Table of Contents
Introduction
This repository contains our continuous monitoring infrastructure (based on a customized Android system dubbed SubCell) for recording fine-grained observability data when users engage with mobile apps using underground cellular networks, as well as our efforts for mitigating Video Stream Stalls (VSSes) in subways on Android devices. We also release a portion of our measurement data (with proper anonymization). Our latest SubCell system is built upon vanilla Android 13/14. Therefore, you’ll be able to run codes in this repo by patching these modifications to proper framework components.
[The entire codebase and sample data are available in our Github repo.]
Codebase Organization
Continous Monitoring Infrastructure
Our modifications to the vanilla Android mainly consist of three parts, as shown in Monitor.
First, we add SubwayDetection.java to detect when a device boards a subway.
Second, once the device is confirmed to be in a subway environment,
SubCell initiates our dedicated data recording service (SubwayDataCollection.java) to log the network states we are concerned with (as listed here).
Third, to monitor the frame rate of the foreground app and detect VSSes, we modify the graphics layer compositor of Android
by adding a timer task (MonitorFrameRate) in SurfaceFlinger.cpp
and updating onPostComposition in Layer.cpp for the statistics of composed layers.
| File | Key Changes | Purpose | Location in AOSP |
|---|---|---|---|
| SubwayDetection.java | A new class | Indentify subway scenarios | packages/services/Telephony/src/com/android/phone/SubwayDetection.java |
| SubwayDataCollection.java | A new class | Collect cross-layer data | packages/services/Telephony/src/com/android/phone/SubwayDataCollection.java |
| SurfaceFlinger.cpp | MonitorFrameRate (added) |
Count the number of composed graphics layers from the foreground app | frameworks/native/services/surfaceflinger/SurfaceFlinger.cpp |
| Layer.cpp | onPostComposition (changed) |
Notify SurfaceFlinger when a graphic layer from the foreground app is composed |
frameworks/native/services/surfaceflinger/Layer.cpp |
System-level Enhancement
We provide our Time-Inhomogeneous State Space Model (TISSM) that leverages a time-varying multi-order state space to finely characterize the underground signal attenuation, enabling Android’s cellular management to find the optimal handover timing. As shown in TISSM_Model, SubCell updates the TISSM-based signal attenuation model when new signal strength measurements are obtained. Leveraging the model, the system then estimates future signal attenuation to determine whether to trigger handover.
App-level Enhancement
We provide our code of the radio/link state collection module in Collection_Module. We encapsulate this functionality into a system-level component named RadioLinkMonitor, which runs within the telephony framework and continuously collects fine-grained radio and link metrics. We expose this module via a binder-based callback interface by extending TelephonyManager to expose a new API registerRadioLinkMetricsCallback. Apps can subscribe to per-second updates by implementing the callback interface and invoking this API.
Platform Requirements
For Android-related modifications, currently our code is run and tested in Android 13 and Android 14 (AOSP). Note that despite quite a number of changes have been made in Android 14 since Android 13, our code is applicable to both given that concerned tracing points remain unchanged.
Data Release
We release a portion of data here, which provides continuous measurements covering entire subway rides.
This dataset includes measurements collected from three subway lines in Beijing, spanning 26.5 hours of subway travel and capturing 676 VSS events.
Each record contains the following information:
| Information | Description |
|---|---|
ts |
Recording time |
rtt |
Round-trip time (ms) |
pkgLossRat |
Packet loss rate |
dlThroughput |
Downlink throughput (MBps) |
ulThroughput |
Uplink throughput (MBps) |
dnsLatency |
DNS query latency (ms) |
dnsSuccessRate |
DNS query success rate |
rsrp |
Reference Signal Received Power |
rsrq |
Reference Signal Receiving Quality |
snr |
Signal-to-Noise Ratio |
mimoLayer |
The MIMO layer |
mcs |
Modulation and Coding Scheme |
rat |
Radio Access Technology, 14 for 4G and 20 for 5G |
cid |
Cell Identity |
label |
The VSS label (1 for VSS and 0 for non-VSS). |
In addition, we provide a complementary dataset which contains data segments around VSS occurrences for reference, including critical network, cellular, and device information:
| Information | Description |
|---|---|
UID |
Unique ID generated to identify a user (cannot be related to the user’s true identity) |
TIME |
UNIX timestamp when recording for this segment starts |
OS |
Android version |
MODEL |
Device model |
APN |
Current access point name |
APP |
The foreground app |
DATA |
Detailed data recorded every second within a short window around each VSS event, with each second’s data arranged in the following order: round-trip time (RTT), packet loss rate, DNS query success rate, DNS query latency, uplink bandwidth, downlink bandwidth, Radio Access Technology (RAT, 14 stands for 4G and 20 stands for 5G), Reference Signal Received Power (RSRP), Signal-to-Noise Ratio (SNR), Cell Identity (CID), Modulation and Coding Scheme (MCS, 4 for QPSK, 16 for 16 QAM, and 64 for 64 QAM), MIMO layers, and the VSS label (1 for VSS and 0 for non-VSS). |
For Developers
Our code is licensed under Apache 2.0 in accordance with AOSP’s license. Please adhere to the corresponding open source policy when applying modifications and commercial uses.