this paper, we propose an adaptive resource-allocation algorithm for orthogonal frequency-division multiple-access (OFDMA) cellular system in which relay stations participate in the channel allocation. This algorithm adjusts the frame structure in time domain adaptively according to the channel state and the real-time system throughput. Simulation results show that the proposed algorithm has higher throughput and lower outage probability.

In this paper, we exploit a novel setting for Cognitive Radio (CR) networks to enable multiple operators to involve secondary users (SUs) as cooperative relays for their primary users. In return, SUs get an opportunity to access spare channels for their own data transmission. Initially, we assume that the CR network supports payment transfer. Then, we formulate the system as a transferable utility coalitional game. We show that there is an operating point that maximizes the sum utility over all operators and SUs while providing each player a share such that no subset of operators and SUs has an incentive to break away from the grand coalition. Such operating points exist when the solution set of the game, the core, is nonempty. Subsequently, we examine an interesting scenario where there is no payment mechanism in the network. This scenario can be investigated by using a nontransferable utility coalitional game model. We show that there exists a joint action to make the core nonempty. A general method with exponential computational complexity to get such a joint action is discussed. Then, we relate the core of this game to a competitive equilibrium of an exchange economy setting under special situations. As a result, several available efficient centralized or distributed algorithms in economics can be employed to compute a member in the core. In a nutshell, this paper constitutes the design of new coalition based dynamics that could be used in future CR networks.

This paper studies the problem of determining locations of Secondary Access Points (SAPs) which belong to two competitive service providers in a certain region. SUs connect to the SAP according to their preferences. A key feature of our modeling approach is that it permits SAPs to set discriminatory powers for SUs, that is, SUs in different locations will be scheduled with different transmission powers. The profit maximizing SAPs compete with each other by setting locations and delivered power schedules to attract SUs, taking into account the impact of the revenue obtained and the power cost incurred. To study this competitive situation, an appropriate extensive form game is defined. The game can be viewed as two-staged. The selection of transmission powers by the SAPs can be investigated for each pair of locations. Then, the problem of location choice can be analyzed, anticipating what transmission powers will be chosen. We make an important observation: the Nash equilibrium of the game exists and the global revenue maximum strategy is a Nash equilibrium strategy. An example demonstrating how to place SAPs in a linear region is presented.

The traditional spectrum auctions require a central auctioneer. Then, the secondary users (SUs) can bid for spectrum in multiple auction or sealed auction way. In this paper, we address the problem of distributed spectrum sharing in the cognitive networks where multiple owners sell their spare bands to multiple SUs. Each SU equips multi-interface/multi-radio, so that SU can buy spare bands from multiple owners. On the other hand, each owner can sell its spare bands to several SUs. There are two questions to be addressed for such an environment: the first one is how to select bands/the owners for each SU; the second one is how to decide the competitive prices for the multiple owners and multiple SUs. We propose a two-sided multi-band market game theoretic framework to jointly consider the benefits of all SUs and owners. The equilibrium concept in such games named core. The outcomes in the core of the game cannot be improved upon by any subset of players. These outcomes correspond exactly to the price-lists that competitively balance the benefits of all SUs and owners. We show that the core in our model is always non-empty. The measurement of price is set to discrete value. Subsequently, the core of the game is defined as discrete core. The Dynamic Multi-band Sharing (DMS) Algorithm is proposed to converge to the discrete core of the game. With small measurement of price, the algorithm can achieve the optimal performance compared with centralized one in terms of total profit of the system.

In this paper we exploit a setting for cognitive radio networks by utilizing cooperation from secondary users (SUs) to assist the transmissions of operators' primary users (PUs). On the other hand, SUs can share the spare spectrum of operators. Such a scenario can be viewed as a market where multiple operators trade their spare spectrum for the assistance of SUs, and multiple SUs trade the transmission energy for access opportunities from operators. We model the system using transferable payoff coalitional game theory. An outcome of a coalitional game is a specification of the coalition that forms and the joint action it takes. We show that the optimum joint action strategy can be obtained as a solution of convex optimization problem. Then, based on dual technique, we show that there is an operating point that maximizes the sum utility over the operators and SUs while providing each player a share such that no subset of operators and SUs has an incentive to break away from the brand coalition.

In this paper, we study the relay assignment problem in cooperative wireless networks with self-interested nodes. Such systems should be organized from the point of view of efficiency, stable and providing consistent incentives to all nodes. We propose a cooperation mechanism which includes the cooperative relationship formation stage and cooperation maintenance stage. The cooperative relationship among the nodes can be modeled as an exchange market game (a special coalitional game) where nodes trade transmission power between each other to get diversity gain. The exchange games have a basic assumption that each node conforms to trade agreement. So that, each agent has the option to trade its good in order to get a better one. In such game, strict core is considered as individual rational, Pareto optimal and relationship-stable solution. A Cooperation Cycle Formation (CCF) algorithm is proposed to get the strict core solution. But, in networks, some deviated nodes may break the cooperation agreement to get more utility gain. Such deviated behaviors in the cooperation cycles can totally destroyed the cooperation relationship. However, the date transmissions in networks have the repeated element (e.g., the data of each user are transmitted in many time slots). Hence, based on the cooperative cycle formed by CCF, we introduce a repeated game model for cooperation maintenance in the second stage. A Dynamic Punishment and Recover (DPR) mechanism is proposed to punish the deviated behaviors and recover cooperation.

Peer-to-peer (P2P) networks are very popular for large-scale data sharing in today's internet. It is naturally envisioned that the file sharing would also be an important application for civil mobile ad hoc networks in the future. The authors propose the cooperative diversity download methods utilising the wireless medium and common file sources in wireless P2P networks. The designed protocols including multi-source opportunistic direct downloading (MODD) and multi-source opportunistic decode-and-forward downloading (MODFD) can adaptively work in different file distribution scenarios. Both of them rely on the cooperation between the sources or receivers and the proposed distributed source-selection schemes. For the situation with one receiver and multi-source, MODD selects ‘best’ source with most favourite channel condition from available m sources. In the situation with multiple receivers and sources, the multiple receivers could form the cooperative pairs to receive the data from the selected source based on decode-and-forward transmission strategy. Compared with MODD, MODFD saves the source-selection time and provides incentives for the cooperation between the peers. A distributed collision solving scheme for the selection schemes is also proposed. The information theoretic analysis of outage probability shows that the proposed downloading methods can both obtain m times diversity gain.

In this paper, we address the incentive-based relay-selection problem over multi-source and multi-relay wireless networks. A two-side market game approach is employed to jointly consider the benefits of all sources and relays. The equilibrium concept in such games is called core. The outcomes in the core of the game cannot be improved upon by any subset of players. These outcomes correspond exactly to the price-lists that competitively balance the benefits of all sources and relays. When the price assumes only discrete values, the core of the game is defined as discrete core. The Distributed Source-Relay Assignment (DSRA) algorithm is proposed for competitive price adjustment and converges to the discrete core of the game. With small enough measurement of price, the algorithm can achieve the optimal performance compared with centralized one in terms of total profit of the system.