Energy Efficient Radio Resource Management Strategies for Green Radio

Transcription

1 Energy Efficient Radio Resource Management Strategies for Green Radio Congzheng Han, Simon Armour Centre for Communications Research, University of Bristol, Bristol, UK Abstract Mobile communication industries are increasingly contributing to the worldwide energy consumption and CO2 emission. This paper addresses a number of ey radio resource management strategies across PHY and MAC layers for reducing base station energy consumption, as measured by a Joules per bit metric. These strategies including power efficient lin adaptation, exploitation of multi-user diversity and trading bandwidth for energy efficiency. By collectively taing advantage of those radio resource management strategies, a multi-user adaptive power and resource allocation algorithm is proposed to ease the power requirements of a base station, whilst maintaining the same levels of service to the user. The scheduling algorithm is applied to an LTE downlin simulator and its performance is evaluated for various traffic load conditions. The results show that the proposed algorithm achieves a significant energy saving (up to 86%) over a conventional non-energy aware resource allocation scheme. Furthermore the energy efficiency performance of various multiple antenna techniques is evaluated along with the impact of control signaling overhead. These multiple antenna schemes are then incorporated into the proposed scheduling algorithm and the additional achievable energy savings are quantified. Index Terms Green Radio, Energy Efficiency, Radio Resource Management 1

2 I. INTRODUCTION The information and communication technology (ICT) industries are responsible for around 2% of the global emissions of carbon dioxide and this figure is expected to reach 3% by 2020 [1]. Within 18 months, the number of worldwide mobile phone subscribers increased from 4 billion to over 5 billion by year 2010 according to Wireless Intelligence [2]. As one branch of the ICT, the energy consumption of mobile cellular networs is increasingly contributing to the global climate change as mobile data traffic continues growing dramatically [3]. In addition, the growing energy costs are becoming a significant OPEX (operational expense) for mobile operators. The concept of green radio is to develop environmentally friendly, low-power and energy efficient solutions for future wireless networs. The goal of the core 5 Green Radio programme of the Virtual Centre of Excellence in Mobile and Personal Communications (MVCE) [4][5] is to achieve a 100-fold reduction in energy use without compromising the quality of service (QoS) of the end user. Reduced energy consumption translates directly to lower carbon emissions and operating expenditure cost for wireless networs. Long term evolution (LTE) is the latest standard in the mobile networ technology tree that previously realised the GSM/EDGE and UMTS/HSPA networ technologies that currently dominate over 85% of the mobile phone global maret [6]. It is introduced as Release 8 in the 3rd generation partnership project (3GPP) [7]. The new evolution aims to reduce pacet delays, improve spectrum flexibility and further reduce the cost for operators and end users. To meet the aggressive performance targets, LTE employs advanced physical layer (PHY) techniques, including orthogonal frequency division multiplexing (OFDM) for downlin transmission and multiple-input and multiple-output (MIMO) techniques [8]. MIMO techniques are nown to be able to improve the data rate, reliability or both over a SISO system. One popular MIMO technique is space-time bloc coding (STBC) which is able to achieve full transmit diversity and enable reliable communication. 2

3 Thus in LTE, an Alamouti [9] [10] based Space-Frequency Bloc Coding (SFBC) technique is adopted in the standard and will be considered in this paper. In LTE, spatial multiplexing (SM) MIMO technique is also considered and its gain is accomplished by simultaneously sending different data streams over the same radio resource, which can dramatically increase throughput and bandwidth efficiency. Previous wor generally focuses on the system reliability and rate [11]. In this paper, the energy efficiency of MIMO and SISO schemes are evaluated and compared and the control signaling overhead is also taen into consideration. Radio resource management (RRM) strategies can improve the utilisation of limited radio spectrum resources and radio networ infrastructure. These strategies allow the system to have control of parameters such as transmit power, channel allocation, modulation order and error coding scheme, etc. As summarised in [12], most of the current literature on joint RRM strategies for general OFDM-based or LTE systems focuses on maximising system capacity for a given QoS requirement and the published wor on energy-aware joint RRM strategies are much sparser. Lin adaptation (LA) is an adaptive RRM scheme that allows transmission parameters to be selected according to current channel conditions. Traditionally the wor primarily focuses either on improving throughput or reducing error rate [13]. Previous study has demonstrated a constant gap between the channel capacity and the maximum efficiency of the adaptive scheme for a target BER [14]. From an energy-saving point of view, as a low order modulation scheme generally requires less transmit power to maintain a specific lin quality requirement, power-aware lin adaptation strategies can help to improve energy efficiency by adopting the most energy-saving transmission mode according to the current quality of a channel. A modulation optimisation strategy for minimising energy consumption is proposed in [15] for a single antenna, coded/uncoded M- QAM system in additive white Gaussian noise (AWGN) channels for short-range communications. Recent wor in [16] demonstrates a global lin adaptation solution to achieve improved energy savings for uncoded M-QAM OFDM systems in frequency selective fading channels. The minimum energy LA 3

4 strategy proposed for MIMO-OFDM in [17] extends the number parameters under the control of the lin adaptation scheme to include not only the modulation order, but also the MIMO scheme and number of antennas. The fundamental bandwidth-power trade-off is presented in [18] and shows the minimum received signal energy per bit for reliable communication over AWGN channels approaches -1.59dB when the transmission bandwidth approaches infinity. This presents the opportunity to trade bandwidth for energy savings. Therefore, if the system is not operating at full load, an opportunity exists to allocate the spare spectrum to users and allow them to switch from highly spectrally efficient modes to modes with lower spectral efficiency to achieve energy savings while maintaining the rate target. Furthermore, if nowledge of the propagation channel is available at the transmitter, channel aware radio resource allocation is another effective RRM scheme and it can be applied to assign different sub-set of sub-carriers to users in an orthogonal frequency division multiple access (OFDMA) system to potentially exploit the multi-user diversity inherent in multi-user systems operating over frequency-selective channels. Previous wor has shown that multi-user diversity can be exploited by effective resource allocation algorithms and this gain can be translated into energy savings [19]. The idea of designing lin adaptation and resource allocation schemes that emphasize energy efficiency is introduced in [20] based on numerical analysis of a single antenna system in a flat fading channel scenario. The contribution of this paper is to combine MIMO and all the potential energy saving RRM techniques and evaluate the achievable energy savings in an LTE downlin context. In this paper, the relationship between spectral efficiency and the required transmit power level measured in Joules per bit for SFBC, SM MIMO and SISO is presented. The analysis shows how much RF energy in Joules is required at the base station for different MIMO scheme to deliver a certain spectral efficiency. This result is particularly important as it indicates the MIMO scheme and MCS level to switch to for most energy savings as load or rate requirement varies. The paper also studies the impact of the control signalling overhead of MIMO schemes on the required energy per bit. 4

5 Furthermore, this paper also evaluates the capabilities of various radio resource management strategies to reduce the total power consumption needed for a wireless system operation based on an LTE downlin simulation. The concept of scaling energy needs with traffic is raised as a way to reduce transmit energy requirement in this paper. As a method based on this concept, the potential energy savings that can be achieved through bandwidth expansion at relative low traffic load are investigated. A novel, adaptive multi-user resource and power allocation strategy is proposed, which taes advantages of various radio resource management strategies including the power-aware lin adaptation, exploitation of multi-user diversity and trading of bandwidth for energy efficiency and combine theses schemes with MIMO techniques to achieve energy reduction while meeting the QoS target. The results show how the RF transmit power varies with traffic load and the important role of the RRM scheme plays in achieving energy savings at the base station. The results also reveal the impact of the adaptive scheme on the amount of allocated resources and transmit power level of individual users. Although the proposed RRM scheme is evaluated in an LTE downlin context, the technique is applicable to centralised systems in general. This rest of the paper is organised as follows. Section II describes the LTE downlin simulator and the considered channel scenario. The advantage of lin adaptation in achieving energy savings as a function of distance is presented in section III. The energy efficiency performances of SISO and MIMO schemes including SFBC and SM at different levels of spectral efficiency are evaluated in section IV with the consideration of control signaling overhead. Section V formulates the problem of energy efficient lin adaptation and resource allocation and introduces an adaptive multi-user resource and power allocation strategy. Together with several conventional RRM strategies, they are applied to the LTE downlin and the potential energy saving gains in varying traffic conditions throughout a day are evaluated based on simulation results. Finally, the paper is concluded in section VII. 5

6 II. SYSTEM AND CHANNEL MODEL DESCRIPTION A. System Model The core of the LTE downlin radio is defined by the conventional OFDM with data transmitted over several parallel narrowband sub-carriers. The use of the narrowband sub-carriers in combination with a cyclic prefix maes OFDM transmission particularly robust to multipath fading inherent in radio propagation. Considering a multi-user scenario, the performance analysis is performed on the downlin of a 3GPP LTE Orthogonal Frequency Division Multiple Access (OFDMA) system. The total system bandwidth is divided into sub-channels, denoted as physical resource blocs (PRBs). PRBs are the smallest elements of resource allocation assigned by the base station scheduler and they are then allocated to different users for multiple access purposes. The ey parameters of the considered LTE OFDMA downlin system are given in Table 1. Instead of feeding bac channel quality indicators (CQI) for all the sub-carriers, a single CQI (based on the average quality of the 12 grouped sub-carriers comprising the PRB) can be fed bac for each PRB and is assumed to be perfectly nown at the BS. Perfect channel estimation is also assumed. The simulation results presented here assume a single, unsectorised cell configuration. A representative sample of the modulation and coding schemes (MCSs) adopted by LTE are used, indicated in Table 2. The effective data rate of a MCS is R = (number of data subcarriers per symbol coding rate number of coded bits per symbol Number of OFDM symbols per time slot)/duration of time slot. B. Channel Model The channel model used in the simulations is the Spatial Channel Model Extension [20] (SCME) Urban Macro scenario, specified in 3GPP [23]. SCME provides a reduced variability tapped delayline model which is well suited for lin level as well as system level simulation. A low spatially correlated channel is assumed for all the users where 10λ spacing at the BS is employed independently and identically distributed (i.i.d.) channel realisations are considered in each 6

7 simulation. The pathloss for an urban macro environment is based on the modified COST231 Hata urban propagation model and is adopted for the considered scenario [23]: where ( db) log ( ) PL 10 d = (1) d is the distance between the BS and user, and is required to be at least 35m. III. ENERGY-AWARE LINK ADAPTATION A. Introduction Lin adaptation schemes allow transmission parameters to be selected according to current channel conditions and wor to date generally focuses on provison of high QoS, which is often considered in the one dimensional form of throughput. From an-energy saving point of view, different modulation schemes offer a tradeoff between the transmit power required to achieve reliable communication and the time/frequency resources required to transmit the same ammount of data. This tradeoff offers an opportunity for adaptation in order to improve energy efficiency and the selection of good modulation parameters for energy efficiency are often different from those for throughput maximisation. B. Performance Analysis Figure 1 shows the simulated performance of individual MCS and the overall system of SISO LTE in the SCME channel. The achievable average throughput shown in Table 2 is given by, Throughput = R(1-PER) where R and PER are the bit rate and the residual pacet error rate for a specific mode respectively. The overall throughput envelope is obtained by using ideal adaptive modulation and coding (AMC) based on the (throughput) optimum switching point. To investigate the performance of employing lin adaptation strategies for energy savings, the pea data rate is defined to be the maximum rate that can be achieved by the highest MCS. The data load factor (LF) is defined as the proportion of the required user rate divided by the pea data rate. 7

8 The transmit energy efficiency of a system can be represented by the Energy Consumption Rate (ECR) metric given by [24]: PT P ECR = = (J/bit) (2) RT R where P and R are the required average transmit power and data rate respectively and they can be either obtained or derived based on Figure 1. T is the period of time of transmission over which the ECR is calculated. The amount of energy savings achieved by a LA scheme can be evaluated using the energy reduction gain (ERG): ERG ECR TEST ( LA) = 1 % (3) ECR REF where ECR REF denotes the ECR of the reference system (considered to be operating at a fixed transmit power of 46dBm in this paper) and ECR TEST (LA) is the ECR of the considered system operating at whatever transmit power is required to match the reference system). Under different loading conditions, the performance of transmit energy reduction gain is evaluated in terms of users distances as shown in Figure 2. This figure assumes that all receivers are at the same distance or that only a single receiver is supported. In the more realistic case, the actual energy saving will be the average of that achieved for each user. Nevertheless, this figure serves to illustrate the relevant issues and the multi-user case is considered later in the paper. By adapting the MCS level to the current user s channel condition, the lin throughput performance presented in Figure 2 has shown that the required SNR can be significantly lowered if a lower rate requirement is applied and this decrease in required SNR corresponds to the same amount of power saving at the transmitter. For example, at a fixed BS-MS distance of 100m, as the system load decreases from 100% to 25%, 70% of ERG is achieved by adapting to a lower spectral efficient mode which requires less transmit power. For a maximum transmit power of 46dBm, under different loading conditions, the operating ranges of the system are the values where the 8

9 curves intercept the horizontal axis in Figure 2. As expected, as the load decreases, the system switches to more power efficient transmission modes and is therefore able to support a longer range. IV. ENERGY EFFICIENCY OF SISO AND MIMO In the LTE standard, an Alamouti based Space-Frequency Bloc Coding (SFBC) technique is specified [9][10]. This transmit diversity based method does not provide a linearly increasing channel capacity as the number of transmit and receive element grows simultaneously. Only one data stream (or layer) is transmitted for 2 or even 4 antennas. System capacity is another important issue in wireless communications apart from transmission reliability. Spatial Multiplexing (SM) is another MIMO technique considered in this paper. By transmitting independent symbols over different antennas as well as over different symbol times, SM is capable of increasing the capacity of the MIMO channels linearly with respect to the minimum number of available transmit and receive antennas [28]. Conventional wor generally focuses on comparison of SISO and MIMO schemes in terms of reliability, coverage and data rate for a fixed transmit power level [11]. In this section, for specific data rates, the energy efficiency of these schemes employing different levels of modulation order and coding rates are evaluated. A. Energy Efficiency Performance Evaluation without Consideration of Overhead In this section, results are present without consideration of transmission overhead. Table 2 shows the maximum spectral efficiency η of SISO, 2x2 SFBC and SM OFDM systems employing different MCSs and the lin throughput performances of the overall systems are presented in Figure 1. Note that for the same modulation order and coding rate, SM achieves double spectral efficiency compared to SFBC and SISO. 1) Energy Consumption for Fixed Rate Requirement 9

10 Based on the results shown in Figure 1, the ECR of the single-user MIMO and SISO schemes for a range of spectral efficiency levels are derived and presented in Figure 3. (Note that the BS-MS distance is assumed to be 100m in order to obtain absolute ECR values but that the relative ECRs of the different MIMO/SISO cases hold true for other ranges.) The lower the ECR, the higher the energy efficiency is. For more details, Table 3 presents the most energy efficient MCS level adopted by different schemes and the corresponding ECR values for a specific spectral efficiency target. For any spectral efficiency requirement higher than bits/s/hz, the ECR of SM is lower than that of SISO and it shows SM is more energy efficient. SFBC is the most energy efficient scheme due to diversity gain when the target spectral efficiency is lower than 3.78 bits/s/hz. For any higher spectral efficiency requirement, the SM scheme becomes the most energy efficient and it is also the only scheme that can support a spectral efficiency greater than 4.32 bits/s/hz. The results show that MIMO with lower modulation achieves better energy efficiency than SISO with higher modulation. For low to medium spectral efficiency requirement, SFBC is generally preferred to SM as a result of diversity gain. In addition, this energy saving achieved by MIMO becomes more significant as the spectral efficiency requirement increases. 2) Energy Consumption for Varing Traffic Loads Figure 4 shows average traffic loads during one day [25]. This data was supplied by Vodafone and shows average traffic over 28 days in the London area. The average hourly traffic profile is normalised to its busiest hourly traffic load. In this paper, the busiest hourly traffic load is normalised to be 100% of system capacity. The pea load is equivalent to the maximum capacity of a SISO system. The traffic loading condition varies significantly over time, with the lowest load around 6am and nearly full load around 9pm. The average traffic load throughout the day is around 66.4% of the maximum. In addition to SM and SFBC scheme, adaptive MIMO is also considered and it selects the most energy efficient MIMO scheme between SFBC and SM based on the channel conditions. The ERG 10

11 of 2x2 SFBC, SM and adaptive MIMO over the SISO reference scheme for the traffic loading cycle profile in Figure 5 is presented in Figure 5. The average ERG values over the 24 hour cycle are listed in Table 4 and can be obtained from: 1 24 ERGREF, h ERGTEST. h ERG 24 = % (J/bit) (4) 24 ERG h= 1 REF, h The results are consistent with the ones presented in Table 3 showing that SFBC is more energy efficient in low traffic load conditions and SM is preferred when the traffic load is above 90%. Employing adaptive MIMO allows an average ERG over SISO of over 81%. B. Energy Efficiency Performance Evaluation with Consideration of Overhead During the LTE downlin transmission, control signalling overhead generally taes approximately 5-30% of the total sub-frame [26]. Assuming Control Channel Element (CCE) aggregation of 8, the control signaling size for a SISO and MIMO systems are 9.01% and 13.81% respectively [26]. Control signaling overhead is transmitted either at pea power or average power. For a CCE size of 8, the ECR of different schemes considering overhead transmission at pea power (46dBm / 40W) is presented in Table 5 as case 1. The ECR of the schemes when overhead is transmitted at the same power level as the user data is also listed in the Table 5 as case 2 for comparison. The power level of the overhead shows a significant impact on the ECR of all schemes at low spectral efficiency range as the power required by transmitting user data is relatively low. As a result, the ECR of all schemes for low spectral efficiency transmission is significantly increased. Furthermore, as MIMO has a larger overhead than SISO, it requires more energy for transmitting control signaling overheads. As result, for a CCE size of 8, the switching point between SISO and SM increases to 0.468bps/Hz and 0.883bps/Hz for pea power and average power transmission of overhead. However, this shift of switching point due to overhead is not very significant for both cases because of the low transmit power requirement for the low spectral efficiency range. The 11

12 switching spectral efficiency level of SFBC and SM remains the same as both MIMO schemes experience the same amount of overhead. V. POWER CONTROL, LINK ADAPTATION AND RESOURCE ALLOCATION IN A MULTI-USER SYSTEM A. Resource Allocation for OFDMA An OFDMA system with an overall target rate of Rtotal is considered. Users are uniformly distributed in a single cell and thus experience different SNR, based on their relative location from the serving base station (BS). No inter-user interference is considered. With the channel information of a user, the transmission power allocated to the c th physical resource bloc (PRB) of th user is expressed as: P, c f ( M, PER ) = (J/bit) 2 (5) α, c where α, c is the channel gain factor. α, c can be expressed as: where 2 H PL G G 2, c Tx Rx α, c = (6) NP H, c is the normalised frequency response averaged across all the sub-carriers in PRB c. PL is the path loss and NP is the noise power. gains, respectively. ( M PER ) G Tx and G Rx are the transmit and receive antenna f, is the required received power with unity channel gain for achieving M bits per PRB per time slot at below or equal to the target PER. f ( M, PER ) { f ( D() i, PER ), K, f ( D( I), PER )}, where i is the MCS index. The information of the supported amount of bits D ( i) and minimum received SNR requirements for the 9 MCSs can be obtained from Figure 1. In LTE, it is suggested that all PRBs allocated to one user in a sub-frame use the same modulation and coding scheme (MCS) [22]. As a result, the transmission power allocated to a user depends on selected MCS scheme i and the average channel 12

13 conditions to the user : α across C number of PRBs assigned to that user, and P ( i, ) C C 2 ( D( i), PER) α ( C ) C is the set of PRBs assigned f = (7) and if the antenna patterns are assumed to be isotropic in the considered scenario, α becomes: PL 2 α H (8) ( C ) = α, c = C c C C NP c C, c Multi-user diversity (MUD) refers to the increase in overall multi-user capacity achieved via an opportunistic resource allocation strategy for which the scheduler assigns resources according to the users instantaneous channel conditions in time, frequency or/and space domain. Within one cell, users near the BS generally have a higher received SNR than the users near to the boundary of the cell. If the data rate requirement of each user is the same, a user near the cell boundary may need a higher level of transmit power to achieve the same rate with those users near the BS. Previous results of LA have shown that switching to a lower spectrally efficient MCS level is able to save energy. Therefore, if spare spectrum is available, bandwidth can be traded for energy savings. The amount of energy savings that can be achieved through trading spectrum depends on the individual user s current received signal level, fast fading channel condition and the operating MCS level. Compared to random or fixed assignment of the available spare spectrum, dynamically allocating the spectrum to the right users can help to achieve further energy saving. B. Adaptive Multiuser Resource and Power Allocation (MRPA) A multi-user resource and power allocation scheme is proposed here for which the objective is defined as minimising the total transmit power requirement P T subject to a given user s rate requirement: 13

14 P T = min K = 1 P ( i, C ) = min K = 1 C f ( D() i, PER) 2 α ( C ) (9) subject to: R = C M for all (10) The resource allocation here jointly considers fair resource allocation, lin adaptation and bandwidth expansion. In order to benefit from trading bandwidth for energy reduction, the proposed algorithm is based on the fact the overall system traffic is not fully loaded, which implies that all the users rate requirement can be met if transmit power is high enough to support the highest MCS level with the highest spectral efficiency. At the initial stage, the algorithm starts with the most spectrally efficient MCS level which can deliver the highest amount of bits in one PRB. Each user is then assigned to a best remaining PRB (highest channel gain) in turn, and this process terminates once all the users are assigned the minimum required number of PRBs to meet their rate requirement when operating at this most spectrally efficient MCS level. The required transmit power is then calculated. There are C PRBs available for transmission, C = { 1, K, c, K,C} denoted C. The details of this initial stage are described as follows:, and each user is assigned a number of PRBs Step 1) set K P T = P = 1 = 0, D( I) M =, = 0, and = I. Ran users in descending order U i in terms of average channel gain Step 2) for = 1 to K if U < R C α = α, c and { α K α < Kα K} C c= 1 <. find * c = arg maxα c C 2, c 14

15 end if end for loop U = U D( I) + C = C { c * } and C= C \{ c * } Step 3) if U >= R for each user, calculate P, go to the next bandwidth scaling process else go bac to step 2 end if In the previous process, the target of the scheduler is to transmit as much data as possible on each PRB to meet users rate requirements without considering the transmit power level. The following bandwidth scaling stage intends to fill the un-allocated PRBs and assign these resources to the users which can achieve the most transmit energy reduction through trading bandwidth. It lowers the received SNR requirement by using a lower modulation and coding scheme through assigning more PRBs. The second stage is described as follows: Step 1) K ( is a set of users ( ( K K ) such that i > 1 (users are not currently assigned the lowest MCS level) for each K ( Calculate the additional number of PRBs b required to support a lower MCS = ceil( R D i 1 ) C level while meeting the rate target, ( ) b 15

16 Step 2) find user Find b PRBs with the highest channel gains from the remaining unallocated PRBs to form a set C ˆ = { cˆ, K, cˆ,, cˆ } cˆ n 1 n K b { cˆ, K, cˆ, K, c } 2 1 ˆ, c : C = n b 1 c C ˆ, c C\C, where arg max ˆ 2 2 α, α α, m = 1, K, n 1 + =, cˆ, ˆ m c m 1 Calculate the savings in power P = P ( i, C ) P ( i 1, C Cˆ ) end for * that achieves the highest positive power reduction by using a lower MCS scheme through bandwidth expansion * = max P arg K (, P * > 0 Step 3) if C * C ˆ * C * =, C= C C ˆ * and i * i * 1 M * = D( *) i C Ø and K ( : P > 0 \ = go to step 1 else scheduling process finishes end if To illustrate how the MRPA algorithm wors, an example is given in Figure 6. A small number of users and PRBs are used in the example and transmit power and MCS selection are described qualitatively in order to eep the illustration simple. There are 5 PRBs in total, and 2 users in the system are each requesting 1/5 of the maximum system throughput. User 1 has much worse channel condition than user 2, and to achieve the same rate target, user 1 therefore requires much higher transmit power. At stage 1, each user is allocated a PRB in turn (in a manner which exploits 16

17 nowledge of the channel quality to achieve multiuser diversity) and transmits using the highest MCS level which requires the highest transmit power. User 1 has the priority to be be allocated its best PRB during this process due to its poor channel condition. At the end of the stage, the users have been allocated 2 out of a maximum 5 PRBs, and this gives an opportunity to allocate additional PRBs to users and switch to a lower MCS level to save energy. This happens during stage 2 when, at every iteration, an additional PRB is allocated to a user. In this example, since User 1 is currently operating at higher transmit power, it is allocated an additional PRB at iteration 1 and its MCS and transmit power consequently reduce. At the second iteration, User 1 s transmit power is now lower than that of User 2, and so the best remaining PRB is allocated to User 2 to reduce transmit power and allow a lower MCS to be adopted by User 2. At the third iteration, User 1 s transmit power is now higher again so the final available PRB is allocated to User 1. C. Other Resource Allocation Strategies In addition to the proposed MRPA algorithm, three other scheduling algorithms are also considered in this paper for comparison purposes. 1) Scheme 1 All users transmit at the highest MCS level which supports the highest rate. Each user is assumed to require a certain number of PRBS, C R D( I) =, which represents the minimum amount of spectrum resources needed to achieve the target rate. The scheme 1 is similar to the stage 1 of the MRPA but the PRBs are assigned to users on a round-robin approach. In this scenario, the BS only allows KA C of the total PRBs to be used for this transmission. Scheme 1 is an example of a fixed scheduling approach which does not consider lin adaptation, bandwidth scaling or multi-user scheduling strategies. The transmit power required based on scheme 1 is considered as the power consumption of the reference system. 17

18 2) Scheme 2 ( Bandwidth Expansion Only) Compared to scheme 1, scheme 2 utilises all the available spectrum resources regardless of the system load conditions. All the available PRBs are fairly distributed to all the users based on a round-robin approach. When the system is not fully loaded, each user is given more PRBs than needed to allow a more energy efficient MCS to be adopted while meeting the rate requirement at the same time. 3) Scheme 3 (MUD Only) The allocation procedure of scheme 3 is the same as the stage 1 of MRPA scheduling algorithm. Similar to the scheme 1, each user is assigned A PRBs, but users suffering from severe path loss have the priority in the selection process to select their best PRBs for transmission. Compared to scheme 1, scheme 3 exploits spectral multi-user diversity by assigning the best PRB to each user. D. Performance Evaluation The maximum load can be supported when the BS is transmitting at the highest MCS level with a pea overall rate R pea of 43.2 Mbps for a SISO system. There are 5 users uniformly distributed in a single cell and each of them requests 20% of the overall system data rate simultaneously. The maximum range is assumed to be 300m. 1) SISO Employing Various Resource Allocation Strategies The LA relies on a predefined table of operational SNR thresholds for each MCS level produced based on the PER requirements of a SISO OFDM system. Based on this LA looup table produced from an AWGN channel, the average energy reduction gains achieved by the SISO employing various radio resource allocation schemes over the reference scheme 1 are calculated for different time of the day based on traffic load conditions. Figure 7 shows the ERG achieved by different schemes without PER constraint. 18

19 The trends of the ERG curves show that the spare spectrum resources in low loading conditions generally help to save transmit energy compared to the highly loaded conditions. When the load is only 25%, scheme 2 achieves nearly 90% energy saving over scheme 1 by assigning the remaining spectrum to users and allowing them to operate at a MCS level with low spectrum efficiency and low energy consumption. In a fully loaded system, as expected, scheme 2 achieves no energy saving over the reference scheme as no spare resources are available. Note that the slight lower loading condition does not necessarily improve energy saving achieved by scheme 1 when the change in load is not significant. This is because the transmit power required by the reference scheme also varies with the load, and relative energy saving in percentage may not be a representation of the exact amount of energy saving. By just exploiting the multi-user diversity, scheme 3 shows over 65% of energy reduction at low loading condition (25%). Even when the system is fully loaded, over 45% of energy reduction can still be achieved due to multi-user diversity gain. The MRPA scheme, which benefits from both multi-user diversity and bandwidth expansion, achieves an overall ERG of 95% at a loading condition of 25%. The average ERG24 of different schemes in one day is presented in Table 6 and the MRPA scheme achieves an average ERG24 of 79% over the reference scheme throughout the day. As expected, when the system is fully loaded, the energy reduction gain achieved by the MRPA scheme only comes from exploitation of multi-user diversity. Figure 8 presents an example of CDF of ERG over 100 different cell configurations when the system is half loaded. For each configuration, the result is averaged across 500 SCME channel instances, updated after each subframe. The CDF plots show that 85% energy reduction is achieved by the MRPA over 95% of the time in the considered scenario. 2) Sensitivity Analysis of ERG Values This section will investigate the sensitivity of ERG values to some ey parameters including the assumptions of LA and PER constraints. 19

20 a) Sensitivity of ERG to PER constraints When a PER constraint of 1% enforced for QoS purposes, the ERG averaged across 24-hour within one day for both scenarios are listed in Table 6 showing negligible differences between a system with and without PER constraint. b) Sensitivity of LA assumptions If the generation of predefined LA tables is based on the SCME channel model, the average ERG performances are presented in Table 6. Due to the fading, the SNR thresholds for each MCS level in a SCME channel model are generally higher than the case of an AWGN model and the difference in SNR thresholds between adjacent MCS levels is larger. Therefore, based on LA table produced in SCME channel model, scheme 3 ( Bandwidth Expansion ) yields greater benefits under LA table based on SCME channel model and the MRPA scheme achieves 5% increase in overall ERG 24. 3) MIMO Employing Various Resource Allocation Strategies The proposed adaptive power and resource allocation algorithm can also be applied to a 2x2 MIMO system. The achievable rate of SM MIMO system is doubled compared to a SISO or SFBC system for the same modulation order and coding rate. An adaptive MIMO scheme is also considered and it selects less energy consuming MIMO scheme for a specific throughput requirement. To meet a PER requirement of 1%, by adopting a single-user SISO system as the reference scheme, the ERG achieved by SISO, SFBC, SM and adaptive MIMO employing MRPA scheduling algorithm is presented in Figure 9. The statistics of the ERG values of different scenarios can be found in Table 7. All MIMO schemes employing MRPA achieves 30% more energy reduction compared to SISO employing MRPA. SFBC is preferred to SM for low to medium traffic load, and SM achieves more energy saving for high traffic load conditions. Applying MRPA to an adaptive MIMO system achieves the highest ERG and achieves an average saving of 97.80% 20

21 4) Energy Saving Over Spectrum Allocation This section will study the relationship between achieved energy saving and the allocated spectrum resources by the proposed MRPA algorithm. For an overall SISO system load of 22% (at 6am according to Figure 4), 5 users requesting an equal share of the overall system rate are considered for resource allocation. The path losses experienced by users are presented in Table 8 and a user suffering a large path loss generally requires a high transmit power level to meet a specific rate target compared to users experiencing less path losses. At stage 1 of the MRPA algorithm, each user is assigned the minimum required number of PRBs and a high MCS level of 8. At stage 2, the algorithm iteratively assigns extra PRBs to the user who achieves the most energy saving by operating at a lower MCS level and this process is presented in Figure 10. The width of the columns represents the number of extra PRBs allocated to a user at stage 2 of the scheduling process and Figure 10 also shows the corresponding ERGs achieved by operating at a lower MCS level over the require transmit power at stage 1. As more PRBs are assigned to users, the energy saving becomes less significant showing a diminishing effect on energy saving through spectrum allocation. VI. CONCLUSIONS This paper investigates the capabilities of various radio resource management strategies to improve transmit power efficiency. By assigning resources to user s instantaneous channel conditions in time, frequency or / and space domain, multi-user diversity can be exploited and the performance gain can be translated into energy savings. For a specific rate requirement, employing energy-oriented lin adaptation strategy allows the system to adapt the most energy efficient MCS to the current user s channel condition to reduce the transmit power requirement. Low spectrally efficient MCS with lower modulation order and stronger coding rate is more energy efficient. As a 21

22 result, if the system is not fully loaded, while meeting the QoS requirement, transmit energy reduction can be achieved by assigning the spare spectrum resources to a user and operating at a lower spectral efficient MCS. A novel adaptive resource and power allocation algorithm has been proposed. By exploiting the advantage of lin adaptation, multi-user diversity and allocation of spare spectrum, the algorithm shows 79-86% energy reduction averaged over an example daily loading cycle. This energy reduction gain may vary with parameters such as PER constraint requirement, LA assumptions and choice of reference scheme. The paper also examines the energy consumption of a 2x2 MIMO with lower order modulation and SISO with higher order modulation and the results reveal that MIMO is more energy efficient for the same spectral efficiency requirement. For the considered scenario, 72% energy saving is achieved by using SFBC over SISO for a spectral efficiency requirement of 1 bits/s/hz and this saving becomes more significant for higher spectral efficiency level. Although overhead in MIMO is greater than SISO, results show SISO is only preferred for LTE when the required spectral efficiency is very low. These results show the great potential of radio resource management strategies in reducing energy consumption, in line with the targets of the Green Radio research programme (Core 5) of Mobile VCE. Future wor will consider extending the model to account for non-linear efficiencies of the power amplifier. ACKNOWLEDGEMENTS The authors wish to acnowledge the financial support of the Core 5 Green Radio Programme of Mobile VCE and than all partners to the project for their discussions and feedbac on this wor. 22

23 REFERENCES [1] ICT and CO2 Emissions, Parliamentary Office of Science and Technology, Dec [2] Over 5 Billion Mobile Phone Connections Worldwide, BBC News, 9 July [Online]: [3] G. P. Fettweis, E. Zimmermann, ICT Energy Consumption Trends and Challenges, in Proc. Of 11th International Symposium on Wireless Personal Multimedia Communications, Lapland, Finland, Sept [4] Mobile VCE Vision group, 2020 Vision: Enabling the Digital Future, December [5] C. Han et al., Green Radio: Radio Techniques to Enable Energy Efficient Wireless Networs, IEEE Communication Magazine, special issue on Green Communications, Vol. 49, No. 6, pp , June [6] Long Term Evolution (LTE): A Technical Overview, Motorola, June 2010, [Online] rs/wireless%20operators/lte/_document/static%20files/6834_motdoc_new.pdf [7] 3GPP; Technical Specification Group Radio Access Networ; (E-UTRA) and (E-UTRAN); Overall Description; Stage 2 (Release 8), 3GPP TS V8.8.0, March [Online] [8] Q. Li, et al., MIMO Techniques in WiMAX and LTE: a Feature Overview, IEEE Communications Magazine, Vol. 48, No. 5, May 2010, pp [9] S. Alamouti, A Simple Transmit Diversity Technique for Wireless Communications, IEEE JSAC, Vol. 16, No. 8, pp , [10] E. Dahlman, S. Parvall, J. Sold, P. Beming, 3G Evolution: HSPA and LTE for Mobile Broadband, Academic Press, [11] K. Peppas, et al., System Level Performance Evaluation of MIMO and SISO OFDM-based WLANs, Wireless Networs, Vol. 15, No. 7, Oct. 2009, pp [12] K. Pedersen, et al., An Overview of Downlin Radio Resource Management for UTRAN Long-Term Evolution, IEEE Communications Magazine, Vol. 47, No. 7, July 2009, pp [13] S. Catreux, V. Erceg, D. Gesbert, R. W. Heath, Jr., Adaptive Modulation and MIMO Coding for Broadband Wireless Data Networs, IEEE Communications Magazine, Vol. 50, No. 6, June 2002, pp

24 [14] A. J. Goldsmith, S-G. Chua, Variable-Rate Variable-Power MQAM for Fading Channels, IEEE Trans. Commun., Vol. 45, No. 10, Oct. 1997, pp [15] S. Cui, A. J. Goldsmith, A. Bahai, Energy-constrained Modulation Optimisation, IEEE Transactions on Wireless Communications, Vol. 4, No. 5, Sept. 2005, pp [16] G. Miao, N. Himyat, G. Y. Li, Energy-Efficient Lin Adaptation in Frequency-Selective Channels, IEEE Trans. on Communications, Vol. 58, No. 2, Feb pp [17] H. S. Kim, B. Daneshrad, Energy-Constrained Lin Adaptation for MIMO OFDM Wireless Communication Systems, IEEE Trans. on Wireless Communications, Vol. 9, No. 9, Sept. 2010, pp [18] S. Verdu, Spectral Efficiency in the Wideband Regime, IEEE Trans. on Information Theory, Vol. 48, No. 6, June 2002, pp [19] C. Han, K. C. Beh, M. Nicolaou, S. Armour, A. Doufexi, Power Efficient Dynamic Resource Scheduling Algorithms for LTE, IEEE VTC 2010 Fall GreeNet Worshop, Ottawa. [20] S. Zhang, Y. Chen, S. Xu, Improving Energy Efficiency through Bandwidth, Power, and Adaptive Modulation, 1st GreeNet Worshop, Sept [21] D. S. Baum, J. Hansen, and J. Salo, An Interim Channel Model for Beyond-3G Systems: Extending the 3GPP Spatial Channel Model (SCM), VTC Spring [22] S. Sesia, I. Toufi, M. Baer, LTE The Long Term Evolution: From Theory to Practice, John Wiley & Sons Ltd., [23] Spatial Channel Model for MIMO Simulations, 3GPP TR V9.0.0, Dec [Online] [24] B. Badic, T. O'Farrell, P. Losot, J. He, Energy Efficient Radio Access Architectures for Green Radio: Large versus Small Cell Size Deployment, IEEE VTC Fall, September [25] J. He, T. O'Farrell, Boo of Assumptions, v.1.6.0, Core 5 Research Programme: Green Radio, September [26] R. Wang, J. Thompson, H. Haas, Practical Issues of Time-Domain Base Station Sleep Mode Design, Interim Contributory Report, Mobile VCE Core 5 Green Radio, July [27] R. Love, R. Kuchibhotla, A. ghosh, R. Ratasu, B. Classon and Y. Blanenship, Downlin Control Channel Design for 3GPP LTE, IEEE Wirelss Communications and Networing Conference, Las Vegas, USA, 2008, pp [28] P. W. Wolniansy, G. J. Foschini, G. D. Golden, R. A. Valenzuela, V-BLAST: An Architecture for Realizing Very High Data Rates Over the Rich-scattering Wireless Channel, 24

25 Bell Laboratories, 1998 URSI International Symposium on Signals, Systems, and Electronics,

26 Table 1: Parameters for LTE OFDMA Downlin Transmission Bandwidth 20 MHz Time Slot/Sub-frame duration 0.5ms/1ms Sub-carrier spacing 15Hz Sampling frequency MHz (4x3.84MHz) FFT size 1024 Number of occupied sub-carriers 601 Number of OFDM symbols per time slot 7 CP length (µs/samples) (4.69/72)x6, (5.21/80)x1 BS Tx Power 46dBm (40W) Propagation Model SCM Urban Macro Noise Power -104dBm User Equipment Noise Figure 6dB Table 2: Modulation and Coding Schemes (10Mz) MCS Modulation Coded bits per symbol Coding Rate Bit Rate R (SISO, SFBC/SM) Spectral Efficiency η (SISO, SFBC/SM) 1 QPSK 2 1/3 5.6/11.2 Mbps 0.560/1.120 bps/hz 2 QPSK 2 1/2 8.4/16.8 Mbps 0.840/1.680 bps/hz 4 QPSK 2 3/4 12.6/25.2 Mbps 1.260/2.520 bps/hz 3 16 QAM 3 1/3 11.2/22.4 Mbps 1.120/2.240 bps/hz 5 16 QAM 3 1/2 16.8/33.6 Mbps 1.680/3.360 bps/hz 6 16 QAM 3 3/4 25.2/50.4 Mbps 2.520/5.040 bps/hz 7 64 QAM 4 3/ /60.48 Mbps 3.024/6.048 bps/hz 8 64 QAM 4 3/4 37.8/75.6Mbps 3.780/7.560 bps/hz 9 64 QAM 4 6/7 43.2/86.4 Mbps 4.320/8.640 bps/hz 26

27 Table 3: ECR for Different Spectral Efficiencies Achieved by SISO and MIMO SISO SFBC SM η Mode ECR Mode ECR Mode ECR ( µ J / bit ) ( µ J / bit ) ( µ J / bit ) 1 bps/hz 16QAM 1/ QAM 1/ QPSK 1/ bps/hz 64QAM 3/ QAM 3/ QAM 1/ bps/hz 64QAM 6/ QAM 6/ QAM 3/ bps/hz Not supported Not supported 64QAM 3/ bps/hz Not supported Not supported 64QAM 6/ Table 4: Minimum, Maximum and Mean ERG of Various MIMO Schemes over SISO Averaged Across 24-hour within a Day Scheme 2x2 SFBC Only 2x2 SM Only Adaptive MIMO ERG Min. Value 66.46% 54.30% 66.46% Max. Value 90.02% 96.40% 96.40% Mean Value ( ERG 24 ) 80.35% 76.50% 81.67% Table 5: ECR Achieved by SISO, SFBC and SM with the Consideration of Overhead with Pea Power (case 1) and Average Power (case 2) Transmission (CCE =8) SISO (ECR µ J / bit ) SFBC (ECR µ J / bit ) SM (ECR µ J / bit ) η Case 1 Case 2 Case 1 Case 2 Case 1 Case 2 1 bits/s/hz bits/s/Hz bits/s/hz Table 6: Average ERG 24 of Different Schemes over Scheme 1 during One Day ERG Scheme 2 Scheme 3 MRPA 24 AWGN with No PER Constraint 50.24% 55.15% 79.76% AWGN with PER 1% Constraint 50.57% 55.53% 77.71% SCME with PER 1% Constraint 70.03% 55.47% 85.96% 27

28 Table 7: ERG of SISO and MIMO Employing MRPA over Reference Scheme SU SISO in SCME Channel ERG Min Value Max Value Mean ( ERG 24) SISO MRPA 50.91% 83.20% 64.23% SFBC MRPA 94.31% 98.60% 97.26% SM MRPA 86.75% 98.93% 94.01% Adaptive MIMO MPRA 94.31% 99.52% 97.80% Table 8: An Example of 5 Users Employing the Proposed MRPA at a System Load of 22% (at 6am based on Figure 4) User Index Path Loss (db) No. of Assigned PRBs At Stage Operating MCS Level at Stage No. of Assigned PRBs by MRPA Operating MCS Level of MRPA

29 Throughput (Mbps) QPSK 1/3 QPSK 1/2 QPSK 3/4 16QAM 1/3 16QAM 1/2 16QAM 3/4 64QAM 3/5 64QAM 3/4 64QAM 6/7 SISO and SM switching point SISO Lin Throughput SFBC Lin Throughput SM Lin Throughput SFBC and SM switching point SNR (db) Figure 1. Simulated Lin Throughput for all the considered MCSs of SISO, 2x2 SFBC and SM OFDM 29

30 Energy Reduction Gain (ERG) 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% LF = 25% LF = 50% LF = 75% LF = 100% BS- MS Distance (m) Figure 2. Energy Reduction under Various User Locations and Fixed Data Rate 30

31 6 5 SISO 2x2 SM 2x2 SFBC Energy Consumption Rate (µj/bit) SISO and SM switching point SFBC and SM switching point Spectral Efficiency (bits/s/hz) Figure 3 Energy Consumption Rate of a Single-user SISO, 2x2 SFBC and SM 31

32 Figure 4. Normalised Traffic Load in 3G Networ in London (24 Day Average) [25] 32

33 Figure 5. ERG of SFBC, SM and Adaptive MIMO over SISO 33

34 Figure 6. An Example of Assigning PRBs to Users based on MRPA Algorithm 34

35 Figure 7. Average ERG achieved by Scheme 2, 3 and the MRPA over the Reference Scheme at different hour of the day based on LA in AWGN Channel with no PER constraint 35

36 Figure 8. CDF of ERG Achieved by Different Schemes over Reference Scheme for a System Load of 50% in AWGN Channel with PER constraint of 1% 36

37 Figure 9. Average ERG of SISO, SFBC, SM and Adaptive MIMO Employing MRPA over SU-SISO for the Same Traffic Load Condition 37