Computational Resource Constrained Multi-Cell Joint Processing In Cloud Radio Access Networks

With the advent of the mobile Internet era,the current wireless access network is facing more and more challenges.In the next generation of mobile communication 5G,a green evolution-oriented cloud radio access network(C-RAN)architecture is proposed,which has the characteris-tics of centralized,collaborative and cloud computing.The centralized baseband processing can greatly reduce the number of base stations needed to cover the same area.The coordinated wire-less remote module and antenna can improve the system spectrum efficiency.The cloud computing base station virtualization technology can reduce the cost,share the processing resources and re-duce Energy consumption,improve infrastructure utilization.However,centralized processing and large-scale cooperative transmission in C-RAN actually require global coordination,scheduling and control,and their computational complexity usually increases with the polynomial of the net-work scale.In future C-RANs,due to the high data rate requirements,dense radio remote head(RRH)is densely deployed,the communication resources of the entire network will become very limited and the global coordination of server computing overhead can no longer be be ignored.As the size of the network grows dramatically,the computational power of servers is becoming another important factor to consider when deploying C-RAN networks.Therefore,it is of great theoretical and practical value to research communication and computational resource allocation algorithms in cloud radio access networks to improve system throughput and reduce transmission delay,thus improving user quality of service(QoS).First of all,this thesis analyzes the problem of joint processing of multi-cell clusters con-strained by centralized computing resources in cloud radio access network.In the C-RAN ar-chitecture,the centralized joint processing can have more significant performance gains than the conventional per-cell processing by avoiding inter-cell interference,but due to the huge computa-tional complexity brought to the overall network,full collaboration is not feasible.In this thesis,different cells are clustered,and then ZF combined with precoding in the same cluster before sig-nal transmission.At the same time,this thesis uses the CPU floating operation points(FLOPs)to describe computational resources and correlates them directly with the computational complexity of signal processing algorithms to model the problem as a combinatorial optimization problem,Since the combinatorial optimization problem is an NP-hard problem,we consider that in order to maximize the total data rate of the entire system,the objective function and computational resource constraints are redescribed by relaxation and successive convex approximation(SCA)techniques.With the special structure of this problem,a low complexity RRH clustering and power alloca-tion algorithm is proposed and solved iteratively.This paper also proves the convergence of the proposed iterative algorithm and the simulation evaluation of the algorithm to show its throughput performance superior to the traditional algorithm.Next,this thesis studies the multi-cell clustering joint processing problem of distributed com-puting resource constraints in cloud wireless access network.Unlike previous centralized C-RAN architectures,multiple compute-server building baseband units(BBUs)in a distributed C-RAN perform both signal processing and joint transmission at the same time,and different matching methods between the RRH and the BBU can also result the difference between fronthaul link se-lection.Distributed processing can greatly reduce the burden on the backhaul link and avoid the high complexity of signal processing in a centralized baseband.In this paper,the baseband signal quantization process on the forward link is described using computational resources.Considering the constraints of nonlinear computational resources in this combinatorial optimization problem,this paper maps the problem of maximizing the throughput into a combinatorial auction problem(CAP).In order to deal with this problem is the nature of NP-hard,this thesis proposes a greedy al-gorithm based on the WISP method and obtains the feasible solutions of the combinatorial auction problem by using the properties of distributed C-RAN.Weighted independent set problems(WISP)in the local a improved algorithm to gradually improve performance.We also use the simulation results to prove the convergence of the algorithm and its superiority over other algorithms.Finally,this thesis optimizes the time-delay of the multi-cell cluster processing with resource constraints.In a cloud radio access network,it is also crucial to consider network congestion per-formance when a burst of traffic arrives,especially for delay aware services.Therefore,it is crucial to study how to reduce communication delay and improve user QoS.In this thesis,we consider the queuing delay based on the combinatorial optimization model to maximize the throughput,and transform the delay-aware optimization problem into the corresponding Markov decision process.Due to the difficulty of solving the problem that the number of exponential classes of the corre-sponding Bellman equations is hard to solve directly,this thesis adopts a stochastic approximation method:an online Q-learning method to obtain the desired objective function and finally achieves the convergence through the distribution iteration.