Delay-optimal Random Access For Massive Machine-type Communications

In recent years,as the basic technique for connecting everything in 5G,machine-type commu-nication(MTC)attracts lots of attention from academia and industry.Different from the human-type communication(HTC),there will be massive number of devices existing in a single cell for MTC.The uplink connecting requests for MTC will be much more than HTC's.However,when massive connecting requests are simultaneously inserted into the random access,serious access congestion can not be avoided in the existing cellular system.Therefore,we need some access control mechanism specially tailored for the MTC.Many schemes are proposed to alleviate the ac-cess collision by dispatching access requests to different slots at access control layer or expanding the access capacity at physical layer.However,besides the consideration of throughput from the perspective of physical layer and access control layer,delay is another one key performance.As a matter of fact,lots of MTC applications,such as industrial control and real-time monitoring have high requirement for delay.Hence,it is necessary to optimize the delay performance for MTC.Motivated by this,this thesis presents some research on the delay-aware massive random access for MTC.Firstly,this thesis considers a general scenario for the random access of MTC.Dependent on whether a MTC device is capable of detecting the channel state information(CSI),the thesis divides the discussion into two branches.For the branch that MTC devices are not capable of de-tecting the CSI,each device makes access decision according to some policy by using the expected value of CSI as the estimation.In addition,each device transmits data with constant power.for the branch that MTC devices are capable of detecting the CSI,the access decision and power control decision of each device is not only related to the queue state information(QSI),but also the C-SI.Both of the branches formulate the delay-aware random access problem as the infinite horizon average cost Markov decision process(MDP).With the property in the probability that a MTC device successfully accesses the base station(BS),this thesis decomposes the Bellman equations,i.e.,the global optimal conditions into per-device fixed-point equations.Then,by applying the primal-dual decomposition to the per-stage optimization problem,this thesis gives the BS-assisted distributive structure.The BS broadcasts the expected simultaneous access and the Lagrange mul-tiplier used for controlling the simultaneous access.Based on the information broadcast by the BS,each device makes access decision according to some policy.Finally,hierarchical stochastic learning algorithm is proposed to update the related parameters.Different from the first branch,when considering the situation that CSI is detectable,this thesis decreases the dimension of the state space by defining conditional actions,and applies the virtual continuous time approximation to obtain the asymptotically optimal power control policy.Secondly,this thesis extends the general case into two specific situations:one is adopting non-orthogonal multiple access(NOMA)in the data transmission and the other is applying ener-gy harvesting(EH)to the energy supply.In the former situation,since the successful decoding requires strict power allocation in the received power domain by the BS in NOMA,the decom-posed per-device fixed-point equations are coupled not only by random access procedure but also the BS's central data transmission schedule.To handle the central schedule coupling,this thesis proposes the concept of dividing the received power domain into multiple levels and adding the level information as the system state.Then this thesis proposes a design that the BS maintains the proportion of successful data transmission after successful random access with different received power level and different schedule gain.Each device uses the proportion information broadcast by the BS for the future estimation to decouple the central schedule between devices.In the lat-ter situation,since the energy harvested is stochastic,the energy queue state information(EQSI)should be considered in the system state except the data queue state information(DQSI)and CSI.To obtain the policy which maps from the system state to the access and power control,with the random access part decoupled,this thesis adopts the first-order Taylor expansion to approximately obtain the closed-form solution for the asymptotically optimal access and power control which are related to the potential functions.