Leveraging a dataset of almost half a billion packets with high-precision packet times and sizes, we extract characteristics of the bursts emitted over Starlink’s Ethernet interface. The structure of these bursts directly reflects the physical layer reception of OFDMA frames on the satellite link. We study these bursts by analyzing their rates, and thus indirectly also the transition between different physical layer rates. The results highlight that there is definitive structure in the transition behavior, and we note specific behaviors such as particular transition steps associated with rate switching, and that rate switching occurs mainly to neighboring rates. We also study the joint burst rate and burst duration transitions, noting that transitions occur mainly within the same rate, and that changes in burst duration are often performed with an intermediate short burst in-between. Furthermore, we examine the configurations of the three factors burst rate, burst duration, and inter-burst silent time, which together determine the effective throughput of a Starlink connection. We perform pattern mining on these three factors, and we use the patterns to construct a dynamic N-gram model predicting the characteristics of the next upcoming burst, and by extension, the short-term future throughput. We further train a Deep Learning time-series model which shows improved prediction performance.

While Internet Service Providers (ISPs) have traditionally focused on marketing network throughput, it is becoming increasingly recognized that network latency also plays a significant role for the quality of experience. However, many ISPs lack the means to continuously monitor the latency of their network. In this work, we present a method to continuously monitor and aggregate network latency per subnet directly in the Linux kernel by leveraging eBPF. We deploy this solution on a middlebox in an ISP network and collect an extensive dataset of latency measurements for both the internal and external parts of the network. We find that our monitoring solution can monitor all subscriber traffic while maintaining a low overhead of only around 1% additional CPU utilization. Our analysis of the latency data reveals a wide latency tail in the last-mile access, which grows during busy periods in the evening. Furthermore, we dissect the external network latency and uncover the latency profiles for the most popular autonomous systems.

Although Starlink has been deployed for several years, a detailed understanding of system internals is still lacking. In this work we employ precise per-packet timestamps obtained from a hardware-timestamp capable NIC connected to a Starlink terminal. We find that Starlink frame timing details are readily observable at the network layer by analyzing the packet timing patterns. Based on a one-week measurement campaign we collect around half a billion of packet size and timing observations. Processing these observations yields 2.3 million transmission bursts. To learn details on the radio resource management we develop a methodology to infer the effective physical layer sending rate. Our findings show that although Starlink throughput can vary widely over multiple time-scales, there are a small number of fundamental physical layer transmission rates. We employ Gaussian Mixture Modeling to determine 14 such fundamental transmission rates, and relate the obtained rates to previous knowledge of the Starlink OFDMA frame structure. Our empirical observations provide an excellent match for a radio resource configuration where a Starlink frame employs 1000 sub carriers and 287 symbols per frame for user traffic transmission, which for uniform 4-QAM modulation yields a base rate of 430.5 Mbps. This physical layer base rate appears to mostly be varied by multiples of 27 Mbps, in several instances likely by modifying the modulation of a subset of the symbols in multiples of 18 symbols.

Abstract Network latency is a critical factor for the perceived quality of experience for many applications. With an increasing focus on interactive and real-time applications, which require reliable and low latency, the ability to continuously and efficiently monitor latency is becoming more important than ever. Always-on passive monitoring of latency can provide continuous latency metrics without injecting any traffic into the network. However, software-based monitoring tools often struggle to keep up with traffic as packet rates increase, especially on contemporary multi-Gbps interfaces. We investigate the feasibility of using eBPF to enable efficient passive network latency monitoring by implementing an evolved Passive Ping (ePPing). Our evaluation shows that ePPing delivers accurate RTT measurements and can handle over 1 Mpps, or correspondingly over 10 Gbps, on a single core, greatly improving on state-of-the-art software based solutions, such as PPing.

The ability to locate a radio transmitter can be useful in many contexts, and a range of localization methods have been proposed. In the context of cellular networks, the position of the base station is known to the operator and regulator, but often this knowledge is not publicly available. For the problem of base station localization, several approaches have been examined in the literature, and several public services exists which estimate the position of cellular infrastructure based on measurement data collected from cellular users. In this work we present sector fitting, a new approach for locating sectorized transmitters based only on observations of the positions and sector identifiers as reported by the cellular UEs. Sector fitting defines a sectorization model which is applied over a search grid to obtain a cost matrix, which is then merged over multiple frequencies to arrive at the best base station location estimate. An extensive evaluation of sector fitting is carried out, using a large data set of observations from train-mounted LTE modems. The results show that sector fitting outperforms the other applicable localization methods. Furthermore, an iterative grid search approach is examined and demonstrated to achieve the same localization accuracy as a full search while drastically reducing the computational cost. Finally, three downsampling methods are evaluated with the results showing that a trade-off can be made to further reduce computational cost, but with slightly worse localization accuracy in most cases.

Numerous approaches exist for locating cellular base stations using UE measurements. Many of these employ signal strength as a basis to determine the position of the base station. However, partly due to the challenging radio environments encountered in the real world, these approaches often achieve limited accuracy. Here we present a novel localization method named sector fitting, which employs fusion of geometrically-based scores to locate the base station based only on the geographical distribution of the measurements and their cell identity. The method is evaluated by positioning LTE eNodeBs using a large set of measurements collected by modems onboard trains. The results show that sector fitting produces considerably more accurate position estimates than any of the considered alternative methods for the subset of eNodeBs it is applicable to.

Current LTE networks provide a large fraction of the mobile communication needs. One recent application area that have attained additional interest is the provision of mobile communication services to train passengers. To allow more efficient use of network resources and better onboard communication experience, onboard traffic aggregation can be performed. In this work we examine a large-scale operational data set from a router-based LTE traffic aggregation system mounted onboard more than 100 trains belonging to a major Swedish train operator. We use both deep learning (DL) with Deep Neural Networks and traditional machine learning (ML) with Random Forests to examine an observed association between train velocity and achieved throughput, which curiously varies over different radio conditions. More than 37000 train journeys are analyzed to explore for structure and learn potential explanatory features. The results indicate that the association has a limited presence on a per cell basis, and that there is only a limited amount of learnable structure per cell. A resampling evaluation shows that the association becomes apparent when cell measurements are aggregated at an order of tens to a hundred cells.

Providing reliable internet connectivity to train passengers can be handled with onboard aggregation routers that use multiple external antennas to simultaneously convey user traffic over multiple links. This work studies the operational characteristics of a large-scale deployment of such a system. The examination focuses on how train velocity is associated with achieved link throughput, and how various interaction effects influence the relationship. A large data set collected over three years is analyzed, indicating that there is a systematic association between train velocity and link throughput that varies over the radio conditions and which is also linked to differences between operators.

The problem of locating radio devices has been addressed by a variety of methods. In the cellular setting, most of the focus have been on locating user equipment (UE). This work focuses on the inverse problem, i.e. locating the eNodeB based on received signal strength (RSS) measurements collected by UEs. We perform a comprehensive evaluation of six variations of two RSS-modeling based localization approaches. Furthermore, two methods for fusing the location estimates of the individual cells were also examined. The evaluation is done using a manually created ground truth data set for eNodeB positions, and a large measurement data set comprising of more than four million observations collected from cellular modems onboard Swedish trains. The best localization accuracy was obtained by one of our proposed variations of logloss fitting using geographic aggregation with highest mean RSRP as the reference point selection criteria. When combined with centroid-based fusion of the individual cell estimates, a median eNodeB localization error of 433 m was obtained, which is a considerable improvement over the second-best approach which achieved a median error of 674 m. The centroid-based fusion approach was found to consistently outperform the DPD fusion approach, which in turn had a better localization error distribution than obtained for solitary cells.