Dynamic Bandwidth Management for Energy Savings in Wireless Base Stations Anton Ambrosy, Michael Wilhelm, Wieslawa Wajda, and Oliver Blume Bell Labs Germany, Alcatel-Lucent Deutschland AG Stuttgart, 70435, Germany Email: {Anton.Ambrosy, Michael.Wilhelm, Wieslawa.Wajda, Oliver.Blume}@alcatel-lucent.com Abstract—The daily traffic variation over time and space requires dimensioning of base stations for high data transmission peaks. This means, that for a major part of the day power is wasted due to over provisioning. To save energy at lower traffic load scenarios we propose a stepwise adaptation of the bandwidth usage. LTE offers a wide spectrum of possible channel bandwidth usage, specified in steps of 1.4, 3, 5, 10, 15 and 20 MHz. A stepwise adaptation of the maximum number of resource blocks that are used during each LTE subframe can be applied to adapt the power consumption to the average traffic load. With the higher channel bandwidth more resource blocks can be utilized and more traffic load can be supported. The energy saving potential for macro base stations has been calculated for adaptive power amplifiers to be up to 20.9% and 25% along a day, based on daily dense urban and rural traffic profiles, respectively. Keywords: energy efficiency, LTE, macro base station, bandwidth, bandwidth adaptation, scheduler, power amplifier, power model, EARTH

I.

INTRODUCTION

Mobile data traffic strongly varies over time and space, which requires dimensioning of the network deployment for the highest data transmission peaks. This implies that for a major part of the day the capacity is over provisioned. Several approaches are possible to reduce the power consumption at low load, summarised and quantified recently by the EARTH project [1, 2], e.g. reducing the number of sectors or MIMO antennas, and switching off relays or capacity cells. These approaches require a dynamic network reconfiguration. Another strategy is a dynamic resource management that plays on the LTE resource utilisation, e.g. by micro Discontinuous Transmission (DTX) or by reducing the numer of resource elements [2]. The straight forward way for reducing the number of resource elements is by reducing the bandwidth. LTE offers a wide spectrum of possible channel bandwidth usages, specified in steps of 1.4, 3, 5, 10, 15 and 20 MHz [3]. With the higher bandwidth more resource blocks can be utilized and more traffic load can be supported. The downlink transmission range of a BS depends on the spectral power density, i.e. the transmit power per LTE OFDM resource element. Thus both the throughput and the power consumption of the radio BSs grow with the used transmission

bandwidth. This means that a broader bandwidth, i.e. more resource blocks, can be offered with increasing traffic load and a smaller bandwidth, i.e. less resource blocks, with decreasing traffic load. The main advantage of this approach is that the total transmitted maximum output power can be decreased when less resource blocks are used for scheduling user data without impacting the range of the cell. First preliminary results on such promising energy saving strategies have been already reported in [4]. The investigations were based on a single cell and were using simplified calculations without consideration of user distributions, user mobility and inter-cell interference effects. In that study green scheduler strategies have been investigated for Bandwidth adaptation (BW), Capacity adaptation (CAP) and DTX. These approaches (as explained in section III) need enhanced hardware features, especially a power amplifier (PA) which allows either to adapt the operation point according to the data traffic along a day or to deactivate the PA when no data are transmitted. In this paper we now present results from full system level simulations, which consider user distribution, user mobility and inter-cell interference. We focus on the technique of BW adaptation in this paper. II.

TRAFFIC LOAD

The traffic load of cells α = f(t,s) varies in time t and space s and is connected to the population density. The project EARTH has found that the daily variation of the number of active users is analogous to the daily variation of the traffic [5, 9] (see Figure 1). In practice, even in busy hour the cell is not fully loaded to avoid blocking. The maximum resource utilisation at busy hour is typically in the range of 60% for macro BSs in dense urban areas compared to less than 30% in rural areas. For an LTE macro cell deployment with the typical inter-site distance of 500 m in dense urban areas and 1732 m in rural environments, this corresponds to the high data traffic load scenario in [9] of up to 120 Mbps/km2 in dense urban and 4 Mbps/km2 in rural deployment areas, respectively. Therefore, for a large part of the day the traffic load is low or very low. The consumed energy could be lowered significantly when the system can adapt its capacity to the actual traffic load. Figure 2 shows for example the dense urban scenario with a maximum resource utilisation of 60% and how the bandwidth

usage can be assigned to the daily traffic load profile of a macro BS with a maximum bandwidth of 10 MHz. This pattern allows for time slots with the assignments of capacity limits corresponding to different bandwidth usages as specified in LTE. The average data rate during night is low enough that a bandwidth of 1.4 MHz can satisfy the user requirements for a period over almost 5 hours.

its fixed operation point (OP) designed for low distortion amplification [10]. At present, this DC power supply is fixed independently of the traffic load and thus, for a major part of the day, power is wasted. From an energy consumption perspective, it is more advantageous to tune the amplifiers’s saturation point based on the load and the channel conditions [6], i.e. changing the transceiver’s OP to provide a ‘just good enough’ signal to noise and distortion ratio performance.

Data traffic daily profile for Europe

BW adaptation can be achieved by coordinating the scheduling characteristic with an adjustment of the PA operating point to the required output power. Apart from the PA load the BW adaptation also changes the amount of signal processing operations in the baseband board; thus additional savings are achieved.

100%

80%

Traffic

60% EARTH reference values Profiles

40%

20%

120

High 30

Mid Low

6 Dense urban traffic peaks [Mbps/km2]

0% 0h

1h

2h

3h

4h

5h

6h

7h

8h

9h 10h 11h 12h 13h 14h 15h 16h 17h 18h 19h 20h 21h 22h 23h Time EARTH reference profile for Europe

Figure 1. Illustration of data traffic average daily profile α(t) in Europe[5] Capacity Limits

18

BW = 10.0 MHz

Average data rate [Mbit/s]

16 14 12 10

BW = 5.0 MHz

8 6

BW = 3.0 MHz

4

In current 3GPP standards an adaptation of the bandwidth needs to be signaled to the mobile terminals in a procedure based on a reconfiguration of cell parameters. In the future a much faster approach may be possible, using fast carrier aggregation procedure, which is currently under discussion in 3GPP. An alternative approach is CAP adaptation, which is introduced in [4]. CAP adaptation always uses the full channel bandwidth of a cell and the number of pilots remains unchanged. However, the adaptation to lower load is performed by scheduling only a fraction of the LTE subcarriers (i.e. limiting the number of scheduled LTE PRBs). This approach reduces the RF transmission power similarly to the BW adaptation approach, however it does not allow to reduce the number of reference symbols. On the other hand, this approach is transparent to the mobile terminals and maintains the full frequency diversity. The CAP adaptation will be not examined here in closer detail, because system level simulations are currently ongoing.

BW = 1.4 MHz

2 0 0

4

8

295 min @ 1.4 MHz

12 Time [h]

16

20

24

150 min @ 10.0 MHz

Figure 2. Offered data rate per cell of a macro BS for BW adaptation (coloured boxes) according to the daily traffic profile (blue line)

III.

BANDWIDTH ADAPTATION

BW adaptation is based on the adjustment of the bandwidth to the required traffic load. Depending on traffic load the LTE, bandwidth can be stepwise downscaled so that lower numbers of physical resource blocks (PRBs) are allocated. The LTE specification requires sending reference signals (“pilots”) for channel quality measurements in every PRB. Therefore, with BW adaptation also less reference signals are used, this additionally reduces the power consumption of a BS. State-of-the-art (SOTA) PAs usually show a linear increase of power consumption over load. In this case the overall energy consumption for a given data volume is independent of the data rate used when sending the data. However, large savings become possible for the BW adaptation technique when it is supported and coordinated with an adaptive BS hardware, especially by an adaptive PA. In macro BSs the PA is the component with the highest energy consumption. Even when no data is transmitted the PA consumes DC power for holding

IV.

POWER MODEL

The energy saving potential of the above approach has been evaluated by using a detailed power model, which is a function of traffic load. The model is described in the EARTH Energy Efficiency Evaluation Framework (E3F) [5, 9]. A 10 MHz macro cell typical radiates up to 40 W output power out of the PA. The characteristic of a Doherty PA based on theoretical evaluations has been used as base line for our investigations. This state-of-the-art (SOTA) PA consumes over 100 W DC input power at full load, see Figure 3. For BW adaptation we introduce an adaptive PA with energy saving features [6]. The lower curves show the improvement of the power characteristic when adapting the OP (OP2-OP8) by changes of its DC supply voltage. The use of a lower voltage limits the maximum output power due to saturation of the PA. A further improvement of the PA characteristics regards its performance when no data is sent at all. If the PA supports a fast deactivation of the amplifier stage its power consumption can be reduced to only 4.4 W. This gives the possibility to enable micro sleep modes even during single OFDM symbols (duration of 1/14 ms) with no transmission. This feature has been discussed for micro DTX in [4, 11]. Next to the PA, the EARTH power model comprises the baseband processing and the RF signal generation, which

consume about Pother = 25 W per antenna of a SOTA BS. Further, the efficiency η of DC/DC conversion, AC/DC power supply, and cooling causes additional power consumption. For SOTA the total overhead power is around 70 W and nearly independent of the load. Fig 4 shows the additional power consumption in blue (light and dark) on top of the consumption of the PA. In the following we compare the power consumption of a SOTA BS to an adaptive BS. First, applying the best possible OP, the light orange area in Figure 4 is the saving potential from using BW adaptation and an adaptive PA. Second, we consider the saving that can be achieved by applying improvements of the other BS components according to EARTH results [6]. This basically turns the overhead proportional to the load and no active cooling is required. The total power consumption of a cell then is Pcell = (PPA+Pother)/η with Pother ~ 0.25*PPA for the SOTA, i. e. at all load levels the PA consumes 65% of the total power. These figures are given for illustration, in the simulation the detailed EARTH model is used. The saving from the improved hardware efficiency over SOTA is indicated as the light blue area in Fig. 4..Note that the proportional overhead reduces further when using the adaptive power amplifier, this indirect saving is not indicated in Fig. 4. 110 90

OP2

70

The scheduling strategy for BW adaptation is illustrated in Figure 5 for an example resource utilisation where 40% of the available resources are used by data. In this case the scheduled user data, control signaling and reference symbols are shown for a standard LTE radio frame. For the used example a bandwidth of 5 MHz already meets the load requirements for the BW adaptation approach. The transmitted maximum average output power of 40W can be decreased because less than 50% of the resource blocks are used for scheduling user data and signaling. Then the OP of the adaptive PA can be modified by reducing the supply voltage so that the power amplifier operates more closely to its most efficient OP (i.e. OP4 in Figure 3, with up to 20W of RF output power). In addition the reference signal overhead is reduced compared to a bandwidth usage of 10 MHz. BW adaptation thus plays on the frequency axis of the resource management. For comparison, Figure 6 shows an

OP1=SOTA

80 OP3

60

1.0

OP4

50

0.9

OP5

40

0.8

OP6

30

OP7

20

OP8

10

PA Deactivation

0 0

5

10

15 20 25 RF Output Power [W]

SCHEDULER

The scheduler sets the principal physical parameters for the transmission at each subframe and, in consequence, decides about the radiated power. For the BW adaptation modes the scheduler aims to minimise empty resources. This is achieved by setting a maximum resource utilisation and by equal utilisation of PRBs within this limit over a given time period. Depending on this maximum utilisation an optimised PA characteristic with limited output RF power can be used.

30

35

40

Load per Symbol

Power Consumption of PA [W]

100

V.

0.7 0.6 0.5

Saturation limit in OP4

0.4 0.3 0.2 0.1

Figure 3. PA characteristics over RF output [6], depending on OP settings (OPi with i=1,…,8) Overhead Power Amplifier PA Sleep mode

OFDM Symbols along 10 ms

Figure 5. Scheduling for BW adaptation: Resource utilisation per symbol at 40% user load 1.0

150

0.9 0.8

100

Load per Symbol

Power Consumption per Cell [W]

200

0.0

50

0.7 0.6 0.5 0.4 0.3 0.2

0 0%

10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Load (as fraction of resource elements)

Figure 4. SOTA power consumption per cell over the scheduled load and the saving potential of the Adaptive PA and of improved hardware efficiency

0.1 0.0 OFDM Symbols along 10 ms

Figure 6. Scheduling for miro DTX : Resource utilisation per symbol at 40% user load

alternative scheduling strategy for the same resource utilization of 40% user load, which aggregates scheduled resources on the time axis to create as much as possible empty symbols which can be used for deactivation of the power amplifier (micro DTX [11]). In [4] we have shown that BW adaptation has higher potential especially for cells with lower average load, e.g. rural cells. VI.

SYSTEM LEVEL SIMULATIONS

System level simulations have been performed with a hexagonal cell layout containing 19 macro base stations each serving 3 cells as defined in [7]. For all simulations wrap around has been applied to avoid edge effects. Other common parameters are the usage of a carrier frequency of 2 GHz, a 3D antenna model for SISO, a video streaming application with constant bit rate of 2 Mbps per user, as well as shadowing with a correlation distance of 50 m and a correlation of 0.5 between sites. Furthermore, an inter-site distance of 500 m and 1732 m and a user mobility of 3 km/h and 30 km/h have been implemented for dense urban and rural scenarios, respectively. Active users are dropped on and move over the simulation playground according to the varying traffic demand over a day. The power consumption per area unit of a cell has been calculated by applying the EARTH Energy Efficiency Evaluation Framework (E3F) [5, 9] for each transmitted OFDM symbol. This includes the daily EARTH traffic load model for dense urban and rural scenarios as well as the EARTH power model for macro BSs with Remote Radio Heads according to the year 2012.

The calculated power consumption per area unit along a day (a) as well as the daily energy savings as function of the peak system throughput (b) are depicted in Figure 7 for dense urban scenarios. In case of low peak system throughput daily energy savings up to 26% can be achieved by adapting the OP of the PA. When also other components of the BS (as the base band processing, the RF part, the limited efficiency of DCDC conversion and ACDC power supply) are able to adapt to the lower load, much higher energy savings up to 70% can be achieved compared to SOTA. Further, in this case a passive cooling has been considered which alone reduces the power consumption by around 10%. The simulation results for rural scenarios are illustrated in Figure 8 for the power consumption per unit area (a) and the daily energy savings vs. peak system throughput (b). These results are very similar compared to the dense urban scenario, but due to the 12-fold larger cell size they show a much lower power consumption per area and a much lower peak system throughput. Even for rather high data traffic profiles still significant savings can be achieved. Daily energy savings have been evaluated for macro BSs under the load scenarios defined in section II. Energy savings of 20.9% along a day can be achieved in dense urban scenario at a peak system throughput of 120 Mbps/km2 under consideration of adaptive improvements in the PA, respectively of 53.6% for adaptability in all components including the PA. For the rural scenario the daily energy consumption can be reduced by 25% and 64.7% at a peak system throughput of 4 Mbps/km2. 0.16

For 120 Mbps/km² Dense Urban

2.0

1.5

(a) 1.0 SOTA

0.5

BW Adaptation (PA only) BW Adaptation (all components)

0.0 0

4

8

12 Time [h]

16

20

Power Consumption per Area [kW/km2]

(a)

Power Consumption per Area [kW/km2]

2.5

0.04

SOTA BW Adaptation (PA only) BW Adaptation (all components)

0.00 4

8

12 Time [h]

16

20

24

80

60 50 40

BW Adaptation (PA only)

70

BW Adaptation (all components)

Dense Urban

(b)

30 20 10

Daily Energy Savings [%]

Daily Energy Savings [%]

0.08

0

BW Adaptation (PA only)

70

(b)

0.12

24

80

For 4 Mbps/km²

Rural

BW Adaptation (all components)

60 50 40

Rural

30 20 10

0

0 0

40

80 120 160 200 240 Peak System Throughput [Mbps/km 2]

280

Figure 7. Power consumption per area unit along a day for peak system throughput of 120 Mbps/km2 (a) and daily energy savings vs peak system throughput (b) for dense urban scenario

0

4 8 12 16 Peak System Throughput [Mbps/km2]

20

Figure 8. Power consumption per area unit along a day for peak system throughput of 4 Mbps/km2 (a) and daily energy savings vs peak system throughput (b) for rural scenario

VII. INTERFERENCE COORDINATION Dynamic bandwidth management can be applied to further improve the performance in terms of energy savings by minimization of inter-cell interference. This is achieved by coordinating the unused part of the bandwidth between next neighbour BSs. As a limiting case, we consider a scenario which completely avoids interference of next neighbours. Therefore, system level simulations have been performed using a reuse factor of 3 for the above determined daily traffic profiles in dense urban and rural scenario in combination with the BW adaptation approach. For these investigations a full reuse scheme with a factor of 3 has been considered, i.e. an allocation of 3 x 10 MHz bandwidth according to 3 neighbour cells. Figure 9 illustrates the daily energy savings as function of the peak system throughput for the dense urban scenario (a) and rural scenario (b). In both cases the standard LTE reuse scheme with factor of 1 (1 x 10 MHz) is compared to a full reuse scheme with factor of 3 (3 x 10 MHz). The results show that much higher energy savings can be achieved if a reuse scheme with factor 3 is applied. The daily energy savings are increased for dense urban scenarios at a peak system throughput of 120 Mbps/km2 from 20.9% for reuse 1 to 30.8% for full reuse 3, if improvements of the PA only are considered. Similar improvements can be observed for the rural scenario at a peak system throughput of 4 Mbps/km2. In this case daily energy savings are increased from 25% for reuse 1 to 28.5% for full reuse 3. 35

Daily Energy Savings [%]

(a)

BW Adaptation (Reuse 1, 1x10MHz) BW Adaptation (Reuse 3, 3x10MHz)

20 Dense Urban

15

VIII. CONCLUSION AND FURTHER WORK

0 0

40

80 120 160 200 240 Peak System Throughput [Mbps/km 2]

280

35 30

Daily Energy Savings [%]

When spectrum is limited to 10MHz a reuse scheme of 3 allows for only 3MHz per cell, which would not be sufficient during the more busy hours (see Figure 2). We here propose a more practical and realistic scheme based on a partial reuse scheme with a rollover. In this scheme each cell of a radio access network is assigned a preconfigured third of the resource blocks for scheduling. However, when the preconfigured resources are exhausted, also the resources assigned to the next neighbour cells may be employed up to the maximum bandwidth usage of 10 MHz. The benefit of the scheme is that resources can be used dynamically, while the assignment of the preconfigured fraction of the resource is static and no signaling between cells is required during operation. The assignment can preferably be done using a distributed SON algorithm [12].

10 5

(b)

In conclusion, especially in dense urban scenarios much higher daily energy savings are achievable when a full reuse scheme with factor of 3 is implemented. For realistic deployments a full reuse scheme with factor of 3 is not applicable, because the available spectrum of operators is often restricted. However, this scheme is used as upper bounding scenario showing the limit that can be achieved by avoiding inter-cell interference from direct neighbours.

In this allocation schemes interference is not completely eliminated at higher load. It is therefore expected that the achievable savings are between the reported values of the standard LTE reuse scheme with factor of 1 and the full reuse scheme with factor of 3. System level simulations with the partial reuse scheme are planned for future investigations.

30 25

Higher daily energy savings can be achieved when considering higher peak system throughputs. Improvements of all components show also much higher savings: 64.9% for full reuse 3 compared to 53.6% for reuse 1 in dense urban scenario and 68.2% for full reuse 3 compared to 64.7% for reuse 1 in rural scenario.

BW Adaptation (Reuse 1, 1x10MHz)

25

BW Adaptation (Reuse 3, 3x10MHz)

20 Rural

15 10 5 0 0

4 8 12 16 Peak System Throughput [Mbps/km2]

20

Figure 9. Energy savings for different reuse schemes based on improvements by the PA only for dense urban scenario (a) and rural scenario (b)

Current wireless LTE macro cell networks cannot adapt their capacity dynamically to the varying traffic load. Bandwidth adaptation is a promising method for reducing the energy consumption during low traffic periods. The BW adaptation approach requires either a fast carrier aggregation procedure, which is currently not standardised but under discussion in 3GPP, or a much slower procedure based on a reconfiguration of cell parameters. System level simulations have been performed to evaluate the power consumption of macro BSs in dense urban and rural scenarios. In this study the user distribution, user mobility and inter-cell interference have been considered. Energy savings along a day have been calculated for SOTA BSs and the proposed BW adaptation approach by using the EARTH power model. Furthermore, impacts by inter-cell interference on the power consumption of BSs have been investigated and a promising new partial reuse scheme with a factor of 1 has been proposed for LTE. In this case the maximum number of resources is splitted equaly for 3 neighbour cells. Only if the preconfigured resources are exhausted, the resources of the

next neighbour cells may be employed up to the maximum bandwidth usage. We have demonstrated that significant energy saving is possible without impacting the system performance. Using the BW adaptation approach highest energy savings can be achieved during night at lowest traffic load. Taking into account only the features of an adaptive PA energy savings of up to 20.9% and 25% can be achieved for dense urban and rural environments, respectively, for the BW adaptation approach along a day. If improvements of all components of a BS are considered much higher energy savings can be achieved. The daily energy consumption can be reduced by 53.6% and 64.7% for a peak system throughput of 120 Mbps/km2 and 4 Mbps/km2 in dense urban and rural scenario, respectively. We have discussed how interference limitation by reuse factor can further improve the energy efficiency. A full reuse scheme with factor of 3 shows much higher energy savings for dense urban scenarios. At a peak system throughput of 120 Mbps/km2 energy savings of 30.8% for full reuse 3 can be achieved compared to 20.9% for the standard LTE reuse scheme with factor of 1, if improvements of the PA only are considered. The discussed partial reuse scheme with factor of 3 will lead to energy savings, which are between the reported values of the standard LTE reuse scheme with factor of 1 and the full reuse scheme with factor of 3. This will be studied in future work. The findings of all investigated configurations are summarized in Table I. TABLE I.

COMPUTED DAILY ENERGY SAVING FOR BW ADAPTATION Adaptive PA, reuse 1

Adaptive PA, reuse 3

Adaptive BS, reuse 1

Adaptive BS, reuse 3

Dense Urban @ 120 Mbps/km2

20.9%

30.8%

53.6%

64.9%

Rural @ 4 Mbps/km2

25.0%

28.5%

64.7%

68.2%

Daily saving

Further energy savings are expected also when combinations with other green scheduling strategies are considered. Promising solutions are combinations of procedures based on BW adaptation with micro DTX, which benefits from the deactivation of the PA when no user data is transmitted. Another interesting use case is in dynamically managed heterogeneous networks, where the BW of macro cells can be adapted when additional small cells are activated. During low traffic times small cells are switched off and the affected macro cells return to higher bandwidth to accommodate the traffic. With an integrated solution 50%-70% of total saving will be achievable in future green wireless LTE deployments [8]. We believe that BW adaptation will be a major contributor to this. Its application is most promising in rural cells but also very attractive in dense urban areas.

ACKNOWLEDGMENT The research leading to these results has received funding from the European Community's 7th Framework Programme FP7/2007-2013 under grant agreement n° 247733 - project EARTH. REFERENCES [1]

“EARTH (Energy Aware Radio and neTwork tecHnologies)”, EU funded research project FP7-ICT-2009-4-247733-EARTH, Jan. 2010 to June 2012, see https://www.ict-earth.eu [2] István Gódor (Ed.), “Final Report on Green Network Technologies”, public deliverable D3.3 of EARTH, June 2012 [3] 3GPP TS 36.104 V8.12.0 (2011-06), Base Station (BS) radio transmission and reception (Rel. 8) [4] A. Ambrosy. O. Blume, H. Klessig, W. Wajda, “Energy saving potential of integrated hardware and resource management solutions for wireless base stations”, PIMRC Toronto, September 2011 [5] M.A. Imran and E. Katranaras (Eds.) “Energy efficiency analysis of the reference systems, areas of improvements and target breakdown”, , public deliverable D2.3 of EARTH, November 2010 [6] R. Torrea (Ed.), “Most Promising Tracks of Green Radio Technologies“,public deliverable D4.1 of EARTH, December 2010 [7] H. Holtkamp and A. Ambrosy (Eds.), “Definition and parametrization of reference systems and scenarios”, public deliverable D2.2 of EARTH, June 2010 [8] M. Olsson, A. Feshke, L. Hevizi, A. Vidacs, I. Godor, P. Fezekas, M. Imran, Y. Qi, O. Blume, “Integration strategy of EARTH Energy Efficiency Enablers”, FN&MS Berlin, July 2012 [9] G. Auer, V. Giannini, I. Godor, P. Skillermark, M. Olsson, M.A. Imran, D. Sabella, M.J. Gonzalez, C. Desset, and O. Blume, “Cellular Energy Efficiency Evaluation Framework”, GreeNet workshop, Budapest, in Proceedings of VTC-Spring, 2011, pp. 1-6 [10] M. Gonzales, D. Ferling, W. Wajda, A. Erdem, P. Maugars, “Concepts for Energy Efficient LTE Transceiver Systems in Macro Base Stations", FN&MS Warsaw, 2011 [11] P. Frenger, P. Moberg, J. Malmodin, Y. Jading, and I. Gódor, “Reducing Energy Consumption in LTE with Cell DTX”, VTC-2011Spring, Budapest, Hungary, May 2011 [12] C.G. Gerlach, I. Karla, A. Weber, L. Ewe, H. Bakker, E. Kuehn and A. Rao, “ICIC in DL and UL with network distributed and self-organized resource assignment algorithms in LTE”, Bell Labs Technical Journal Vol 15(3), December 2010, Special Issue: Self-Organizing and SelfOptimizing Networks, pp. 43-62