Telecommun Syst DOI 10.1007/s11235-017-0385-1
Joint power allocation and relay selection strategy for 5G network: a step towards green communication Akshita Abrol1 · Rakesh Kumar Jha1 · Sanjeev Jain2 · Preetam Kumar3
© Springer Science+Business Media, LLC 2017
Abstract Green communication has emerged as the most important concept for the next generation networks. Along with improved data rate and capacity, the upcoming 5G networks aim at improving energy efficiency without compromising on the user experience. In this paper, we have used amplify and forward relays in the heterogeneous network topology consisting of low power and high power nodes. A three layered system model for power optimization is discussed using a relay selection strategy for power optimization with the aim to improve energy efficiency of the network. Further, we have used Hidden Markov Model for training and maintaining of base station, relay and SCA with the aim of probabilistic power allocation to client nodes in order to solve the power optimization problem. We have also used adaptive modulation schemes for lowering the power consumption of the network to meet our goal of green communication for the next generation network. Keywords 5G · Energy efficiency · HMM · Green communication · Relay · Small-cell
B
Rakesh Kumar Jha
[email protected] Akshita Abrol
[email protected] Sanjeev Jain
[email protected] Preetam Kumar
[email protected]
1
Department of Electronics and Communication Engineering, Shri Mata Vaishno Devi University, Katra, J&K 182320, India
2
Department of Computer Science and Engineering, Shri Mata Vaishno Devi University, Katra 182320, J&K, India
3
Department of Electrical Engineering, Indian Institute of Technology, Patna, Patna 801106, India
1 Introduction In the recent future, green communication has emerged as the primary objective in the next generation networks. Mobile operators have been reported to be among top most energy consumers [1,2]. This calls for an urgent need to shift the focus from pursuing improved data rate and capacity to improved energy efficiency. In order to meet the demand of high data rate, increased capacity and better QoS, a holistic approach for power optimization is required. The energy efficiency of 5G network is expected to be increased 100X in International Mobile Telecommunications (IMT)—Advance and future IMT [3]. The motivation for pursuing green communication comes from the increased energy cost being experienced by telecom operators as well as the realization of social responsibility of reducing carbon footprint. This has led to a joint effort both on the industrial as well as academic front for developing energy efficient techniques. Various projects including EARTH [4], ‘green radio’ [5] and so on have gained importance. The third generation partnership project’s (3GPP) long term evolution-advanced (LTE-A) [6] and IEEE 802.16 standard [7] are also working to optimize the power consumption of the next generation wireless networks. Power optimization of wireless network is all the more important from the users’ perspective. The radiative power need to be kept under control keeping in mind the health issue of the users. Secondly, the next generation networks will support IoT which will involve unbounded and everywhere access to information along with ever increasing energy hungry applications. All this needs to be supported with power constrained devices. However, the major concern is the battery capacity which is only increasing at 1.5× per decade. So, for meeting the demands of the future networks, shifting towards energy efficient communication is imperative [8,9].
123
A. Abrol et al.
1.1 Background Some techniques for improving EE of the next generation networks have been discussed in [10]. Lately, relays have attracted a lot of attention as a method of improving EE. This is due to the fact that it also increases the coverage and reliability of transmission in a wireless network [11,12]. The next generation networks uses the concept of separate setups for indoor and outdoor communication [13] by employing small cell access points to reduce the power consumption by the base station. In order to further improve the EE of such networks, relays are used for transmission to small cells located far away from the base station. This architecture has already been proposed in [10]. The next generation networks are expected to have different types of nodes acting as relays which includes fixed power relays, user equipment as relays and high mobility relays [14]. There can be various relay selection policies including channel gain based relay selection [15], least distance relay selection [16], highest transmission rate selection [17], signal to noise ratio (SNR) based relay selection [18] and so on. Reduced channel estimation overhead for high mobility scenario and delay minimization for time critical applications can be used in the next generation scenario. In this paper, our goal is to concentrate on improving the EE of the network so we adopt the scheme focusing on green communication for relay selection. As we know relay is receives the signal and retransmit the signal to one or more than one users. In current scenario cooperative relaying is playing important role in wireless communication through cooperative relays we can enhanced the throughput and coverage and at the same time decrease the delay in the network. Buffer-aided is one of the methods has been applied in cooperative operation in which improves outage probability, throughput and power optimization [19, 20]. As we know buffer capacity is very minimum at relays so spoofing and jamming attack easily possible at relay nodes. Through proper relays selection policies we can enhanced the secrecy rate at the relays node [21,22]. Different relaying techniques are available among which amplify-and-forward, decode-and-forward and compressand-forward are the most widely used. In Amplify and Forward (AF), the relay retransmits amplified version of signal received from source to the destination. In Decodeand-Forward (DF), the relay transmits the re-encoded signal to the destination while in Compress-and-Forward (CF), the relay transmits the compressed signal to the destination. In this paper we use AF scheme due to its easy implementation as well as higher energy efficiency as compared to other schemes. Power consumption in a wireless network is mainly of two type: static and dynamic [23]. The static power is fixed and is consumed in site cooling equipment, network signal
123
processing, etc. On the other hand, dynamic power is the power used for transmission which is dynamically adjusted according to channel conditions. Hence, most of the work is focused on minimizing this dynamic power. In the existing literature, power allocation minimization problem in a multiuser scenario is focused on throughput maximization [24], QoS enhancement [25], coverage expansion [26] and so on. No research is available for energy efficient small cell and relay-assisted networks concentration on dynamic power allocation in multiuser scenario for the next generation networks. Energy harvesting is key solution for future scenario network and through this we can overcome the problem related environment and it can be achieved through cell association of users with small-cell base stations (scBS) [27], Energy harvesting is very important in health monitoring system due to live data processing and in this case battery backup is very crucial. Different sensors located on the penitent body required continuous power backup and it can achieve when energy beam transmitted adaptively towards the sensors than battery life of the sensors and performance of real time health monitoring system can be improved [8]. 1.2 Contribution In this paper, we have focused on green communication by considering the power allocation problem along with relay selection strategy in a relay-assisted small cell based multiuser network. The system consists of a three-layered model for power optimization with low powered and high powered and relay nodes. In order to provide optimum power allocation to all the nodes, we have used Hidden Markov model to calculate probability of each state according to various observations. Based on relay selection strategy for green communication and power allocation using Hidden Markov model, we have proposed an algorithm for energy efficiency optimization by decreasing total power consumed. The allocation problem is solved by applying Hidden Markov model to train and maintain the base station, relay as well as small cell access point. The parameters considered for power allocation are type of user, distance, SNR and application demanded by the client. Further, we have used adaptive modulation schemes for different users depending on their distance from the source. The rest of the paper is divided into the following sections. The second section contains the three-layered system model for power optimization of the next generation network. It provides detailed mathematical analysis of the work done. Section 3 describes the pseudo code of the algorithm implemented for power optimization along with the flow of the steps used in the process. Section 4 includes the simulation parameters and the detailed analysis of the results obtained. The conclusion and future research scope is discussed in Sect. 5.
Joint power allocation and relay selection strategy for 5G network: a step towards green…
2 Relay-assisted small cell-based heterogeneous network This section describes the system model and mathematical analysis for power optimization of next generation network. 2.1 System model for power optimization We have considered a cooperative wireless network for the next generation scenario. It involves peer to peer relaying consisting of three types of relays namely fixed power relay, mobile relay and user equipment relay. The system model is a three-layered architecture consisting of low power as well as high power nodes as shown in Fig. 1. Each layer corresponding to a different zone of power used for transmission. The first layer corresponds to the zone where base station transmits to the nodes. The second layer represents the region where power allocation is done through relay while the third layer represents the region of transmission and power allocation by the Small access point (SCA). For power optimization of network, we have adopted two approaches. Firstly, we have used a relay selection strategy for green communication in the next generation network and secondly, we have used Hidden Markov Model (HMM) for power allocation to all the nodes considering next generation wireless network. The various parameters based on which the power allocation using HMM has been done are as follow:1. 2. 3. 4.
Type of node Application demanded by the client node Signal-to-noise ratio at the client node Distance from the client node
Depending on these parameters, we have applied HMM for training and maintenance of the base station, relay and SCA to allocate the power to active clients. The allocation by base station and relay is based on the type of node i.e. relay, mobile station or SCA. In simpler terms, it represents the number of clients further being served by the client node which maybe nil in case of user equipment and can vary to a greater number in case of SCA or relay. The other parameters considered are distance, SNR and application demanded by the client node from the allocating node. The Quality of Service (QoS) of each client is maintained while allocating the power through HMM. The allocation of power by the SCA is however done differently. The power inside the small cell is allocated depending on the application demanded by the client node. The other parameters were found to be of least significance inside the small cell. Hence, the allocation largely depends on the application demanded. Each application demands a different QoS class. We have considered five different QoS classes as in 802.16e which are as follows:-
• Unsolicited Grant Service (UGS): it comprises of real time data streams of fixed size data packets which are issued at periodic intervals. • Extended Real-time Polling Service (ertPS): it comprises of real time service flows that generate variable sized data packets which are issued at periodic intervals. Its application includes VoIP. • Real-time Polling Service (rtPS): it comprises of real time data streams with variable sized data packets which are issued at periodic intervals. Its application includes MPEG video. • Non-real-time Polling Service (nrtPS): it comprises of delay tolerant data streams with variable sized data packets for which a minimum data rate is required. • Best Effort (BE): it comprises of data streams for which no minimum service level is required and therefore may be handled on a space available basis. Based on the QoS required for each application, power is allocated to the nodes inside the small cell. Further, each transmission is done using adaptive modulation and coding techniques. Depending of the SNR and distance, different modulation techniques such as QAM, QPSK and BPSK have been used to save more power. 2.2 Mathematical analysis for power optimization We assume a simple next generation scenario consisting of a source X s , a cluster K consisting of N number of half-duplex Amplify-and-Forward (AF) relays Yr where r ∈ {1, N }, two small cells one in Layer 1 and other in Layer 2 each having M active users demanding various applications and l mobile user clients spread out both in Layer 1 and Layer 2 of Fig. 1. There doesn’t exists a direct link but a link through a relay and or a SCA between each source and destination pair. All wireless links are assumed to have Additive White Gaussian Noise (AWGN) channel fading. It exhibits frequency non-selective Rayleigh block fading having complex Gaussian distribution with zero mean and unity variance γ 2 . The channel gains gi, j are assumed to be exponentially distributed where i ∈ {s, r, sca} and y ∈ {r, sca, d}. The variance of thermal noise ℵo at the receiver end is also assumed to be AWGN. We have used a pilot-symbol aided system to determine the channel state information (CSI). The CSI has been estimated at the relay for the source to relay link and at the SCA for the relay to SCA or direct source to SCA link considering the power constraint for pilot at both in order to obtain optimum power utilization. The signal to noise ratio from one node to another is defined as
S N Ri, j
2 Ps gi, j = γ2
(1)
123
A. Abrol et al.
C5
* # C1 C4$
C2%
@ C3
DESTINATION INTERFERENCE SIGNAL DATA SIGNAL
SCA1
FIXED RELAY LAYER 1 LAYER 2
SCA2
LAYER 3 Xs MOBILE RELAY
gs,rj SOURCE
Receiving Relay
gs,ri gri,rj
gs,d
@
rtPS
$
ertPS
*
nrtPS
%
UGS
#
BE
Ys,r
UE RELAY
Transmitting Relay
gri,d Ys,d Yr,d
Fig. 1 System model for power optimization
Here Ps denote the power of the source and the relay is assumed to have out-band relaying. The received signals at relay, SCA and destination nodes are given as Ys,r =
Ps gs,r X s + ℵs,r
(2)
Yr,sca/d = βr gr,sca/d Ys,r + ℵr,sca/d
(3)
Ysca,d = gsca,d Yr,sca + ℵsca,d
(4)
In Eq. (3), βr represents the amplification factor for AF relay. Considering relay power to be represented by Pr , the value of the amplification factor is given by √
βr =
Pr 2 Ps gs,r + γ 2
(5)
√
Pr Ps gr,sca/d gs,r X s + ℵr,sca/d 2 Ps gs,r + γ 2 (6)
123
γ
2
2 Pr gr,sca/d = 1+ γ2 2 Ps gs,r + γ 2
(7)
Now, from (6) and (4), we get √
Ysca,d =
Pr Ps gr,sca/d gs,r gsca,d X s 2 Ps gs,r + γ 2
+ ℵsca,d
(8)
Here ℵsca,d is the AWGN with variance γ 2 given by γ
Using (2) and (5) in (3), we get Yr,sca/d =
Here, ℵr,sca/d is the AWGN with variance γ 2 while ℵr,sca/d is the AWGN with variance γ 2 given by
2
2 2 Pr gr,sca/d = 1 + gsca,d γ2 1+ 2 2 Ps gs,r + γ
(9)
The relay selection for green communication is performed by reception at minimum signal to noise ratio threshold S N Rth by maintaining the desired quality of service for a particular application by using the following logic
Joint power allocation and relay selection strategy for 5G network: a step towards green…
Rselected at S N R=S N Rth = arg min Ps∗ + Pr∗
(10)
In order to minimize the dynamic power consumption of the network, hidden markov model is applied at the base station, relay as well as at small cell access point. Let us consider a general case of Markov model in our system with S distinct states T1 T2 . . . TS where S = 4 for our scenario of power allocation. Here, power allocation represents a stochastic process. The system changes its state according to the probability associated with each transition. The time interval can be represented by t and the state at a particular instant as X t . The exact description of the process of power allocation requires the knowledge of all predecessor states. So the required conditional probability will be P (X t |X t−1 , X t−2 , . . . , X 1 )
(11)
But in this paper we have used first order Markov assumption and truncated the description to only the current and predecessor state. So using this simplifying assumption, the conditional probability is given by P (X t |X t−1 )
(12)
Similarly, the probability of a certain chain of sequence {X 1 , X 2 , . . . , X t } is given by P (X 1 , X 2 , . . . , X t ) =
t
P (X n |X n−1 )
n=1
(13)
To map the power allocation process in terms of HMM, we have considered four parameters namely type of node, distance, SNR and application which represent stochastic process. So now the power allocation problem is a doubly embedded stochastic process which consists of a hidden stochastic process which can be observed or worked out from the observation of the mentioned parameters. Let one such observation be represented by Yt at a particular instant t. Then the conditional probability according to Bayer’s rule is given by P (X t |Yt ) =
P (Yt |X t ) P (X t ) P (Yt )
(14)
If we omit the probability P (Yt ), we get the measure of probability which is proportional to probability and referred to as likelihood L. The likelihood for n such observations can be represented in (15) as L (X 1 , . . . , X t |Y1 , . . . , Yt ) =
t
P (Yn |X n )
n=1 t
·
n=1
P (X n |X n−1 )
The parameters initialized for characterizing the power allocation process in terms of an HMM are as follows:1) The number of states of the model represented by S and given by T = {T1 T2 . . . TS }. We have considered four states i.e. S = 4 and includes base station, relay, small cell access point and clients represented as T = {B S, SC A, Relay, Client} The number of observations per state is represented by Q and given for jth parameter by U j = U1 U2 . . . U Q . The work done considers three types of users namely relay, SCA and mobile station which are categorized depending on the nodes served by them as already mentioned. The application demanded by the users are of five types as represented below. The distance between the BS and client node has discrete values varying from 35 to 500 m while between SCA and client node varies from 3 to 40 m. Similarly, the SNR has discrete values varying from 0 to 40. In this paper we have optimized the power with three layers approach where these layers is directly depends on distance from BS and other mentioned parameters like relays, mobile relays, UE relay, C1 to C5 etc. is appeared in different layers. From our system model we can easily observed that BS have assigned highest power to those relays whose location is appeared inside SCA because here one SCA is serving more numbers of clients i.e. C1 to C5 and at the same time one SCA can support up to 40m distance of communication. After this, maximum power has allocated by the BS to those relays whose positions are located at the cell boundary because at this location non-unformal distribution is appeared i.e. Null area is appeared. Since in 5G scenario D2D (Device to Device communication) is one of the most important communication so in this our proposed model UE relays also playing important role in power optimization. In this condition BS has assigned minimum power to UE because here one UE relays serve only one other UE. So order of power allocation from BS to different SCA, UE relays as follows: B S → SC A → UEVANET → UED2D Based on the performance of individual clients, the observations are divided into low (L), medium (M) and high (H) groups for both distance and SNR. This can be determined by applying clustering algorithms. So we have UU ser T ype = {Relay, SC A, Client} U QoSClass = {r t P S, nr t P S, er t P S, U G S, B E} UU ser Distance = {L d , Md , Hd } UU ser S N R = {L s , Ms , Hs }
123
A. Abrol et al. 0.10
0.05
Table 1 Transition probabilities Demand
0.05 RELAY
BS 0.25
0.05
0.6
0.05
0.3
0.05 C1/C2
SCA 0.85
0.05
0.85
Fig. 2 State diagram for power optimization
2) The set of probability measures represented as{A, B, π }. A represents the probability distribution of state transition, B represents the probability distribution of observation symbols and π represents the initial guess of state distribution. The initial guess of parameters must be chosen carefully. Each base station, relay and SCA will be trained and maintained by HMM to allocate power dynamically to all users. The steps involved are initialization of the HMM parameters, forward procedure and backward procedure. The details of these steps can be found in [28]. As discussed above, to find the observations pertaining to individual clients, clustering algorithms like K-means clustering. In our work, BaumWelch algorithm is used to estimate the initial guess of parameters{ A, B, π }. It considers initial distribution of probabilities to be uniform. This means for 4 states present, the probability of each state will be 0.25. The algorithm then converges to the nearest local maximum value of the likelihood function. This leads to accurate learning of the model and the process doesn’t affect the online processing of clients nodes’ request. 2.3 Transition probability matrix for power allocation The model used in our paper represents a special case of HMM in which every state can be reached from every other
123
Response BS
SCA
Relay
Client
Base Station (BS)
0.05
0.60
0.25
0.10
SCA
0.05
0.05
0.05
0.85
Relay (R)
0.05
0.55
0.10
0.30
Client (C1/C2)
0.05
0.05
0.05
0.85
state. This is represented in the form of a state diagram in Fig. 2. The transition probabilities from one state to another represent the probability of power allocation by that state to rest of the states. For instance, while considering power allocation by base station to the nodes, it is more likely to allocate maximum power to SCA rather that client node as SCA serves other client nodes as well. In the same way, the probabilities for transition from one state to another have been listed in Table 1. SCA has maximum probability of power allocation to a client node while relay is more likely to allocate more power to SCA It should be noted that client itself is very unlikely to allocate power being a power constrained node to SCA or relay or base station. It has maximum probability of power allocation only to another client node. This is possible in the case of device-to device communication only. Probability of each state has been calculated on the basis of stochastic process. All the information has been mentioned in Table 1 is related with system model Fig. 1. All the probability has assigned on the basis of real time deployment in current scenario of 5G Wireless Communication Network. Since BS have maximum probability to communicate with SCA because SCA can handle many users up to 40 m. After this BS have assigned the probability of communication with relays because relays overcome the non-uniformity at the cell boundary and it can serve more than one users with amplify or amplify and forward methods. After this BS can communicate with client directly but its probability is lesser because in this condition BS will serve one by one users and it is not best condition for power optimization. Finally BS will never want to communicate himself so its probability is minimum. Similarly other process will complete as Table 1.
3 Realization and representation of objective This section comprises of pseudo code of the proposed scheme for understanding and ease of its implementation. It consists of initialization of parameters, realization of the scenario, placement and selection of relay and application of Hidden Markov Model (HMM) at base station, relay and SCA for optimum power allocation.
Joint power allocation and relay selection strategy for 5G network: a step towards green…
Pseudo Code of Algorithm for Power Optimization Initialization Generate random users and SCA Create Small Cell with 3 users each Calculate D for SNR < SNRTH at P=Pmax Choose relays at radius D around BS for i=1: no of users Generate channel conditions (i) end for if (D > Dthreshold) for i = 1: no of relays Calculate Psr Calculate Prd Calculate Pt end for Find the relay with min power usage (ii) end if Initialize conditional probability of power for nth instant (iii) P(pown | pown-1 | pown-2,…..,pow1) History Created = (Number of Instant)Set of Power Element-1 Calculation of joint probability of current and past observations using Markov Assumption (iv) P(pow1,… pown) = P(pown | pown-1) Compute log likelihood of power using Bayer’s rule No of users = BS users (v) for k=1:2 # HMM for base station and relay (vi) for i=1: no of users for j=1: no of users if node is serving other nodes (vii) if SNR(i)>SNR(j) if D(i)>D(j) if Ap(i)>Ap(j) Allocate power end if else Allocate power else Allocate power else Allocate power end for end for no of users = relay users (viii) end for no of users = SCA users (ix) for i=1:no of users for j=1:no of users Based of application allocate power (x) end for end for
123
A. Abrol et al.
START
A
INPUT PARAMETERS
SET SNR THRESHOLD FOR PARTICULAR APPLICATION
FOR i=1:R
INITIALIZATION
# R represents total number of relays
GENRATE RANDOM USERS AT AREA OF 500m AROUND BASE STATION
# Generation of Random Users TRUE
CALCULATE SOURCE TO RELAY POWER
IF ALL USER DISTANCE<400
FALSE
GENERATE SMALL CELLS WITH RADIUS 40m
CALCULATE RELAY TO DESTINATION POWER CALCULATE TOTAL POWER FOR TRANSMISSION FIND THE INDEX OF RELAY WITH MIN TRANSMISSION POWER
FALSE
IF DISTANCE REQUIREMENT SATISFIED
# Relay with minimum power is selected
Fig. 3 Flowchart for relay selection
# Check for all distance constraints
TRUE CALCULATE DISTANCE ‘D’ FOR SNR
PLACE RELAYS AT RADIUS ‘D’ FROM BASE STATION
FOR i=1: U
The flow of the power optimization approach is shown in figure for better understanding of the pseudo code. Figure 4 shows the initialization of parameters and random deployment and realization of the scenario consisting of base station, small cells, relays and mobile stations shown in Fig. 1. Figure 3 shows the process of relay selection for power optimization of the network depending on SNR threshold. Figures 5 and 6 show the application of Hidden Markov Model at base station relay and SCA for probabilistic power allocation to their client nodes. It shows the various parameters depending on which power is allocated which include distance, SNR, number of users served by each node and application demanded by the client node. Hidden Markov model at SCA allocates power to nodes depending on only the application demanded by the client node as other parameters are insignificant inside small cell. Further, depending on distance of each client node, adaptive modulation schemes are allocated to the nodes for power optimization.
123
# SNR_TH represents SNR threshold # Pmax represents the maximum power constraint
# U represents all nodes
GENERATE CHANNEL CONDITIONS
COMPUTE ATTENUATION FOR ALL LINKS
IF (BOUNDAR Y CONDITION EXISTS)
# Check if relay is required for transmission FALSE
TRUE A
Fig. 4 Flowchart for initialization
B
Joint power allocation and relay selection strategy for 5G network: a step towards green…
B
C
INITIALIZE CONDITIONAL PROBABILITY OF POWER FOR CLIENTS
FOR i =1:n
INITIALIZE TRAINING DATA
# Computation of Probabilities for Power Allocation using Hidden Markov Model
# n represents number of mobile nodes
FOR j =1:n
COMPUTE INITIAL GUESS OF PARAMETERS
IS NODE SERVING OTHER NODES
IMPROVE GUESS OF PARAMETERS
# Using Bayer’s Rule COMPUTE LOG LIKELIHOOD
C
POWER ALLOCATION BY SCA TO USERS BASED ON APPLICATION USING HMM
PLOT AND STORE RESULTS
YES
CHECK SNR CONDITION
# Power Allocation using HMM at SCA based on Application demand
NO
NO
# Check all the conditions for power allocation
YES
CHECK DISTANCE CONDITION
# Output
END
YES
DOES APPLICATION REQUIRE MORE POWER
NO
Fig. 5 Computation of probability using HMM YES
4 Simulation parameters and results This section discusses the various parameters used for the simulation of scenario in Fig. 1. The scenario shows the base station communicating with 5 actively distributed users which include 2 mobile station, 2 SCA and 1 Relay which is selected according to the destination for optimum power usage. Each SCA has 5 active clients distributed within its 40 m radius. The performance of users considering various parameters at randomly generated user locations has been evaluated. The channels are modelled according to 3GPP LTE standard and the path loss models for BS and SCA considered are different from each other. The parameters used in simulation are listed below in Table 2.
ALLOCATE POWER TO CLIENT
CHECK DISTANCE CONDITION
# Adaptive Modulation Scheme ALLOCATE MODULATION SCHEME TO CLIENT QAM/QPSK/BPSK
Fig. 6 Flowchart for power allocation using HMM
123
A. Abrol et al.
S. no
Parameter
1
Macro cell radius
0.5 km
2
Small cell radius
0.040 km
3
Carrier frequency
2.4 GHz
4
Bits transmitted
103
5
Minimal user distance from SCA
0.003 km
6
Minimal user distance from BS
0.035 km
7
SNR threshold
2.5 dB
8
Path loss exponent
2
9
Number of users in small cell
5
10
Number of MS
10
11
Number of relays
3
12
Number of small cells
2
13
Path loss at distance ‘d’ from BS
148.1 + 37.6 log d dB
14
Path loss inside small cell at ‘d’
127 + 30 log d dB
The results and the inferences obtained from the simulation of the given system model for power optimization in Fig. 1 done in MATLAB using standards given in [29] have been discussed further. The power we have focused on is the dynamic power consumption of the network as static power consumption is mostly constant. So our proposed scheme focusses on minimizing the dynamic part of the power consumption to improve the energy efficiency.
45 40
POWER ALLOCATED
The graph shown in Fig. 7 represents the relay selection for green communication. The highlighted relays are selected for transmission in each iteration. The relays are selected depending upon the required SNR threshold for minimum power in order to maximize the energy efficiency of the network. Let us consider that there are 3 relays in real time deployment scenario and it will execute for 3 iterations. Once the process execute for power allocations to all three relays for three iterations. Here Red color shows power allocated to relay 1 in 3 different iterations. Green Color shows power allocated to relays 2 in 3 different iterations. Similarly Yellow color shows power allocated to relay 3 in 3 different iterations. On the basis of performance analysis observed that Relay 1 consumes the least amount of power 7.63 W in third iteration for providing the required QoS to a particular destination. Hence, Relay 3 is selected for this particular transmission. Similarly relays 2 consumes the least amount of power 6.94 W in first iteration for providing the required QoS to a particular destination. Hence, in this case Relay 1 is selected for this particular transmission and Relay 3 consumes the least amount of power 7.48 W in 3 iterations. Hence we have concluded that for green communication relay 2 shows best performance in 1 iteration.
10
8.97 8.39 6.94
8.227.459.21
7.638.27.48
5 0 1
2 Iteraon/ Relays
3
Fig. 7 Comparison between power levels of relays for selection of best relay for a particular transmission
4.1 Relay selection
123
Value
POWER CONSUMED (W)
Table 2 Simulation parameters
PROPORTIONAL POWER ALLOCATION POWER ALLOCATED BY HMM AT BS
35 30 25 20 15 10 5 0
1
2
3
4
5
CLIENTS
Fig. 8 Comparison of proportional power allocation and power allocated by base station trained using Hidden Markov Model
4.2 HMM at base station The graph in Fig. 8 shows a comparison of power allocated by the base station to mobile station, relay and small cell access points. It can be observed from the graph that equal power of 20 W is allocated to all the users irrespective of their demanded QoS or channel conditions or distance or
Joint power allocation and relay selection strategy for 5G network: a step towards green… Table 3 Power allocated by BS using HMM User index
User distance
Power allocated (Watt)
4
338
42.307
2
5
230
31.1237
5
3
92
17.0245
7
ertPS
SCA
QPSK
2
37
6.4694
28
ertPS
Relay
QAM
1
38
1.5061
38
nrtPS
MS
QAM
20
POWER ALLOCATED
18
User SNR
14 12 10 8 6 4 2 6
User type
Modulation scheme
rtPS
SCA
BPSK
rtPS
MS
BPSK
It should be noted that we have considered the worst case scenario where the same relay node serves both mobile station and SCA. In case it transmits to one of the only, proportional power allocation scheme allocates entire 20 W to one user. In this case HMM will save tremendously. By using HMM for power allocation in the case considered, there is a power saving of 0.1 W. As well as it leads to allocation of more power to SCA because it has to serve more nodes at least one of which demands a higher power using application. Both use QAM modulation scheme as shown in Table 4 because the distance at which the nodes are present is less than 40 m.
POWER ALLOCATION BY HMM AT RELAY PROPORTIONAL POWER ALLOCATION
16
0
QoS class (application)
7
CLIENTS
Fig. 9 Comparison of proportional power allocation and power allocated by relay trained using Hidden Markov Model
the number of clients a node is further serving if proportion allocation model is followed. On the other hand, by using HMM for power allocation, power is allocated depending on the parameters mentioned. So, in this particular iteration it leads to a power saving of 1.6 W. The saved power will vary with each iteration depending on the users’ demand and channel conditions. The results have been tabulated in Table 3. It is clear from the table that maximum power is allocated SCA present far away with lower SNR because it has to serve more nodes at least one of which demands an application requiring high power consuming QoS. The modulation schemes are allocated to users depending on their distance from the base station. BPSK having least bit error rate of the three is used for transmission to the nodes located at maximum distance from base station while QAM is used for nearby nodes. 4.3 HMM at relay The graph in Fig. 9 shows a comparison of power allocated by the relay to mobile station and a small cell access point. It can be observed from the graph that equal power of 10 W is allocated to both if proportion allocation model is followed.
4.4 HMM at SCA The graph in Fig. 10 shows the comparison of power allocated to all the nodes in the small cell using Hidden Markov Model at SCA saves about 0.46 W power as compared to proportional power allocation model in which each node consumes 4 W of power irrespective of the QoS required for the application demanded. Node 11 consumes maximum amount of power owing to its demand of an application belonging to QoS class requiring rtPS. While nodes 8, 9 and 10 consume least power because of lower QoS requirement. Node 12 also consumes significant amount of power due to demand for application belonging to ertPS class of service. All nodes use QAM as shown in Table 5 because the nodes lie inside small cell at a distance less than 40 m from the SCA. 4.5 Comparison of results Figure 11 shows the comparison of results obtained for four different cases. It can be observed that power consumed by the client node increases as the SNR threshold required for the application increases for every case. Use of proportional power allocation consumes the maximum amount of power out of the four compared. Using relay selection lowers the total power consumed by some amount. The use of Hidden Markov Model for power allocation lowers the consumed power by considerable amount and using relay selection (RS)
123
A. Abrol et al. Table 4 Power allocated by relay using HMM User index
User distance
Power allocated (Watt)
User SNR
QoS class (application)
6
34
13.8
19
7
21
6.1
28
15
User type
Modulation scheme
rtPS
SCA
QAM
nrtPS
MS
QAM
35
PROPORTIONAL POWER ALLOCATION POWER ALLOCATION BY HMM AT SCA
PROPORTION POWER ALLOCATION WITHOUT RS PROPORTIONAL POWER ALLOCATION WITH RS HMM POWER ALLOCATION WITHOUT RS HMM POWER ALLOCATION WITH RS
POWER CONSUMED (dB)
POWER ALLOCATED
30
10
5
25 20 15 10 5 0
0 8
9
10
11
0
12
5
10
15
20
25
30
35
40
SNR (dB)
CLIENTS
Fig. 10 Comparison of proportional power allocation and power allocated by SCA trained using Hidden Markov Model
Fig. 11 Power consumed for transmission to client node versus SNR 60
in combination with HMM produces the best results in terms of power saving. Hence, the graph shows that power consumed by applying the proposed scheme of probabilistic power allocation using the Hidden Markov Model along with relay selection leads to lowest total power consumption. The graph in Fig. 12 shows the comparison of power consumed by client nodes with respect to the distance of the node from the base station. The proportional power allocation scheme leads to greater power consumption as compared to allocation using Hidden Markov model both with and without using relay selection. It can be clearly seen from the graph, the proposed scheme of using relay selection along with HMM for power allocation outperforms others in terms of power consumption especially as the distance of nodes from the base station increases.
POWER CONSUMED
50
40
30
20 PROPORTION POWER ALLOCATION WITHOUT RS PROPORTIONAL POWER ALLOCATION WITH RS HMM POWER ALLOCATION WITHOUT RS HMM POWER ALLOCATION WITH RS
10
0
0
50
100
150
200
250
300
350
400
450
500
DISTANCE (m)
Fig. 12 Power consumed for transmission to client node versus distance between base station and client node
Table 5 Power allocated by SCA using HMM User distance
User distance
Power allocated (Watt)
User SNR
QoS class (application)
User type
Modulation scheme
11
27
14.9
21
rtPS
MS
QAM
12
22
4.5
23
ertPS
MS
QAM
9
16
0.1
27
BE
MS
QAM
8
15
0.03
36
BE
MS
QAM
10
15
0.01
39
BE
MS
QAM
123
Joint power allocation and relay selection strategy for 5G network: a step towards green… 80
PROPORTIONAL POWER ALLOCATION WITHOUT RS PROPORTIONAL POWER ALLOCATION WITH RS HMM POWER ALLOCATION WITHOUT RS HMM POWER ALLOCATION WITH RS
PROPORTIONAL POWER ALLOCATION WITH RS
70
HMM POWER ALLOCATION WITHOUT RS
30
HMM POWER ALLOCATION WITH RS
60
EE (bits/J-S)
POWER CONSUMED
35
PROPORTIONAL POWER ALLOCATION WITHOUT RS
50 40 30
25
20
15
20 10
10 0 2
4
6
8
10
12
14
16
NO OF CLIENTS
5 100
150
200
250
300
350
400
450
500
DISTANCE (m)
Fig. 13 Comparison of power consumed with respect to the node density
Fig. 14 Comparison of EE with respect to the distance between the base station and client node
The graph in Fig. 13 shows the comparison of total power consumed for transmission with respect to the client density. It can be seen from the graph that as the density of nodes increases the power consumed has the tendency to increase. The proposed scheme which uses Hidden Markov model along with relay selection produces the best results in terms of power even when the client node density increases. Higher the node density, more is the power saving by use of dynamic power allocation. The overall power saving by the proposed scheme for dynamic power allocation has been represented in the form of Table 6. The total power saved is 2.16 W by the use of Hidden Markov model for training and maintaining of base station, relay and SCA. The graph in Fig. 14 shows the comparison of energy efficiency of the network with respect to the distance between the base station and the client node. It is clear from the graph that the EE decreases as the distance between the base station and the node increases. The relay selection scheme alone also provides improvement in EE as represented in the graph. However, our proposed scheme of joint relay selection and probabilistic power allocation outperforms the others and provides a considerable improvement in EE. The results show that the proposed scheme provides a considerable improvement in power saving of the network
which leads to energy efficiency improvement of the next generation wireless network. Hence, by using the proposed scheme we can solve the problem of power optimization of the 5G networks. Our proposed network is very useful in real time deployment in industry for D2D (Device to Device) Communication and Ultra Dense Network (UDN). Power optimization will be the prime factor in Next Generation Network (NGN).
5 Conclusion and future scope Power optimization in 5G networks is the need of the hour from the distributors’ as well as consumers’ point of view. A need to reduce the radiation power levels as well as the carbon footprint is being felt worldwide. This has encouraged pursuing of green communication for wireless communication networks as well. The next generation wireless networks aim at improving the energy efficiency of the 5G networks by saving maximum amount of power under QoS constraint. The major concern is to improve EE without compromising on the user experience. In this paper, we have focused on optimizing the power in two steps by using joint relay selection and power allocation approach. The selection of relay has been done keeping in
Table 6 Total power by using HMM Power saving (W) Ps = PP A − PH M M
Power consumed by all users (W) Proportional power allocation PP A
Hidden Markov Model PH M M
Base station
20*5 = 100
98.4
1.6
Relay
10*2 = 20
19.9
0.1
SCA
4*5 = 20 19.54 Total power saving = PS = 1.6 + 0.1 + 0.46 = 2.16 W (32 dBm)
0.46
123
A. Abrol et al.
mind the goal of green communication and we have used a probabilistic power allocation strategy at base station, relay as well as SCA to allocate power to nodes depending on different parameters. Hidden Markov model has been used for the purpose of training and maintaining of base station, relay and SCA for minimizing the dynamic power consumption of the client nodes. Adaptive modulation schemes also help in saving power. The results obtained show considerable improvement in EE of the network and lower power consumption with our proposed approach as compared to proportional allocation strategy. The saved power can be used to serve the high power demanding nodes as an extension of this work for improving their QoS as well as the tradeoff between other parameters and energy efficiency can be analyzed for the next generation networks Hence, using the proposed approach we have provided a solution to the power optimization problem of the next generation networks. Acknowledgements The authors gratefully acknowledge the support provided by 5G and IoT Lab, DoECE, and TBIC-Shri Mata Vaishno Devi University, Katra, Jammu.
13.
14.
15.
16.
17.
18.
19.
20.
References 1. Laurent He’rault, Holistic approach for future energy efficient cellular networks, CEA-LETI, 11th Leti annual review, June 2009. 2. Feng, D., Jiang, C., Lim, G., Cimini, L. J., Feng, G., & Li, G. Y. (2013). A survey of energy-efficient wireless communications. IEEE Communications Surveys & Tutorials, 15(1), 167–178. 3. IMT vision-framework and over-all objectives of the future development of IMT for 2020 and beyond, document Rec. ITU-R M.2083-0, September 2015. 4. EARTH. Energy aware radio and network technologies project. [Online]. Available: https://www.ict-earth.eu/default.html. 5. Han, C., et al. (2011). Green radio: Radio techniques to enable energy-efficient wireless networks. IEEE Communication Magazine, 49(6), 46–54. 6. 3GPP TR 36.814 v.9.0.0 “Technical specification group radio access network; evolved universal terrestrial radio access (EUTRA); Further advancements for E-UTRA physical layer aspects,” March 2010. 7. IEEE standard for local and metropolitan area networks part 16: Air interface for broadband wireless access systems amendment 1: Multi-hop relay specification, IEEE Standard 802.16j, June 2009. 8. Mesodiakaki, A., Adelantado, F., Alonso, L., & Verikoukis, C. (2017). Joint energy and spectrum efficient user association in millimeter wave backhaul small cell networks. IEEE Transactions on Vehicular Technology, 66(2), 1810–1821. 9. Mesodiakaki, A., Adelantado, F., Antonopoulos, A., Di Renzo, M., Alonso, L., & Verikoukis, C. (2016). Energy and spectrum efficient user association in 5G heterogeneous networks. In IEEE PIMRC. 10. Abrol, A., & Jha, R. K. (2016). Power optimization in 5G networks: A step towards green communication. IEEE Access, 4, 1355–1374. 11. Pabst, R., et al. (2004). Relay-based deployment concepts for wireless and mobile broadband radio. IEEE Wireless Communications Magazine, 42(9), 80–89. 12. Salem, M., Adinoyi, A., Yanikomeroglu, H., & Falconer, D. (2010). Opportunities and challenges in OFDMA-based cellular relay net-
123
21.
22.
23.
24.
25.
26.
27.
28.
29.
works: A radio resource management perspective. Transactions on Vehicular Technology, 59(5), 2496–2510. Gupta, A., & Jha, R. K. (2015). A survey of 5G network: Architecture and emerging technologies access. IEEE Access, 3, 1206–1232. Nomikos, N., Skoutas, D. N., & Makris, P. (2014). Relay selection in 5G networks. In Proceedings of the international wireless communications and mobile computing conference (IWCMC) (pp. 821–826). Bletsas, A., Shin, H., & Win, M. Z. (2007). Cooperative communications with outage-optimal opportunistic relaying. IEEE Transactions on Wireless Communications, 6(9), 3450–3460. Zou, Y., Zhu, J., Zheng, B., & Yao, Y.-D. (2010). An adaptive cooperation diversity scheme with best-relay selection in cognitive radio networks. IEEE Transactions on Signal Processing, 58(10), 5438–5445. Altubaishi, E., & Shen, S., (2011). Variable-rate based relay selection scheme for decode-and-forward cooperative networks. In Proceedings of the IEEE wireless communications and networking conference (pp. 1887–1891). Jing, Y., & Jafarkhani, H. (2009). Single and multiple relay selection schemes and their achievable diversity order. IEEE Transactions on Wireless Communications, 8(3), 1414–1423. Nomikos, N. K., Charalambous, T., Krikidis, I., Skoutas, D. N., Vouyioukas, D., Johansson, M., et al. (2015). A survey on bufferaided relay selection. Communications Surveys & Tutorials. doi:10. 1109/COMST.2015.2512441. Nomikos, N., Vouyioukas, D., Charalambous, T., Krikidis, I., Makris, P., Skoutas, D.N., Johansson, M., & Skianis, C. (2013). Joint relay-pair selection for buffer-aided successive opportunistic relaying. Transactions on Emerging Telecommunications Technologies. Nomikos, N., Nieto, A., Makris, P., Skoutas, D. N., Vouyioukas, D., Rizomiliotis, P., et al. (2015). Relay Selection for Secure 5G Green Communications. Journal Telecommunication Systems, 59, 169–187. doi:10.1007/s1235-014-9890-7. Chopra, G., Jha, R. K., & Jain, S. (2017). A survey on ultra-dense network and emerging technologies: Security challenges and possible solutions. Journal of Network and Computer Applications, 95(1), 54–78. doi:10.1016/j.jnca.2017.07.007. Miao, G. W., Himayat, N., & Li, G. Y. (2010). Energy-efficient link adaptation in frequency-selective channels. IEEE Transactions on Communications, 58(2), 545–554. Phan, K. T., Le-Ngoc, T., Vorobyov, S. A., & Tellambura, C. (2009). Power allocation in wireless multi-user relay networks. IEEE Transactions on Wireless Communications, 8(5), 2535–2545. Liu, Y., & Petropulu, A. P. (2013). QoS guarantees in AF relay networks with multiple source-destination pairs in the presence of imperfect CSI. IEEE Transactions on Wireless Communications, 12(9), 4225–4335. Sadek, A. K., Han, Z., & Liu, K. J. R. (2010). Distributed relay-assignment protocols for coverage expansion in cooperative wireless networks. IEEE Transactions on Mobile Computing, 9(4), 505–515. Parzysz, F., Di Renzo, M., & Verikoukis, C. (2017). Poweravailability-aware cell association for energy-harvesting small-cell base stations. IEEE Transactions on Wireless Communications, 16(4), 2409–2422. Rabiner, L. R. (1989). A tutorial on hidden Markov models and selected applications in speech recognition. Proceedings of the IEEE, 77(2), 257–286. Further advancements for E-ULTRA physical layer aspects (Release 12), document 3GPP TS 36.942, September 2014.
Joint power allocation and relay selection strategy for 5G network: a step towards green… Akshita Abrol (S’16) received the B.E. degree in Electronics and Communication Engineering from Jammu University, Jammu and Kashmir, India, in 2013. She is currently pursuing the M.Tech degree in Electronics and Communication Engineering at Shri Mata Vaishno Devi University, Katra, Jammu and Kashmir, India. Her research interest includes the emerging technologies of 5G wireless communication network and Green Communication. Currently she is doing his research work in Power Optimization of 5G networks. She is working on MATLAB for Wireless Communication. Miss. Abrol received the Teaching Assistantship at the Ministry of Human Resource Development from 2014 to 2016. Rakesh Kumar Jha (S’10M’13-SM’15) received the B.Tech degree in Electronics and Communication Engineering in Bhopal, India, the M.Tech degree from NIT Jalandhar, India, and the Ph.D. degree from NIT Surat, India, in 2013. He is currently an Assistant Professor with the Department of Electronics and Communication Engineering, Shri Mata Vaishno Devi University, Jammu and Kashmir, India. He is carrying out his research on wireless communication, power optimizations, wireless security issues, and optical communications. He has authored over 30 international journal papers and more than 20 international conference papers. His area of interest is wireless communication, optical fiber communication, computer networks, and security issues. Dr. Jha’s concept related to router of wireless communication has been accepted by the International Telecommunication Union (ITU) in 2010. He received the Young Scientist Author Award from ITU in 2010, the APAN Fellowship in 2011 and 2012, and the Student Travel Grant from COMSNET in 2012. He is a Senior Member of the Global ICT Standardization Forum for India, Society for Industrial and Applied Mathematics, the International Association of Engineers, and the Advance Computing and Communication Society.
Sanjeev Jain born at Vidisha in Madhya Pradesh in 1967, obtained his Post Graduate Degree in Computer Science and Engineering from Indian Institute of Technology, Delhi, in 1992. He later received his Doctorate Degree in Computer Science & Engineering and has over 24 years’ experience in teaching and research. He has served as Director, Madhav Institute of Technology and Science (MITS), Gwalior. Presently, he is working as a vice chancellor at Shri Mata Vaishno Devi University, Katra. Besides teaching at Post Graduate level Professor Jain has the credit of making significant contribution to R&D in the area of Image Processing and Mobile Adhoc Network. He has guided Ph.D. Scholars and has undertaken a number of major R&D projects sponsored by the Government and Private Agencies. His work on Digital Watermarking for Image Authentication is highly valued in the research field. Preetam Kumar received the Ph.D. degree in the area of Wireless Cellular Communications from Indian Institute of Technology (IIT) Kharagpur, India. He is Senior Member of IEEE. Currently, he is an Associate Professor in the Department of Electrical Engineering, Indian Institute of Technology Patna, India. He has more than 15 years of teaching research and industry experience. His research interests are Physical Layer Issues in Wireless Communications, 5G and Digital Communication Systems. He is a regular reviewer of premier IEEE journals and conferences. He is also the Editorial Board member of Springer’s International Journal of Wireless Personal Communications.
123