Uplink Power Control in LTE – Overview and Performance Principles and Benefits of Utilizing rather than Compensating for SINR Variations Arne Simonsson and Anders Furuskär Wireless Access Networks, Ericsson Research [arne.simonsson, anders.furuskar]@ericsson.com anders.furuskar]@ericsson.com
—Uplink —Uplink power control is a key radio resource Abstract management function. It is typically used to maximize the power of the desired received signals while limiting the generated interference. This paper presents the 3GPP long Term Evolution (LTE) power control mechanism, and compares its performance to two reference mechanisms. The LTE power control mechanism constitutes of a closed loop component operating around an open loop point of operation. Specifically, the open loop component has a parameterized fractional path loss compensation factor, enabling a trade-off between cell edge bitrate and cell capacity. The closed-loop component can be limited to compensate for long-term variations, enabling fast channel quality variations to be utilized by scheduling and link adaptation. Simulation results indicate that the LTE power control mechanism is advantageous compared to reference mechanisms using full path loss compensation and SINR balancing. The fractional pathloss compensation can improve the cell-edge bitrate and/or the capacity with up to 20% while at the same time battery life time is improved. The fast SINR balancing closed loop mechanism performs poorly at high load since it does not utilize the link adaptation and the full link performance capability in LTE. E-UTRA; LTE; power power control; uplink uplink
I.
I NTRODUCTION
Power control is a crucial radio network function in cellular systems. This paper describes the LTE power control for the Physical Uplink Shared Channel (PUSCH) [1], discusses different applications of it, and evaluates its performance for different parameter settings. The focus is on the benefit of fractional pathloss compensation, first proposed in [2]. A performance performance comparison comparison to an SINR balancing p ower ower control control scheme is also included. The paper is outlined as follows. In Section II power control is discussed both in general and specifically for LTE as well as the studied power control principle. The simulation assumptions are described in Section III followed by results in Section IV. Finally, conclusions are drawn in Section V. I I.
POWER CONTROL
Power control refers to setting output power levels of transmitters, base stations in downlink and mobile stations in uplink, with an objective to improve system capacity, coverage
978-1-4244-1722-3/08/$25.00 ©2008 IEEE.
and user quality (data rate or voice quality), and to reduce power power consumption. consumption. To reach these th ese objectives, objectives, power-control power-control mechanisms typically aim at maximizing the received power of desired signals, while limiting the generated interference. In the downlink, a simple and efficient power control strategy, used in most recent system concepts, is to transmit with a constant output power. Often the maximum base station power power is used. Variations Variations in channel channel conditions conditions and interference levels are adapted to by means of scheduling and link adaptation rather than with power control. This strategy obviously maximizes the received power. The generated interference power is instantaneously high, but the interference energy generated for a given amount of data transferred is minimized by maximizing the data rate, thus minimizing the transmission time, through scheduling and link adaptation. The LTE uplink power control may be considered as a means to apply this downlink concept in the uplink direction.
A. Uplink Power Power Control in in LTE The LTE uplink is orthogonal, meaning that there is, at least in the ideal case, no interference between users in the same cell but only interference between cells. The amount of interference generated to neighbor cells depends, among other things, on the mobile-terminal position, or more specifically the pathgain from the terminal to these cells. Terminals close to neighbor cells generate more interference than terminals far away. For a given generated interference level in a neighbor cell, terminals far away from that cell may hence transmit with a higher power than terminals near the cell. Further, there is a correlation between being close to the serving cell and being far away from neighbor cells. All these characteristics are utilized in the LTE uplink power power control. The orthogonal orthogonal LTE uplink uplink allows allows multiplexing multiplexing of terminals with different received uplink power within the same cell. On the short term scale this means that peaks in multipath fading may be utilized through scheduling and link adaptation to increase the data rates, rather than compensated for by reducing power. On the long-term, the received power target is further set based on the pathgain to the serving cell, so that terminals that generate little interference may have a larger received power target. The LTE uplink power control is based on both signal strength measurements done by the terminal it self (open-loop
1
power control) as well as measurements by the base station, used to generate transmit power control commands that are subsequently fed back to the terminals as part of the downlink control signaling (closed-loop power control). The fractional path loss compensation is done in the open-loop but controlled with a factor Į by the network [1]. In more detail, the openloop component of the LTE power control is defined by: PSD = P0 + Į • PL [dB]
(1)
where PSD is the transmitted power spectral density, PL is the estimated pathloss, and P0 is a parameter used to control the SNR target (see [3] for this relationship). There are several features in LTE to support closed loop control [1]. A fast 2-bit Transmit Power Control (TPC) f(i) that can be sent in each uplink scheduling grant to the UE controlling each subframe i enabling up to 1kHz update rate. This TPC is relatively to the open loop setting and can be either accumulated as in UTRA or absolute. There is also Transport Format (TF) selection dependent power offset ǻTF (TF(i)) where ǻTF is a table configured by higher layers with one entry for each transport format TF . This can also be used to control the power each scheduling grant by scheduling smaller transport formats reducing the transmitted power. In addition to this P 0 can be individually controlled for each UE P 0UE enabling a slow power control adjustment.
B. Power Control Algorithms Studied in this Paper A set of simple basic uplink power control principles are studied. They represent different usage and parameter settings of the LTE power control mechanism. 1) No Power Control (No PC) Fixed transmission power, the UE power is set to P = P max, where Pmax is the maximum UE power. Used as a reference case. This can be applied in LTE by setting Į=0 and P0 = Pmax. 2) Open Loop Power Control, Į=1 Open loop with a fixed received SNR target. Į=1 and P0 = SNR target +Pnoise, where Pnoise is the noise power level, SNR target is a targeted received power level relative to the noise floor. Note that the desired SNR target with this algorithm must include a margin for expected interference. Values of SNR target between 0 and 30dB have been simulated. 3) Open Loop Power Control, Į=0.7 Open loop with fractional path loss compensation. Į = 0.7 and P0 = SNR target +Pnoise. All the compensation factors Į in the LTE standard (0.4, 0.5, 0.6, 0.7, 0.8, & 0.9) have been simulated in combination with SNR target between 0 and 30dB. Į=0.7 was found to give a good trade-off between cell-edge bitrate and capacity, as will be shown in the results. 4) Closed Loop Power Control Fast SINR balancing closed loop. This is based on algorithm 2, but with individual UE specific compensation factors targeting a desired effective SNR; SINR target. The open loop component is set as algorithm 2, Į=1 and P0 = SNR target +Pnoise , where P0 defines the initial power only. Effective SNR (after antenna combination) including interference is measured
TABLE I.
SIMULATION PARAMETRS Traffic Models
User distribution
Uniform
Data generation
On-off with activity factor f ; 20, 40, 60, 80, 100% Radio Network Models
Distance attenuation
L = 35.3+37.6·log(d), d = distance in meters
Penetration loss
20dB
Shadow fading
Log-normal, 8 dB standard deviation
Multipath fading
SCM, suburban macro
Cell layout
Hexagonal grid, 3-sector sites, 57 sectors in total
Cell radius
167m (500m inter-site distance) System Models
Spectrum allocation
0.2 & 10MHz at 2GHz 1&50 resource blocks, 12&600 subcarriers
UE power class (P max)
250mW
Max antenna gain
15dBi
Modulation and coding
QPSK & 16QAM, continious coding
Overhead
28% for reference signals and L1/L2 control channels (5 symbols per TTI for data)
Receiver
MMSE with 2-branch receive diversity
for each UE and compared with the desired SINR target. The difference is adjusted by sending TPC to the UE. An ideal closed loop is simulated resulting in an ‘upper bound’ for this type of algorithm. In line with this assumption perfect interference knowledge and an ideal update rate are assumed. The closed loop compensation is repeated until power levels converge. III.
SIMULATION ASSUMPTIONS AND MEASURES
Models and simulation parameters are according to the 3GPP evaluation criteria case 1 [4] including sub-carrier modeling of OFDM and Spatial Channel Model (SCM). A selection of simulation parameters is listed in Table 1. Handover margin, delay and measurement error are modeled by randomly selecting among cells within 3 dB from the best based on downlink path gain excluding multipath fading. Control channels are assumed to be error-free, but their overhead is taken into account. Note that the presented results are for relative comparison only and do not give correct absolute LTE performance. No measurement or power setting error is included. Static snapshot simulations have been used. In each iteration of the simulation, terminals are randomly positioned in the system area, and the radio channel between each base station and terminal antenna pair is calculated according to the propagation and fading models. To study different system load levels, terminals are randomly selected to be transmitting with an activity factor f ranging from 20 to 100%. In active cells transmitting users are selected independently of channel quality. Based on the channel realizations and the active interferers, a signal-to-interference and noise ratio (SINR) is
2
100
100
90
90
80
80
70
70
] 60 % [ . F . 50 D . C
] 60 % [ . F . 50 D . C
40
40
30 20 10 0 -20
30
No PC OL PC SNRtarget :8dB :1 OL PC SNRtarget :5dB :0.7
10
CL PC SINRtarget :0dB -15
-10
-5
0 5 10 Effective SNR [dB]
15
20
25
0
30
Figure 1. Effective SNR distribution. 10MHz & f=100%, optimized for celledge bitrate
IV.
NUMERICAL R ESULTS
The power control algorithms have been evaluated in three different scenarios, a wideband (10MHz) allocation targeting either maximizing mean or cell-edge performance, and a narrowband (0.2MHz) allocation.
A. Targeting High Cell-Edge Bitrate in 10MHz 10 MHz have been simulated, modeling a high data rate service where all 50 resource blocks are scheduled to a single user in each cell. Power control targets have been scanned to find the parameter setting giving highest cell edge bitrate at full load ( f =100%). The best setting was found to be: •
SNRtarget=8dB for open loop with Į = 1,
•
SNRtarget = 5dB for open loop with Į = 0.7,
0
5
10 15 Active Radio Link Bitrate [Mbps]
20
25
18
] s p b 16 M [ ) n a 14 e m d n 12 a c r e p 10 h t 5 ( e 8 t a r t i B k 6 n i L o 4 i d a R e 2 v i t c A
(2)
where, U(f) is the total number of active users for activity factor f , and N cell =57, the number of cells in the system. This assumes that user are scheduled an equal amount of time. The mean and the 5th percentile of the active radio link bitrate are used as measures of average and cell-edge user quality respectively. Note that as the activity factor increases, individual user bitrates decrease because of increased interference and thereby decreased SINR. The served traffic however increases as the number of active users increase.
CL PC SINRtarget :0dB
Figure 2. Bitrate distribution. 10MHz & f=100%, optimized for cell-edge bitrate.
calculated for each link and receive antenna. Then, using a receiver model, an effective SNR (after antenna combining) is calculated per resource block. Following this, using the mutual information model of [5], the effective SNR values are mapped to active radio link bitrates Ru, for each active user u. Note that Ru is the bitrate that user u gets when scheduled. Active base stations and users differ between iterations, and statistics are collected over a large number of iterations. For each activity factor, the served traffic per cell T(f) is calculated as the sum of the active radio link bitrates for the active users:
T(f) = Ȉu=1U(f) Ru / N cell
No PC OL PC SNRtarget :8dB :1 OL PC SNRtarget :5dB :0.7
20
0
No PC OL PC SNRtarget:8dB
:1
OL PC SNRtarget:5dB
:0.7
CL PC SINRtarget:0dB
0
2
4
6 8 Served Tra ffic [Mbps]
10
12
14
Figure 3. Mean (circles) and cell-edge (triangles) bitrate vs served traffic, 10MHz scheduled optimized for cell-edge bitrate
•
SINRtarget = 0dB for closed loop.
The resulting effective SNR and bitrate ( Ru) distributions are shown in Fig.1 and Fig.2 respectively. The fast ideal closed loop balances all users on the target of 0dB except for the 5 th percent worst. This shows that the selected target is the optimal for cell-edge bitrate as it is defined to be measured at the 5 th percentile. The open loop algorithms also decrease the variance of the SNR compared to the reference without power control. Since the open loops do not compensate for interference and fast fading there is still a significant variance of effective SNR. Note that the median SNR with the open loop algorithms are higher than with the closed loop and in the range of the reference without power control. In Fig.3 the resulting cell-edge and average active radio link bitrate are shown as a function of the served traffic per cell (T ). The markers are the simulated activity factors ( f ). Note that for the closed loop the cell-edge and average bitrates are almost equal. All power control algorithms improve the cell edge bitrate significantly compared to using no power control. However, the SINR balancing closed loop algorithm results in
3
15
18
] s p b 16 M [ ) n a 14 e m d n 12 a c r e p 10 h t 5 ( e 8 t a r t i B k 6 n i L o 4 i d a R e 2 v i t c A
14 13 ] 12 s p b M11 [ c i f f 10 a r T d e 9 v r e S 8
7 OL PC :1 OL PC :0.7 CL PC
6 5
0
5
10 15 20 SNRtarget /SINRtarget [dB]
25
30
Figure 4. Served traffic per cell for different SNR and SINR targets. 10MHz and f=100%
low served traffic. And even though the 5 th percentile bitrate is highest at 100% activity (right most markers) it performs worse in general. This type of SINR balancing closed loop is good for circuit switched radio channels but does not perform well for packet switched channels with link adaptation such as PUSCH in LTE. The link adaptation range is not utilized as seen in Fig.2. As seen in Fig. 3, the two open loop algorithms improve the cell edge bitrate almost equally. This improvement comes at the cost of average bitrate reduction. The fractional compensation open loop performs in general best with the same cell edge performance as with full compensation but around 20% higher average bitrate. This is also seen in Fig.2 where the fractional open loop performs equal to or better than the fully compensating open loop.
B. Targeting High Capacity in 10MHz As shown above, utilizing mobiles in good radio conditions and link adaptation improves the served traffic and the mean bitrate. To further investigate this power control targets have been scanned to find the parameter setting giving highest mean bitrate as a function of served traffic at high load, f 60%. The results are shown in Fig. 5. The best setting was found to be: •
SNRtarget=20dB for open loop with Į = 1,
•
SNRtarget = 15dB for open loop with Į = 0.7,
•
Effective SINRtarget = 13dB for closed loop.
The higher open loop targets are due to that interference margin must be included. This optimization is almost the same as maximizing served traffic at full utilization ( f =100%), as seen in Fig. 4, for open loop the later results in some dB:s higher targets. The open loop algorithms can achieve around 20% higher capacity than the closed loop. In Fig. 5 it is seen that with this parameter setting all three algorithms improve both average bitrate and cell edge bitrate at the same time compared to the reference with constant power. This is since the inter cell interference is reduced. Both average and cell edge bitrate are improved by reduction of interference
0
No PC OL PC SNRtarget:20dB
:1
OL PC SNRtarget:15dB
:0.7
CL PC SINRtarget :13dB
2
4
6
8 10 Served Tra ffic [Mbps]
12
14
Figure 5. Mean (circles) and cell-edge (triangles) bitrate vs served traffic. 10MHz scheduled optimized for mean bitrate 100 No PC 90
OL PC SNRtarget:20dB :1
80
OL PC SNRtarget:15dB :0.7 CL PC SINRtarget :13dB
70 ] 60 % [ . F . 50 D . C
40 30 20 10 0
0
0.05
0.1
0.15 Power [W]
0.2
0.25
Figure 6. Transmission power distribution. 10MHz & f=100%. Optimized for mean bitrate.
and using a more efficient link quality range where link adaptation is active. The closed loop performs best at low load since it adapts to interference. However, at higher load the open loop algorithms perform better for cell edge users. One reason for this is that SINR balancing costs radio network capacity, as seen in Fig. 4, moving the closed loop to the left in Fig. 5. This results in that even though the mean bitrate at 100% utilization is higher than with the open loop it is lower for the same served traffic. The open loop algorithms perform equally regarding average bitrate and capacity. However, the fractional compensating open loop shows 20% better cell edge bitrate, see Fig. 5. The reason for this is that the fractional compensation has a lower SNR target resulting in lower average transmission power as shown in Fig. 6. Lower transmission power decreases the interference which the cell edge users gain from and enables a lower SNR target for the same average bitrate. This is also another reason why the closed loop performs worse at higher load. Higher power is used when compensating for interference variations. The fully
4
100
] s 0.45 p b M [ ) 0.4 n a e m0.35 d n a 0.3 c r e p h 0.25 t 5 ( e t a r 0.2 t i B k 0.15 n i L o i 0.1 d a R e 0.05 v i t c A
0 0.05
No PC OL PC SNRtarget :20dB :1
90
OL PC SNRtarget :15dB :0.7
80
CL PC SINRtarget:13dB
70
No PC OL PC SNRtarget:20dB :1
] 60 % [ . F . 50 D . C
OL PC SNRtarget:15dB :0.7 CL PC SINRtarget:13dB
40 30 20 10
0.1
0.15 0.2 0.25 Served Tra ffic [Mbps]
0.3
0
0.35
0
0.05
0.1
0.15 Power [W]
0.2
0.25
Figure 7. Mean (circles) and cell-edge (triangles) bitrate vs served traffic. 0.2MHz.
Figure 8. Transmission power distribution. 0.2MHz & f=100%.
compensating open loop is between the closed loop and the fractional compensating open loop.
control rate of the terminal can be from a simple open loop only solution to a fast one millisecond closed loop control.
C. Results for 0.2MHz Bandwidth The 10MHz examples above model a high data rate service with data to fill all the 50 resource blocks. For lower data rates fewer resource blocks per TTI will be scheduled. Also, when there are more users for the scheduler to select among there are scheduling principles that share the resource blocks in frequency domain rather than in time domain. It is therefore of interest to see the performance also for narrowband scheduling. In Fig. 7 the results are shown for single resource block scheduling per TTI that is 0.2MHz bandwidth. The same parameter settings as in section B that optimized mean bitrate for 10MHz are used. The 0.2MHz results in Fig. 7 show a similar relationship between the algorithms as the 10MHz results in Fig. 5 but with more emphasis. This is since with the maximum power of 250mW the power spectrum density increases with decreased bandwidth resulting in higher interference levels and larger impact of power control. This is also reflected in the power distributions in Fig. 8. Comparing with the power distributions in Fig. 6 for 10MHz one can see that there as expected is a smaller fraction of mobiles that are power limited with 0.2MHz. The closed loop still results in a large fraction of power limited mobiles resulting in a much higher interference level which clearly hits the cell edge users as seen in Fig. 7. Again the fractional path loss compensation performs best at high load. V.
CONCLUSIONS
The uplink power control in LTE is flexible, simple and robust. It consists of a closed loop component operating around a reference obtained by parameterized open loop. It enables a variety of implementations with different objectives supporting different deployment scenarios and services. The network
A capacity improving feature is the fractional path loss compensation of the open loop. It enables a trade-off between cell edge bitrate and cell capacity. It has clear advantages compared to traditional full compensation open or closed loop. Simulation results indicate that the fractional compensation can: •
improve the cell-edge bitrate with up to 20% for a given average bitrate
•
improve the average bitrate with up to 20% for a given cell-edge bitrate
•
improve the capacity with up to 20%
at the same time the power consumption is reduced. The fractional compensation is configurable with a simple broadcast factor Į used by the UE in the open loop algorithm. In comparison, a traditional fast SINR balancing closed loop mechanism, which is possible to realize with the LTE power control frame work, performs best at low load but poorly at high load since it does not utilize the link adaptation and the full link performance range of LTE. R EFERENCES [1]
3GPP “E-UTRA Physical layer procedures”, TS 36.213 V8.1.0
[2]
J.F. Whitehead, “Signal-Level-Based Dynamic Power Control for Cochannel Interference Management”, VTC 1993.
[3]
3GPP, R1-074850, “Uplink Power Control for E-UTRA – Range and Representation of P0”, Ericsson.
[4]
3GPP, “Physical Layer Aspects for Evolved UTRA “, TR 25.814, V7.0.0.
[5]
K. Brueninghaus et al., “Link Performance Models for System Level Simulations of Broadband Radio Access Systems”, in proceedings of IEEE PIMRC 2005.
5
