Characterizing Power Consumption of Dual-Frequency GNSS of a Smartphone
aa r X i v : . [ ee ss . SP ] A p r Characterizing Power Consumption ofDual-Frequency GNSS of Smartphone
Bikram Karki and Myounggyu WonDepartment of Computer Science, University of Memphis, TN, United States { bkarki, mwon } @memphis.edu Abstract —Location service is one of the most widely usedfeatures on a smartphone. More and more apps are built based onlocation services. As such, demand for accurate positioning is everhigher. Mobile brand Xiaomi has introduced Mi 8, the world’sfirst smartphone equipped with a dual-frequency GNSS chipsetwhich is claimed to provide up to decimeter-level positioningaccuracy. Such unprecedentedly high location accuracy broughtexcitement to industry and academia for navigation research anddevelopment of emerging apps. On the other hand, there is asignificant knowledge gap on the energy efficiency of smartphonesequipped with a dual-frequency GNSS chipset. In this paper, webridge this knowledge gap by performing an empirical studyon power consumption of a dual-frequency GNSS phone. Tothe best our knowledge, this is the first experimental study thatcharacterizes the power consumption of a smartphone equippedwith a dual-frequency GNSS chipset and compares the energyefficiency with a single-frequency GNSS phone. We demonstratethat a smartphone with a dual-frequency GNSS chipset consumes37% more power on average outdoors, and 28% more powerindoors, in comparison with a singe-frequency GNSS phone.
Index Terms —Dual-frequency GNSS, Mobile Computing, En-ergy Efficiency
I. I
NTRODUCTION
Location service is one of the most widely used features ona smartphone. According to the Ericson mobility report, morethan 4.8 billion smartphones equipped with Global NavigationSatellite System (GNSS) chipsets are active worldwide in2018 [1]. More than 50% of smartphone apps exploit locationinformation [2] such as navigation, games, social media,dining service, to name a few. As the number of locationservice-based apps increases significantly, demand for higherpositioning accuracy is ever higher. Unfortunately, provisionof high localization accuracy has been limited and availableonly to professional and government use mostly due to thehigh cost and military security issue [3].However, a widespread use of smart devices is expandingthe possibility of providing highly accurate positioning serviceto various apps on general user’s smartphones and conse-quently resulted in the introduction of the world’s first smart-phone equipped with a dual-frequency GNSS chipset, XiaomiMi 8 [4]. Xiaomi Mi 8 is equipped with a Broadcom’s dual-frequency GNSS chipset (BCM47755) [5] and is capable ofreceiving L1/E1 and L5/E5 signals from GNSS satellites [3].More and more smartphones such as the Huawei Mate 20,Google Pixel 4, Lenovo Z6 Pro, LG G8, Samsung S10, andSony Xperia 1 are being equipped with dual-frequency GNSSchipsets. With these phones that claim to provide decimeter- level positioning accuracy, a new dawn in location-based appsand navigation research is on the horizon.One of the critical challenges for these dual-frequencyphones is the energy efficiency. Although recent dual-frequency GNSS chipsets for these phones are equipped withadvanced low-cost antennas that provide improved duty cy-cling to reduce the power consumption, dual-frequency GNSSchipsets require higher chipping rates and more processing,resulting in higher receiver power consumption to achieve highpositioning accuracy. Numerous works have been performedon effectively utilizing dual-frequency GNSS phones focusingon achieving high positioning accuracy [6][7][8][9]. However,how much power is consumed to achieve such high locationaccuracy is largely unexplored.In this paper, we perform the first empirical study on powerconsumption of a smartphone equipped with a dual-frequencyGNSS chipset. To the best of our knowledge, this is the firstwork that characterizes the power consumption of a dual-frequency GNSS phone in comparison with a smartphoneequipped with a single-frequency GNSS chipset. For thisstudy, two phones are selected, i.e., a currently available dual-frequency GNSS phone, Xiaomi Mi 8, and a single-frequencyGNSS phone, Xiaomi Redmi Note 7, from the same vendor forfair comparative study. In particular, we measured the powerconsumption exclusively for updating positions performed bythe GNSS modules of the phones by ruling out the energyconsumption incurred by other hardware/software componentsof the phones. Experiments were performed both indoors andoutdoors to understand the effects of different environmentson the energy efficiency. We demonstrate that a phone with adual-frequency GNSS chipset consumes 37% more power onaverage for updating positions compared with its counterpartequipped with a single-frequency GNSS chipset outdoors, and28% more power in an indoor environment. Concretely, thecontributions of this paper are summarized as follows. • We perform the first empirical study on the powerconsumption of a smartphone equipped with a dual-frequency GNSS chipset for both indoor and outdoorenvironments. • We demonstrate that Xiaomi Mi 8 with a dual-frequencyGNSS chipset consumes 37% more power on averageoutdoors, and 28% more power indoors, compared withXiaomi Redmi Note 7 equipped with a single-frequencyGNSS chipset. • We present a useful reference on the energy efficiency of dual-frequency GNSS phone to facilitate research onmobile computing and navigation that exploits those dual-frequency GNSS phones to achieve higher positioningaccuracy.This paper is organized as follows. In Section II we reviewthe background on GNSS concentrating on dual-frequencyGNSS. We then explain the experimental settings and themethodology for measuring power consumption of both dualand single-frequency GNSS phones in Section III. The resultsare analyzed in Section IV, and we conclude in Section V.II. B
ACKGROUND
A. Global Navigation Satellite System (GNSS)
GNSS is a system of satellites that provides time andlocation information anywhere on or near the Earth whenan unblocked line of sight to four or more GNSS satellitesis available [10]. There are two major GNSS systems thatcover the entire world. Global Positioning System (GPS) is themost widely used system developed by US. It has at least 24GNSS satellites. Globalnaya Navigatsionnaya SputnikovayaSistema (GLONASS) is a navigation system developed byRussian consisting of 31 GNSS satellites. There are two othersystems with global coverage that are under development:BeiDou and Galileo. Beidou is a Chinese navigation systemthat has 22 satellites. While global coverage is not providedyet, it is already used in Asia-Pacific region. Galileo is theEuropean navigation system consisting of 18 satellites. Fullglobal coverage by Galileo is expected in 2020.GNSS satellites transmit radio signals over two or morefrequencies in L band, i.e., the operating frequency range of1-2 GHz. Fig. 1 shows the frequencies used by different GNSSsystems. Radio signals transmitted from GNSS satellites carryranging codes and navigation data which are used to calculatethe coordinates of satellites and the distance between a satelliteand a receiver. The binary phase shift keying (BPSK) [ ? ] isused to modulate these signals. L5B2aE5a B2IL3E5b L2C L2 B3 E6 B1I L1 C/AE1 L1 . M H z . M H z . M H z M H z M H z . M H z . M H z . M H z . M H z M H z M H z BeidouGPS GalileoGLONASS
Fig. 1. GNSS frequencies for different navigation systems.
A GNSS navigation message conveys various informationsuch as the position and velocity of satellites, clock, satelliteorbit shape, etc.
The navigation messages are transmitted ata slower rate than the ranging codes. For example, receivinga whole navigation message takes 30 seconds for GPS [11].The message consists of two types of data: Almanac andEphemeris. The almanac data contain the coarse orbital pa-rameters of all satellites and information about ionosphericdelay corrections. Receivers use the almanac data to searchsatellites. Especially, receivers that support only a single GNSS frequency use the ionospheric delay data to correct the delay.To transmit the whole almanac data, 25 navigation messagesare needed, and it takes about 12.5 minutes to complete thetransmission. In contrast to the almanac data, the ephemerisdata contain highly precise orbital parameters of satellites andclock correction information. The ephemeris data is used tocalculate the positions of satellites precisely. Each satellitebroadcasts its own ephemeris data every 30 seconds.A smartphone is equipped with a GNSS/navigation chipsetto receive GNSS signals. It is kind of a blackbox that producesthe user position, velocity and time (PVT) as well as infor-mation about tracked satellites. Fig. 2 shows a block diagramof a typical GNSS receiver. The GNSS antenna is used tocapture GNSS signals in L band. The RF front-end takes theRF signals as input from the antenna and performs down-conversion to reduce the cost. And then, the analog to digitalconverter (ADC) digitizes the signal. The baseband processingmodule performs several signal processing tasks to acquire andtrack the signals. The acquisition task determines satellites thatare in view and can be tracked. The tracking stage is usedto update dynamically the code delay and carrier frequencyof the signal in order to track the signal correctly. The PVTprocessing block combines the information from the basebandprocessing block to derive a solution ( e.g., PVT).
Antenna RF Front End
Raw Data and Navigation Message
RFFront-end Analog toDigitalConverter Baseband Processing PVTProcessingLocal Oscillator Input/OutputUser InterfaceFilteringPre-Amplifier
Fig. 2. Block diagram of a typical GNSS receiver.
B. Dual Frequency GNSS
Dual-frequency GNSS receives two different radio signalsat different frequencies from each satellite to provide moreaccurate positioning. Most of currently available devices uti-lize a single narrow band (L1/E1) with low sampling rates.Recently, the mass market introduce products that supportdual wide band (upper L band and partial lower L band)with high sampling rates. However, due to the high cost,use of these dual-frequency GNSS devices has been limitedto professional and governmental users. In 2017, Broadcomintroduce the first low-cost dual-frequency GNSS chipset,BCM47755, specifically designed for smartphones [5]. In2018, u-blox launch their dual-frequency GNSS chipset, F9chip [12], and STMicroelectronics introduce the Teseo receiverthat supports L1/L2 or L1/L5 frequencies [13]. Intel [14] andQualcomm [15] also start production of their dual-frequencyGNSS chipsets in 2018.With the growth of the mass market for dual-frequencyGNSS chipsets, the first smartphone, Xiaomi Mi 8, equippedwith a dual-frequency GNSS chipset, Broadcom’s BCM47755,s introduced in 2018. This smartphone supports two frequen-cies (L1+L5) and is capable of tracking and processing GPSL1 C/A, GPS L5, GLONASS L1, Galileo E5a and QZSS L5,Galileo (GAL) E1, BeiDou (BDS) B1, GLONASS L1, andQZSS L1 signals.These smartphones equipped with a dual-frequency GNSSchipset enjoy significant advantages. While enhanced posi-tioning accuracy by directly estimating the ionosphere delayis the most significant benefit, the dual-frequency GNSSimproves robustness against jamming and provides access toadvanced satellite navigation technologies such as PPP [3] andRTK [16], which are currently available for only specializedreceivers.More precisely, errors from different sources influence thepositioning accuracy such as imprecise information receivedfrom a satellite ( e.g., on-board clock, ephemeris, etc. ), at-mosphere effects ( e.g., ionosphere and troposphere), receivernoise, and multipath effect [17]. Among these various sourcesof errors, the ionosphere causes the biggest delay [2]. Theionosphere is a layer of the Earth’s atmosphere extending from60km up to 2,000km that contains a high concentration of freeelectrons and ions that can reflect radio waves. The impact ofthe layer is measured based on Total Electron Content (TEC)which is defined as the number of electrons in a tube of a 1m cross section between two points, i.e., a receiver and a satellite.Thus, the contribution of the ionosphere can be written basedon TEC as the following. I p = 40 . · T ECf . (1)Here f is the carrier frequency, and . is the TEC parameterwhich depends on the location of receiver, the intensity of solaractivity, and the hour of day. A dual-frequency GNSS chipsetcan eliminate the impact of the ionosphere based on a pseu-dorange ρ calculated on frequency f and a pseudorange ρ calculated on frequency f as follows. Details on calculatinga pseudorange can be found in [18]. ρ ∗ = f ρ − f ρ f − f . (2)Here ρ ∗ the pseudorange without the ionosphere effect.There are challenges, however, to achieve enhanced accu-racy. A dual-frequency GNSS impacts the design of the re-ceiver, i.e., the antenna, RF front-end, and baseband processingblocks should be replicated to handle the additional frequency,leading to higher power consumption. Specifically, improvedpositioning accuracy is available with antennas with improvedduty cycling for reducing power consumption. Utilization ofdual bands requires higher chipping rates and more processingpower resulting in degraded energy efficiency. This paperaims to characterize such extra power consumption for adual-frequency GNSS phone by comparing that of a single-frequency GNSS phone. III. S YSTEM S ETUP
A. System Settings
We select a currently available dual-frequency GNSS phone,Xiaomi Mi 8, and its single-frequency counterpart, XiaomiRedmi Note 7 for this experimental study. Xiaomi Mi 8 isequipped with a dual-frequency GNSS chipset, BroadcomBCM47755, and Xiaomi Redmi Note 7 has a single-frequencyGNSS integrated into its Qualcomm Snapdragon 660 proces-sor. Both phones have similar hardware specs as they arefrom the same vendor. They are equipped with Snapdragonprocessors, 16M color capacitive touchscreens, Adreno GPUs,and a similar set of sensors. We installed the same OS,Android Pie 9 (API level 28) on both phones. To cut offthe effect of Assisted GPS [19] ( i.e., a technique that utilizesthe information from cell towers for faster position update),the sim cards of both phones were removed and Wi-Fi wasdisabled. An app was created that requests for a positionupdate every second. All other apps including backgroundprocesses were all disabled, and the brightness level of screenwas kept to minimum.
Fig. 3. Experimental setup.
We used the Monsoon power monitor to measure powerconsumption [20]. We deployed the system both indoors andoutdoors. Fig. 3 displays the system settings for outdoordeployment. The probes of the power monitor were connectedto the battery terminals of smartphone, providing current tothe phone. A laptop was connected to the Monsoon powermonitor through USB to measure the current drawn and thevoltage at a rate of 5KHz in real time.We also confirm that both phones receive signals from asufficient number of visible satellites. Figs. 4 and 5 display thenumber of visible satellites over time after the app is startedoutdoors and indoors, respectively. It shows that the number ofvisible satellites quickly increases when the app is started, andthen the phones see about 18-22 satellites outdoors and about14-18 satellites indoors. We observe that a sufficient numberof satellites were available in both environments, althoughthe number of visible satellites was smaller in the indoorenvironment.
B. Methodology
In this section, we present details on measuring the powerconsumption of the phones for updating positions. After GPS
100 200 300 400 500
Time (sec) N u m be r o f S a t e lli t e s Mi 8Redmi Note 7
Fig. 4. The number of visible satellites outdoors.
Time (sec) N u m be r o f S a t e lli t e s Mi 8Redmi Note 7
Fig. 5. The number of visible satellites indoors. is switched on, it takes some time to complete the first positionfix. This is called the time to first fix (TTFF) [21]. There arethree different scenarios for TTFF. If GPS has been turned offfor a long time and/or has moved a long distance, GPS does nothave the almanac, ephemeris, time and position information. Inthis case, which is called the cold start, TTFF can be very largewhich may take several minutes. When only the ephemerisdata is not available, which is called the warm start, TTFFcan be significantly reduced as short as 30 seconds. If all thedata are available, which is called the hot start, TTFF becomesminimal taking only 0.5 to 20 seconds.Figs. 6 and 7 show the power consumption of Mi 8 andRedmi Note 7. Both phones consume power independent ofthe GPS activity before the app is started. This is the baselinepower consumption. Once the app is started, consumed powerquickly increases for a short moment to load the app. Andthen, both phones consume relatively higher energy for TTFFcompared to regular position update. The results show thatTTFF for both phones were different. In fact, it is known thatdifferent GNSS chipsets have varying TTFF. Once the firstposition is fixed, both phones use power to update position. Inthis experiment, we focus on measuring power consumptionfor this regular position update, which accounts for the majorpart of power consumption for many apps based on location
Time (sec) P o w e r ( m W ) TTFFBaselineApp Start Position Update
Fig. 6. Power consumption of Xiaomi Mi 8.
Time (sec) P o w e r ( m W ) App Start TTFF Position UpdateBaseline
Fig. 7. Power consumption of Xiaomi Redmi Note 7. service.
Time (sec) P o w e r ( m W ) Fig. 8. Background power consumption of Xiaomi Redmi Note 7.
A challenge is to measure only the consumed powerfor updating positions, excluding other sources of powerconsumption such as background kernel processes, sensors,network modules, and screen. A tricky part is that thesephones consume different amounts of power for these non-GPS activities. Fortunately, we found that the baseline powerconsumption of the two phones was relatively stable whene disable all background processes, disconnect networkservices such as Wi-Fi, and minimize the screen brightnessto minimum as shown in Fig. 8. Given the stable baselinepower consumption, we simply subtract the baseline powerconsumption from measured power consumption and obtainthe “pure” power consumption used for updating positions.More precisely, we set a one second interval within whicha position update is performed, measure accumulated powerconsumption during this period, and then subtract it by thebase line power consumption, obtaining consumed power fora single position update.IV. R
ESULTS
A. Outdoor Experiments
Time (sec) P o w e r ( m W ) Fig. 9. Power consumption of Xiaomi Mi 8 for position update.
Time (sec) P o w e r ( m W ) Fig. 10. Power consumption of Xiaomi Redmi Note 7 for position update.
Figs. 9 and 10 display the power consumption of Mi 8 andRedmi Note 7 for updating positions, respectively. As shown,a peak is observed every second a request for a position updateis sent to the phones. The results also show that Mi 8 witha dual-frequency GNSS chipset consumes more power thanRedmi Note 7 with a single-frequency GNSS chipset. Fig. 11more clearly shows the difference in power consumption aswe align the graphs exactly with the time when a request forposition update is sent.
Time (sec) P o w e r ( m W ) Mi 8Redmi Note 7
Fig. 11. Comparison of the power consumption of Xaiomi Mi 8 and XiaomiRedmi Note 7 for a one second interval.
100 200 300 400 500
Consumed Energy (mJ) CD F Mi8Redmi Note 7
Fig. 12. Cumulative distribution function graph of the consumed energy ofXiaomi Mi 8 and Xiaomi Redmi Note 7 for updating positions.
We then calculate the “pure” power consumption for posi-tion update by subtracting with the baseline power consump-tion. We repeat a 5 minute measurement 5 times for eachsmartphone. Fig. 12 displays the cumulative distribution func-tion (CDF) plots of power consumption for position updateof both smartphones. The results demonstrate that the averagepower consumption for Mi 8 and Redmi Note 7 was 318mJ( ± ± B. Indoor Experiments
We measure the power consumption of Mi 8 and RedmiNote 7 used for updating positions in an indoor environment, i.e., inside an apartment. Fig. 13 shows the results. As shown, i8 Redmi Note7050100150200250300350 C on s u m ed E ne r g y ( m J ) OutdoorIndoor
Fig. 13. Power consumption of Xiaomi Mi 8 and Xiaomi Redmi Note 7 inindoor and outdoor environments.
Carrier-to-noise-density ratio (dBHz) CD F Mi 8 (outdoor)Mi 8 (indoor)Redmi Note 7(outdoor)Redmi Note 7(indoor)
Fig. 14. Signal strength of GNSS signals ( i.e., carrier-to-noise-density) forMi 8 and Redmi Note 7 for indoor and outdoor environments. the average power consumption of Mi 8 indoors and outdoorswas 321mJ ( ±
28) and 318mJ ( ± ±
23) and 232mJ ( ± ONCLUSION
We presented the first empirical study on the power con-sumption of a dual-frequency GNSS smartphone. The mea-sured power for a dual-frequency GNSS phone was compared with a single-frequency counterpart from the same vendor. Wedemonstrated that the dual-frequency phone consumed 37%more power on average for position update compared with thesingle-frequency phone outdoors, and 28% indoors. We expectthat the results will be a useful reference for academia andindustry in developing mobile applications exploiting locationservice based on dual-frequency GNSS.R
Understanding GPS: principles and applica-tions . Artech house, 2005.[3] U. Robustelli, V. Baiocchi, and G. Pugliano, “Assessment of dualfrequency gnss observations from a xiaomi mi 8 android smartphoneand positioning performance analysis,”
Electronics
Sensors , vol. 19,no. 9, p. 2189, 2019.[7] A. Elmezayen and A. El-Rabbany, “Precise point positioning usingworlds first dual-frequency gps/galileo smartphone,”
Sensors , vol. 19,no. 11, p. 2593, 2019.[8] B. Chen, C. Gao, Y. Liu, and P. Sun, “Real-time precise point positioningwith a xiaomi mi 8 android smartphone,”
Sensors , vol. 19, no. 12, p.2835, 2019.[9] Z. Niu, P. Nie, L. Tao, J. Sun, and B. Zhu, “Rtk with the assistance ofan imu-based pedestrian navigation algorithm for smartphones,”
Sensors ,vol. 19, no. 14, p. 3228, 2019.[10] B. Hofmann-Wellenhof, H. Lichtenegger, and E. Wasle,
GNSS–globalnavigation satellite systems: GPS, GLONASS, Galileo, and more .Springer Science & Business Media, 2007.[11] D. Wells, N. Beck, A. Kleusberg, E. J. Krakiwsky, G. Lachapelle,R. B. Langley, K.-p. Schwarz, J. M. Tranquilla, P. Vanicek, andD. Delikaraoglou, “Guide to gps positioning,” in
Canadian GPS Assoc
Sensors , vol. 19, no. 19,p. 4302, 2019.[17] J. Sanz Subirana, J. Juan Zornoza, and M. Hern´andez-Pajares, “Gnssdata processing. vol. i: Fundamentals and algorithms,”
Noordwijk:European Space Agency Communications , 2013.[18] P. D. Groves,
Principles of GNSS, inertial, and multisensor integratednavigation systems . Artech house, 2013.[19] F. S. T. Van Diggelen,
A-gps: Assisted gps, gnss, and sbas
Proc. of the ION ITM , pp. 25–27,2010.[22] A. T. Loai, A. Basalamah, R. Mehmood, and H. Tawalbeh, “Greenerand smarter phones for future cities: Characterizing the impact of gpssignal strength on power consumption,”