2017 Selective Green Device Discovery for Device to Device

}@gmail.com Abstract The D2D communication is expected to improve devices’ energy-efficiency, which has become a major requirement of the future wireless network. Before the D2D communication can be performed, the device discovery between devices must be done. The previous works usually only assumed one mode of device discovery, i.e. either use network-assisted (with network supervision) or independent (without network supervision) device. Therefore, we propose a selective device discovery for device-to-device (D2D) communication that can utilize both device discovery modes and maintain devices’ energy-efficiency. Different from previous works, our proposed method selects the best device discovery mode to get the best energy-efficiency. Moreover, to further improve the energy-efficiency, our proposed method also deployed in D2D cluster with multiple cluster heads. The proposed method selects the most suitable mode using thresholds (cluster energy consumption and new device acceptance) and cluster energy expectation. Our experiment result indicates that the proposed method provides lowest energy consumption per new accepted device while compared with schemes with full network-assisted and independent device discovery in low numbers of new device arrival (for the number of new devices arrival = 1 ~

Due to the usage of different modes of D2D communication, selection method to select different D2D modes are needed to adapt to different conditions. There are several works that presented various selective modes for D2D. In [21], the proposed selection of different D2D resource allocation mode is presented. In [22], Floating Band D2D is proposed to select between in-band and out-band D2D communication. In [22], an adaptive mode for D2D in bursty traffic condition is proposed. The adaptive mode assigns best D2D mode based on traffic load and channel occupation time [22]. In [23], a distance-based method to select between the cellular mode and the direct D2D mode is proposed.
However, as the best of our knowledge, the selection procedure for device discovery that considers energy consumption is still underdeveloped. Nevertheless, D2D communication will better meet users' demand if it can be deployed with or without network support. There is a need to develop a selection mechanism to select best modes that can satisfy energy-efficiency, as mentioned in the vision of ultra-dense next generation wireless network [24]. Previous works usually more focused on either network-assisted [15,25] or independent device discovery [14,26,27].
The contributions of our work can be presented as follows. First, we propose a selective green device discovery method, which is a threshold-aware approach to select the best device discovery mode based on cluster's previous energy consumption. Based on several parameters and condition, during the time period, the method may change its nature from eNB-assisted to independent and vice versa. Therefore, the proposed adaptive approaches are expected to take both the advantages of the network-assisted and independent device discovery. Second, the proposed method is implemented in a multi-CH scenario using the cluster head rotation method. Third, we conducted simulations to examine the performance of the proposed method.
The rest of this paper is organized as follows. Section II explains the system model that is being used. In Section III, different modes of device discovery in D2D are defined. In Section IV, the full scope of the proposed method is presented. The results are presented and discussed in Section V. Finally, the conclusion of this paper is drawn in Section VI.

System Model and Formulation
In this work, an LTE-A single cell area with an outdoor dense environment is assumed. A single BS is located at the center of the cell. The orthogonal frequency division multiple access (OFDMA) in the downlink (DL) is assumed for radio access [28]. All D2D devices are capable of communicating in both short range (SR) link for inter-device communication and long range (LR) link in the BS -device communication [28]. The devices which act as CHs is transferring the data content to its respective CM. The parameters and simulation assumptions of this work, which is inspired from [11,20,22] are presented in Table 1. The D2D communication in this work is assumed as data content transfer through shortlink communication between CHs and CMs [11]. The devices are assumed to be spread randomly in the cell area with uniform distribution. All devices are assumed to have similar maximum battery capacity (2900 mAh) and sufficient battery level for performing D2D communication (at least 10% of battery capacity). Moreover, all devices are also capable of forming clusters and can act either CH or CM.
Finally, the energy efficiency of the proposed method is measured as the average energy consumption for discovering a new device. Inspired from [18], we assume that energy efficiency is defined as the size of distributed data content per energy consumed. Therefore, the energy efficiency of a D2D session of the cluster can be expressed as is the device discovery energy consumption and is the number of accepted new devices of d a D2D session of the cluster .

Device Discovery Modes
As described in [15] there are four basic steps in the discovery process. First, in "discovery request step", the new device asks permission to announce its discovery signals [15]. Second, in "information step", the network informs monitoring devices about the new device and assigned resource [15]. Third, the discovery signals are transmitted [15]. Finally, in "report match" step, monitoring devices report the network about the discovery of new device [15].
In this work, the network-assisted device discovery is assumed to provide full network assistance and supervision. As presented in Figure 2, the model steps in [15] (Proximity Service discovery of 3GPP [29]) are adapted for network-assisted device discovery to meet the multi-CH D2D clustering scenario our previous work [19]. The PC5 radio interface is employed for interdevice communication [15,29]. First, the new device sends its discovery request to the network via the PC3 radio interface. Second, the network informs devices in the cluster about the new devices and assigned spectral resource via the PC3 radio interface. Third, the new device transmits the discovery signals to existing cluster devices via the PC5 radio interface. Fourth, the cluster devices report the discovery match to the network via the PC3 radio interface. On the other hand, for the independent device discovery, CHs is tasked to take network roles in device discovery. As presented in Figure 3, the model steps in [17] (Proximity Service discovery of 3GPP [29]) are modified for independent device discovery to meet the multi-CH D2D clustering scenario in our previous work [19]. The PC5 radio interface is also utilized for inter-device communication [15,29]. In step 1.a, the new device sends its discovery request to CH via the PC5 radio interface. The CH informs acceptance of the new device to the network via the PC3 radio interface in step 1.b. In step 2.a, the network informs resource allocation to CHs via the PC3 radio interface. The CH forwards information about the new device and assigned the spectral resource to CMs via the PC5 radio interface. In step 3, the new device transmits the discovery signals to existing cluster devices via the PC5 radio interface. In step 4, the CMs report the discovery match to the CH via the PC5 radio interface.
The independent-device discovery allows the CH to take the responsibility in the process of device discovery [19]. However, in this work, due to its compliance to the inband D2D (LTE-A), it is assumed that the network still taking roles for resource assignment (which resulting in step 1.b and 2.a) [19]. Therefore, we formulate device discovery energy consumption in this work as follows. On the one hand, inspired from [11,19,28,30], the cluster energy consumption for networkassisted device discovery can be calculated as where is the new device that selected to join to the cluster in a particular round of cluster head rotation, is the discovery content, indicates CH and CM, and indicates BS. The first part indicates Step 2, the second part indicates Step 3, and the third part indicates Step 4 from Figure 2. Additionally, inspired from [11,19,28,30], the energy consumption of the new device for network-assisted device discovery can be defined as where the first part indicates Step 1, the second part indicates Step 2, and the third part indicates Step 3 from Figure 2. On the other hand, the cluster energy consumption for independent device discovery can be calculated as (4) The CH energy consumption (denoted as ), inspired from [11,19,28,30], can be calculated as where indicates all new devices, indicates only selected new device, indicates CM, indicates CH, and indicates BS. The first part indicates Step 1.a, the second part indicates Step 1.b, the third part indicates Step 2.a, the fourth part indicates Step 2.b, and the fifth part indicates Step 4 from Figure 3. The CM energy consumption (denoted as ), inspired from [11,19,28,30], can be calculated as where the first part indicates Step 2.b, the second part indicates Step 3, and the third part indicates Step 4 from Figure 3. Additionally, inspired from [11,19,28,30], the energy consumption of the new device for independent device discovery can be defined as where the first part indicates Step 1.a, the second part indicates Step 2.b, and the third part indicates Step 3 from Figure 3.

Proposed Method
The proposed selective method is assumed to be deployed in massive device scenario with outdoor ultra-dense network (UDN) condition. In this condition, the device discovery will be frequently performed. Thus, the deployment of energy-efficient device discovery is crucial. The D2D devices also work alongside cellular (LTE-A) devices, although has their own dedicated spectral resource. Additionally, the devices are assumed to be static during the D2D communication. We assume a future UDN scenario, where each device can utilize cellular and D2D communication. However, in this work, we only focus on D2D communication deployment in each device. The data content, which is content that several users has interests in, triggers the initiation of D2D communication ( Figure 4). The data content is also assumed to be transferred via file transfer protocol (FTP). The data content can be defined as video data or other multimedia contents for multicast and broadcast multimedia service (MBMS) [18]. Additionally, the devices can freely form a cluster to distribute data content. Moreover, the multi-CH D2D data transfer, i.e. cluster head rotation method, is utilized. The cluster head rotation, which is a further enhancement of the D2D clustering method, enables multiple CHs to distribute data content sequentially. The cluster head rotation also contains a selection mechanism to select content-distributing CHs.
In Figure 4, based on [18], the D2D framework which implemented in this work is presented. First, D2D communication is assumed to be initiated by devices' common interest of a data content [28,31]. Second, we assume to use both network-assisted and independent D2D that utilize licensed spectrum [18]. Third, we assume both a priori and a posteriori discovery. Accordingly, new devices can join the cluster after and before a particular data transfer in the round . Because of the utilization of clustering (for cluster head rotation method), we assume both long and short range link in this work [18,31]. Long range link is used to communicate between BS and CH. Short range link is used to communicate between CH and CMs.
Moreover, dedicated resource and static devices (no mobility of devices) are assumed for D2D devices [18]. Furthermore, there is no link adaptation that utilized for D2D communication in this work [18]. Next, fixed transmit power is assumed in this work [18]. Finally, for data distribution in D2D communication, multi-CH method (cluster head rotation method [20]) is deployed [18]. In this work, this process (from the D2D initiation until the cluster head rotation) is regarded as one D2D session. In this paper, our objective is to propose an energy-efficient selective method that can both utilize network-assisted and independent device discovery. The process is based on the adaptive method for bursty traffic in [22]. However, the method in [22] is aimed to adapting the D2D resource model according to delay traffic in bursty traffic model scenario. Departing from [22], our method considers D2D cluster energy-efficiency as a constraint to select the status of network support (independent or network-assisted) in device discovery. Therefore, we propose the addition of a selection function in the D2D implementation framework.
In Figure 5, the proposed selection process for the device discovery mode is presented. The process can be further described as follows. First, the previous device discovery mode is identified. Next, the threshold of energy consumption is calculated and compared with previous energy consumption . Moreover, the expected energy consumption ,is calculated. The result is used to determine which device discovery mode that can produce least energy consumption. Finally, the device discovery mode for the next session is determined. Figure 5. Selection process of device discovery mode As presented in the Figure 5, thresholds of cluster energy consumption and rejected devices are utilized. The energy consumption threshold can be expressed as (8) where is energy consumption in cluster at time and is threshold variable for energy consumption. Moreover, the threshold of rejected devices can be defined as (9) where is new devices that coming to join the cluster at the present time and is threshold variable for the rejected devices.Furthermore, the selection of mode with minimum expected energy consumption, which is based on minimum delay selection in [22], can be expressed as where ,and indicates the expectation of cluster energy consumption and the selection prior to the expectation of cluster energy consumption. The proposed method is expected to handle the dynamic of cluster device energy consumption and the possibility of battery drain in massive devices scenario. The expected cluster energy consumption ,is intended to ensure the cluster achieves most efficient energy consumption from both device discovery modes. As presented in Figure 2, the device discovery and data transfer procedure are deployed. In the simulation, we assume that there is no condition change for following device discovery and data transfer procedure.

Results and Discussion
In Figure 6, the device discovery energy consumption of the cluster per accepted new device is presented. In the figure, we can examine the energy consumed for accepting a new 1673 device to join the cluster. For all scenarios, multi-CH clustering with the number of CHs = 5 is deployed. The number of initial device in the cluster is 10. According to our result, the cluster spends 0.0215 J in average to accept a new device for the proposed method. For the fully independent discovery, the cluster spends 0.0230 J in average to accept a new device. The cluster spends 0.0217 J in average to accept a new device for the proposed method for the fully network-assisted discovery. Additionally, the result shows that the proposed selective method has the best performance (lowest device discovery energy consumption per new accepted device) for low number of new device arrival (for the number of new devices arrival = 1 ~ 3). For number of new devices arrival , the energy consumption of the proposed method is lower than independent mode only.  Figure 7, the ratio of selected by the selected device discovery mode is presented. In the figure, the portion of both network-assisted and independent mode is presented. According to our result, the independent mode is selected more compared to the network assisted mode. Using simulation parameters and assumptions as presented in Table 1, the independent mode was selected in 62% of data transmit session on average. On the other hand, the networkassisted mode was selected in 38% of data transmit session on average. In Figure 8, the ratio of new devices acceptance is presented. In the figure, the portion of accepted new devices is compared with rejected new devices. According to our result, average new device acceptance ratio is 89%. In the proposed method, new device acceptance is determined by battery level. Therefore, in this simulation, battery level of new device battery level (denoted by ) is a critical factor in device acceptance. Moreover, the result shows that there is no significant difference for every addition of device per cluster. We observe that this due to the fact that battery level is used as the acceptance criteria. Figure 8. New devices acceptance

Conclusion
In this paper, an energy-efficient selective method of device discovery for D2D communication is proposed. The proposed method is able to meet the research objective by providing a selection to utilize network-assisted and independent device discovery. A multi-CHs D2D method for content distribution, i.e. cluster head rotation method, is also implemented in this work. According to the simulation result, the proposed method proved to have decent energy efficiency compared with a scenario that only deploy network-assisted or independent device discovery. For accepting new device, compared to fully independent and fully networkassisted discovery, the proposed method has lowest device discovery energy consumption (compared to fully independent and fully network-assisted discovery) in low numbers of new device arrival (for the number of new devices arrival = 1 ~ 3).
For future works, a selection mechanism to select the degree of network support in network-assisted device discovery will enhance the selection process. Deployment of indoor scenario within a small coverage area, an effort to comply with ultra dense network consideration, will also be considered. The complexity reduction for selection method will also be considered.