A Survey on LTE/LTE-A Radio Resource Allocation Techniques for Machine-to-Machine Communication for B5G Networks

Machine-to-Machine (M2M) communication refers to autonomous communication among devices that aims for a massive number of connected devices. M2M communication can support ubiquitous communication and full mechanical automation, and it will change everything from industry to ourselves. Recent developments in communication technology make Long Term Evolution (LTE)/Long Term Evolution-Advance (LTE-A) a promising technology for supporting M2M communication. LTE can support the diverse characteristic of M2M communication due to its IP connectivity, coverage area, and scalability. Therefore, the LTE schedulers should satisfy the need for M2M communication. Motivated by these facts, in this paper, we present a survey on the classification of LTE / LTE-A scheduling methodologies from the perspective of M2M communication. We classify the schedulers based on their objectives, such as energy efficiency, spectrum efficiency, group-based, and Quality-of-Service (QoS) support for Machine Type Communication Devices (MTCDs). We also highlight the scope of future research direction for the scheduling work.

• Wide range of delay and throughput requirements. • Vast variety of applications promulgate for a more extensive range of service requirements. In this study, we focus on the LTE/LTE-A scheduling approaches proposed by various endeavors and present a detailed literature survey of the proposed scheduling schemes from the perspective of M2M communication. All the work related to scheduling schemes is broadly classified into four categories, namely Efficiency Focused, Group Based, QoS Focused, and Hybrid or Multi-Objective. Classification of scheduling schemes as shown in Fig. 9.
The rest of the paper is organized as follows. Section II presents an overview of the M2M communication system. Section III explains about LTE Network and scheduling process. Section IV provides a detailed survey of existing LTE scheduling schemes. Section V highlights the future research direction for resource scheduling and Section VII finally conclude the paper.

II. BACKGROUND
In this section, we provide brief backgrounds on M2M communication, system architecture, and applications of M2M. We also brief the enabling technologies for the M2M area network and M2M standardization efforts given by worldwide standardization bodies.

2) E-HEALTH
This type of application is useful in monitoring a person's health remotely [43]. Usually, a person wears sensors like a smartwatch, which monitors a person's health, such as blood pressure monitoring, heart rate monitoring, etc.; due to limited resources and processing power, these sensors send the gathered data to the MTCG, usually a smartphone [44]. The MTCG collects the data and sends it to the remote server of the e-Health service via eNB, where health care professionals analyze the data and act accordingly. These applications work like Body Sensor Network except the communication is bidirectional [45]- [47]. Applications that fall in this category are as follows.
• Remote health monitoring • Fitness information monitoring

3) SMART METERING
M2M applications that fall in this category are responsible for the efficient use of water, electricity, and gas through smart metering devices. These meters are an essential part of the smart grid [48], [49]. The applications in this category are as follows [50], [51].

4) SMART CITY
This category of M2M application belongs to the applications that are developed and deployed to make easy and smooth access to service to the citizen and save energy and cost [52], [53]. Sensors are deployed across the city, and they will send data to the gateway, and then it is forwarded to the MTC server [46], [47], [54] for further analysis. Some of the significant applications are as follows.
• Smart vehicle parking • Smart city-waste management • Smart street lights • Pollution control • Smart traffic management 5) SMART HOME These applications are developed to provide comfortable living to home users with remote detection and execution of particular tasks such as Smart Wi-Fi Plug [55], [56]. Some of the significant M2M applications for the smart home are as follows [46], [47], [57].
• Remote control of appliances • Water and gas leakage detection • Security system's remote monitoring The applications mentioned above are not the only applications of M2M. The application domain of M2M is far more significant and vast than specified in this section.

C. WIRELESS TECHNOLOGY LANDSCAPE FOR M2M AREA NETWORK
The realization of a stable M2M system requires reliable and robust communication among MTCDs, For that, it is desirable to integrate the cellular network infrastructure to the M2M system [3], [58]. The cellular network provides reliable and stable transmission with comprehensive coverage and also supports the mobility of devices. As most of the devices in M2M are stationary and periodically send small data packets using only cellular networks for communication among MTCDs is not efficient because the cellular system is primarily developed for H2H communication with the support of mobility of devices and large data packets transmission [9], [59]- [61]. Besides the cellular network, there are various other wireless technologies available for M2M communication such as WAN, PAN, Bluetooth, Low-Power WiFi, ZigBee, UWB. UWB provides a high data rate for indoor communication such as surveillance systems [62]. However, most of the MTCDs are battery-operated, so it is beneficial to use low power technologies for M2M communication like Bluetooth 4.0, ZigBee (IEEE 802.14q), WAN (IEEE 802.11ah). Low -Power Wide Area Network (LP-WAN) provides not only small power communication but also a wide distribution of MTCDs in a large area [9], [33], [63].

D. M2M COMMUNICATION STANDARDIZATION EFFORTS
Various research exercises on M2M were carried out by various endeavors and establishments across the globe. In the past, many telecommunication operators proposed commercial solutions for M2M communication in different parts of the world. These solutions were application-specific and were introduced as vertical M2M architecture or tools for a particular M2M application [64], [65]. For the scalable development of M2M communication, standardization bodies felt the requirement for generic horizontal M2M architecture as a common platform for M2M application development. In Aug 2010, ETSI published the first listing of general provisions for M2M service, followed by the functional requirements for M2M communication [59], [66]. A cooperative effort of seven standardization institutes of the world created a unique partnership for the standardization of M2M in 2012 called oneM2M. In the 3 rd Generation Partnership Project (3GPP) release 8 and 9, the first standard was included to support M2M communication. More specific methods have been proposed in 3GPP release 12 for M2M communication such as privacy, power control, group management, and service maintenance [9], [67], [68].

E. WHY IS M2M SCHEDULING IN LTE IMPORTANT?
LTE is a cellular communication standard that offers high bandwidth, mobility as well as flexibility to accommodate for varying requirements of UEs. M2M communication typically consists of low bandwidth, bursts of data with a different set of QoS requirements from their H2H counterparts. M2M communication is predominantly uplink-based and VOLUME 9, 2021 contends with the uplink H2H traffic. There is a variety of mission-critical applications of M2M communication that need to be prioritized over H2H, and this contention is an issue that the uplink scheduler has to manage [69], [70].
As the M2M communication system has a massive number of devices with varying QoS requirements, therefore designing scheduling schemes that support machine type communication over the LTE network is a challenging task [71]. A massive number of MTC devices are infrequently sending varying sizes of data packets; the LTE bandwidth offers a limited number of physical resources and is optimized for H2H communication [72], [73]. Therefore, it is required to design a solution for M2M communication, which gives optimum utilization of the available physical resources while satisfying unique QoS requirements of M2M communication.
The following constraints should take into consideration to design scheduling schemes for M2M communication [74]; (i) Scaling of Scheduling, as the M2M communication system has a massive number of connected MTCDs, (ii) Standardization and backward compatibility of the scheme, (iii) Capacity limitation (Power and Processing) of MTCDs, and (iv) QoS requirements of MTCDs [75], [76].

III. LTE COMMUNICATION
LTE is a cellular communication standard for mobile devices. It provides an efficient high-speed transmission up to 50 Mbps data-rate in uplink direction and 100 Mbps data-rate in downlink direction at a reduced cost per bit. The LTE communication is based on Orthogonal Frequency Division Multiplexing (OFDM) technology and uses Orthogonal Frequency Division Multiple Access (OFDMA) modulation scheme for downlink transmission and Single Carrier -Frequency Division Multiple Access (SC-FDMA) modulation scheme for uplink transmission. Moreover LTE provides better resource sharing and lower interference than the previous generations of cellular communication [74], [77].

A. LTE NETWORK ARCHITECTURE
The LTE network architecture consists of evolved Node Base (eNodeB or eNB), UEs, and core-network, called System Architecture Evolution (SAE), as described in Fig. 5. The SAE primarily consists of the following components: the Mobility Management Entity (MME), the Serving-Gateway (S-GW), the Packet Data Network Gateway (PDN-GW/ P-GW), and the Home Subscriber Server (HSS). The SAE core, known as Evolved Packet Core (EPC), provides multiple services like authentication, mobility management, setting up of bearers, and control of QoS parameters [78], [79]. The eNodeB is responsible for connecting the UEs to this core network. UE or user equipment is a user or a machine that connects to the eNodeB to access the network [80]- [82].
The primary duties of an eNB are radio resource management, encryption and compression of IP data packets, and selection of an MME. The eNB performs radio resource scheduling at its MAC. The MME is part of the core network that deals with user authentication, session management, and mobility management. This entity keeps track of the user device, and there can only be one MME connected to a device at a time. The S-GW takes care of the data packet routing, forwarding, and manages mobility between LTE and other networks [83]- [85]. This component also allows for the replication of user data for lawful interception. The P-GW provides a facility to connect the UE with an external network i.e., the Internet. The P-GW also enforces the charging policy and allows for packet analysis or interception. The HSS is the master database for all users, and it is typically stored in a single node. This component helps in the authentication and authorization of users for the services offered by the network [34], [80].

B. LTE PACKET SCHEDULER STRUCTURE
In any cellular communication technology such as LTE, multiple devices contends for limited resources offered by the network operator's infrastructure. The distribution of these resources to numerous devices over the radio channel consists of assigning time slots and frequency channels to these devices. This process of assigning time and frequency to devices is called radio resource scheduling. The algorithms used for this allocation are instrumental in providing optimal services to end-users and applications [74], [77]. Packet scheduling is performed by the packet scheduler at eNB. Fig. 6 shows the functional diagram of the LTE packet scheduler. The scheduler performs the scheduling task in both the time domain and frequency domain. In the first phase, Time Domain Packet Scheduler (TDPS) scheduler selects a sufficient number of devices that can be assigned the resources [86], [87]. The selection of the devices by the TDPS scheduler is performed based on some criteria such as device priority, channel quality, Buffer Status Report (BSR), etc. After selecting eligible UEs, TDPS passes the list (list of RNTIs) of selected UEs to the Frequency Domain Packet Scheduler (FDPS) for the further resource allocation process. The FDPS assigns the physical resources to UEs as per the device and channel status and device requirements [71], [80], [88].
LTE packet scheduler also performs the task related to the Hybrid Automatic Repeat Request (HARQ) management for failed packet transmission, link adaptation based on the packet's feedback (ACKs/NACKs) and Channel Quality Index (CQI) to adjust the transmission rate, transmit power level, Modulation and Coding Schemes (MCS) for error-free transmission. Packet scheduler also receive inputs about QoS, BSR, and Medium Access Control (MAC) & Radio Link Control (RLC) information to perform efficient scheduling decisions [74], [80].

C. LTE RADIO FRAME STRUCTURE
LTE communication uses OFDMA in the downlink and SC-FDMA in the uplink channel. The data in both the uplink and downlink is transmitted as frames of 10ms duration, as shown in Fig. 7. Each frame is further divided into 10 subframes of length 1ms each. The duration of the subframe is known as the transmission time interval (TTI), and each such subframe is further divided into two slots of 0.5ms duration each. The resource units are allocated in slots of 0.5ms long in the time domain and 180KHz bandwidth in frequency domain [89]- [91]. The block of 0.5 ms long in the time domain and 180KHz wide in the frequency domain is called Physical Resource Block (PRB). The PRB is the minimum unit that can be allocated to a UE, and resources are allocated in multiple PRBs. Each PRB is a grid of 12*6 or 12*7 Resource Element (RE) comprises of 12 subcarriers of 15KHz each in the frequency domain and 6 (extended CP) or 7 (Normal CP) symbols in the time domain [74], [80], [92].
The number of PRBs in uplink is in the range of 6 to 100 depending on the bandwidth. LTE provides a facility of flexible bandwidth from 1.4 MHz to 20 MHz. Thus the bandwidth of 1.4 MHz provides 6 PRBs of 0.5ms*180KHz, and bandwidth 20MHz provides 100 PRBs of 0.5ms*180KHz. The most basic modulation unit is a resource element, a single block of 12*7 grid of a PRB, and contains one symbol of 15khz. Each resource element may contain 2 or more bits depending upon the modulation and coding scheme (2 bits in QPSK, 4 bits in 16QAM) [31], [93], [94].

D. LTE RESOURCE SCHEDULING
The LTE packet scheduler is a MAC layer functionality of eNB, which is responsible for packet scheduling and physical resource sharing decisions. When a UE sends a scheduling request (SR) to eNB, the packet scheduler assigns the required resources to that UE based on the received VOLUME 9, 2021 information from UE and network such as BSR, a sounding reference signal (SRS), and available resource [95]. Physical resource scheduling decisions depend on various attributes such as QoS attributes, CQI, fairness, energy efficiency, spectral efficiency as per the objective of the scheduling objective. Fig. 8 shows schematic diagram of LTE scheduling process [96]- [98].
Whenever a UE or a machine has data in its buffer for transmission, it sends a BSR packet as an uplink resource scheduling request to the eNB over Physical Uplink Control Channel (PUCCH). BSR reporting interval can be configured to send periodically or data availability in UE's buffer [99]- [101]. Upon receiving scheduling requests over PUCCH from UE, the scheduler allocates m available physical resources to n requesting UEs, using a specified algorithm according to received reference signals and algorithm's objective and sends scheduling grant information in Downlink Control Information-0 (DCI0) format to UE. The scheduling grant contains information about MCS, frame number, transmit power, etc. If a UE i received scheduling grant in n nt frame, then the UE i can sends data in (n + 4) th frame over Physical Uplink Shared Channel (PUSCH) [102]- [106]. PUCCH and PUSCH are logical channels in the MAC layer, as shown in Fig. 8. LTE resource scheduling is divided into two categories. The first is dynamic or channel-dependent, and the other is static or channel-independent scheduling. Dynamic scheduling considers the channel quality between UE and eNB in scheduling decisions. Whereas static scheduling does not consider CQI in scheduling decisions. In LTE, 15 CQIs are defined by 3GPP, ranging from 1 to 15. CQI values are utilized to decide the modulation and coding rate for the transmission to achieve a lower Bit Error Rate (BER). Table 1 shows the 4 bit CQI table for LTE [107], [108].
LTE scheduling mechanism follows some constraints enforced by the LTE standard itself or by the limitation of the communicating UE. Some scheduling constraints and limitations are as following [80], [102], [109].
• Resource Exclusiveness and Contiguity:-A single PRB can only be allocated to a unique UE, and allocation of PRBs should be continued in the frequency domain for where R i,j are PRB allocated to UE i and R i ,j are PRB allocated to the UE i where R i,j refers to PRB j is allocated to UE i.
• Power Constraint:-The power assign to the transmission channel P TX ,i should be more than the minimum power P min,j required for transmission with assigned resources i and cannot exceed the total power P T ,i of UE i.
A UE i should have equal power level for all PRBs allocated to the that UE in a TTI as follows.
• Throughput Constraint:-Achieved throughput for a UE i should be greater than minimum required Guaranteed Bit Rate (GBR) T GBR,i and should be less than the maximum allowable throughput T max in j th TTI [80].
where n RE,i is the number of resource elements allocated to UE i, and b sym is bits per symbol.
• Delay Constraint:-Achieved packet delay D i for UE i should be less than the delay budget D B,i .
• Maximize Connected UEs:-The scheduling schemes should maximize the connected UE UE N ,K in k th TTI. • Maximum Resource:-The allocated resource R max,i to UE i cannot exceed the total available resources R max,ava .

E. CHALLENGES IN LTE SCHEDULING
There are many factors based on which the quality of allocation can be ascertained; namely, fairness, throughput, the fulfillment of QoS requirements, handling the massive number of devices, prevention of starvation, etc. [71], [110], [111]. There is a growing need to connect machines that communicate without human intervention. These machines communicate predominantly in the uplink direction and have a different set of QoS requirements than the H2H communication. The scheduling of M2M alongside H2H communication is a challenging problem [72], [112], [113].

IV. CLASSIFICATION OF SCHEDULING TECHNIQUES
In this section, we discuss selected physical resource scheduling techniques in the LTE network for M2M and H2H communication. All the scheduling techniques are classified depending on the focused objective of the scheduling. Fig. 9 shows the taxonomy of resource scheduling techniques.
A. EFFICIENCY-BASED TECHNIQUES M2M devices have low processing capabilities and limited battery life. The majority of these devices are deployed with fixed batteries. Some of these devices also serve real-time mission-critical applications. Due to this nature of M2M communications, suitable radio resource allocation techniques are required to provide the energy efficiency and throughput needed for these devices. These can be classified into two categories based on what they optimize: namely energy efficient, spectrally efficient, and throughput focused resource scheduling techniques [114]- [117].

1) ENERGY EFFICIENT TECHNIQUES
These algorithms aim to minimize energy consumption in MTCDs individually or for the whole network. One approach is to give preference in the allocation of resource blocks to the MTCD that has the best channel quality to increase throughput and decrease power consumption. The transmission rate depends on the quality of the channel between the UE and eNB [103], [118], [119]. For the poor channel condition, a sufficient data rate can be achieved using high transmit power (P tx ) and using MCS such as 16QAM/64QAM. Thus the transmit power P tx of UE can be seen as one of the link adaptation schemes. Some applications can work efficiently with flexible data rates. For such cases, the energy efficiency of a UE can be increased with a lower data rate. In practice, the data rate of radio-link between the UE and eNB is controlled by Modulation and Coding rate [106], [120], [121]. Fig. 10 shows the relationship among the transmit power P tx , data rate T tx , and transmission channel quality. The LTE network provides two methods to control the UEs' power consumption: ''Closed-Loop Power Control'' and ''Open-Loop Power Control.'' The Closed-Loop power control mechanism uses the feedback provided by the UEs to the eNB through sounding reference signals [122]. These SR signals give information about the path gain and shadowing through the path between the UE and the eNB and are used to calculate the Signal-to-Inference plus Noise Ratio (SINR), which is employed to make decisions about the MCS and transmit power required for the transmission. MCS affects the amount of information transmitted per transmission and the power consumption [73], [106]. The power-control for the PUCCH is defined as follows.
And the power-control for PUSCH is defined as follows.
where; P Tx max,c is per-carrier maximum transmit power, P PUSCH ,c is allocated power for PUSCH over carrier c, P 0,PUCCH is the target received power, P PUCCH is allocated power for PUCCH, P 0,PUSCH is cell-specific parameter, PL DL is downlink path loss, α is partial path-loss compensation, Format is P tx power offset, and δ is explicit power-control commands. The MCS shows the requirements of different transmit power level for the different MCS. The term 10. log(M ) reflects the power required per resource block. Larger resource block size required more power to transmit data. Thus, (9) and (10) show that the transmission rate and resource block size affect the power consumption of transmitter [73].
Discontinuous Reception (DRx) scheme is also used to improve the UE's energy efficiency, in which UE only consumes power when it is transmitting or receiving data to or from eNB; otherwise, it gets in sleep mode and consumes very little power. A UE can be configured to monitor PDCCH discontinuously. DRx can be set in two ways Long DRx and Short DRxas per the requirements of the UE's application. DRx scheme maintains various timers such as on duration timer, DRx inactivity timer, and DRx re-transmission timer. DRx mechanism set the time offset and set or reset the timers as per the configurations [73], [109]. A typical DRx scheme with PDCCH reception is shown in Fig. 11 -Rekhissa et al. [123] proposed two energy-efficient uplink allocation strategies in H2H / M2M co-existence scenario by modifying Carrier-By-Carrier (CBC) and Recursive Maximum Expansion (RME) algorithms for UEs as well as MTCDs. The authors define UE metrics for each RBs and allocate RBS such that RB i is allocated to the UE, which has the highest metric for the i th RB. CBC approach consists of choosing the best CQI and allocating the corresponding RB to the MTCD, this process of allocation is repeated until no one UE is remaining or all RBs are assigned. RME recursively expands allocation toward the left/right of an allocated RB for a device.
Azari et al. [124] model energy consumption and network lifetimes based on transmission and circuit energy consumption and proposed an algorithm that maximizes network lifetime by allocating devices that have the most effect on network lifetime first. The authors define network lifetime as shortest, longest, and average individual lifetime and expected lifetime metric as follows.
where E(t) is the remaining power, T is reporting interval, ε s is static power use, and ε d is the average power of UE.
The authors modeled the problem as Min-Max optimization and proposed variations of this algorithm that work with limited Channel State Information (CSI). Results demonstrate that the spectral efficiency and energy efficiency show an inverse trend, i.e., Increasing data sent in each resource block increases spectral efficiency and decreases energy efficiency.
In [125], the authors modeled machines' energy consumption as a constraint minimization problem and defined it as follows.
where E n (f ) is a function of the device's power consumption in data transmission state and signal transmission state. When the device is not in any transmission state, it goes into a sleep state. The authors consider devices as sensory nodes and proposed two energy-efficient scheduling algorithms.
The first algorithm is used when the distance between eNB and device is long, it schedules data that has short deadlines first and tries all possible allocations of RBs to reduce power consumption. The second algorithm minimizes the number of active sub-frames to achieve efficiency and is used when the distance between the device and eNB is short. The algorithms were compared with EDF, WF 2 Q, and Chen's algorithm and were found to consume less energy with satisfactory fairness and scheduling success ratios.
In [61], Azari et al. proposed an energy-efficient scheme to enhance the lifetime of the network using an optimum size cluster. The expected lifetime of a cluster is defined as the ratio of remaining power to the average power consumption of nodes in each duty cycle and is expressed as follows.
where E 0 is the remaining power, g is cluster size, E h is the average power consumption of cluster head, and E m is the average power consumption of other devices in the cluster. The authors reduce the energy consumption by selecting an optimum value for g. And proposed a distributed clustering scheme for massive M2M devices by modeling power consumption and creating clusters of optimal size. The authors also proposed a lifetime-aware scheduling technique that maximizes network lifetime. Results indicate that this technique consumes less energy than standard scheduling schemes.
In [126], the authors proposed an energy-efficient scheduling technique for small data transmissions in an LTE network. The proposed algorithm selects an optimal MCS according to the payload size to achieve energy efficiency. The authors defined the energy efficiency of a UE as the ratio of the number of transmitted payloads bits L by UE to the energy consumed by the UE in the transmission E T and are calculated as follows.
The simulation results show that their approach maximizes the battery lifetime of MTCDs. The authors also suggested a simple PRB allocation in which all necessary PRBs are allocated to send the entire packet, which maximizes energy efficiency. In [107], the problem is reduced to an NP-hard mixed-integer linear fractional programming problem consisting of MCS assignment, allocation of resources, and power, and data scheduling. The authors achieved the global optimum using Charnes-Cooper transformation and another technique called Glover linearization scheme to obtain the global optimum. The authors compared the performance of their technique with Greedy and EES [125], and results indicate that their technique outperforms both in terms of packet dropping ratio and provides optimal energy efficiency when the number of resource blocks is limited.

2) THROUGHPUT AND SPECTRAL EFFICIENT TECHNIQUES
The spectral efficiency of a radio link is defined as the achieved data rate is over a fixed channel bandwidth. Spectral-efficiency is also termed as normalized throughput and measured in bits/second/Hz [131], [132]. Spectral efficiency and throughput efficiency are proportionally related to each other. It can be enhanced through resource optimization, spectrum sharing among multiple users [133], optimum allocation of MCS, and optimum resource grid size. These strategies are also related to the device's utility. The parameters which broadly affect the spectral efficiency of a radio link are MCS and SNR. Higher MCS and SNR give higher spectral efficiency [73], [134], [135]. 3GPP defines 15 MCS indexes ranging from 1 to 15. For a selected MCS index, a considerably sufficient SNIR is required at the receiver to maintain the BER acceptable. In LTE with Adaptive Modulation and Coding (AMC) the normalized throughput (T N ) is given as follows.
Theoretical throughput for LTE network can be defined as follows.
T the (Bits/ms/TTI ) = PRBs * REs * S TTI * Bits MCS (16) where PRBs is the number of resource blocks in a given bandwidth, REs is the number of resource elements in a resource block, and S TTI is the number of slots per TTI, and Bits MCS is the number of bits per symbol. Fig. 12 shows the relationship among the SNR, MCS, and Throughput. Multiple approaches have been proposed for resource scheduling to increase throughput and spectralefficiency. For example, Yaacoub et al. [127] proposed a resource scheduling scheme to increase spectral efficiency and define UE's utility as a function of achievable throughput and the number of PRBs as follows.
where R k and I RB,k are achievable throughput and allocation of RBs to the user k. The authors proposed an algorithm that greedily allocates each RB to a corresponding user, which causes a maximum increase in throughput. In [108], Lin et al.
proposed a channel aware and buffer aware technique that sets priorities based on CQI and BSR values. The technique performs better compared to proportional fairness, purely opportunistic, and round-robin against fairness, throughput, and packet loss probability. Alawi et al. [45] proposed a scheme to meet the minimum rate and delay requirements of the user by considering MAC and physical layer information. The authors used the game-theoretic approach by applying two  cooperative games nontransferable utility (NTU) and Transferable utility(TU). The authors use the Nash bargaining solution for NTU and a coalition-based method for solving TU.
Similarly, Wang et al. [128] proposed a Nash Bargainingbased game-theoretic model for optimal resource allocation to maximize throughput as per the QoS of UEs and MTCDs. The authors divide the problem into two sub-problems. The first is channel allocation and the second sub-problem is power allocation. The authors model the channel allocation problem as a matching problem between UEs and MTDCs, where MTCDs (max 2) share channels with a UE. The authors use an Exhausted algorithm and KM algorithm to solve this matching problem to maximize unit and system earnings. According to the paper, maximizing unit earning and system earning is equivalent to reducing interference in the common channel. The power allocation problem is solved by restricting the power of MTCDs to a threshold value. The UEs are allocated power to maximize throughput using the Lagrangian multiplier method. Safdar et al. [129] proposed another approach where the authors use both cooperative and non-cooperative games for the femtocell environment.
Tseng et al. [130] proposed a genetic-algorithm-based technique that uses binary bit chromosome mutation based on fitness values. The fitness value is defined using resource block pairs. The selection procedure for next-generation parents is carried out using two methods: roulette wheel selection, which is similar to the Russian roulette in which larger blocks are allocated to chromosomes with larger fitness values, and Tournament selection method in which random mating is performed, and the best performing offspring becomes the subsequent parent. The procedure continues until either the desired convergent rate is obtained or the number of generations reaches a threshold number. The authors compare their algorithm against the random allocation method, and their algorithm performed better in terms of throughput and packet service rates over the range of N u (number of users).

B. GROUP-BASED TECHNIQUES
M2M communication has a massive number of connected devices and diverse traffic patterns. In some cases, data (i.e., sensory data) need not send immediately to the server due to high correlation or delay tolerance nature [136]. So the data can be preprocessed at intermediary nodes (i.e., MTCGs) to reduce traffic and energy consumption of MTCDs through data aggregation and preprocessing [30], [31], [137]. For example, a temperature sensor sending temperature readings (T S ) every 30 seconds, but the temperature up to a threshold (T thr ) is acceptable. Then it is unnecessary to send readings if T s < T thr . These intermediary nodes are known as MTCGs or Aggregators [138], [139]. The main objectives of group communication in the LTE network are as follows.
• Data Aggregation:-When multiple MTCDs transmit small packets and delay-tolerant co-related data then the data packets from different MTCDs can be aggregated at an intermediary node and sends together to save bandwidth and to reduce frequent scheduling [30].
• Connectivity Support:-Gateways, which have the dual connectivity facility, can provide connectivity support to non-LTE devices [140].
• Reduced Network Traffic:-Unnecessary data transmission can be avoided through preprocessing of data at the gateway to reduce the network traffic, for example sending only the temperature readings which are above the threshold value [86].
• Reduced Energy Consumption:-Energy consumption can be optimized by limiting the frequency of packet transmission and reducing transmission-time [93].
• QoS Support for MTCDs:-Gateway-based communication can provide QoS support to MTCDs through preprocessing and intelligent decision approach [31]. The grouping of devices is generally based on the characteristics and requirements of devices, Which can be classified as the following criteria [89], [141].
• QoS Requirements of Devices:-The MTCDs having the same QoS requirements can be grouped to support a QoS-aware scheduling decision [142]. Periodic, Burst, Frequent) can be grouped to avoid the frequent scheduling task [143].
• Payload Size:-To support the data aggregation approach MTCDs can be grouped based on the payload size (i.e, Small/Medium/Large).
• Physical Layer Parameter:-MTCDs can be grouped based on the physical layer properties (i.e., Channel Quality, Tx power) to support better resource utilization.
• Locality of Device:-MTCDs can be grouped based on the distance from different gateways. This approach can improve energy efficiency through short-range communication. There are two types of group communication first is relay-based and cluster-based. In relay-based communication, relay serves as an eNB for the devices that are not directly able to communicate to the eNB. A UE can with higher capacity work as a relay. In cluster-based communication, a UE act as a cluster head and is responsible for forwarding data from UE to eNB and vice-versa [30], [141]. Fig. 13 shows typical group communication paradigm in LTE network. The research community has extensive work in group-based resource scheduling for the LTE network, mostly focused on energy consumption and data traffic.
In [141], Songsong et al. proposed a proportional fairness algorithm using user grouping. The algorithm groups devices based on the number of carriers they can be assigned. All the carriers having the same bandwidth are grouped into L aggregated carriers. Each aggregated carrier has V resource blocks. The power allocated to a carrier in a group carrier is given as P c = V /P T . Results indicate that this algorithm provides better fairness than proportional fairness (PF) with degradation in throughput. In [93] Ho et al. proposed an energy-conserving 2-hop transmission-based allocation scheme. The author defines the maximum achievable bit rate for each sub-carrier as follows.
The author get power for the group coordinator by maximizing the Eq. (18) through iterative process. The authors also propose optimal number of coordinators in a group. Gotsis et al. [89] proposed a queue-aware QoS based scheduling technique. The UEs are grouped in L clusters based on the individual QoS requirements of UEs. Physical resources are shared among clusters rather directly to individual UE. The authors defined a probabilistic model such that the probability of violation δ of maximum delay threshold is given as follows.
where, W q is the experienced packet delay and θ is the QoS factor of the group. Xu et al. [30] propose a group-based scheme for Random Access (RA) and uplink scheduling procedure. A UE (M ) can be a member of group (k) if it satisfy the following condition - where n TB is the number of resource blocks allocated for transmission, C TB , and C MTCD are the capacity of resource block and UE, respectively. The authors consider it a priority for a group of UE for uplink scheduling. The authors also propose a group paging scheme for resource allocation that improves delay and access probability. Frank et al. [86] defined a scheme to reduce the effects of cell-edge interference, which involves multiple adjacent base stations to communicate multi-cellular CSI reports through a fast back-haul network to a central scheduling unit. The authors proposed an interference aware uplink scheduling algorithm based on a proportional fairness approach to avoid inter-cell interference.
Hsu et al. [31] proposed an enhanced cooperative access class barring and traffic adaptive radio resource management (ECACB + TARRM) for M2M devices. This technique builds upon enhanced cooperative access class barring (CACB) to which the authors add support for UEs and TRRM. TRRM implies that UEs and MTCs use different PRBs, preambles, and MTC devices, which are clustered based on their data rates and random access rates. To determine better parameters for access class barring, the number of MTC devices is used as a factor over the factors used by CACB, that connect to an eNB.
Bayat et al. [67] proposed a distributed coalition forming algorithm that involves the rules called ''merge-andsplit.'' The authors used data aggregation for machine-type devices that has different delay requirements and proposed a game-theoretic approach using coalition games. The algorithm allows MTCDs to organize into groups independently with each group head handling data to and from the machines in each group.

C. QUALITY OF SERVICE BASED TECHNIQUES
The M2M communication is different from the regular H2H communication in terms of the number of connected devices, packet size, frequency of data transmission, a broad range of delay budgets, throughput requirements, priority, etc. ETSI defines QoS class identifier for M2M communications in LTE [110], [144], [145], as shown in Table 4. As the M2M communication system has a massive number of devices with varying QoS requirements, therefore designing scheduling schemes for M2M communication over the cellular network is a challenging task [71], [146]. A massive number of MTC devices are infrequently sending varying sizes of data packets; the LTE bandwidth offers a limited number of physical resources and that are optimized for H2H communication [72]. Therefore, it is required to design a solution for M2M communication, which gives optimum utilization of the available physical resources while satisfying unique QoS requirements of M2M communication [98], [147], [148].
AI-Rawi et al. [149] proposed an opportunistic channel adaptive radio resource scheduling algorithm for dynamic traffic patterns based on the buffer sizes of users. The scheduler estimates the expected rate µ n,i of device i if the resource block n is allocated to the device with given estimated throughput of x i as follows.
The scheduler maximizes the Eq. (21) to find optimal allocation. The authors evaluated pruning, which is a process of recovering weaker bands for use by other UEs. The authors evaluated the effect of delays in receiving buffer information and concluded that it would result in better fairness, whereas limited buffer information would lead to inefficient resource usage. Delgado et al. [82] defined a utility function (R k |S k ) as a function of throughput R k and set of allotted resources S k of device k and maximize the utility of the device for optimal  allocation of resources as follows.
The authors proposed two algorithms that aim to reduce delay while having a minimum throughput constraint. The authors use two highly scalable greedy heuristics based on the problem. Afrin et al. [26] defined an urgency metric U i for the device i as a function of deadline d i and BSR index B i as follows.
The devices are selected based on urgency metric U i to improve the satisfaction of delay requirements. This approach allows the eNB to know the age of the oldest packet in their buffer using a new MAC control field in the MPDU. Afrin et al. [151] proposed a buffer-based adaptive semi-persistent scheduling (SPS) scheme, which does not have the same overheads as dynamic scheduling while offering the same flexibility. The authors examine the influence of semi-persistent scheduling on the QoS and compare it with fixed allocation SPS schemes. Safa et al. [150] proposed a technique that aims to satisfy delay requirements of M2M devices. The authors defined a QoS aware allocation metric γ c i (t) of a UE i for QoS introducer α i (t) of UE as follows.
Their technique gives priority to the UEs having high priority data while not starving others. Mata et al. [76] proposed a genetic algorithm based approach to optimize for video streaming with focus on video chat. The authors defined PF metric λ m n (t) as a ratio of instantaneously achieved data rate r m n (t) to the long-term rate R n (t) for a user n with assigned resources m over a period of time t as follows.
The authors also defined a metric based on the HoL of the packets that resides in the UE's buffer. Maia et al. [112] proposed an extension to the QoS classes for M2M in two groups, event-based and time-based. The authors try to control the effect of M2M communication on H2H communication by calculating the current demand of resources for H2H communication as a ratio of average resource allocated to the average buffer size and expressed as follows.
Based on the current demand, resources are shared among H2H and M2M devices. Maia et al. [113] proposed a genetic algorithm-based method and introduced a new scheme of initialization, crossover, mutation, and a QoS aware fitness function. Kumar et al. [99] proposed a multi-class scheduler for MTCDs with different delay requirements. The authors classify M2M data into periodic update (PU) and event-driven (EU). The authors aimed to maximize utility by using heuristics and a sigmoid-based utility function for each device type. Their algorithm prioritizes ED data over PU data as long as PU deadlines are being met and makes sure that congestion due to failed updates is reduced. Kumar et al. [100] proposed a delay in optimal scheduling strategy in which multiple M2M devices communicate with an application server from multiple M2M aggregators. Giluka et al. [87] proposed a classification and prioritization scheme of M2M and H2H service flows based on QoS requirements. In a given class, H2H devices are given higher priority, while a limit is set a maximum limit for the assignment of radio resource blocks assigned to MTCDs. The authors define a utility of a QoS class C i as follows.
where S(H ) and S(M ) are the satisfiability function of H2H and M2M communication requests. Agdhmadi et al. [152] proposed a scheme to provide QoS to Guaranteed Bit Rate (GBR) services based on QCI and using priorities for M2M. Erpek et al. [153] proposed a scheme that gives priority to delay-bound traffic over delay-sensitive traffic. The authors implement a utility proportional fairness policy based on the same. Brown et al. [68] proposed a predictive resource allocation scheme using on Maximum Likely-hood Estimation (MLHE) and defined MLHE as follows. where, The authors used inter-sensor propagation time to determine when it will reach downstream sensors. This approach allows sensors to send fewer scheduling requests that reduce traffic and delay. Abdelsadek et al. [22] proposed a scheme, which considers the scheduler as M/D/1 queues model. The achieved throughput for the UEs u is given as follows.
where λ u and µ u are the average arrival rate and average service rate, respectively. The authors improved the computational efficiency of the optimization problem. Alaa et al. [35] proposed a non-preemptive queuing model and investigated the scheduling performance for different QoS classes of M2M and H2H devices with dynamic access grant time interval scheduling. The authors use the M/G/c/c model to improve bandwidth utilization and QoS satisfaction. Karadag et al. [98] proposed an optimization approach for MTCD transmissions considering the repetitive nature of these transmissions. The authors proposed semi-persistent scheduling and implement using the Depth-First approach and minimum frequency-fit approach to reduce the use of frequency bands used by MTCDs while maintaining delay requirements. The authors proposed a heuristic algorithm in polynomial time with fixed priority assignments to solve this problem.
Ouaissa et al. [116] proposed a hybrid model of RR, First Maximum Expansion, and Maximum Throughput. Abdalla et al. [2] proposed a technique that aims to retain the Quality-of-Experience (QoE) of UEs while processing the message requests of M2M devices. The authors proposed a new set of QCIs for M2M to ensure end-to-end QoS. Hassebo et al. [91] proposed a technique that aims to manage QoS requirements of M2M devices along with massive access while protecting H2H devices from a lapse in service quality. The authors used a semi-persistent approach for scheduling a large number of MTCDs while using typical dynamic scheduling for H2H devices.

D. HYBRID/MULTI-OBJECTIVE TECHNIQUES
In this category, we consider the scheduling algorithms which focus on more than one objective in combination with any scheduling objective like a priority, QoS, throughput, energy, and fairness. For example, Elhamy et al. [84] proposed a technique called ''BAT '' that aims to balance throughput and delay requirements. This technique is a hybrid technique that allocates M2M resources and UE at the same time interval using an RME like expansion method. Maia et al. [113] proposed a technique that aims to reduce congestion, satisfy QoS requirements, and ensure fairness of the allocation of M2M devices while minimizing the effect on H2H traffic. Their algorithm uses a state transition function with three states to evaluate the probability of allocation to M2M devices while optimizing for the said factors. Kwan et al. [101] proposed the classic throughput and fairness balancing approach called PF scheduler, which aimed to improve the fairness of Max-Rate scheduler while taking some loss in throughput. AlQahtani et al. [54] proposed a scheduling technique that borrows from RR and Best-CQI (B-CQI) to solve the fairness and throughput trade-off. RR provides ideal fairness, whereas B-CQI provides high data rates with weak fairness. Results by testing against RR and B-CQI indicated that the technique provides a balance between fairness and throughput. AlQahtani et al. [60] proposed a queuing-model based access strategy in the case of H2H and M2M coexistence. The authors evaluated the system performance using a continuous-time Markov-Chain model. Results indicate that this technique increases the overall resource utilization while decreasing blocking probability. Mardani et al. [114] proposed a technique that aims to minimize energy consumption while maintaining QoS requirements of H2H devices. The authors used a fuzzy logic-based controller that anticipates and manages uncertainties and obtained an optimal bandwidth ratio for each type of service flow. Mardani et al. [115] proposed a technique that aims to maximize throughput, satisfy power budget constraints, and statistical QoS delay requirements. The authors defined the problem as a mixed-integer non-linear problem and proposed a solution using Lagrange multipliers. Aijaz et al. [154] proposed a technique that aims to minimize energy consumption and proposed statistical QoS provisioning for M2M and H2H devices. The authors defined the problem as a Mixed Integer Programming problem of maximizing effective energy efficiency in bits-per-joule capacity. The authors solved this using the Canonical Duality Theory. The authors also proposed another approach, where the authors proposed two low complexity heuristic techniques for the same.
Dawaliby et al. [81] proposed a technique that aims to maximize throughput and reduce delays in the case of LTE-M protocol. The authors model the problem to the 0/1 knapsack problem. Dawaliby et al. [109] proposed a technique that aims to minimize energy consumption while maintaining QoS requirements of M2M devices. The authors employ a cross-layer scheme using a Memetic-based algorithm the authors consider the QoS requirements while minimizing the energy consumption using discontinuous reception switching. Kalil et al. [94] evaluated a genetic algorithm that considers multiple constraints for the uplink scheduling problem. This approach is evaluated against the optimal allocation binary-integer programming problem (BIP). It offers comparable performance for low-population (<300 UEs) to the optimal solution while having comparatively lower time complexity.
Tagarian et al. [155] proposed a technique that aims to minimize energy consumption while maintaining delay QoS requirements of machines. The authors used a gateway-based approach where the use of clustering manages massive access. The optimization problem is solved using genetic algorithms for maximizing energy efficiency. To manage delay, the authors used an existing scheduling approach. Fagan et al. [85] applied a deep learning approach for downlink scheduling. The data set is derived using a genetic algorithm over many simulated random UE data reports and used to train the deep learning network. This approach allowed for approximating the genetic algorithm schedule without the delay of a genetic algorithm. Comsa et al. [75] proposed a scheduling scheme using the q-learning method to dynamically adjust the fairness and system capacity trade-off each transmission time interval. The proposed algorithm decides allocation using CQI for each class of users.
Chen et al. [70] proposed a heuristic technique that aims to minimize the energy consumption of MTCDs while guaranteeing the QoS. The authors minimized MTCDs' energy consumption using lower modulation and coding and spatial reuse. Abrignani et al. [23] considered the problem of improving throughput while reducing resource usage and minimizing Inter-Cell Interference (ICI) in the case of a densely populated heterogeneous network. The authors modeled the problem to a Mixed Integer Linear Programming (MILP). The author's employed a heuristics approach to solving MILP. The algorithm was compared against RR and found to perform better in terms of throughput.
Hamdoun et al. [90] considered an evolutionary game approach to preserve UE's QoS while preserving the battery life of Machine Type Devices (MTDs). Here MTDs are in the same group share spectrum with a UE, which is matched VOLUME 9, 2021 to it. The MTDs switch dynamically from non-cooperative to cooperative strategies. Results indicate that this adaptive technique performs better than a fixed discrete power allocation strategy as well as a non-cooperative strategy in terms of power consumption as well as QoS satisfaction. Salam et al. [156] proposed a technique to improve outage probability, energy efficiency, and system capacity called cooperative data aggregation scheme, which employs fixed data aggregators and mobile data aggregators. These aggregators serve M2M devices with varied QoS requirements. The authors also considered parameters of queuing delay and the number of devices not served in a class. Edema et al. [83] presented a study on existing Fixed access grant time interval (AGTI) and Time-controlled scheduling and proposed a dynamic AGTI scheme based on M2M and H2H traffic intensities focusing on resource utilization and QoS satisfaction. Lin et al. [111] proposed a technique that aims to minimize energy consumption while maintaining QoS requirements of M2M devices. Their algorithm used the concept of Multi-access edge computing. The authors considered packet processing time and travel time in latency calculation.

V. FUTURE RESEARCH DIRECTIONS
Although in the last decade, extensive research has been carried out by endeavors from industry and academia in the field of radio resource scheduling in LTE networks. Researchers have proposed various techniques to improve the efficiency of the scheduling process and optimizes the radio resources. With the emergence of M2M communication and recent developments in cellular technology, i.e., 5G-NR, there are lots of opportunities to work on the radio resource allocation problem. In this section, we provide future research directions for radio resource scheduling from the M2M perspective.

A. EFFICIENCY FOCUSED
Lots of research work exists in the literature about the efficiency-based approaches of resource scheduling in the LTE network for both H2H and M2M communication. As found in the literature, the spectrum is a scarce resource, and most of the MTCDs are battery-operated. The M2M communication system has a massive number of devices. Therefore, there is still a scope of resource allocation schemes for Non-Orthogonal Multiple Access (NOMA) technology, which provides different power labels in resource allocation. NOMA is suitable for devices having various power labels and the co-existence of H2H and M2M communication. Spectrally efficient resource allocation has the opportunity to work with cooperative communication and cognitive radio communication.

B. GROUP-BASED
Cooperative communication and cognitive-radio communication support group communication and inter-cluster/group communication. There is a need to optimize the resource allocation process to work with cooperative and cognitive-radio communication. There is also the possibility of resource allocation with mmWave beamforming to allocate resources to a group of devices efficiently. The massive-Multiple Input Multiple Output (massive-MIMO) is futuristic technology; there is a need to optimize the resource allocation for the MIMO communication model.

C. QOS FOCUSED
Although the QoS classes are well defined in the literature, they all are statically defined. So the resource allocation methodologies can be improved to support the dynamic QoS classes. mmWave can support QoS support with beamforming and relayed communication.

VI. LEARNINGS FROM STUDY
The study provides a comprehensive foundation of radio resource scheduling techniques in the LTE environment from the perspective of M2M communication. This study helps us identify the gaps in existing research work for potential future research in M2M radio resource scheduling and highlights the primary methodologies employed by researchers. Through this literature review, we get an overview of the fundamental architecture and the physical layer concepts of the LTE network. Also, it helps us to understand the basic architecture of M2M communication.
This review helps to understand the following critical aspects of the LTE radio resource allocation process.
• Scheduling Metrics -The scheduling metrics have a vital role in preferring or selecting a particular UE over others while assigning resources to the UEs.
• Scheduling Objectives -The study provides an insight into the scheduling objectives like efficiency, QoS, etc. that were focused on through previous research works.
• Scheduling Methodologies -The study helps to understand the pros and cons of previously employed approaches like game theory, queuing theory, etc. and provides direction towards the possibilities of implementing new methodologies in resource scheduling.
• Constraints and Limitations -In this literature review, we find out the constraints and limitations of LTE (i.e. resource exclusiveness and contiguity) and MTCDs (i.e. power constraint), that draws a boundary for the resource scheduling process.
• Parameters That Affect Scheduling Performance -By comparing different scheduling work, we find out the parameters, i.e., MCS, number of PRBs, etc. that affect the performance of the scheduling methodology.
• Current State of LTE Resource Scheduling -This study helps us get an integrated and synthesized overview of the current state of the LTE scheduling.

VII. CONCLUSION
The study aimed to provide a comprehensive survey and classification of the scheduling strategies proposed by researchers for LTE / LTE-A networks from an M2E perspective. We classified scheduling strategies based on their aim for which the scheduling strategies are optimized. Most of the work in efficiency-based approaches focused on the improvements in energy efficiency and throughput efficiency. However, some work is also focused on spectral efficiency and interference avoidance. In group-based scheduling approaches, most of the work focused on reducing uplink traffic and network coverage to achieve energy efficiency, spectrum reuse, interference control, and QoS support. In QoS-based scheduling approaches, the authors focused on the QoS of devices, mostly delay, and throughput. Some techniques supported the device's QCIs. Some authors worked on defining QCIs for devices based on statistical priorities. In hybrid / multi-objective approaches, authors focused on satisfying multiple objectives in the combination of any purposes mentioned earlier. The authors implemented different modeling approaches, like game-theoretic, graph-theory, genetic-approach, constraint optimization, queuing-theory, greedy-approach, heuristic-methods, weighted-sum, deeplearning, neural-network, and fuzzy-logics, to achieve their objective of the work. He is also an Adjunct Professor with Newcastle University, U.K.; Asia University, Taiwan; King Abdulaziz University, Jeddah, Saudi Arabia; Charles Darwin University, Australia; and the University of Petroleum and Energy Studies, Dehradun, India. He has published more than 400 technical research articles (DBLP: https://dblp.org/pers/hd/k/Kumar_0001:Neeraj) in top-cited journals and conferences which are cited more than 19 000 times from well-known researchers across the globe with current H-index of 76 (Google Scholar). He has guided many research scholars leading to Ph.D. and M.E./M.Tech. His research is supported by the funding from various competitive agencies across the globe. He has also edited/authored ten books with international/national publishers like IET, Springer, Elsevier, and CRC. MAMOUN ALAZAB (Senior Member, IEEE) received the Ph.D. degree in computer science from the School of Science, Information Technology and Engineering, Federation University of Australia. He is currently an Associate Professor with the College of Engineering, IT and Environment, Charles Darwin University, Australia. He is also a cyber-security researcher and practitioner with industry and academic experience. He has more than 300 research papers in many international journals and conferences, and 9 authored/edited books. His research interests include cyber-security and digital forensics of computer systems with a focus on cybercrime detection and prevention. He is the Founding Chair of the IEEE Northern Territory (NT) Subsection.