Characterizing Energy Efficiency of Wireless Transmission for Green Internet of Things: A Data-Oriented Approach
aa r X i v : . [ c s . I T ] A p r Characterizing Energy Efficiency of WirelessTransmission for Green Internet of Things: AData-Oriented Approach
Hong-Chuan Yang,
Senior Member, IEEE and Mohamed-Slim Alouini,
Fellow,IEEE
Abstract
The growing popularity of Internet of Things (IoT) applications brings new challenges to thewireless communication community. Numerous smart devices and sensors within IoT will generate amassive amount of short data packets. Future wireless transmission systems need to support the reliabletransmission of such small data with extremely high energy efficiency. In this article, we introduce anovel data-oriented approach for characterizing the energy efficiency of wireless transmission strategiesfor IoT applications. Specifically, we present new energy efficiency performance limits targeting atindividual data transmission sessions. Through preliminary analysis on two channel-adaptive transmis-sion strategies, we develop several important design guidelines on green transmission of small data.We also present several promising future applications of the proposed data-oriented energy efficiencycharacterization.
Index Terms
Internet of Things, wireless communications, energy efficiency, fading channels, adaptive transmis-sion.
This work was supported in part by an NSERC Discovery Grant.H.-C. Yang is with the Department of Electrical and Computer Engineering, University of Victoria, Victoria, BC V8W 2Y2,Canada (e-mail: hy@uvic,ca).M.-S. Alouini is with the Computer, Electrical, and Mathematical Sciences and Engineering (CEMSE) Division, King AbdullahUniversity of Science and Technology (KAUST), Thuwal 23955, Saudi Arabia (e-mail: [email protected]).
I. I
NTRODUCTION
The Internet of Things (IoT) will dramatically change the way we interact with the world.IoT extends the Internet to our daily objects, such as appliances, cameras, lights, displays, andvehicles, by equipping them with micro-controllers, communication, and networking capability.Such extension will transform our daily life and enable many new applications, ranging fromhome automation and traffic management, to smart grids and mobile health-care [1]. Furthermore,these connected devices will generate a large amount of data, the timely processing of whichwill bring huge social and economical benefits. Meanwhile, several technical challenges need tobe addressed to realize the full potential of IoT. For example, the IoT needs to support diverseapplication scenarios, which typically have diverse service requirements [2]. The provision ofIoT functionality will definitely increase the overall system cost, the justification of whichrequires suitable business model. Furthermore, the communication and networking functionsof IoT devices will necessarily consume extra energy. As the number of connected devices andsensors within the IoT will be enormous, the overall energy consumption of future IoT couldbe prohibitive with conventional transmission strategies. As such, there is a pressing need fordeveloping green IoT technologies.Wireless transmission is the idea choice for connecting IoT devices. Therefore, designinghighly energy efficient wireless transmission strategies will be essential to the realization ofgreen IoT. Energy efficiency has always been a serious concern for wireless systems sincewireless devices typically have limited energy supply. Various advanced transmission technolo-gies, including channel adaptive transmission [3] and cooperative relay transmission [4], [5], aredeveloped and deployed to support high data rate wireless services with low energy cost. Thesetransmission technologies were typically designed with the goal of enhancing or approaching thecapacity limits of wireless channels for a given transmission power, as the energy efficiency isusually quantified as the ratio of channel capacity over transmission power [6], [7]. On theother hand, most existing metrics characterize energy efficiency in an average sense. Suchcharacterization can not provide useful guidelines to the energy efficiency improvement forindividual transmission sessions over IoT, which usually occur in a sporadic fashion.The IoT introduces a paradigm shift to wireless communications. Most IoT applications entailsquick information exchanges from smart devices/sensors. These machine-type terminals willsporadically access the networks for the transmission of short packets that contains metering data, status information, and remote commands. These transmission sessions will have muchshorter duration than conventional traffics, such as phone calls and video streaming. Conventionaltransmission system design typically adopts a c hannel-oriented approach assuming a consistentand continuous data traffic and improves the average channel quality with advanced transmis-sion technologies. Meanwhile, such approach ignores the specifics of individual transmissionsessions, such as the traffic characteristics and the prevailing channel/network condition. Whenthe transmission sessions are short, the energy efficiency achieved by individual sessions willvary dramatically as the result of the channel variation. To further improve the energy efficiencyof wireless transmission systems, especially for IoT applications, we need to optimally designthe transmission strategies from the perspective of individual transmission sessions.Motivated by this intuition, we propose a novel data-oriented approach for the energy effi-ciency optimization of wireless transmission strategies for IoT applications. Specifically, whena certain amount of data is available for transmission, we optimally decide the transmissionstrategy for the highest possible energy efficiency. The transmission strategy will be adjusted foreach data transmission session according to the traffic characteristics and the channel/networkconditions. Intuitively, we expect that the average energy efficiency of wireless transmissionwill be further enhanced if the transmission strategy is optimized for each transmission session.In this article, we present an initial investigation on this data-oriented approach for developingenergy efficient transmission strategies. In particular, we introduce two new data-oriented energyefficiency performance metrics targeting at individual data transmission sessions. We illustratetheir analysis on two popular channel adaptive transmission strategies over fading channels.Finally, we discuss several promising future research directions with the data-oriented approachfor green wireless transmission system design and analysis.II. C ONVENTIONAL ENERGY EFFICIENCY METRICS
Energy efficiency metrics are essential to the analysis and design of green communicationsystems. They help assess and compare the energy consumption of different designs and providelong-term research goals. The energy efficiency metrics for wireless communication systems canbe generally classified into two categories: i) network-level metrics and ii) link-level metrics.Network-level metrics characterize the energy efficiency of the whole system with the considera-tion of service coverage. Examples include the ratio of coverage area to site power consumptionwith the unit of km /Watt [8] and the average power usage per service data rate per coverage area in Watt/bps/m [9]. As many factors, including equipment choice, network structure, andfacility arrangement, affect the energy consumption of a wireless network, these network-levelmetrics can not provide direct guidelines to green design of wireless transmission system.Link-level metrics typically focus on the energy efficiency of a particular transmission link andquantify the efficiency of the transmission system in achieving a certain transmission rate withrespect to resource utilization. For example, the achieved data rate per unit power consumption,with unit of bits/s/Watt or equivalently bits/Joule, is a widely used energy efficiency metric [7].This metric was applied to the tradeoff analysis among different system design parameters [6].The radio efficiency metric in m · bit/s/Watt [10] considers both transmission rate and transmissiondistance. With the application of Shannon capacity formula, the upper bounds of these energyefficiency metrics can be evaluated. On the other hand, these metrics are typically defined forconstant channel realization with fixed transmission power and as such can not directly apply tofading wireless channels with time-varying channel gains.We can generalize most link-level metrics to fading wireless channels by applying the ergodiccapacity concept. Ergodic capacity characterizes the largest possible average transmission ratethat a wireless channel can support. Using ergodic capacity, we can evaluate the average energyefficiency of wireless transmission over fading channels. In particular, the ergodic capacity wasutilized to evaluate the area spectral efficiency of cellular systems [11]. The metric was latergeneralized to quantify the energy efficiency of point-to-point transmission with the considerationof affected area [12]. Meanwhile, these ergodic capacity based metrics on can only characterizethe energy efficiency of wireless transmission in an average sense. The resulting analysis is gen-erally applicable to conventional continuous data traffics. The IoT involves numerous machine-type terminals that generates sporadic small data packets. The energy consumption of individualdata transmission session for these small data varies dramatically with the prevailing channelrealization. The realization of green IoT relies heavily on the energy efficiency improvement forshort transmission sessions.To further enhance the energy efficiency of wireless system for ‘small data’ transmission,we need to study wireless transmission technologies from a new perspective. In this article, wefollow a data-oriented approach and propose to characterize the energy efficiency of wirelesstransmission for the perspective of individual data transmission sessions. More specifically, weraise the following fundamental questions: Given a certain amount of data to be transmitted,what is the probability that the amount of energy required for its successful transmission is greater than a threshold level? Given the amount of available energy at transmitter, what is thelargest amount of data that can be transmitted over the wireless channel reliably? The answers tothese questions will provide the valuable design guidelines for the energy-efficient transmissionof small data. In the following, we introduce two data-oriented energy efficiency metrics toaddress these design questions.III. M INIMUM ENERGY CONSUMPTION
The fundamental service requirement of green IoT applications is to reliably transmit a certainamount of data to its destination over a given channel in a highly energy-efficient manner. Wedefine a data-oriented energy utilization metric, namely minimum energy consumption (MEC),as the minimum amount of energy required to transmit a certain amount of data over a wirelesschannel. Let H denote the amount of data to be transmitted. The MEC will be a function of H , denoted by E min ( H ) . For a given H value, MEC will vary with the transmission power, thechannel bandwidth, the channel realization, and the adopted transmission strategy. To illustratefurther, we consider the MEC analysis for two adaptive transmission strategies over a point-to-point wireless link. We assume that the channel introduce flat fading. The noise spectral densityat the receiver over the channel bandwidth B is N , which leads a noise power of N B . A. Continuous rate adaptation
We first consider the continuous rate adaptation with constant power (CRA) transmissionstrategy. Specifically, the transmitter adapts the transmission rate with the channel conditionwhile maintaining constant transmission power P t [13]. For the small data scenario, where H isrelatively small, data transmission will typically complete in a channel coherence time. Applyingthe Shannon capacity formula, the maximum instantaneous data rate for reliable transmission isequal to B · log (1 + P t g/N B ) , where g is the instantaneous channel power gain. The minimumtime duration to finish data transmission is determined as H/ ( B log (1 + P t g/N B )) . We canthe calculate MEC as the product of the transmission power and the minimum transmission timeas E min ( H ) = P t H/ ( B log (1 + P t g/N B )) , which varies with the instantaneous channel gain g . To address the earlier design questions, we define the energy outage rate (EOR) as theprobability that MEC for a certain amount of data is greater than a threshold energy amount. Inparticular, EOR is mathematically defined as EOR = Pr[ E min ( H ) > E th ] , where E th denotes −3 −2 −1 Energy Threshold, J E ne r g y O u t age R a t e B = 200 KHz, P t = 300 mWB = 400 KHz, P t = 300 mWB = 200 KHz, P t = 100 mWB = 400 KHz, P t = 100 mW Fig. 1. Energy outage rate of CRA over slow Rayleigh fading channel ( H =
50 kB, and g = -10 dB). the energy threshold. Equivalently, EOR can be calculated as the probability that the per-bitenergy consumption is greater than a threshold value E th /H . The EOR for data transmissionwith CRA within a channel coherence time can be calculated as EOR cra = F g (cid:20) N P t /B (cid:18) exp (cid:0) ln(2) HP t BE th (cid:1) − (cid:19)(cid:21) , (1)where F g ( · ) denotes the CDF of the channel power gain g . As such, EOR serves as an statisticalcharacterization for the energy efficiency experienced by individual data transmission sessionwith CRA.Fig. 1 plots the EOR of CRA transmission over slow Rayleigh fading channel as the functionof the energy threshold E th for different transmission parameter settings. We set the data amountto 50 kB and the average channel power gain to -10 dB. We can see that the EOR for all casesdecreases with the energy threshold. Larger channel bandwidth help reduce the EOR for the sametransmission power level, as expected by intuition. Meanwhile, for the same channel bandwidth,larger transmission power leads to larger EOR. Typically, larger transmission power help improve the received SNR for the same channel realization, which allows for higher transmission ratewith CRA and in turn reduces the time duration to finishing data transmission. The transmissiontime reduction is, however, in logarithm with respect to P t . As such, the MEC increases with P t , which leads to high EOR. B. Continuous power adaptation
Power adaptation is a popular adaptive transmission strategy. Here, we consider the continuouspower adaptation with constant rate (CPA) transmission strategy. In particular, the transmitteradapts the transmission power with the channel condition while maintain a constant receivedSNR, denoted by γ c , under the peak power constraint P max (also known as truncated channelinversion [13]). Mathematically speaking, the transmission power P t is set to γ c N B/g when g ≥ g T = γ c N B/P max , and 0 otherwise. Such transmission strategy can support error freetransmission at rate B log (1 + γ c ) when g ≥ g T . The MEC with CPA transmission can becalculated as E min ( H ) = γ c N g H log (1 + γ c ) , (2)when g ≥ g T . We can see that MEC is inverse proportional to the channel gain g for CPA, whereasfor CRA, MEC is approximately proportional to / log ( g ) . Power adaptation can achieve betterenergy efficiency than rate adaptation at the cost of a certain probability of transmission outage.Note that when g < g T , the transmitter with CPA will hold the transmission until the channelcondition improves, which may cause long delay.The EOR of CPA transmission can be calculated as the probability that E min ( H ) given in (2)is greater than the energy threshold E th . Noting that the transmission will be held when g < g T ,and as such, no transmission energy is consumed, the EOR for a certain amount of data withCPA can be evaluated as EOR cpa = (cid:20) F g (cid:18) γ c N E th H log (1 + γ c ) (cid:19) − F g ( g T ) (cid:21) / (1 − F g ( g T )) . (3)Fig. 2 illustrates the EOR performance of CPA over slow Rayleigh fading channels. Inparticular, we examine the effect of peak transmission power and target received SNR duringtransmission. We can see that maintaining a higher target received SNR with CPA leads to largerEOR. This can be explained by noting from Eq. (2) that the MEC with CPA will increase with γ c . Another way to appreciate this behavior is to note that higher γ c implies larger transmissionpower during transmission on average. We also observe from Fig. 2 that larger peak transmission −3 −2 −1 Energy Threshold, J E ne r g y O u t age R a t e γ c = 16 dB, P max = 6 W γ c = 16 dB, P max = 3 W γ c = 10 dB, P max = 6 W γ c = 10 dB, P max = 3 W Fig. 2. Energy outage rate of CPA over slow Rayleigh fading channel ( H =
50 kB, B =
200 kHz, and g = -10 dB). power results in larger EOR, especially when the energy threshold is large. With CPA, larger P max will lead to larger probability of transmission for the same target SNR. While leading tolonger delay, smaller P max will ensure that the system transmit only over more favorable channelcondition and as such reduce the energy consumption. We conclude that different P max valueslead to different tradeoffs between energy efficiency and transmission delay.IV. M AXIMUM INFORMATION DELIVERY
Most IoT devices are running on stringent energy budget. Many devices will be poweredby energy harvesting from the ambient environment. Therefore, the efficient utilization of thelimited energy resource for data transmission is of critical importance for IoT devices. In thissection, we characterize the energy efficiency of individual data transmission session from theinformation delivery perspective. In particular, we define maximum information delivery (MID)as the maximum amount of information that can be reliably transmitted with a given amountof energy. Such characterization would be instrumental to the energy provision design for IoT devices. Mathematically, we denote MID by H max ( E ) , which is a function of the availableenergy amount, denoted by E . Here, MID will depends on the channel bandwidth, the channelrealization, and the adopted transmission strategy. Note that MID can be applied to evaluate thebits/joule metric as H max ( E ) /E . We illustrate the MID analysis again by considering CRA andCPA transmission strategies over a point-to-point link for small data transmission scenario. A. Continuous rate adaptation
With CRA transmission, the transmitter can transmit continuously for
E/P t time period,where P t is the transmit power. If the amount of energy E is relatively small and E/P t isless than a channel coherence time T c , then the MID of CRA can be calculated as H max ( E ) =( E/P t ) B log (1+ P t g/N B ) bits. The bits/joule energy efficiency becomes B log (1+ P t g/N B ) /P t ,which is changing with the instantaneous channel gain g . In particular, H max ( E ) is approximatelyproportional to log ( g ) for large g . When E is large and E/P t spans multiple T c ’s, the MIDwith CRA is determined as H max ( E ) = P Ni =1 T c B log (1 + P t g i /N B ) , where N is the numberof T c ’s and g i is the channel power gain during the i th T c . Here we assumed block fadingchannel, where the channel gain remains constant for one T c and changes to an independentvalue afterwards.Since MID is generally varying with the channel realization, we define the information outagerate (IOR) as the probability that MID for a given amount of energy E is less than a thresholdentropy value, denoted by H th . Mathematically, IOR is given by Pr[ H max ( E ) < H th ] . Apparently,the IOR analysis requires the statistics of H max ( E ) , which depends on the channel bandwidth,the channel statistics, and the adopted transmission strategy. For example, when E is small and E/P t is less than T c , the IOR with CRA can be calculated as IOR cra = F g (cid:20) N P t /B (cid:18) exp( ln(2) H th P t EB ) − (cid:19)(cid:21) . (4)For the scenario that E/P t involves multiple T c ’s, the IOR will be equal to the probability thatMID is less than H th , the evaluation of which will requires the distribution of the sum of N independent random variables. Further investigation of IOR for CRA transmission will be aninteresting topic for future research.Fig. 3 plots the IOR of CRA transmission as the function of threshold entropy H th for differenttransmission parameter settings over slow Nakagami fading channel. The amount of energyavailable for transmission usage is 80 mJ, which we assume can only support a transmission −4 −3 −2 −1 Threshold Entropy, bits I n f o r m a t i on O u t age R a t e B = 200 KHz, P t = 300 mWB = 400 KHz, P t = 300 mWB = 200 KHz, P t = 100 mWB = 400 KHz, P t = 100 mW Fig. 3. Information outage rate of ORA over slow Nakagami fading channel ( E =
80 mJ, m =
2, and g = -10 dB). duration of one T c . We can see that the IOR for all cases decrease with the threshold entropy.Larger channel bandwidth helps reduce the IOR for the same transmission power level, asexpected by intuition. On the other hand, similar to EOR performance, IOR increases withlarger transmission power for the same channel bandwidth. This is due to the fact that thetransmission time is reducing linearly with transmission power whereas the transmission rate isincreasing in logarithm with respect to P t . B. Continuous power adaptation
We now consider the MID analysis for CPA transmission strategy. Specifically, the transmitpower is adaptively set to maintain a constant receive SNR of γ c while satisfying the peak powerconstraint P max . As such, the transmission rate is fixed at B log (1 + γ c ) with transmit power γ c N B/g when g ≥ g T and equal to zero otherwise. Assuming slow fading environment where E can only support transmission over one channel coherence time, i.e. Eg/ ( γ c N B ) < T c , the −3 −2 −1 Threshold Entropy, bits I n f o r m a t i on O u t age R a t e γ c = 16 dB, P max = 5 W γ c = 16 dB, P max = 1 W γ c = 10 dB, P max = 5 W γ c = 10 dB, P max = 1 W Fig. 4. Information outage rate of CPA over slow Nakagami fading channel l ( E =
80 mJ, B =
200 kHz, m =
2, and g = -10 dB). MID can be calculated as H max ( E ) = Egγ c N log (1 + γ c ) . (5)We can see that the MID with CPA is linearly increasing with channel power gain g . Essentially,CPA transmission achieves higher energy efficiency than CRA at the cost of a certain probabilityof transmission outage.The IOR with CPA transmission can be calculated as the probability that H min ( E ) is less than H th . Noting that no transmission power will be consumed over a coherence time if the channelpower gain g is less than g T , the IOR for a certain amount of energy with CPA can be evaluatedas IOR cpa = (cid:20) F g (cid:18) γ c N E H th log (1 + γ c ) (cid:19) − F g ( g T ) (cid:21) / (1 − F g ( g T )) . (6)Fig. 4 illustrates the IOR performance of CPA over slow Nakagami fading channels. We againexamine the effects of peak transmission power and target received SNR during transmission. Wecan see that maintaining a higher target received SNR with CPA leads to larger information outage rate. This behavior can be explained by noting that higher γ c implies larger transmission powerduring transmission on average. We also observe from Fig. 4 that the peak transmission powerlevel has minimum effect on IOR performance unless the entropy threshold is very small. Similarto EOR performance, the IOR performance degrades slightly when P max increases. Smaller P max will ensure that the system transmits only over more favorable channel condition and reduce thetransmission power consumption on average.V. F URTHER CONSIDERATIONS
The above proposed data-oriented metrics characterize the energy efficiency performance ofindividual data transmission sessions over fading wireless channels. In particular, MEC prescribesthe smallest amount of energy required for transmitting a certain amount of data over fadingchannels, whereas MID signifies the largest amount information that can be transmitted witha given amount of energy. Given the time-varying nature of wireless fading channels, theseperformance limits are described in a statistical sense, in terms of EOR and IOR, respectively.By specifying the best possible performance for individual transmission session, these limitswill provide valuable guidelines to the development of practical energy-efficient transmissionstrategies for IoT applications.In previous sections, we illustrate the energy efficiency analysis of continuous rate and contin-uous power adaptive transmission strategies based on MEC and MID metrics. Both transmissionstrategies assume a certain channel state information (CSI) at the transmitter. The energy con-sumption associated with CSI acquisition was neglected in the analysis. When the amount ofdata is small, as in the ‘small data’ scenario for IoT applications, the extra energy needed forCSI provision at the transmitter may be comparable to the transmit energy consumption. Furtheranalysis on the overall energy consumption at the transmitter will be instrumental, especially forthe comparison with transmission strategies requiring no CSI at the transmitter.Adaptive modulation and coding (AMC) and automatic repeat request (ARQ) are two practicalrate-adaptive transmission strategies that explore limited feedback from the receiver. AMC adaptsthe transmission rate for a certain reliability requirement whereas ARQ enhance the reliabilitywith retransmission [13]. With the proposed data-oriented energy efficiency metrics, we cancompare the energy efficiency of AMC and AQR on the common ground of energy consumptionper transmission session. Such study will generate new design insights on energy efficienttransmission strategies for the limited CSI at the transmitter scenario. The power consumption at the transmitter includes transmission power and circuit power.The circuit power consumption is typically negligible compared with transmission power forconventional high power wireless transmission over long distance scenarios. Meanwhile, manyIoT devices can not afford high transmission power. In such scenarios, the circuit power mayeven dominates the overall power consumption [14]. Furthermore, to maintain the same outputpower, the power consumption of RF amplifier may vary with the chosen modulation scheme,as different modulation scheme will lead to different RF amplifying efficiency [15]. As such,the energy efficiency analysis with these practical considerations will entail new challenges.Energy harvesting is an essential technology for green IoT and will provide IoT deviceswith eternal power supply. Meanwhile, the amount of energy that can be harvested over acertain time period varies considerably. The MID analysis together with the energy arrivalprocess characterization will be essential to the successful design of the energy-aware schedulingalgorithms. The general design goal is to ensure that the IoT devices will have sufficient energyto complete their transmission with high probability. With the data-oriented energy consumptionanalysis, we can analyze and compare the performance of different scheduling algorithms fordiverse target applications. VI. C
ONCLUDING R EMARKS
In this article, we present a novel data-oriented approach for energy efficiency charaterizationof wireless transmission systems. We target at the small data transmission scenario for IoTapplications. In particular, we introduce two data-oriented performance limits on energy efficiencyfor arbitrary wireless data transmission. As their initial application, we analyze two channeladaptive transmission strategies and examine the effects of system parameters on their energyefficiency performance. We observe that the data-oriented approach can bring interesting newinsights on green wireless transmission over fading channels. This article serves as an initialintroduction to the data-oriented approach for green wireless transmission design. There are manyimportant aspects to be addressed, including the limited and no CSI at transmitter scenarios. Weexpect that the data-oriented perspective will stimulate promising novel design of green wirelesstransmission strategies for IoT applications.R
EFERENCES [1] F. Xia, W. Wang, T. M. Bekele, and H. Liu, “Big scholarly data: A survey,”
IEEE Transactions on Big Data , vol. 3, no. 1,pp. 18–35, Mar. 2017. [2] A. Zanella, N. Bui, A. Castellani, L. Vangelista, and M. Zorzi, “Internet of things for smart cities,” IEEE Internet of ThingsJournal , vol. 1, no. 1, pp. 22–32, Feb 2014.[3] S.-G. Chua and A. J. Goldsmith, “Adaptive coded modulation for fading channels,”
IEEE Trans. Commun. , vol. 46, no. 5,pp. 595–602, May 1998.[4] A. Sendonaris, E. Erkip, and B. Aazhang, “User cooperation diversity-part I: System description,”
IEEE Trans. WirelessCommun. , vol. 51, no. 11, pp. 1927–1938, Nov. 2003.[5] J. N. Laneman, D. N. C. Tse, and G. W. Wornell, “Cooperative diversity in wireless networks: Efficient protocols andoutage behavior,”
IEEE Trans. Inf. Theory , vol. 50, no. 11, pp. 3062–3080, Nov. 2004.[6] Y. Chen, S. Zhang, S. Xu, and G. Y. Li, “Fundamental trade-offs on green wireless networks,”
IEEE CommunicationsMagazine , vol. 49, no. 6, pp. 30–37, June 2011.[7] M. H. Alsharif, R. Nordin, and M. Ismail, “Survey of green radio communications networks: techniques and recentadvances,”
Journal of Computer Networks and Communications , Sept 2009, pp. 1–5.[10] L. Zhao, J. Cai, and H. Zhang, “Radio-efficient adaptive modulation and coding: Green communication perspective,” in , May 2011, pp. 1–5.[11] M. S. Alouini and A. Goldsmith, “Area spectral efficiency of cellular mobile radio systems,” in , vol. 2, May 1997, pp. 652–656 vol.2.[12] L. Zhang, H. C. Yang, and M. O. Hasna, “Generalized area spectral efficiency: An effective performance metric for greenwireless communications,”
IEEE Transactions on Communications , vol. 62, no. 2, pp. 747–757, February 2014.[13] A. Goldsmith,
Wireless Communications . Cambridge University Press, 2005.[14] S. Cui, A. J. Goldsmith, and A. Bahai, “Energy-constrained modulation optimization,”
IEEE Trans. Wireless Commun. ,vol. 4, no. 5, pp. 2349–2360, Sep. 2005.[15] T. H. Lee,