106 UDC 621.39 On the performance measures of LTE radio access procedure under massive M2M communications Ekaterina G. Medvedeva* , Alexey V. Chukarin* , Vladimir V. Rykov*† , Yuliya V. Gaidamaka*‡ * Peoples’ Friendship University of Russia (RUDN University) 6 Miklukho-Maklaya St, Moscow, 117198, Russian Federation † Department of Applied Mathematics and Computer Modeling Gubkin Russian State University of Oil and Gas 65 Leninsky Prospekt, Moscow, 119991, Russian Federation ‡ Federal Research Center “Computer Science and Control” of the Russian Academy of Sciences (FRC CSC RAS) 44-2 Vavilov St, Moscow, 119333, Russian Federation Email: medvedeva_eg@rudn.ru, chukarin_av@rudn.ru, vladimir_rykov@mail.ru, gaydamaka_yuv@rudn.ru Providing the evolution from current wireless systems to fifth generation (5G) network is to support massive Machine-to-Machine (M2M) wireless communications in radio access network. Performance analysis of the random access channel (RACH) is a top issue within the M2M-connection in LTE networks, because prior the data transmitting, the session initiation procedure, which perform the connection initiation for user equipment, could overload the channel dealing with burst arrival of connection requests. The purpose of this paper is to continue the analysis of RACH initiation procedure using discrete Markov chain model, and to investigate the dependence of average delay time from preamble processing time. The simulation model is obtained, which allows for estimating the influence of preamble collision on the success access connection initiation in radio access network. Key words and phrases: LTE-advanced, 5G, machine-type communications, random access channel, collision, access success probability, access delay, Markov chain, session initiation procedure, mathematical model. Copyright © 2018 for the individual papers by the papers’ authors. Use permitted under the CC-BY license — https://creativecommons.org/licenses/by/4.0/. This volume is published and copyrighted by its editors. In: K. E. Samouylov, L. A. Sevastianov, D. S. Kulyabov (eds.): Selected Papers of the 12th International Workshop on Applied Problems in Theory of Probabilities and Mathematical Statistics (Summer Session) in the framework of the Conference “Information and Telecommunication Technologies and Mathematical Modeling of High-Tech Systems”, Lisbon, Portugal, October 22–27, 2018, published at http://ceur-ws.org Medvedeva E. G. et al. 107 1. Introduction The basic concept of the transition from modern wireless systems to 5G-technologies is to support the massive machine-to-machine(M2M)and Internet-of-things (IoT) de- vices’ connections and still provide a number of promising highly demanded services. According to ETSI [1], potential M2M devices and applications capable of generating and transmitting data autonomously in IoT network are: 1. intelligent devices; 2. smart city; 3. intelligent networks; 4. e-health; 5. connected cars; 6. smart households and energy management; 7. remote industrial process control. At the same time, the main tasks for observing the required performance indicators are the ability to scale the network, improve energy efficiency and reduce the cost of sensory user devices. Such technologies as Radio Frequency Identification, Zigbee, Bluetooth Low Energy and Low-Power WiFi, which typically implement unlicensed frequency bands, and operate on low power consumption and short transmission range, are designed to support M2M applications. The disadvantage of using such technologies are excessive interference between devices in the coverage of the unlicensed spectrum, which reduces the reliability of these systems, and complicating of initiating access to the radio environment increases the connection delays [2]. To address these issues in the development of IoT, the application of low-power technologies, such as Low Power Wide Area (LPWA), Sigfox, Long Range (LoRa), Weightless and Long Term Evolution (LTE), is recommended, and LTE cellular technol- ogy is the most suitable solution due to the wide coverage in the existing infrastructure, security, licensed spectrum and easier maintenance [3]. One possible solution to the LTE network scalability problem is based on an analysis of the RACH connection initiation procedure [4, 6, 7]. For a number of scenarios of M2M-interconnection the access delay for user equipment (UE) dominates, exerting a significant load on the channel even before the actual data transfer begins [8]. This problem appears at peak times, in the case of simultaneous activation of a large group of devices, for example, when sensors are reconnected after a power outage [9]. This burst arrivals can initiate RACH’ overload for a long period of time. In paper [4] modelling of session initiation procedure provided the opportunity to implement the RACH parameters to increase the probability of a successful connection (access success probability) and to reduce the average access delay [5]. In [6] the dependence of the collision probability on the number of M2M-devices was investigated in the conditions of rapidly growing M2M traffic and high demand of UEs to a single base station (BS). Here the approach with state-dependent arrival and service rates can be used [12]. The purpose of the current work is to continue an analytical model’s development for evaluating performance measures, provided possible retransmission of three messages (Msg1, Msg3, Msg4), which we describe in Section 3. To verify the obtain results we build the simulation model of session initiation procedure, and in Section 4 present the part of simulated programme code. Section 5 is provided with numerical analysis of the dependence of different preambles processing time on average RA procedure delay and comparison of analytical and simulation methods. 2. Random Access procedure The basic Random Access (RA) procedure, which initiate the connection between UE and eNB, consists of four steps and can be divided into two stages: a link synchronization step (Msg1, Msg2) and a service transfer (Msg3, Msg4). 108 APTP+MS’2018 UE eNB RA Resource selection TRA _REP  (k  1) Msg 1 Preamble successful 5 TRAR  2 transmitted Msg 2 UE monitors RAR during interval t W RAR 5 0  t  WRAR 6 ms Msg 3 THARQ  4 The HARQ HARQ ACK Retransmission of Msg 3 Msg 4 TM 4 , GAP  1 and Msg 4 can be up N HARQ times THARQ  4 Completion RA procedure HARQ ACK Figure 1. Sequence of messages’ transmission in RA procedure. It begins with transfer from UE to eNB the Msg1 (Preamble Transmission), and selection one from the set of 64 preambles [10, 13]. Chosen index of preamble request differentiates multiple devices. When two or more UEs select a same RA preamble, a collision occurs and all UEs should retransmit Msg1. Further the UE receives response – a message RAR (Random Access Response, Msg2) – from the eNB. If UE does not receive the response Msg2, user’s transmitter increases the power and repeats the preamble transmission over the time interval, following which UE answers Msg3 (Connection Request). Then, the automatic acknowledgement HARQ ACK (Hybrid Automatic Repeat Request Acknowledgment) allows to protect the signaling message transmission. If the Msg3 is successfully transmitted and processed, the eNB responds with a Msg4 (Connection Response). If the UE does not receive from the eNB the Msg4, the Msg4 message will be sent again in specified time interval. By exceeding the Msg1 transmission counter the connection initiation procedure is considered unsuccessful. In case of exceeding number of Msg3/Msg4 transmission, the procedure starts from the new Msg1 preamble transmission, in case the maximum number of preamble transmission preambleTransMax is not reached. The example of complete successful connection initiation is presented on Fig. 1. 3. Mathematical model In this work we extend the previous results, presented in [11]. We build the mathe- matical model of RACH procedure, taking into account the possibility of retransmission of messages (Msg1 or preamble, Msg3 and Msg4) between UE and BS with limited number of retransmissions. Let us introduce probabilistic events 𝐴𝑖 ={Msg(i) is successfully transmitted}, in- verse events 𝐴𝑖 ={Msg(i) is unsuccessfully transmitted}, and denote the corresponding probabilities P({𝐴𝑖 })=1 − 𝑝𝑖 and P({𝐴𝑖 })=𝑝𝑖 , 𝑖 ∈ {1, 3, 4}. We consider discrete-time Medvedeva E. G. et al. 109 Markov chain {𝜉𝑖 , 𝑖 = 0, 1, . . . , (1 + 𝑁3(︁+ 𝑁4 )𝑁1 } over)︁ the state space 𝒳 = {(x, y, z) ∈ 𝑥1 𝑥2 𝑥3 𝑥4 𝑥5 Z3 ⋃︀ (0, 0, 0)}, so that: {(x, y, z) = 𝑦1 𝑦2 𝑦3 : 𝑥1 = 0, 𝑁1 , 𝑦1 = 0, 𝑁3 , 𝑧1 = 𝑧1 𝑧2 0, 𝑁4 , 𝑥4 ≤ 𝑦3 ≤ 𝑁3 𝑥4 , 𝑥5 ∈ {0, 1}, 𝑦2 ∈ {0, 1}, 𝑧2 ∈ {0, 1}, }, where: 𝑥1 is total number of transmitted Msg1, 𝑥2 is the number of successful transmitted Msg1, (0 ≤ 𝑥2 ≤ 𝑥1 ), 𝑥3 is the number of times the counter 𝑁3 is reached when transmitting Msg3, 𝑥4 is the number of times the counter 𝑁4 is reached when transmitting Msg4 (0 ≤ 𝑥3 , 𝑥4 ≤ 𝑥2 ), 𝑥5 is an indicator denoting the current state of the last transmitted Msg1, which equals to 1 in case of successful current transmission, 0 – in case of collision, 𝑦1 is total number of transmitted Msg3 after last successful Msg1’s transmission, 𝑦2 is the number of successful transmitted Msg3 after last successful Msg1’s trans- mission, 𝑦3 is the number of transmitted Msg3 (both successful and unsuccessful) followed with blocking of all Msg4 transmitted 𝑁4 times, 𝑧1 is total number of transmitted Msg4 after last successful Msg3’s transmission, 𝑧2 is an indicator denoting the current state of the last transmitted Msg4, which equals to 1 in case of successful transmission, 0 – in case of unsuccessful transmission. Statement 1. The probability 𝑃 (x, y, z) of visiting the state (x, y, z) from the initial state (0, 0, 0) is determined with (1): 1 −𝑥2 𝑁3 𝑥3 −𝑥4 +𝑦3 +(𝑦1 −𝑦2 )𝑢(𝑦2 𝑧2 ) 𝑃 (x, y, z) = 𝑝𝑥 1 (1 − 𝑝1 )𝑥3 +𝑥4 +𝑥5 𝑢(𝑧2 ) 𝑝3 × (1 − 𝑝3 )𝑥4 +𝑦2 𝑢(𝑦2 𝑧2 ) 𝑝4 𝑁4 𝑥4 +(𝑧1 −𝑧2 )𝑢(𝑧2 ) (1 − 𝑝4 )𝑧2 𝐶𝑥𝑥13−1 +𝑥4 𝑥3 𝐶𝑥3 +𝑥4 × ⎛ ⎞𝑢(𝑥4 −1) 𝑦3 −𝑥4 𝑁4 −1 (−1)𝑖 𝐶𝑥𝑖 4 𝐶𝑦𝑥34−𝑖𝑁 ⎜ ∑︁ ⎟ ×⎜ ⎝ ⎟ 3 −1 ⎠ , (1) 𝑖=0 where 𝑢 (𝑥) is Heaviside function. The multipliers with the Heaviside function in the exponents allow to ignore redundant retransmissions that arise if the connection initiation procedure is not successful. We denote the set of states 𝒳𝑠 = {(x, y, z) : 𝑥5 = 𝑦2 = 𝑧2 = 1}, leading to successful session initiation and the set 𝒳𝑓 = {(x, y, z) : 𝑥1 = 𝑁1 , 𝑥3 + 𝑥4 = 𝑥2 , 𝑧2 = 0} of failed initiation procedure states. Then access success probability and access failure probability are derived with (2) and (3) respectively: ∑︁ 𝜋𝑠 = 𝑃 (x, y, z) , (2) (x,y,z)∈𝒳𝑠 ∑︁ 𝜋𝑓 = 𝑃 (x, y, z) . (3) (x,y,z)∈𝒳𝑓 Statement 2. For 𝑁1 ∈ {1, 2, 3} expression (2) can be obtained in closed form (4): 4𝑁1 − 6 [︂(︂ (︁ )︁)︂𝑁1 𝜋𝑠 = 1 − 𝑝1 + (1 − 𝑝1 ) 𝑝𝑁3 3 + (1 − 𝑝 3 )𝑝 𝑁4 4 − 𝑁1 (1 − 𝑝3 ) (︂ (︁ )︁)︂𝑁1 ]︂ 3 −1 − 𝑝3 𝑝1 + 𝑝𝑁 3 (1 − 𝑝1 ) 𝑝3 + (1 − 𝑝3 )𝑝𝑁 4 4 , (4) 110 APTP+MS’2018 Figure 2. Time intervals for message transmission depending on probabilistic events. ∑︀ Since the normalization condition (x,y,z)∈𝒳 𝑃 (x, y, z) = 1, the access failure probability could be obtained as 𝜋𝑓 = 1 − 𝜋𝑠 . On Fig. 2 the scheme of transitions between events of procedure with instructions of corresponding time intervals is presented. Total delay of the transition from the initial state (0, 0, 0) to the state (x, y, z) is the sum of the time intervals of the transmission corresponding messages, and described by formula (5): 𝑑 = 𝑑(x, y, z) = 𝑥3 (△1 + 𝑇𝑅𝐴𝑅 + 𝑊𝑅𝐴𝑅 + △2 + 𝑁3 (𝑇𝐻𝐴𝑅𝑄 + 𝑇𝑀3 ) + + 𝑥4 (△1 + 𝑇𝑅𝐴𝑅 + 𝑊𝑅𝐴𝑅 + △2 + 𝑇𝐻𝐴𝑅𝑄 + 𝑇𝐴𝑀4 + 𝑁4 (𝑇𝐻𝐴𝑅𝑄 + 𝑇𝑀4 )) + (𝑦3 − 𝑥4 )(𝑇𝐻𝐴𝑅𝑄 + 𝑇𝑀3 ) + 𝑥5 𝑢(𝑧2 )(△1 + 𝑇𝑅𝐴𝑅 + 𝑊𝑅𝐴𝑅 + △2 ) + (𝑥1 − 𝑥2 ) × × (△1 + 𝑇𝑅𝐴𝑅 + 𝑊𝑅𝐴𝑅 + 𝑊𝐵𝑂 ) + 𝑦2 𝑢(𝑦2 𝑧2 )(𝑇𝐻𝐴𝑅𝑄 + 𝑇𝐴𝑀 4 ) + (𝑦1 − 𝑦2 )𝑢(𝑦2 𝑧2 ) × × (𝑇𝐻𝐴𝑅𝑄 + 𝑇𝑀3 ) + (𝑧1 − 𝑧2 )𝑢(𝑧2 )(𝑇𝐻𝐴𝑅𝑄 + 𝑇𝑀4 ) + 𝑧2 𝑇𝐻𝐴𝑅𝑄 . (5) To find average access delay 𝑑 the formula (6) is presented. ∑︀ (x,y,z)∈𝑋𝑠 𝑃 (x, y, z) 𝑑 (x, y, z) 𝑑= . (6) 𝜋𝑠 Medvedeva E. G. et al. 111 4. Simulation model This section presents the simulation model of RA procedure. To verify the results obtained using the formulas from Section 2, the code for the simulation model was written, which is a simulated attempt of initiation connection between UE and the BS. The main part of the code is the imfn function. The input values for this function are as follows: 1. the maximum number of retransmissons 𝑁1 , 𝑁3 , 𝑁4 ; 2. collision probability 𝑝1 ; 3. probability of unsuccessful transmissions of Msg3 and Msg4; 4. time intervals vector. After setting the value to variables using given time interval vector, we build the matrix, which contain the final state that the system transit to after the connection attempt is completed, and the time it takes for one attempt to initiate a connection. imfn=function(N1,N3,N4,p,p34,app,time){ del1=time[1]; del2=time[2] Trar=time[3]; Wrar=time[4] Wbo=time[5] Tm3=time[6]; Tm4=time[7] Tam4=time[8]; Tharq=time[9] m=matrix(0,app,14) colnames(m)=c("x1","x2","x3","x4","x5","y1","y2", "y3","z1","z2","t","c1","c3","c4") Next is the for loop, which execute set number of iterations. The more iterations are set, the better simulation is obtained. The body of the for loop starts with another while loop, which iterates until the connection is initiated or the counter 𝑁1 is exceeded. The body of the for loop begins with the adding to the matrix element value △1 , which corresponds to the duration, required for the current attempt – the time interval determined for RA Resource seletion before sending the message Msg1. Next, the sample function returns one value (0 is the collision of the Msg1, 1 is the successful transmission of the Msg1), which is then added to the value of the matrix element, corresponding to the given total number of Msg1 retransmission. Appropriate time intervals are added to the total duration (𝑇𝑅𝐴𝑅 , 𝑊𝑅𝐴𝑅 ). for(i in 1:app){ while(m[i,1]