Académique Documents
Professionnel Documents
Culture Documents
Battery Life of
Smartphones
and Tablets
Image licensed by Ingram Publishing
A Practical Approach to
Optimizing the LTE Network
T
his article provides a practical approach to opti- perceptibly degrading the user experience. The connected
mize the long-term evolution (LTE) network to and idle-mode discontinuous reception (DRX) mechanism
extend the battery life of LTE devices such as smart- along with the user inactivity timer (UIT) are assessed
phones and tablets. The majority of the commercial- and evaluated by means of field testing in a live LTE net-
ly deployed LTE networks are optimized for data dongles work to identify the optimal deployment scenario for the
where the battery consumption is not a major concern. most common data applications in a commercially
With the introduction of LTE smartphones and tablets and deployed LTE network.
other handheld LTE devices, battery consumption has The new-generation mobile communication system,
become a major bottleneck to the end user. In this article, LTE, aims to enable users with a new mobile experience,
we analyze, evaluate, and validate the key relevant features providing higher data rates and lower latencies that can
and parameters that can potentially reduce the battery transform the overall industry into a new wireless ecosys-
consumption and, hence, extend the device usage time. tem of smartphone devices and applications. However,
The objective of this article is to obtain the optimum smartphone applications put disproportionate pressure
parameters setting for relevant LTE features to efficiently on networks due to their typical behavior of many connec-
deliver non-real-time applications with bursty traffic over tions with low transmitted data volume per connection. In
the LTE network by providing power savings without addition, the energy demands of battery-powered devices
to serve these transaction behaviors for each application
Digital Object Identifier 10.1109/MVT.2014.2311571 may cause the battery to drain quickly, even without di-
Date of publication: 3 June 2014 rect interaction from the end user. The development of
System Access
PLMN Selection/ Connected
(Registration or Data
Initial Camping Mode
Transfer)
Procedures
Besides the RRC idle- and connected-mode DRX mecha- S rxlev= Q rxlevmeas -(Q rxlevmin +Q rxlevminoffset) - Pcompensation. (1)
nisms, there is another mechanism that can be used in the
LTE network to specify a timer that can detect data inac- In RRC-connected mode, the C-DRX functionality is con-
tivity while in connected mode to enforce a release to the trolled by the RRC parameters as shown in Figure 4 [2], [3].
RRC connection and move the UE to idle mode [2]. The There are two states in C-DRX mode: active duty state and
latter is a complementary procedure to bridge the benefits sleep duty state. During the active duty state, the UE moni-
of being in RRC idle and/or connected mode along with the tors the PDCCH for the possible DL transmission from the
DRX procedures used in each state. Figure 2 summarizes network. While in sleep duty state, the UE will switch off its
the interactions between the three mechanisms. transmitter/receiver to save power. The switching between
the said C-DRX states depends not only on the timers, which
Key Parameters Overview are the on-duration timer, the DRX inactivity timer, and the
The DRX, C-DRX, and the state transition key RRC layer DRX retransmission timer, but also on other special situa-
parameters are shown in Figure 3. This article studies tions, such as the hybrid automatic repeat request (HARQ)
the performance of a subset of these parameters to find buffer and the random access channel (RACH) procedure
the optimum balance between power saving and the during the contention resolution timer.
other aspects of latencies and call performance. There The main parameters that are controlling the C-DRX
are several other factors that can be taken into account operation are summarized as follows [4], [9], [10]:
to find the optimum balance between these objectives, DRX cycles: the UE maintains two DRX cycles called
such as the short DRX cycle and the long DRX cycle, which
the nature of the traffic transferred between the eNB have different durations. The short DRX cycle is
and the UE, which is based on the application type optional, and, if this is configured, the UE starts with
the network topology and architecture in terms of the a short DRX cycle (2640 subframes) when it enters
RFs deployed and the different technologies [second the DRX mode. When the configurable short DRX
generation/third generation (3G)] that the UE is timer expires, the UE switches to the long DRX cycle
instructed to measure periodically (102,560 subframes). The optimal periodicity of the
the network vendor implementation for the interac- DRX cycle depends on quality-of-service require-
tion between the underlying procedures of different ments, especially on the maximum latency. The long
LTE protocol layers. DRX cycle impacts other features such as automatic
In idle mode, and during the wake-up cycles, the UE may neighbor relation (ANR). A short sleep time in DRX
choose not to measure neighbors when the S rxlev 2 S intrasearch has a negative impact on the ANR measurements. A
[2]. S rxlev is calculated as in (1) with the parameters shown large value of the long DRX cycle extends the channel
in Table 1. Pcompensation used in all testing is 0 dB quality indicator (CQI) and the sounding reference
onDurationTimer
DRX Cycle Default Paging Cycle
UIT drxInactivityTimer
Intrafreq Cell Qrxlevmin
drxRetransmissionTimer
Reselection Qrxlevmin offset
Parameters Sintrasearch longDrxCycle
shortDrxCycle
drxShortCycleTimer
signal (SRS) reporting periods and, therefore, allocations. The on-duration timer is a part of a DRX
decreases the throughput and deteriorates the UL cycle and allows the network to define a window for
synchronization, respectively. reaching each UE such that the multiuser priorities can
The on-duration timer: This timer specifies the number be considered. It also allows the network to tightly
of consecutive PDCCH subframes (1200) during which share the time among DRX users and to effectively con-
the UE shall monitor the PDCCH for possible figure control channel resources (particularly UL
C-DRX State
Subframe 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2
On Duration
DRX
Inactivity
When DRX Inactivity Expires, Short DRX Timer When Short DRX Timer Expires,
Start Short DRX Timer UE Enters Long DRX Cycle
DRX Inactivity
7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2
On Duration On Duration
UE Sends UE Sends UL UE Receives
UE Decodes UE Receives UE Retransmits
SR on Data on PUSCH DL ACK
PDCCH with UL DL NACK Again on PUSCH
PUCCH
Grant
Figure 4 The C-DRX operation in RRC-connected mode. PUCCH: Physical uplink control channel; PUSCH: physical uplink shared channel;
ACK: acknowledgment; and NACK: nonacknowledgment.
1,400
1,180.98
1,200
1,000
Current (mA)
800 734.51
641.78 628.76 641.07 634.59
600 501.71 488.86
400
200
6.57
0
Test Test Test Test Test Test Test Test Test
Case 1 Case 1 Case 1 Case 2 Case 3 Case 4 Case 5 Case 6 Case 7
Figure 5 The battery current consumption in a near-cell stationary scenario. FTP: file transfer protocol.
and 2 in Figures 5 and 7, it is observed that these two test In addition, during idle-mode wake-up cycles, the UE
cases, where the C-DRX feature is deactivated, provide the will be required to perform intrafrequency neighbor cell
maximum current consumption for ping-type applications measurements in far-cell RF conditions when compared to
by an increase of ~14% in the current consumption. This is near-cell RF conditions for cell reselection purposes. This
mainly due to the fact that the UE stays in connected mode also explains the difference in current consumptions for
for an extended amount of time without any C-DRX sleep du- test cases 1 and 2 in different RF conditions. The results
ties between the ping packets. This is a strong reason that imply that optimizing cell reselection parameters can
the UIT alone may not introduce significant battery savings have an impact on the idle-mode current consumption.
for bursty applications. In this case, imple-
menting C-DRX combined with a shorter
UIT produces the optimum battery sav- 900
ings, which is fully exercised in test case 7. 776.43 783.84
800
The signaling load due to a shorter UIT is 656.49 650.92 649.27
700
Current (mA)
14 25
formed by loading different Web pages
12
20 that have different types of content
10
8 15 and Java applications multiple times
6 and by clearing the cache between
10
4 each run. Figure 9 demonstrates the
5 ping latency for different test cases. It
2
0 0 is shown that the ping packet delay
Test Test Test Test Test Test Test Test
(or RTT) is impacted when the C-DRX
Case 1 Case 1 Case 2 Case 3 Case 4 Case 5 Case 6 Case 7
Ping TestIdle Connected Mode feature is enabled. The impact is ~10
FTP
Download ms of extra RTT compared to when
Expected Battery Life
Connected the C-DRX feature is deactivated. This
Mode % of Battery Consumption by Modem Activity
is due to the C-DRX mechanism decid-
ing how the parameters allow a fast
Figure 8 The expected battery life for ping and FTP applications. adaptation to monitor the PDCCH
when the traffic activity starts.
Consequently, cell reselection and RACH performance It is evident that test case 7 still provides graceful deg-
shall be monitored so as not to negatively impact the per- radation in terms of ping RTT even with the aggressive
formance of these important procedures. sleep duty settings for higher power saving purposes.
In Figure 8, it is shown that test case 7 provides the best This is evident especially for higher ping packet sizes for
battery life and lowest battery usage by modem activity. which this test case can provide acceptable latency and
For continuous file download activities, the battery life can minimum current consumptions. Because of the small
sustain for ~5 h while the screen is on. Test case 7 shows increase in average RTT, the impact on the end-user
~8% of battery consumption produced by modem activities experience is almost negligible with the non-real-time
(i.e., PDCCH monitoring and measurements) because of the applications considered in this article. For real-time ap-
frequent modem sleep duty cycles, while test case 1 shows plications, the packet losses and delays may be strictly
20% of the battery consumption coming from continuously tied to the performance of the C-DRX parameters and
monitoring the PDCCH and measurements in the absence of shall be evaluated separately in future work.
modem sleep duty. It is worth noting that, if a longer period To further analyze the results in Figure 9, test case 4
such as 640 ms was used, it would probably consume less yields higher latency compared to test case 7, although
energy, but it would also engender a longer delay. the latter has a higher long DRX cycle because test case 4
has a lower DRX inactivity timer (60 ms) compared to test
Impact of DRX on Latency case 7 (i.e., 100 ms). The ping intervals were kept small,
In this section, we evaluate the impact of C-DRX on ping and, hence, with a higher DRX inactivity timer, the UE goes
latency and Web page loading time (which can also be to inactive mode for less time, which keeps the ping la-
expressed as user session throughput). The ping latency tency lower with test case 7 in this scenario. This is also in
70
64
55 32 B 640 B 1,460 B 5,000 B
60 56
52
50 48 47
50 49 51
50 44
42 45 42 44 42
38
RTT (ms)
40 37 37
29
30
19
20
13 13 10 11 9 12
8 7 6 6 8
10
0
DRX Off Test Test Test Test Difference Difference Difference
(Test Case 3 Case 4 Case 5 Case 7 (Best (Worst (Average
Cases 1 with DRX On- with DRX On- with DRX On-
and 2) DRX Off) DRX Off) DRX Off)
0 0
YouTube.com Apple.com Facebook.com Google.com
DRX Off (Test Cases 1 and 2) Test Case 4
Test Case 5 Test Case 7
Test Case 3 Web Site Average Data Size
Figure 10 The Web page loading time for common Web sites.
line with test cases 3 and 5. In test case 3, this timer is set the UIT, multiple sync sessions consisting of one or up
to 3 ms, which yields a latency of 64 ms (higher than test to five data transactions can occur, triggering RRC state
cases 4 and 7). This is true for high ping packet sizes such transitions. A data transaction takes the UE from idle to
as 5,000 B. For a lower ping packet size such as 320 B, we connected state and back to idle, as shown in Figure 11,
can observe that test case 4 outperforms test case 7, and depending on the configured UIT.
this is now due to the higher long DRX cycle in test case 7. In a further look into the Facebook client sync pattern,
The Web page loading time is depicted in Figure 10 for the application is evaluated for 22 h and the patterns are
different Web sites. In general, there is no major perfor- observed as summarized in Table 4, which are collected
mance degradation of Web page loading time between using application-profiling tools. In another case of a
the test cases with the C-DRX feature activated or deac- typical loaded scenario where multiple applications are
tivated, and a maximum of 1-s delay is encountered, with required to sync for updates on the Android operating
an average of 0.5 s. In addition, the Web page loading time system (OS), the nature of sync patterns become different,
clearly depends on the contents and the nature of the as shown in Table 4, which are collected using the same
Web page server. As shown in Figure 10 and Table 3, high- application-profiling tool. The table shows the sync pat-
er Web site data contents in kilobytes can be loaded very terns of 18 active applications and widgets in the Android
quickly for the Apple.com Web site compared to You- OS for a 1-h duration.
Tube.com or Google.com. The user session throughput is For such applications, C-DRX with test case 7 will help
maintained the same among all test cases without any im- minimize battery consumption without a noticeable impact
pact when the C-DRX is activated. These results demon- on latency as shown in previous sections. Figure 12 shows
strate that the C-DRX, with a good battery consumption the significant improvement in battery life with test case 7
configuration as in test case 7, provides a good end-user over other test cases, especially when the DRX feature is
experience with an extended battery life.
Application Traffic Behavior Table 3The average user session DL throughput in Mb/s for Web browsing
and DRX Impact of common Web sites.
It is important to evaluate the C-DRX in Average User Session DL Throughput (Mb/s)
terms of smartphone/tablet applications
behavior. This section provides an over- YouTube.com Apple.com Facebook.com Google.com
view of some applications behavior and DRX OFF
their impact on battery life, with differ- (Test cases 1 0.386 1.946 1.110 0.484
and 2)
ent C-DRX parameter configurations.
An Android client sync pattern is illus- Test case 3 0.358 1.894 1.154 0.571
trated in Figure 11. Sometimes Android Test case 4 0.431 1.867 1.254 0.374
client sync sessions cannot be finished
Test case 5 0.385 1.849 1.155 0.503
within one data transaction. And when
Test case 7 0.373 1.852 1.114 0.508
intervals between packets are larger than
Client Initiated
Client Initiated
Client Initiated
Client Initiated
Client Initiated
16 s 2s 2s 12 s 2s RRC-Idle
Camped
30 s 60 s 60 s ~5 min
RRC-
Time Connected
DL Volume (B) 2,538 748 782 794 742
UL Volume (B) 2,156 748 668 888 668
Table 4 A summary of the Facebook client sync pattern and the fully loaded application sync pattern.
257.53
259.16
300
233.39
250 sults show that with the optimized DRX parameter settings,
Battery Life (h)
42.16
38.2
50
Conclusions
0 In this article, we have analyzed the DRX mechanism for
Test Test Test Test Test Test Test
Case 1 Case 2 Case 3 Case 4 Case 5 Case 6 Case 7 power saving of the UE battery in the LTE system. It has
been demonstrated that the DRX is tightly connected
Figure 12 The expected battery life for the Facebook application with the parameter settings and all other RRC proce-
and the loaded application sync patterns scenarios. dures such as idle-mode DRX and state transitions