DRaMA: Device-specific Repetition-aided Multiple Access for Ultra-Reliable and Low-Latency Communication

Transcription

1 DRaMA: Device-speciic Repetition-aided Multiple Access or Ultra-Reliable and Low-Latency Communication itaek Lee, Sundo im, Junseok im, and Sunghyun Choi Department o ECE and INMC, Seoul National University, Seoul, orea {ktlee, sdkim, jskim4}@mwnl.snu.ac.kr, schoi@snu.ac.kr Abstract In Fith-Generation (5G) New Radio (NR), satisying a scenario called Ultra-Reliable and Low-Latency Communication (URLLC) is one o main goals. URLLC requires very high reliability with low latency, ms user plane latency and % reliability. In Long Term Evolution-Advanced (LTE- A) system, however, latency requirement required by URLLC cannot be satisied because resource allocation process increases latency. In this paper, we propose DRaMA, a grant-ree uplink transmission scheme or low latency required in URLLC. With its own preamble and repetition pattern, a device repeatedly transmits the same packet using dierent requency resources across consecutive mini-slots right ater sending the preamble. We develop a mathematical model o collision probability to design repetition patterns, which determine the device-speciic requency resources. We demonstrate that DRaMA satisies URLLC requirements in most cases by using system level simulation. I. INTRODUCTION With the advent o the Internet o Things (IoT) era, cellular systems are evolving in dierent directions than beore. In addition to the peak data rate, which is the most important actor in current mobile communication system, so called Long Term Evolution-Advanced (LTE-A) system, there are other actors to consider such as latency and connection density or cellular IoT. In particular, low latency is emerging as a very important issue in the IoT applications. Thereore, in Fith- Generation (5G) New Radio (NR) deined by 3GPP, satisying low latency becomes a major goal. A service that requires very low latency and high reliability in 5G is called Ultra- Reliable and Low-Latency Communication (URLLC). Speciically, URLLC services require ms user plane latency and 0 5 Packet Error Rate (PER), i.e., % reliability []. Examples o URLLC applications requiring very low latency and high reliability include actory automation, process automation, and smart grid [2, 3]. These applications eature intermittent traic with very small packet size and low latency requirement. For example, in actory automation scenario, i an unintented misbehavior o a machine is detected, it needs to be reported to a controller immediately. Such occasional problems can be reported with very small amount o data, but should be reported as quickly as possible. Current LTE-A systems have unavoidable latency because device should be allocated resource to transmit uplink data. In particular, LTE-A device should perorm our-step resource allocation process or uplink transmission, thus incurring high latency [4]. Thereore, it is not appropriate to use grant-based uplink transmission through our-step resource allocation process or latency-sensitive URLLC applications. A possible way to reduce latency or such an environment is grant-ree uplink transmission without resource allocation process. Unlike the grant-based transmission scheme in which orthogonal resources are used by each device or uplink transmission, each device perorms a contention-based transmission in a grant-ree uplink transmission scheme. Uplink transmissions o multiple devices using same resource can lead to a collision, thus resulting in transmission ailure. Thereore, grant-ree uplink transmission scheme works well only when the collision problem is resolved. That is, the collision probability should be kept very low. In this paper, we propose a grant-ree Medium Access Control (MAC) protocol, called Device-speciic Repetitionaided Multiple Access (DRaMA), which supports low latency uplink transmission. Each device that perorms uplink transmission in DRaMA is assigned ) an orthogonal preamble in advance or device identiication and 2) a repetition pattern or perorming uplink data transmission. A device using DRaMA transmits uplink data repeatedly using a predetermined repetition pattern right ater the preamble transmission without a resource allocation. The repetition pattern determines which requency resources to use or each transmission. Apparently, the probability o successul packet transmission increases by repeatedly transmitting the same packet in multiple mini-slots. In this manner, the collision problem is resolved so that grantree uplink transmission becomes easible. Also, DRaMA can operate robustly to the channel error by using repetition transmission. In addition, the perormance o DRaMA is urther enhanced by designing near-optimal repetition patterns in terms o the collision probability. The rest o the paper is organized as ollows. In Section II, we introduce resource allocation process in LTE-A and minislot in 5G. In Section III, we summarize the related work. Section IV provides a detailed description o the proposed DRaMA protocol. In Section V, PER analysis o DRaMA is presented, and then, how to design a near-optimal repetition pattern is presented. Ater evaluating the perormance in Section VI, we conclude the paper in Section VII.

2 II. PRELIMINARIES In the current LTE-A system, device must be allocated resources rom evolved Node B (enb) or uplink transmission. The resource allocation process consists o our steps, and device can perorm uplink data transmission only ater the ollowing resource allocation process. ) A device sends Scheduling Request (SR) message to its enb via Physical Uplink Control Channel (PUCCH). 2) The enb sends an uplink grant to the device via Physical Downlink Control Channel (PDCCH). 3) The device sends a Buer Status Report (BSR) to enb via Physical Uplink Shared Channel (PUSCH). 4) The enb allocates PUSCH resources to the device or the data transmission. 5) The device sends uplink data to the enb via the allocated PUSCH resources. The our-step process is required to allocate uplink resources to each device. However, the our-step process causes high latency, and hence, it is not suitable or URLLC services. A method o reducing latency at physical layer is being considered. 3GPP has agreed on using shortened Transmission Time Interval (TTI) that is minimum scheduling unit o transmission [5]. In the current LTE-A, one TTI corresponds to 4 OFDM symbols ( ms), whereas or 5G, two OFDM symbol lengths (0.43 ms) are considered one TTI. A scheduling unit composed o two OFDM symbol is called a mini-slot. We use a mini-slot as one TTI or DRaMA to satisy URLLC requirements. III. RELATED WOR There have been several proposals or reducing latency in the cellular systems with eicient resource allocation process [6, 7]. Although these schemes provide scheduling method or low latency, the latency due to the resource allocation process still remains. Contention-based uplink transmission schemes without resource allocation process have been also proposed. In [8], Sparse Code Multiple Access (SCMA) is proposed. SCMA scheme uses sparse codewords so that a multi-user detection based on Message Passing Algorithm (MPA) is enabled. Thereore, SCMA enables contention-based uplink transmission by enabling enb to successully decode data o all devices even i data o multiple devices is transmitted using the same resources. However MPA receiver still has high complexity, despite eorts to reduce the complexity. Thereore, i the number o devices increases, it is not suitable to use SCMA. As a solution to reduce latency, Persistent Scheduling (PS) or Semi-Persistent Scheduling (SPS) may be another option [9]. While SPS and PS have been originally developed to support such applications as VoIP, which transmit periodic packets, it can reduce latency by reducing scheduling overhead. However, since SPS and PS pre-schedule orthogonal resources to devices, it is ineicient or URLLC applications with non-periodic packet arrivals. IV. DRaMA: PROPOSED GRANT-FREE UPLIN TRANSMISSION SCHEME A. System Model In this paper, we consider a situation where tens to hundreds devices perorm uplink transmission when an event occurs. We consider an environment where all devices are in the RRC CONNECTED state [0]. In an example such as a smart grid or a actory automation environment where devices are externally powered and latency is an important actor, the IoT devices can be considered to maintain the RRC CONNECTED state. Thereore, all devices using DRaMA are assumed to be uplink synchronized. We assume that all devices have the same type o traic. Each device has an unique preamble and a repetition pattern or uplink transmission, which are allocated in the initial setup stage. Uplink transmission using DRaMA is composed o two phases: preamble phase and data transmission phase. The preamble phase takes up one mini-slot and is the stage where the device sends its own preamble beore sending uplink data. The preamble and the repetition pattern have a one-to-one correspondence relationship. Thereore, the preamble serves not only as a device ID but also as a pattern ID. Also, the preamble is used as a reerence signal or channel estimation. In the data transmission phase, one mini-slot is required to transmit one packet. Accordingly, the data transmission phase takes up as many mini-slots as the number o repeated transmissions, and each device actually transmits an uplink packet. We reer to the length o the data transmission phase (i.e., the number o repetitions) as T. An uplink transmission process including the preamble phase and the data transmission phase is reerred to as one cycle. Detailed discussion is provided in Section V. B. Proposed Preamble Basically, the preamble used in the proposed scheme has the same principle as random access preamble o LTE-A system. In the LTE-A system, Zado-Chu (ZC) sequence [] is used as the random access preamble. The enb can successully decode all the preambles even i the enb receives multiple dierent preambles simultaneously, since dierent ZC sequences have very low cross correlation. Unlike the random access preamble o LTE-A with a subcarrier spacing o.25 khz, the proposed preamble has the same subcarrier spacing as an OFDM symbol or data transmission, i.e., 5 khz. Thereore, the duration o ZC sequence o the proposed preamble can be shortened by one OFDM symbol duration length or data transmission. In addition, i two OFDM symbols are used, the proposed preamble can be transmitted in a mini-slot duration. We propose a method to increase the number o preambles by using more requency resources up to ull bandwidth and using a smaller value o the cyclic shit. Using the above method, it is possible to generate a much larger number o preambles than 64 in the current LTE-A so that those preambles can be uniquely allocated to devices.

3 Minimum resource unit One Frequency slot Minimum resource unit One Frequency slot A B C D (a) Repetition patterns o devices A, B, C, and D Packet arrival Successul uplink data transmission Preamble transmission Failed uplink data transmission Device A One mini-slot (a) One resource block as one requency slot One mini-slot (b) Two resource blocks as one requency slot Fig.. Examples o repetition patterns. C. Repetition Pattern A repetition pattern used determines which requency resource to use when the device repeats transmission in multiple mini-slots. The minimum unit o requency resource is determined by the required packet size. For the convenience, we reer to the minimum unit o requency resource as a requency slot. One requency slot is composed o one or more RBs depending on the packet size and MCS. In addition, we deine the number o available requency slots in the system bandwidth as F. Fig. shows examples o repetition patterns where one resource block or two resource blocks are used as one requency slot. A repetition pattern is represented by a F T matrix. Each element o the matrix yields i the th requency slot in the t th mini-slot is selected, or 0, otherwise. I multiple devices transmit uplink data using the same requency slot in a speciic mini-slot, a collision occurs and transmission may ail. Thereore, how to avoid collisions should be considered when designing repetition patterns. We deine the similarity o two speciic repetition patterns, that is the number o mini-slots in which the selected requency slots o the two patterns are the same. The similarity o two speciic patterns which are represented by matrices, respectively, P and Q can be expressed as sim(p, Q) =tr(p Q). () Two devices with high similarity patterns have high inluence on each other when transmitting simultaneously. Also, only i the similarity o two patterns is or 0, collision probability is independent in all mini-slots. Obviously, it is best i all devices are assigned patterns with zero similarity, but it might be impossible because the number o available patterns is limited. Thereore, the similarity o all patterns should be minimized while satisying the required number o patterns. D. Uplink Transmission with DRaMA When an uplink packet arrives in a device, the device waits or a nearest preamble phase and transmits its preamble. By sending the preamble, the device inorms the enb o its device ID and the pattern to be used or the uplink transmission. Device B Device C Device D (b) Uplink transmission using DRaMA Fig. 2. Examples o uplink transmission using DRaMA. As described above, the enb can successully decode all the transmitted preambles. Immediately ater transmitting the preamble, the device transmits an uplink packet repeatedly using the predetermined repetition pattern. Fig. 2 shows a simple example o uplink transmission o our devices using DRaMA. In this example, a repetition pattern is represented by a 3 3 matrix. Devices A, B, C, and D use repetition patterns as shown in Fig. 2(a). In Fig. 2(b), the only device A transmits a packet in the irst DRaMA transmission cycle, devices A, B, and C transmit in the second transmission cycle, and devices A, B, C, and D transmit in the third transmission cycle. In the second and third cycle, we ind that devices suer rom collisions. However, each device successully transmits a packet in at least one transmission by using a dierent repetition pattern. V. REPETITION PATTERN DESIGN In this section, we discuss how to design repetition pattern to improve the perormance o DRaMA in terms o the collision probability and packet error probability. It is important to design good repetition pattern rather than repeating transmission randomly, because strictly low PER is required by URLLC. The PER perormance o DRaMA is closely related to the collision probability. We design an near-optimal repetition pattern and a pattern assigning rule. Whenever a repetition pattern is assigned to a new device, the repetition pattern is determined with the rule o minimizing the sum o the collision probabilities. A. Numerical Analysis o PER Perormance We start by numerically analyzing PER perormance o DRaMA. In this work, we assume that packet arrivals o each device ollow a Poisson process. Thereore, the collision

4 probability o the k th device when the k th device uses the th requency slot in the t th mini-slot, φ col,k, is expressed as φ col,k = e (N )λcyc, (2) where N is the number o devices included in the t th minislot and the th requency slot, and λ cyc (/cycle) is the packet arrival rate. λ sec (/s) and λ cyc (/cycle) are the same packet arrival rates in dierent units. Note that λ cyc = T λ sec. That is, the collision probability is equal to the probability that any devices other than the k th device transmits simultaneously with the k th device. Another cause o packet transmission ailure channel error. We deine that φ err,k is channel error probability o the kth device in the t th mini-slot. Since channel error and collision are independent, the probability o packet transmission ailure in one mini-slot is expressed as p err = φ err,k + φ col,k φ err,k φ col,k. (3) Since URLLC requires tight latency constraint, i packet transmission is not successul within latency constraint, it is deined as transmission ailure. That is, assuming that the time duration o s mini-slots is latency constraint, the device should transmit its packet within s mini-slots. I the waiting time to send a preamble ater packet arrival is too long, it is impossible to use all the repetitions. That is, the number o opportunities o transmission repetition depends on the packet arrival time. Assume that t w is the number o mini-slots rom a minislot in which a packet arrives to a mini-slot just beore the preamble phase. That is, t w is the number o mini-slots that a device should wait to send preamble. Then, the number o opportunities, t o, is expressed as t o =max ( 0, min(s t w,t) ). (4) Assuming that i) the transmission ails in the mini-slot in which collision occurs, and ii) all the collision probabilities are independent, the transmission ailure probability P ail,k in one cycle is expressed as P ail,k = T + t w= ( T + t o t= p err ). (5) We design repetition patterns to ensure that the collision probabilities in dierent mini-slots are independent and the sum o the collision probabilities o the devices is minimized by using (2) and (5). B. Frequency Hopping Repetition Pattern We propose a repetition pattern using requency hopping technique. The number o requency slots to jump beore each transmission is constant and a requency slot is determined by using mod(,f) operation. The requency slot used or transmission in the irst mini-slot in a speciic repetition pattern is reerred to as root requency slot and the number o requency slots to jump beore each transmission is reerred to as hopping degree. Root requency slot index has a value rom to F and hopping degree has a value rom 0 to F. Then, a repetition pattern with the root requency slot index r and the hopping degree h, respectively, is expressed as a F T matrix, M (r,h), with the ollowing elements. {, i i = mod(r +(j )h,f)+, m (r,h) i,j = (6) 0, otherwise. Assuming that h has a value rom 0 to F, the number o patterns that make the similarity o all patterns less than or equal to is expressed as # patterns w/ similarity or 0 = F 2 min t T gcd(f, t). (7) I determining F be a prime number, we can make F 2 patterns that make the similarity o all patterns less than or equal to regardless o T. In addition, i we group F patterns with the same h but dierent r into a group, we divide the patterns into F pattern groups. Two patterns in the same group have a similarity o 0, and two patterns in dierent groups have a similarity o, thus the collision probabilities are independent in all mini-slots. At the initial setup stage, dierent patterns are assigned to devices with the ollowing rule. ) One pattern group is selected, and all o the F patterns in the group are assigned to the devices. 2) Another pattern group is selected, and the process is repeated until all devices are assigned a pattern. I arrival rate is very low (e.g., λ sec =/s), the requency hopping repetition patterns satisy the ollowing lemma. Lemma. I we consider assigning a requency slot to a new device in a mini-slot, the sum o the collision probabilities is minimized by assigning a requency slot assigned to the minimum number o devices to the device. Proo. Assume that there exist devices in total, and N ( ) devices are assigned to the th requency slot out o F requency slots. We consider the situation where the ( +) th device is newly included in the th requency slot. Packet arrival rate o each device is λ cyc (/cycle). We deine Φ as the sum o the collision probabilities o devices already included. Besides, we deine Ψ + () as the sum o the collision probabilities o ( +) devices when the ( +) th device is included in the th requency slot. Since only the collision probabilities o the devices that are included in the th requency slot change, Ψ + () is then expressed as Ψ + () =Φ =Φ Φ N +(N + )( e λcyc ) N ( e (N )λcyc ) ++N e N λcyc (e λcyc ) e N λcyc ++(N λcyc )e N λcyc, i λ cyc. (8)

5 I N λ cyc is less than 2, Ψ + () is an increasing unction o N. Since λ cyc is very small, it is reasonable that N λ cyc is less than 2. Thereore, to minimize the sum o the collision probabilities, the ( +) th device should be assigned to the requency slot that is expressed as =argminψ + () F =argminn. (9) F That is, allocating the requency slot containing the smallest number o devices to the new device minimizes the sum o the collision probabilities. Obviously, the requency hopping patterns allocated to all devices by the above-described pattern allocation rule allocate a pattern having a requency slot including the smallest number o devices in each mini-slot to a new device. Also, since the collision probabilities in dierent mini-slots are all independent, such requency hopping patterns minimize the sum o the collision probabilities in all mini-slots whenever a device is assigned a repetition pattern. C. Determining Number o Repetitions The irst thing to notice beore determining the number o repetitions is that simply increasing the number o repetitions cannot be the best way. As the number o repetitions increases, the length o one cycle becomes longer so that average number o devices perorming uplink transmission in a cycle increases and the collision probability also increases. In addition, as the length o one cycle increases, the average waiting time in the Tx buer increases and the average latency increases. Thereore, it is necessary to determine an appropriate number o repetitions considering both latency constraint and collision probability. Since the exact channel is unknown in each transmission, we assume that channel error probability is ixed when determining the number o repetitions. First, we determine expected ailure probability and ind the number o repetitions that make the expected ailure probability to be minimized. The expected ailure probability is determined as P ail, which is the average ailure probability o all devices. In act, ailure probability P ail,k is a monotonic increasing unction o φ err,k, so that any ixed channel error probability does not aect determining the number o repetitions. Thereore, or all T with 2 T F, we determine the number o repetitions to minimize the expected ailure probability. Thereore, the number o repetitions T satisying the above condition is expressed as T =argminp ail. (0) 2 T F In this way, T minimizes the expected ailure probability. VI. PERFORMANCE EVALUATION In this section, we evaluate the perormance o DRaMA via system level simulation using MATLAB. In the simulation, we Collision rate Pseudo-random Frequency hopping (Simulation) Frequency hopping (Analysis) Pseudo-random Frequency hopping (Simulation) Frequency hopping (Analysis) 0-7 (a) Collision rate (b) Packet error rate Fig. 3. Frequency hopping patterns vs. pseudo-random patterns. Packet error rate TABLE I SIMULATION ENVIRONMENTS. Parameter Value Number o devices Cell radius 250 m Carrier requency 2.0 GHz Channel model WINNER+B model [2] Packet arrival Poisson process Maximum device Tx power 23 dbm Packet arrival rate λ sec 0.5, /s assume that the system bandwidth is 20 MHz, supporting up to 00 RBs. A small cell environment suitable or the considered IoT environment with a cell radius o 250 m is assumed. We assume that devices transmit packets with data o 32 bytes that URLLC applications require []. 32 byte packets can be transmitted with 4 RBs using MCS 2 and with 5 RBs using MCS 8 with a mini-slot. Thereore, using the entire 00 RBs, 25 and 20 requency slots can be used with MCS 2 and 8, respectively. Since the largest prime number smaller than 25 is 23, the number F o requency slots in DRaMA requency hopping pattern is 23 using MCS 2. On the other hand, since the largest prime number smaller than 20 is 9, the number o requency slots in DRaMA is 9 using MCS 8. In addition, because URLLC eatures intermittent traic, we assume packet arrival rate o 0.5 /s and /s. The values used in the simulation are summarized in Table I. A. Frequency Hopping Patterns vs. Pseudo-random Patterns We comparatively evaluate the perormance o DRaMA with requency hopping patterns and pseudo-random patterns, based on PER due to the collision and the channel error (assuming ixed channel error probability o 0.00). Fig. 3(a) shows collision probability with varying number o devices using DRaMA with pseudo-random patterns and requency hopping patterns. The collision probability o the devices using requency hopping patterns is lower than the collision probability using pseudo-random patterns. Fig. 3(b) shows PER with varying number o devices. We observe that DRaMA with requency hopping patterns outperorms DRaMA using pseudo-random patterns, since pseudo-random patterns cannot guarantee independent collision probability in mini-slots. I collision probability in each mini-slot is not independent, the perormance is degraded because some devices may aect collisions more.

6 Average latency (ms) Persistent scheduling DRaMA 0 (a) Average latency (b) Packet error rate Fig. 4. DRaMA vs. PS. PER Persistent scheduling DRaMA PER MCS = 8 & λ = 0.5/s MCS = 8 & λ = /s MCS = 2 & λ = 0.5/s MCS = 2 & λ = /s ECDF PER 0-5 (a) Average PER (b) PER o devices (λ =0.5) Fig. 5. Packet error rate o DRaMA = 00 & MCS = 8 = 00 & MCS = 2 = 200 & MCS = 8 = 200 & MCS = 2 = 300 & MCS = 8 = 300 & MCS = 2 B. DRaMA vs. Persistent Scheduling In the URLLC scenarios with event driven packet arrival, SPS that periodically modiies scheduling is not appropriate. Thereore, we compare DRaMA with PS only. Both devices with DRaMA and devices with PS use MCS 2. Fig. 4(a) shows the average latency o DRaMA and PS with varying number o devices. We observe that the average latency o DRaMA remains constant, while the average latency o PS increases with the number o users. Since each device uses orthogonal resources in PS, scheduling period should be increased when the number o devices increases. However, the average latency in DRaMA remains constant regardless o the number o devices because the cycle length is ixed. Fig. 4(b) shows PER o DRaMA and PS with varying number o devices. Unlike devices with DRaMA, devices with PS transmit a packet only once with a given resource. Thereore, it is observed that PS is more vulnerable to channel error than DRaMA. Also, in the case o PS, the scheduling period becomes longer as the number o devices increases so that i scheduling period is greater than or equal to the latency constraint, PER increases sharply. C. Packet Error Rate We evaluate the perormance o DRaMA with various MCS, packet arrival rate, and the number o devices. Fig. 5(a) shows the average PER or various cases. Using (5) and channel error probability according to each MCS, the upper bound o the number o devices satisying URLLC requirement is derived. In the case o λ =, the upper bounds are 23 and 240 when MCS indexes are 8 and 2, respectively. We observe that DRaMA can support URLLC requirements well in cases where the number o devices is less than the upper bound. Fig. 5(b) shows Empirical Cumulative Density Function (ECDF) o PER or each device when the arrival rate is 0.5 /s. It is observed that most devices transmit packets within ms with PER lower than 0 5 except or the case where the number o devices is 300. In the case o 200 devices, only three out o 200 devices do not satisy URLLC requirement. VII. CONCLUSION In this paper, we propose DRaMA, a novel grant-ree uplink transmission scheme or URLLC services. A device using DRaMA can perorm uplink transmission without resource allocation that causes high latency by using a unique preamble and a predetermined device-speciic repetition pattern. We design the repetition patterns o DRaMA with good perormance through collision probability and PER analysis. The perormance o DRaMA is evaluated through system level simulation. In most cases, the results demonstrate that packets can be transmitted within ms with PER lower than 0 5 that is required in URLLC scenario. In addition, it is observed that DRaMA perorms much better when requency hopping repetition pattern is used, rather than pseudo-random pattern is used. Also, DRaMA outperorms PS, since DRaMA satisies very low PER requirement which PS cannot satisy at all. ACNOWLEDGEMENTS This work was supported by Institute or Inormation & Communications Technology Promotion (IITP) grant unded by the orea government (MSIT) ( , Multiple Access Technique with Ultra-Low Latency and High Eiciency or Tactile Internet Services in IoT Environments). REFERENCES [] 3GPP TR 38.93, Study on Scenarios and Requirements or Next Generation Access Technologies, V4.3.0, Aug [2] P. Schulz et al., Latency critical iot applications in 5G: Perspective on the design o radio interace and network architecture, IEEE Communications Magazine, vol. 55, no. 2, pp , 207. [3] I. Al-Anbagi et al., A low latency data transmission scheme or smart grid condition monitoring applications, in Proc. IEEE EPEC, 202, pp [4] 3GPP TS 36.32, Evolved Universal Terrestrial Radio Access (E- UTRA); Medium Access Control (MAC) Protocol Speciication, V4..0, Dec [5] R-62263, Comparison o slot and mini-slot based approaches or URLLC, 3GPP TSG-RAN WG #87bis, Nov [6] J. Brown and J. Y. han, A predictive resource allocation algorithm in the LTE uplink or event based M2M applications, IEEE Transactions on Mobile Computing, vol. 4, no. 2, pp , 205. [7] A. S. Lioumpas and A. Alexiou, Uplink scheduling or machine-tomachine communications in LTE-based cellular systems, in Proc. IEEE GLOBECOM Workshops, 20, pp [8]. Au et al., Uplink contention based scma or 5G radio access, in Proc. IEEE GLOBECOM Workshops, 204, pp [9] 3GPP TR 36.88, Study on latency reduction techniques or LTE, V4.0.0, Jun [0] 3GPP TS 36.33, Evolved Universal Terrestrial Radio Access (E- UTRA); Radio Resource Control (RRC); Protocol Speciication, V3..0, Apr [] 3GPP TS 36.2, Evolved Universal Terrestrial Radio Access (E- UTRA); Physical Channels and Modulation, V3..0, Mar [2] Drat New Report ITU-R M.[IMT.EVAL], Guidelines or evaluation o radio interace technologies or IMT-advanced, Nov