In opportunistic cognitive radio networks, when the primary signal is very weak compared to the background noise, the secondary user requires a long sensing time to achieve reliable spectrum sensing, leaving little time for secondary transmission. To tackle this issue, we propose an active reconfigurable intelligent surface (RIS)-assisted spectrum sensing system, where the received signal strength from the target primary user can be enhanced and underlying interference within the background noise can be mitigated. In comparison with the passive RIS, the active RIS not only adjusts the phase shifts of the reflecting elements but also amplifies the incident signals. Notably, we study the optimization of the reflecting coefficient matrix (RCM) to improve the detection probability given a maximum tolerable false alarm probability and limited sensing time. Then, we show that the formulated problem can be equivalently transformed into a weighted mean square error minimization problem using the principle of the weighted minimum mean square error (WMMSE) algorithm, and an iterative optimization approach is proposed. In addition, to fairly compare passive RIS and active RIS, we study the required power budget of the RIS to achieve a target detection probability under a special case where the direct links are negligible and the RIS-related channels are line-of-sight. The conclusions drawn from this special case are further validated through simulations under more general channel conditions. Furthermore, the effectiveness of the WMMSE-based RCM optimization approach is demonstrated via extensive simulations. The results also reveal that the active RIS can outperform the passive RIS when the interference is relatively weak, whereas the passive RIS performs better in strong interference scenarios due to its ability to support a large number of reflecting elements under the same power budget.

The space-terrestrial integrated network (STIN) is crucial for achieving ubiquitous connectivity in the 6G era. However, leveraging full potential of STIN is challenging due to the distinct characteristics and objectives of constituent networks. Inspired by symbiotic communication (SC), this paper proposes a service exchange-based symbiotic STIN system that optimizes objectives of different networks by exploiting their complementary features. Specifically, the ground network provides task offloading services to the space network, while the space network reciprocates with communication services. To minimize computation delay in the space network and maximize the energy efficiency (EE) of the ground network, we formulate a multi-objective optimization problem (MOOP) that jointly optimizes task offloading, resource allocation, and beamforming. We first transform the MOOP into a single-objective optimization problem (SOOP) via the ε-constraint method and then develop a successive convex approximation (SCA) algorithm to characterize its fundamental performance, which requires future state information. As obtaining such non-causal information is hard, we design a more practical multi-agent reinforcement learning (MARL) algorithm based on insights from the SCA. Besides, to address the challenges of storing multiple MARL policies for different EE-delay trade-offs, we develop a diffusion model-based behavior cloning (BC) algorithm to obtain a general policy suitable for varying trade-offs. Simulation results show that proposed algorithms outperform benchmarks and confirm that the proposed service exchange realizes a symbiotic STIN.

In this paper, we propose a novel integrated sensing and backscatter communication (ISABC) system in which each moving target is attached with a backscatter device (BD) to facilitate simultaneous target identification and parameter estimation. When the base station (BS) transmits signals to its desired user, each BD attached to the target transmits the target identification information to the BS via backscatter communication, which concurrently enhances the echo signal strength. The BS needs to detect the BD symbols and to estimate the target parameters using the echoes. This task, however, is challenging due to the coupling between the BD symbols and the target parameters. To address this issue, we propose a novel iterative detection and estimation (IDE) framework, which involves the following two processes alternately: 1) Utilizing a modified maximum likelihood (ML) estimator to perform parameter estimation with the detected BD symbols; 2) Employing the ML detector to detect the BD symbols with the estimated target parameters. Since the inter-carrier interference (ICI) of the OFDM signal is independent of the BD symbols, but contains the delay and Doppler shift information, we develop a target parameter initialization method using such ICI component to improve the performance of the proposed IDE scheme. Moreover, the closed-form Miller-Chang bound is derived to demonstrate the theoretical performance for the target parameter estimation. Finally, simulation results are provided to validate the effectiveness of the proposed designs.

Cognitive aerial-terrestrial networks (CATN) hold promise in addressing the spectrum shortage challenges posed by thriving aerial networks, where aerial users (AUs) requiring high-quality downlink communications suffer severe interference from numerous terrestrial base stations (BSs). To alleviate such interference, we propose jointly optimizing the user association and coordinated beamforming (CBF) of the terrestrial network, thereby maximizing the sum rate of the secondary terrestrial users (TUs) under the interference temperature constraints of the primary AUs. Traditional iterative optimization schemes are impractical for this problem due to their high computational complexity and information exchange overhead. Although deep reinforcement learning (DRL)-based schemes offer a viable alternative, their performance is sensitive to the weight of the constraint violation penalty in the reward. To overcome these limitations, we propose a safe DRL-based user association and CBF scheme for CATN, which avoids multiple training attempts to find the optimal penalty weight before actual deployment and reduces expenses. Specifically, the studied system is modeled as a networked constrained partially observable Markov game, where each TU agent chooses its associated BS, and each BS agent decides its beamforming vectors, aiming to maximize the reward while satisfying the safety constraints to protect the AUs. Simulation results show that the proposed scheme can achieve a higher sum rate of TUs than a two-stage optimization scheme while the average received interference power of the AUs is generally below the threshold.

Pilot contamination is a tough problem for multi-cell massive multiple-input multiple-output (MIMO) circumstances. This paper proposes a naive Bayes based pilot allocation approach to reduce its negative impact. Specifically, a naive Bayes classifier is formulated by utilizing the large-scale fading factors and channel angle of arrival (AOA) intervals to obtain the pilot allocation relationship between any two terminals in different cells. Then a binary numerical model is adopted to describe this relationship and a pilot allocation method is constructed to mitigate the potential interference. Compared with the existing schemes, simulation results verify the superiority of the proposed approach in improving the system performance with finite number of base station (BS) antennas.

Unmanned Aerial Vehicles (UAVs) ad hoc network, also known as Flying Ad hoc Network (FANET), is expected to be a cornerstone in future-oriented civilian and military fields. To efficiently support the requirements of low-delay and massive connections in FANET, Medium Access Control (MAC) mechanisms employ several methods to promote information interaction between UAVs. Among these, multi-channel scheme is expected to be prominently appropriate for FANET. However, when there are massive nodes with higher channel access occasions, a drawback of this kind of mechanisms is that the data channel and the control channel are scheduled and regulated separately, which will inevitably lead to channel saturation and load imbalance. To cope with this situation, we propose a Preemption-based Multi-channel MAC (PM-MAC) scheme. This scheme leverages the multi-channel mechanism to harness the advantages of both CSMA and TDMA, while addressing the transmission requirements of delay-sensitive and delay-insensitive data. It reduces the transmission latency of urgent data while ensuring a certain level of throughput. Based on the scheme, a Markov model based on preemption is constructed. The performance results show that the proposed PM-MAC yields a substantial improvement in delay and throughput compared with benchmarks.

Multiple Access Control (MAC) mechanism is imperative for Vehicular Ad-hoc Network (VANET) communications, which directly impacts system performance. IEEE 1609.4 protocol is dedicated to address the issue of MAC in VANETs. The protocol divides channels into a Control Channel (CCH) and Service Channels (SCHs), where nodes compete in CCH to gain access to SCHs to transmit data. However, this mechanism fails to guarantee Quality of Service (QoS) performance of heterogeneous services. To cope with this situation, this paper proposes a novel Priority-Aware Hybrid MAC protocol (PAH-MAC) for VANETs, enhancing QoS by introducing differentiated transmission schemes for Vehicular to Vehicular (V2V) and Vehicular to Infrastructure (V2I) communications. For V2V links, emergency data is broadcast directly in CCH to earn the highest channel access priority. For V2I links, service data is further classified into two priority levels. Safety-critical service packets employ a preemption-based TDMA scheme to ensure the opportunity of transmission in SCHs higher than regular service packets. And V2I regular service packets follow a TDMA-based reservation mechanism but will defer transmission if preempted by emergency service packets. This design optimizes spectrum efficiency while guaranteeing QoS for critical V2I services. Furthermore, we develop an analytical framework based on an M/D/1 tandem M/M/4 queueing model and a deferred reservation model, followed by theoretical and simulation-based evaluations. The simulation results confirm the effectiveness of the proposed scheme in terms of throughput and control overhead compared with typical IEEE 1609.4 under high mobility of VANET.

Cognitive radio (CR) has been identified as a highly promising spectrum-sharing technology for enhancing the spectrum efficiency of future wireless networks. Recently, the emerging reconfigurable intelligent surface (RIS) has been integrated into CR networks to further advance spectrum efficiency. However, existing works primarily focus on conventional RIS architectures, which can only adjust the reflection of incident wireless signals on one side of the RIS. As a result, these architectures cannot tune the wireless signals to improve performance across the full 360∘ coverage. Additionally, most existing studies concentrate on enhancing the transmission of the secondary user (SU), with the potential benefits to the transmission of the primary user (PU) remaining unclear. In this paper, we investigate a simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS)-empowered opportunistic CR network, where the STAR-RIS can tune the wireless signals on both sides simultaneously. To unleash the potential of this system, we study the reflection and transmission coefficients (RTCs) optimization problem to maximize the SU’s average achievable rate while ensuring the target sensing performance and meeting the PU’s rate requirement. We then introduce a two-stage RTC optimization framework, which decouples the optimization process into sensing and transmission stages. Specifically, the RTCs in the sensing stage are optimized to improve the PU’s transmission while achieving the desired detection and false alarm probabilities. In the transmission stage, the RTCs are optimized to maximize the SU’s average achievable rate while ensuring that the PU’s rate requirements are satisfied. Furthermore, we propose a single-stage optimization method to reduce the implementation complexity caused by frequent RTC reconfigurations. Extensive simulations demonstrate the performance improvements enabled by STAR-RIS and validate the effectiveness of the proposed RTC optimization methods.

In space-air-ground integrated networks (SAGINs), cognitive spectrum sharing has been regarded as a promising solution to meet the rapidly increasing spectrum demand of various applications, because it can significantly improve the spectrum efficiency by enabling a secondary network to access the spectrum of a primary network. However, different networks in SAGIN may have different quality of service (QoS) requirements, which can not be well satisfied with the traditional cognitive spectrum sharing architecture. To address this issue, in this paper, we propose a hierarchical cognitive spectrum sharing architecture (HCSSA) for SAGINs, where the secondary networks are divided into a preferential one and an ordinary one. Specifically, the aerial and terrestrial networks can access the spectrum of the satellite network under the condition that the caused interference to the satellite terminal is below a certain threshold. Besides, considering that the aerial network has a higher priority than the terrestrial network, we aim to use a rate constraint to ensure the performance of the aerial network. Subject to these two constraints, we consider a sum-rate maximization for the terrestrial network by jointly optimizing the transmit beamforming vectors of the aerial and terrestrial base stations. To solve this non-convex problem, we propose a penalty-based iterative beamforming (PIBF) scheme that uses the penalty method and the successive convex approximation technique. Moreover, we also develop three low-complexity schemes, where the beamforming vectors are obtained by optimizing the normalized beamforming vectors and power control. In addition, we consider the case where only statistical channel state information is available and the case where channel estimation errors exist, and propose the corresponding beamforming schemes. Finally, we provide extensive numerical simulations to evaluate the performance of the proposed beamforming schemes and demonstrate the advantages of the proposed HCSSA compared with the traditional cognitive spectrum sharing architecture.

Cognitive aerial-terrestrial networks (CATNs), which enable aerial and terrestrial networks to share spectrum resources, have been identified as a promising solution to address the spectrum utilization challenges brought by thriving aerial networks. However, during spectrum sharing, aerial users, such as airplanes and flying cars, suffer severe interference from numerous terrestrial base stations (BSs). To alleviate such inter-ference, in this paper, we investigate a coordinated beamforming (CBF) problem, which aims to maximize the sum rate of the secondary terrestrial users while keeping the interference to the primary aerial users below a pre-defined threshold. Since aerial users usually move fast, obtaining real-time global channel state information is challenging. Also, frequently calculating the beamformers with high-complexity algorithms is unaffordable. These issues make traditional iterative optimization algorithms impractical in solving this problem. To address these issues, deep reinforcement learning (DRL) based algorithms can be applied. However, the performance of DRL-based algorithms is sensitive to the penalty weight of a weighted penalty term for violating constraints in the reward function. In this paper, we propose a safe DRL-based distributed dynamic CBF scheme to avoid the tricky adjustment of the penalty weight, where the studied CATN is described as a constrained Markov decision process (CMDP) with cost sets to decouple the objective and constraints. The CMDP is solved by a safe DRL algorithm, which maximizes the reward while satisfying the safety constraints. Simulation results show that the proposed algorithm can achieve a high sum rate of terrestrial users while interference power constraints are generally well satisfied.

Cognitive spectrum sharing has been regarded as a promising solution to improve spectrum utilization efficiency for space-air-ground integrated networks (SAGINs). However, in SAGIN, different networks may have different quality of service (QoS) requirements, which pose challenges to the traditional cognitive spectrum sharing architecture. For example, the aerial network typically has high QoS requirements, which may not be met when it acts as a secondary network. To address this issue, we propose a hierarchical cognitive spectrum sharing architecture (HCSSA) for SAGIN, where the secondary networks are divided into a preferential one and an ordinary one. Specifically, in SAGIN, an aerial network and a terrestrial network share the spectrum of a satellite network. HCSSA gives higher priority to the aerial network by a QoS constraint, while the terrestrial network is the ordinary secondary network without protection. Besides, the satellite terminal requires the received interference to be below a threshold. Subject to these two constraints and the maximum transmit power constraints, we aim to maximize the sum rate of the terrestrial network by optimizing the transmit beamforming vectors of the aerial base station (BS) and the terrestrial BSs. To solve this non-convex problem, we propose an iterative beamforming scheme by exploiting the penalty method and the successive convex approximation scheme. Simulation results show the performance of the proposed beamforming scheme and illustrate the advantages of HCSSA compared with the traditional cognitive spectrum sharing architecture.

The space-terrestrial integrated network (STIN) is a pivotal architecture to support ubiquitous connectivity in the upcoming 6G era. Inter-operator resource and service sharing is a promising way to realize such a huge network, utilizing resources efficiently and reducing construction costs. Given the rationality of operators, the configuration of resources and services in STIN should focus on both the overall system performance and individual benefits of operators. Motivated by emerging symbiotic communication facilitating mutual benefits across different radio systems, we investigate the resource and service sharing in STIN from a symbiotic communication perspective in this paper. In particular, we consider a STIN consisting of a ground network operator (GNO) and a satellite network operator (SNO). Specifically, we aim to maximize the weighted sum rate (WSR) of the whole STIN by jointly optimizing the user association, resource allocation, and beamforming. Besides, we introduce a sharing coefficient to characterize the revenue of operators. Operators may suffer revenue loss when only focusing on maximizing the WSR. In pursuit of mutual benefits, we propose a mutual benefit constraint (MBC) to ensure that each operator obtains revenue gains. Then, we develop a centralized algorithm based on the successive convex approximation (SCA) method. Considering that the centralized algorithm is difficult to implement, we propose a distributed algorithm based on Lagrangian dual decomposition and the consensus alternating direction method of multipliers (ADMM). Finally, we provide extensive numerical simulations to demonstrate the effectiveness of the two proposed algorithms, and the distributed optimization algorithm can approach the performance of the centralized one. The results also reveal that the proposed MBCs can enable operators to achieve mutual benefits and realize a symbiotic resource and service sharing paradigm.

Cooperative spectrum sensing (CSS) is a key enabling technology of cognitive radio networks with multiple secondary users (SUs). In conventional CSS systems, when the primary signals are weak, the SUs require long sensing time to achieve a high detection probability for protecting the transmission of the primary user (PU), leading to little remaining time for secondary transmissions. To address this issue, we propose a reconfigurable intelligent surface (RIS) assisted CSS system, where multiple RISs are employed to improve the CSS performance within limited sensing time. Considering that the dependency of the CSS performance on the received primary signal strengths at the SUs differs across various CSS schemes, the RIS configurations could also be optimized differently regarding these CSS schemes. Motivated by this, we investigate the phase shift matrix (PSM) optimization problems to maximize the cooperative detection probability given a maximum tolerable false alarm probability, and we consider two typical kinds of CSS schemes, namely, data fusion and decision fusion. As it is intractable to directly solve these problems due to the complex expressions of the cooperative detection probability with respect to the PSMs, we show that the solutions can be obtained by transforming these problems into channel gain-related optimization problems. Furthermore, we show that the proposed PSM optimization methods can be extended to the more practical scenarios where instantaneous channel state information (CSI) is unavailable. In such cases, we leverage statistical CSI to improve the CSS performance in the sense of expectation. Subsequently, we conduct a numerical analysis on the number of reflecting elements required to achieve a target detection probability in the statistical CSI case. Finally, simulation results demonstrate that the proposed PSM optimization methods can significantly improve the CSS performance within limited sensing time.

Massive multiple-input multiple-output (MIMO) is a key enabling technology for next-generation communication systems. In massive MIMO cellular networks, coordinated beamforming (CBF), which jointly designs the beamformers of multiple base stations (BSs), is an efficient method to enhance the network performance. In this paper, we investigate the sum rate maximization problem in a massive MIMO mobile cellular network, where in each cell a multi-antenna BS serves multiple mobile users simultaneously via downlink beamforming. Although existing optimization-based CBF algorithms can provide near-optimal solutions, they require real-time and global channel state information (CSI), in addition to their high computation complexity. Due to the non-negligible delay of practical backhaul networks and the high-complexity optimization process, it is almost impossible to apply them in mobile cellular networks. Noting that the considered problem under the practical constraints can be modeled as a networked distributed partially observable Markov decision process, we propose a deep reinforcement learning-based distributed dynamic coordinated beamforming (DDCBF) scheme, which enables each BS to determine the beamformers with only local CSI and some historical information from other BSs. Besides, the beamformers can be calculated with a considerably lower computational complexity by exploiting neural networks and expert knowledge, i.e., a solution structure observed from the iterative procedure of the centralized optimization algorithms. Moreover, we provide extensive numerical simulations to validate the effectiveness of the proposed DRL-based approach. With lower computational complexity and less required information, the results show that the proposed approach can achieve comparable performance to the centralized iterative optimization algorithms.

Spectrum and service sharing among multiple mobile network operators (MNOs) is regarded as a promising technology to construct future mobile networks, because it provides a higher utilization efficiency of the network resources, e.g., spectrum and network infrastructure. However, each individual MNO may not be able to benefit from this technology without joint optimizations. In this paper, we investigate a symbiotic spectrum and service sharing paradigm, which can improve the overall performance of the multi-MNO network while guaranteeing each MNO benefits as well. Particularly, we formulate two user association problems with novel mutual benefit constraints, namely, sum rate maximization and load balancing. Besides, we introduce a sharing coefficient for inter-MNO service sharing, which can be regarded as an inter-MNO service level agreement (SLA) accounting for the regulation of user association behaviors. Then, we propose a successive convex approximation (SCA) based algorithm and a fractional programming (FP) based algorithm to solve the sum rate maximization problems for different inter-MNO sharing strategies. In addition, we also develop a Lagrangian dual decomposition based algorithm for the load balancing problems. Finally, extensive numerical simulations are provided to demonstrate the effectiveness of the proposed algorithms. The results show that mutual benefit constraints can help multiple MNOs realize a symbiotic spectrum and service sharing paradigm. By comparing different inter-MNO sharing strategies, it can also be observed that the highest performance gain can be achieved when the multiple MNOs share their spectrum and service simultaneously.

In opportunistic cognitive radio networks, the secondary user (SU) requires long sensing time to achieve a reliable spectrum sensing performance when the primary signal is very weak, leading to little remaining time for the secondary transmission. To tackle this issue, we propose an active reconfigurable intelligent surface (RIS) assisted spectrum sensing system to enhance the received primary signal at the SU, therefore the required sensing time can be reduced. In contrast to the passive RIS, the active RIS can amplify the incident signal and hence is more efficient in terms of the required reflecting elements as well as power consumption. Particularly, we study the reflecting coefficient matrix (RCM) optimization problem to improve the performance of the active RIS assisted spectrum sensing system. With the knowledge of the spiked model from random matrix theory, the RCM optimization problem can be transformed to an equivalent problem maximizing the largest eigenvalue of the population covariance matrix of the sensing signal samples. Then, we adopt the weighted minimum mean square error (WMMSE) algorithm to obtain the optimal RCM. Besides, we also investigate the minimum power budget for the active RIS to realize a near-1 detection probability under a simplified case, where the direct link does not exist and line-of-sight RIS-related channels are considered. Simulation results show that the active RIS can outperform the passive RIS for the same power budget in the RIS-assisted spectrum sensing system.

To utilize network resources efficiently and reduce network construction costs, inter-operator spectrum and service sharing has been regarded as a promising technique for constructing future multi-operator networks. As the mobile operators are inherently competitors, one necessary prerequisite for inter-operator spectrum and service sharing is to guarantee each operator's revenue, namely, achieving mutual benefits among operators. Noting that the recently proposed symbiotic communication can achieve mutual benefits among different radio systems, in this paper, we investigate inter-operator spectrum and service sharing from a symbiotic communication perspective. Particularly, we propose a mutual benefit constraint to guarantee the revenue of each operator, and we aim to maximize the weighted sum rate (WSR) of the multi-operator network through joint optimization of the user association and beamforming design. Besides, considering the cost incurred by inter-operator spectrum and service sharing, we introduce a sharing coefficient to characterize the revenue of operators. Then, we develop an algorithm based on alternating optimization and successive convex approximation (SCA) methods for the WSR maximization problem. Simulation results demonstrate that the proposed mutual benefit constraints and optimization algorithms can enable operators to achieve mutual benefits.

Inter-operator resource and service sharing, which can utilize the network resources more efficiently and improve users' quality of services, is regarded as a promising method for the operators to construct future multi-operator networks. Particularly, an appropriate user association scheme is quite essential to enhance the network capacity. In this paper, our objective is to maximize the capacity of a multi-operator network by optimizing the user association scheme. Specifically, we investigate three inter-operator sharing scenarios, i.e., service-sharing scenario, spectrum-sharing scenario, and full-sharing scenario. Then, we propose a fractional programming (FP) based algorithm for the user association optimization problem. As inter-operator service sharing enables users to be served by other operators, a sharing coefficient is introduced for each operator to indicate its service level agreement (SLA) for regulating the users' association behaviors. As shown in the simulation results, the user association scheme obtained by the proposed algorithm can achieve a higher capacity than other alternative schemes. In comparison with the scenario without inter-operator sharing, both inter-operator spectrum sharing and service sharing can improve network capacity significantly, and the largest capacity is realized under the full-sharing scenario.

In this paper, we investigate a dynamic coordinated beamforming (CBF) problem to enhance the sum rate of a massive multiple-input multiple-output (MIMO) cellular network. Although existing optimization-based algorithms can provide near-optimal solutions, they require real-time global channel state information (CSI) and have high computational complexity, making them not viable in practical mobile networks. To tackle this issue, we propose a deep reinforcement learning based distributed dynamic CBF framework, which allows each base station (BS) to determine the optimal beamformers with only local CSI and some historical information transferred from other BSs. Besides, the computational complexity is substantially reduced thanks to the exploitation of neural networks and expert knowledge, i.e., a known solution structure that can be observed from a closed-form optimization algorithm. Simulation results demonstrate that the proposed approach can outperform the closed-form optimization methods and achieve comparable performance to the state-of-the-art optimization algorithm.

In this paper, we propose a reconfigurable intelligent surface (RIS) enhanced spectrum sensing system, in which the primary transmitter is equipped with a single antenna, the secondary transmitter is equipped with multiple antennas, and the RIS is employed to reduce the required signal samples while realizing a high detection probability. Without loss of generality, we adopt the maximum eigenvalue detection approach, and propose a corresponding analytical framework based on random matrix theory, to evaluate the detection probability in the asymptotic regime. Besides, the RIS is configured with only the statistical channel state information to avoid realtime channel estimation. With the statistical configuration, the asymptotic distributions of the equivalent channel gains are derived. Then, we provide the theoretical predictions about the number of reflecting elements required to achieve a detection probability close to 1. Finally, we present the Monte-Carlo simulation results to evaluate the accuracy of the proposed asymptotic analytical framework for the detection probability and the validity of the theoretical predictions about the number of REs required to achieve a detection probability close to 1. Moreover, the simulation results show that the proposed RIS-enhanced spectrum sensing system can substantially improve the detection performance.

We consider a homogeneous cellular network where a multi-antenna base station (BS) in each cell transmits messages to its intended user over a common frequency band. To improve the system capacity of this multi-cell multi-input single-output (MISO) interference channel, one of the state-of-the-art algorithms, namely, downlink-beamforming coordination, allows all BSs to cooperate with one another to mitigate the effect of inter-cell interference. However, most existing algorithms are suboptimal and impractical in a dynamic wireless environment, due to the high computational complexity and the overhead involved in collecting global channel state information (CSI). In this study, we exploit deep reinforcement learning (DRL) and propose a distributed dynamic downlink-beamforming coordination (DDBC) method with partial observability of the CSI. Each BS is able to train its own deep Q-network and employs appropriate beamformer depending on its environment, which is observed through a designed limited-information exchange protocol. The simulation results show that the proposed DRL-based DDBC method, with a considerably lower system overhead, achieves a system capacity that is very close to that of the fractional programming algorithm with global and instantaneous CSI measurements. In addition, this work demonstrates the potential of utilizing DRL to solve DDBC problems in a more practical manner.