Fine-grained LTE radio link estimation for mobile phones
Introduction
Can we trust mobile phone data rate measurements? This apparently trivial question is key to evaluate the feasibility of the anticipatory networking [1] paradigm and the related future network solutions [2], [3]. For instance, exploiting achievable rate prediction to optimize mobile applications [4], [5], [6] requires some information exchange between mobiles and base stations so that current decisions (e.g. scheduling, admission control) can be made taking into account the future states of the system. However, while prediction errors have been studied [7], [8], the capability of mobile phones to obtain accurate measurements has never been investigated in mobile networks. In addition to that, many recent studies [9], [10], [11], [12], [13] rely on crowd-sourced datasets to derive their conclusions without questioning mobile phone measurements accuracy and whether it is possible to aggregate them. Although reliable mobile phone applications to measure the network bandwidth exist [14], [15], [16], they focus on end-to-end measurements that do not provide the required level of granularity to enable anticipatory optimization. In fact, while end-to-end data rate is ideal to optimize TCP performance, the resource allocation optimization would rather benefit from the actual radio link data rate between eNodeB and user equipment (UE).
In this paper, we study whether mobile phones can accurately measure LTE radio link data rate and with which granularity (i.e. sampling frequency). To achieve this, we compare the data rate estimates computed at the physical layer of the radio link through a sniffer, at the mobile phone kernel through tcpdump and by a mobile application.
Our study is divided into two measurement campaigns: the first and largest set of experiments consists of burst transmissions, where a small amount of data is sent back-to-back to collect data rate estimates computed by the different entities (i.e., phone, sniffer and server), while in the second set, we evaluate latencies between single data packet transmissions and their corresponding acknowledgments (ACKs). These latencies allow us to study the root-causes of differences among the behaviors of different phones. In all the tests, we compared three mobile phones by different vendors and equipped with different chipsets, first performing the test from the server to the phone and, then, in the opposite direction.
The main findings of our study are the following:
- 1.
Mobile phones achieve accurate (%) and precise (%) data rate measurements with as few as KB in the downlink, where accuracy and precision are related to how close the measurement are to the sniffer ground-truth readings.
- 2.
Uplink measurements are less accurate and less precise (% and % respectively in the worst case), because LTE uplink scheduling delay causes a higher variability in the results.
- 3.
Different chipsets exhibit variable biases and performance, thus requiring dedicated calibration to optimize accuracy.
- 4.
Downlink accuracy and precision are linked to the latency measured on the phone: chipsets providing shorter and more deterministic latencies obtain better estimates.
Further, we provide a thorough guide on how to perform similar measurements. We present the state of the art of tools that can be used to assess the performance of LTE components, such as eNodeBs and UEs, as well as discuss the tradeoffs of each approach.
The rest of the paper provides a survey of related work in Section 2, specifies the measurement setup and the devices involved in Section 3, and discusses the two measurement campaigns in Sections 4 Burst transmissions, 5 Isolated transmissions. Section 6 summarizes the main findings, while Section 7 specifies how to repeat our measurements and discusses a few setup alternatives. Finally, Section 8 concludes the paper.
Section snippets
Related work
A considerable number of recent papers focus on LTE measurements and measurement techniques, but, to the best of our knowledge, none of them rely on accurate LTE scheduling information to validate their findings. Among them, Huang et al. [10] studied LTE performance measured from mobile phone data. In order to obtain a known reference for the results, the authors performed experiments using controlled traffic patterns to validate their findings.
The fraction of LTE resources used for
Setup and definitions
Fig. 1 illustrates our experimental setup, which consists of five entities. The target UE is the mobile device under test which is connected to the target eNodeB. The sniffer is a BladeRF x40 software defined radio [30] that samples and records the LTE signal to be decoded by OWL. The sniffer is shown as connected to the eNodeB-UE link only, but it actually records and decodes all control messages sent by the eNodeB and, thus, it is aware of all of the traffic exchanged in the cell. The server
Burst transmissions
The first measurement campaign has the main objectives of evaluating the accuracy and the precision of data rate estimates obtained by mobile applications, and to analyze the differences in performance obtained by the three phones.
We use the following symbols: and denote durations, transmission sizes, number of packets, and the data rates. All these quantities are easy to compute from the information available in our tests and they do not require complex filtering. In fact, we just
Isolated transmissions
This section details the second set of measurements. The objective of this campaign is to measure phone communication latencies to justify the differences in their behavior. As above, we first illustrate the experiment on a diagram (Fig. 7) and on some trace examples and then we discuss the results.
Summary
Fig. 9 provides a visual summary of the results discussed in the paper. In the figure, one boxplot is shown for each of the main experiments, highlighting the median (central mark of the boxes) and the and percentiles (box edges) of the estimator ratios . At the bottom of the figure we specify the type of used. BA are ratios between data rate measurements performed on bursts (B) at the application (A) and cell estimate. BP are the same but computed by the phone kernel (P), while GA
Measurement setup alternatives
This section describes how to setup a mobile phone measurement environment by providing a few software alternatives and discussing their benefits and their drawbacks. The main objective of the measurements we are going to describe is to provide a flexible environment that allows testing both mobile phones and base stations.
Fig. 10 shows a complete environment exploiting all four software solutions, namely OWL [20], MobileInsight [21], OpenAirInterface [22] and srsLTE [23]. However, in the
Conclusions
In this paper we presented the first experimental evaluation of the accuracy and the precision of LTE data rate measurements performed by mobile phones. To summarize the main finding of the study:
Mobile phones can achieve accurate and precise data rate measurements: we showed that downlink application measurements are both accurate and precise (R1), downlink kernel-level measurements improve the precision, but only slightly (R4), and uplink kernel estimates are accurate and precise, but less
Acknowledgments
This work has been supported by the European Union H2020-ICT grant 644399 (MONROE), by the Madrid Regional Government through the TIGRE5-CM program (S2013/ICE-2919), the Ramon y Cajal grant from the Spanish Ministry of Economy and Competitiveness RYC-2012-10788 and grant TEC2014-55713-R.
References (41)
- et al.
CS2P: Improving video bitrate selection and adaptation with data-driven throughput prediction
- et al.
Lightweight mobile bandwidth availability measurement
Elsevier Comput. Commun.
(2016) - et al.
A survey of anticipatory mobile networking: Context-based classification, prediction methodologies, and optimization techniques
IEEE Commun. Surv. Tutor.
(2017) - NGMN, Next Generation Mobile Networks. URL...
The tactile internet: applications and challenges
IEEE Veh. Technol. Mag.
(2014)- et al.
Anticipatory admission control and resource allocation for media streaming in mobile networks
- et al.
A control-theoretic approach for dynamic adaptive video streaming over HTTP
ACM SIGCOMM Comput. Commun. Rev.
(2015) - N. Bui, J. Widmer, Mobile network resource optimization under imperfect prediction, in: IEEE WoWMoM,...
- T. Mangla, N. Theera-Ampornpunt, M. Ammar, E. Zegura, S. Bagchi, Video through a crystal ball: effect of bandwidth...
- J. Huang, F. Qian, A. Gerber, Z.M. Mao, S. Sen, O. Spatscheck, A close examination of performance and power...
Mobile network performance from user devices: A longitudinal, multidimensional analysis
An End-to-End Measurement Study of Modern Cellular Data Networks
LTE radio analytics made easy and accessible
Cited by (6)
LDRP: Device-Centric Latency Diagnostic and Reduction for Cellular Networks Without Root
2024, IEEE Transactions on Mobile ComputingConnection Type Identification and Uplink Speed Estimation of Malware Infected Hosts
2022, Journal of Information ProcessingExperience: A five-year retrospective of MobileInsight
2021, Proceedings of the Annual International Conference on Mobile Computing and Networking, MOBICOMStudy of Radio Link Losses in Different Mediums and Receiving Stations
2021, 19th IEEE Student Conference on Research and Development: Sustainable Engineering and Technology towards Industry Revolution, SCOReD 2021Device-based LTE latency reduction at the application layer
2021, Proceedings of the 18th USENIX Symposium on Networked Systems Design and Implementation, NSDI 2021