Evaluation of Li-Based Battery Current, Voltage, and Temperature Profiles for In-Service Mobile Phones

A battery’s state of charge or runtime, and state of health or life, will depend on the product’s discharge current over time. For a mobile phone, the discharge current depends on the specific apps that are operated. This paper presents an experimental study to measure and evaluate the operational charge/discharge profile, temperature and terminal voltage of six Android apps; WhatsApp, Facebook, Facebook Messenger, Instagram, Snapchat, and TikTok on smartphones. The results show how the discharge current required by an app’s operation, will affect the battery runtime and life, due to the combined effect of discharge current and temperature.


I. INTRODUCTION
The number of mobile (smartphone) users has surpassed 3.5 billion in 2020, which is 42% of the total global population [1], [2]. These users increasingly demand smartphones with improved functionality and performance to support gaming and social media applications (apps) [3]. This in turn stresses the battery system, which is widely considered as one of the weakest link in a modern mobile phone [4], [5]. In fact, some energy-intensive apps may reduce the battery operation time to as short as several hours for users [6], and were noted to be the likely cause of the Apple IOS slowdown problem in 2017 [7]. Krause et al. [8] showed the negative and highly non-linear impact of C-rate on the life of Li batteries. Dong et al. [9] also noted that accelerated battery aging occurred due to high discharge C-rates.
Reliability issues have also been highlighted with high C-rates as it causes high temperature which accelerates battery aging phenomena and at times causes thermal runaway [9], [10]. A significant drop in terminal voltage is also reported with high C-rates which impacts the smartphone operation [7]. Under high discharge C-rates, for a The associate editor coordinating the review of this manuscript and approving it for publication was Derek Abbott . degraded battery with large internal resistance, heat dissipation becomes a major issue and unexpected smartphone shutdowns have been triggered even when there was substantial charge remaining in the battery [11]. This also results in lower effective battery capacity than the manufacturer's ratings.
Many efforts are made to address this problem of high C-rate in modern smartphones. For instance, some manufacturers suggest improving the algorithms for battery management system (BMS), but it can result in slowing down the operating system [7]. Another suggested method is to improve the system efficiency through low-power processors [5], [8]. Some researchers address the problem of high C-rate by improving battery capacity through improving the chemical operation of the battery by replacing the conventional electrodes of Li battery with the superior capacity (molybdenum disulfide) electrodes with enhanced functional capacity [12], [13].
With regards to in-use operation of smartphones, various studies suggests that traditional testing and qualification of Li-ion batteries may not be able to simulate the actual use of a smartphone in real environment [14], and that manufacturers must address the actual power dissipation of the apps as well as the batteries and smartphones circuitry [15], [16]. While some models have been developed VOLUME 8, 2020 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ to profile power usage of smartphones, they rely on indirect methods to evaluate discharge patterns using aggregated current regimes [14], [17]- [20]. Direct measurement of discharge spike currents have also be conducted, but these rely on external sensors that can add significant measurement errors [17]- [19]. This paper presents the use of built-in mobile sensors to directly measure current discharge profile, terminal voltage, and battery temperature, when the phone is operated using six different apps: i.e., WhatsApp, Facebook, Facebook Messenger, Instagram, Snapchat, and TikTok. Our approach solves the problem of indirect measurements, and parameter averaging [21]. The proposed method uses the manufacturers' provided built-in mobile sensors data through an application program interface (API). We developed a relevant Android package kit (APK) to decipher battery parameters for the apps under investigation. We also provide the algorithm and API that we used (see Appendix) so other researchers can assess other mobile phones.
The rest of this paper is composed as follows. Section 2 summarizes Android-based phones and APIs that collect operation data. Section 3 presents discharge profile analyses of two case-study phones and discusses the impact of the six apps on battery drainage and temperature. Section 4 presents the conclusions.

II. ASSESSMENT OF SMARTPHONES FOR IN-SERVICE OPERATION
The availability of in-service battery data, including current, voltage, and temperature, depends on manufacturer-provided software interfaces as well as the capability of the built-in sensor hardware in a smartphone. Higher resolution of data is key for accurate parameter assessment, for instance, a resolution of microamperes for peak current drawn is more useful as compared to milliampere or ampere level resolution. Similarly, the update time of parameters (minimum interval after which each value is updated in the software APIs) is also important to accurately gauge the parameter variations.
Seven Android-based smartphone manufacturers were evaluated 1 using the developed APK to check the granularity (resolution and update time) of data for discharge current, terminal voltage, and battery temperature through the manufacturer-provided API (software). All the seven smartphone manufacturer support our developed APK, but most of the manufacturers do not provide high-granularity data to evaluate detailed current drawn from the batteries [22]. For instance, the current discharge profile for Samsung J7 is shown in Fig. 1(a), and no change in current was observed with varying app operation, although some variation in temperature was observed.
Similarly, a change in the current for Huawei Nexus P6 was averaged out for about 20 s, not allowing detailed information of current spikes with a particular app operation. Further, the update of voltage or temperature is large and as a result no change was observed. In the case of a Google phone (Pixel 2), the current discharge profile and temperature/voltage profile averaged out after every 2 s and 15 s, respectively, as shown in Fig. 1(c). Like Samsung and Huawei, it also limits the resolution on current drainage values.
The results for all seven manufacturers are summarized in Table 1. Because the highest resolution of electrical current and the fastest update time was measured with the Vivo (V9) and Motorola (Droid Turbo), those two smartphones were selected for further study. Both unused phones under test were fully charged to compare against nameplate battery capacities to check for any unusual degradation, and both phones passed  the test. With regards to the market share, Vivo sold over 102 million phones worldwide in 2018 with a market share of about 7%, whereas Motorola sold over 35 million phones in 2018 [30].
Whereas a smartphone has numerous temperature sensors implemented across the device, battery temperature is typically sensed through a thermistor close to the terminals as shown for a typical assembly in Fig. 2 [31], [32]. In our developed app, the 'EXTRA_TEMPERATURE' API command is used, which specifically measures the battery temperature [33]. Battery current and voltage are generally monitored through a single chip such as Maxim MAX77705C, shown in Fig. 2 [34]. As the sensors are already embedded (built-in) in the mobile phone circuitry, our developed app uses a non-intrusive method to measure battery parameters through the manufacturer-provided APIs [22].

III. RESULTS AND DISCUSSION
To obtain the discharge profile, the Vivo (V9) and Motorola (Droid Turbo) phones were initially charged to full battery capacity (100%), and then each app was run for approximately 12-18 min, followed by the next charge cycle back to full charge. We found that 10+ min is sufficient to test all major functionalities in an app. Subsequently, the phone was discharged using the second app, and so on. During this time, most of the general capabilities of a specific app were tested. For example, in the case of WhatsApp, all features such as texting, photo send/download, voice note send/download, voice call, and video call were tested. The detailed features tested during the experiment for all apps are summarized in Table 2. Throughout the experiment, the experimental conditions were kept consistent to minimize variations from one app to the next. For instance, only the testing app (Snapchat or Facebook or others -only one at a time) along with our developed APK was run. All apps were tested on phones connected through WiFi and the sound volume and screen brightness were kept at 100% throughout the experiment ambient temperature at 26 ± 0.5 • C.
For both phones under test, Fig. 3 shows the operation of various apps (i.e., WhatsApp, Facebook, Facebook Messenger, Instagram, Snapchat, and TikTok) on the current drain, the terminal voltage of the battery, and the battery temperature. The negative current represents charging, whereas the positive current shows discharging ( Fig. 3(a)). Several positive spikes during charging appear which were triggered when the screen was turned on for viewing of the battery status. Terminal voltages ( Fig. 3(b)) show an increasing trend while charging and a decreasing trend during the discharging, as expected. However, the rapid change observed in the current profile during charging is not seen in the voltage profile due to lower update time for the voltage (a software limitation VOLUME 8, 2020  by the manufacturer). Each sensor has a separate resolution and update times as provided by the manufacturer and summarized in Table 1 where, for most of the manufacturers, the update time of the voltage sensor is high (several seconds) as compared to the current sensor (milliseconds).
The result in Fig. 3(a) show that the discharge rate depends on the apps and is also dictated by mobile phone technology (e.g., processor chipset, camera technology, screen type, and resolution). For instance, the processor used in Vivo V9 (Qualcomm MSM8953-Pro Snapdragon 626) is more energy-efficient than the one used in Motorola Droid Turbo (Qualcomm APQ8084 Snapdragon 805) due to improved process technology node from 28 nm (Motorola) to 14 nm (Vivo) [27], [28]. Further, the screen resolution for Motorola is 565 ppi in comparison with 400 ppi for Vivo. A higher resolution requires more pixel per inch and ultimately consumes more energy per inch. So, the behavior of apps vary significantly on various phones and ultimately the same app can perform in a varied manner on different mobile manufacturers [27], [28].
The temperature profile for both phones is shown in Fig. 3(c). The temperature consistently increases during app usage and decreases during the charging process (in comparison with the discharge operation under app usage) [35], [36]. For the Vivo V9 phone, the temperature is fairly consistent during the use of various apps and, during the charging interval, the temperature normalizes to below 32 • C with the highest battery temperature of about 40 • C. However, for the Motorola phone, usage of Snapchat, TikTok, and Facebook Messenger particularly increases the temperature beyond 43 • C and at least 3 • C higher than all other app usage or charging conditions. The amount of current spikes from the battery is also higher for Snapchat, TikTok, and Facebook Messenger for the Motorola phone, which could significantly affect aging and other related issues [37], [38]. To describe the current discharge profile, the mean current information is not sufficient to characterize these apps. In addition, current spike information is also essential as it can have many unintended consequences such as temperature rise and lifetime deterioration.
In order to evaluate this further, we calculate the probability density functions (PDFs) and normal distribution functions (NDFs) of the discharge current profiles for all the apps under investigation in Fig. 4. The PDFs/NDFs facilitates the reader to understand and reconstruct the results for further studies.
A PDF specifies the probability distribution for various levels of current drawn while using various apps. The highest point on the PDF relates to the mean value of current. In the case of the Motorola phone, several apps such as Snapchat have a higher mean value (967 mA) compared to Instagram (572 mA) with results summarized in Table 3. Higher currents drawn from the battery are undesirable from a heat dissipation perspective. In addition, high-frequency current spikes are not suitable for optimum battery operation as they rapidly degrade battery life [37], [39]. The values of standard deviation (SD) along with mean currents is also summarized in Table 3 for reconstruction of discharge profiles.
Experiments showed that the discharge profile for the six apps is significantly different on each phone. With regards to C-rate discharge on phones, higher rates are detrimental for both battery runtime and lifetime. Typically, up to 0.3C is considered a safe limit for accelerated lithium-ion battery degradation [40]- [42]. Therefore, the probability of current spikes above 0.3C is shown in Table 3 to assess the power usage of various apps under consideration. It can be noticed that the performance of energy-intensive apps such as Facebook Messenger and Snapchat vary from Vivo to Motorola. For instance, in the case of Vivo, the mean VOLUME 8, 2020 current and the probability of current spikes above 0.3C for Facebook Messenger are 740 mA and 16.7%, respectively, which is higher than Snapchat with a mean current and probability of current spikes at 705 mA and 2.8%, respectively. Further, higher current spikes may not necessarily result in higher mean currents as seen with Snapchat, which has a mean current of 705 mA with a current spike probability of 2.8% compared with WhatsApp with a mean current of 653 mA and a current spike probability of 5.8%. Higher current spikes are associated with several well-documented issues such as degradation, capacity fade, and premature shutdowns [43]- [45].
With regards to Motorola, the performance of Facebook Messenger is relatively better than Snapchat as well as TikTok. The mean current for Facebook Messenger is 845 mA, compared to 967 mA for Snapchat and 933 mA for TikTok. Similarly, the current spike probability for Facebook Messenger is lower at 17.2% compared to 21.2% for Snapchat and 20.1% for TikTok. Fig. 5 plots the cumulative probability function (CPF) to link the current discharge profile with the probability of occurrence. The CPF also shows a variation in amplitude of the current spikes for all the apps. A spread across the x-axis (current) shows a higher probability of occurrence for a larger spike, which is undesirable. It is therefore evident from Fig. 5 that the performance of the Vivo V9 is better than the Motorola Droid Turbo for the use of apps under consideration.
With regards to the battery capacity and runtime of smartphones, power-hungry apps could reduce the operation time significantly. For instance, for the Motorola Droid Turbo model with a 3900 mAh battery [28], talk time of up to 48 h is suggested by the manufacturer, whereas our data shows that using Snapchat and TikTok for less than 5 h would fully drain the battery. On the other hand, using Instagram for at least 7 h is possible with the in-use discharge pattern. In addition to the average current drawn, C-rating is another critical parameter that plays an essential role in battery life span [41]. Typically, a C-rating of 0.1C to 0.4C for lithium-ion batteries is recommended by many manufacturers [42], [46], [47]. So, it is important to optimize app development to limit the battery current spikes. The C-rating of batteries for the phones investigated in this paper is not specified by the manufacturers, but our analysis shows that for the Motorola phone, current spikes of over 0.6C were observed, whereas the maximum current spikes for the Vivo phone were less than 0.5C. Therefore, manufacturers should provide this criterion while specifying the usage time of the phones under continuous operation.

IV. CONCLUSION
This study monitored and assessed the in-service charge/ discharge profile (e.g. current discharge, terminal voltage, and battery temperature) of Android-based smartphones during active operation of the six most downloaded social media apps of 2019, i.e., WhatsApp, Facebook, Facebook Messenger, Instagram, Snapchat, and TikTok. The in-service charge/discharge profile was retrieved through our developed Android package kit (APK), which accesses the internal sensors through the manufacturer-provided application program interface (API). The interested readers can see our APP program in the Appendix.
As was demonstrated in this study, the discharge rate depended on the apps and their operation, as well as the phone hardware. For instance, the processor used in Vivo V9 (Qualcomm MSM8953-Pro Snapdragon 626) appears to be more energy efficient than the one used in Motorola Droid Turbo (Qualcomm APQ8084 Snapdragon 805). As a result the same app draws less current on Vivo phone as compare to Motorola phone.
This study also found that the operation of an app, for any given phone, will affect its long-term performance and reliability. For instance, Snapchat operation on Motorola was found to be the most energy and temperature intensive with a mean discharge current of 967 mA (for usage interval of approx. 20 min) with temperature rising to 45 • C. The apps Instagram, Facebook, WhatsApp, Facebook Messenger and TikTok showed lower mean discharge currents of 572, 610, 653, 854 and 933 mA, respectively during operation.
In addition to mean discharge current, electrical current spikes in current drawn is useful in modeling battery degradation through accelerated aging processes. For the Motorola phone, results showed that the probability of current spikes beyond 0.3C in Snapchat and TikTok is more than 20%, which is significantly higher than all other apps. For the Vivo phone, the operation of Facebook Messenger was significantly more energy-intensive compared to Snapchat or TikTok. Higher spikes in current results in non-linear temperature rise due to I 2 R dissipation resulting in accelerated aging process [37], [38].
While manufacturers suggest a runtime for smartphone batteries (the time the battery lasts in one full charge), our findings from in-service phone operation confirm that the actual (experimental) runtime is significantly lower than manufacturer-suggested talk time or operation time. In particular, the discharge currents are underestimated for in-service operation resulting in shortened operational duration. Taking the operation of Snapchat on a Motorola phone as an example, the mean discharge current is 967 mA (approx. C/4 ratei.e., runtime to about 4 h), which is significantly smaller than the manufacturer-suggested phone runtime of up to 48 h. While the phrase ''up to 48 hours'' can be the highest limit, it is misleading for many consumers, since the actual runtime will be an order of magnitude lower for Snapchat operation. Similarly, the mean discharge current for Snapchat for Vivo V9 was 705 mA (approx. C/5 rate) with runtime of about 5 hr.
Finally, the information presented in this paper evaluates the discharge profiles for smartphones under different app usage for only two manufacturers. In the future, other phones could be considered. However, not all smartphone manufacturers allow/support measurement of discharge profile in high resolution (milli-second). For instance, Apple does not provide the relevant API, whereas Samsung and Google provides only an average current drainage profile [7], [48]. The average value provided by these manufacturers does not contain information on spike currents or effect on battery drainage by various app features on short time intervals, which are essential to characterize the operation of apps or model degradation based on varying consumer behavior.