Research On RLC Layer Modeling In LTE Network Based On Queuing Theory

The rapid growth of Internet applications and data services promotes development ofLTE. MIIT issued TD-LTE licenses to Mobile, Telecom, Unicom, which identified ourcountry had entered the4G Internet era. In the era of networking, Internet of Things becomesan inevitable trend. As one of the hottest applications in Internet of Things, M2M can providewireless connectivity for machine to machine, machine to man, man to machine. StrategyAnalytics shows78%of M2M terminals will be connected via3G or faster networks by2022.LTE will catch up with3G depending on its advantages and become the main bearer networkof M2M services.ABI predicts that the number of M2M cellular connections will be over450million by2018, whose magnitude will reach hundreds of millions. The radio resources are limited.Rational resource allocation scheme is an effective means to ensure QoS. Communicationsnetwork modeling can provide a theoretical basis for resource allocation by abstracting theactual network to build mathematical model and using the theoretical performance indicatorsto optimization program. Queuing theory is a popular theory of network modeling.The RLC layer of LTE network has cache. According to the indicated resource by theMAC layer, SDUs are segmented, concatenated and retransmitted, which affect the arrivalprocess or the service process of SDUs and the system performance. Therefore, we can modelRLC layer based on queuing theory, and research the effect of segmentation, concatenationand retransmission on system performance to guide resource allocation. Since the probabilityof retransmission is small, i.e., the effect of retransmission on system performance isrelatively small. Therefore we ignored ARQ mechanism. In order to simplify the analysis, wefocused on the segmentation modeling of RLC layer.In this paper, M2M services had fixed resource each scheduling period. The SDUsarrived at the RLC layer with a single Poisson process. Considering two lengths of SDUswhose length distribution was (0,1). The long SDU was two times as long as that short SDU.The service process was equivalent to RLC layer transported SDUs by segmentation eachscheduling period. Therefore, the service time was a scheduling period which was afixed-length distribution. The RLC PDU was a resource granularity level and the same sizewith the short SDU. The size and the number of PDUs were fixed each scheduling period.We focused on the performance indicators (average number and delay) of unequal length SDUs which could guide the allocation of resources. When the remaining PDU wasinsufficient to carry a long SDU, segmentation made one part of long SDU serviced in ascheduling period, and the remaining part serviced the next scheduling period. So the numberof serviced SDUs was unknown. We could’t obtain the performance indicators of the unequalSDUs. To solve this problem, we proposed an equivalent analysis thought which considered along SDU as two short SDUs. We converted the research of performance indicators ofunequal length SDUs to that equal length SDUs. So the number of serviced equal lengthSDUs was fixed each scheduling period. The service process was a fixed length distributionwhose batch size was fixed, and the arrival process was Poisson process whose batch size was(0,1) distribution. The equivalent RLC layer segmentation model was M[L]/D[r]/1. Delay ofunequal length SDUs was equal to the sum of the delay of equal length SDUs and additionaldelay that came from segmentation of long SDUs. we obtained the average delay and thenumber of unequal length SDUs.Since M[L]/D[r]/1is too complex to solve in the time domain, so we examined the matterfrom mathematical view. The probability distributions of batch Poisson arrival and fixedlength service has a special form in frequency domain. By probability generating functionmethod combined with iterative technique and Rouché’s Theorem, we converted complextime-domain queuing model into the frequency domain to simplify the process of solving.M2M service characteristics (i.e., the arrival intensity and length distribution of RLCSDUs) and the number of resource that is indicated by MAC layer (i.e., RLC PDUs) have acertain impact on system performance. By software simulation we found that for the arrivalintensity and length distribution, there is a number of RLC PDUs make the average numberand delay of SDUs minimum, which can be used as theoretical basis of radio resourceallocation in LTE network.The solving methods of queuing model affect the accuracy of performance evaluation.We solved D-ON-OFF/Geo/1and D-ON-OFF/Geo/1/N by the probability generating functionmethod, the maximum entropy method and matrix geometric method. We evaluated thecomplexity of these three methods through derivation.Comparing the performance indicatorsof these three methods with simulation model, we evaluated the accuracy of these threemethods. The section laid the foundations of mathematics for the next chapter.