CRAWDAD dataset eurecom/elasticmon5G2019 (v.2019-08-28)

Berkay Koksal; Robert Schmidt; Xenofon Vasilakos; Navid Nikaien

4G and 5G RAN monitoring data collected using the ElasticMon 5G monitoring framework over FlexRAN

Ten dataset files containing 4G/5G MAC, RRC and PDCP statistics and monitoring data, grouped into two versions of 5 datasets each: raw statistics and processed monitoring data. Raw datasets are recorded using ElasticMon v0.1, a prototype version of a monitoring framework extension of the FlexRAN 5G programmable platform for Software-Defined Radio Access Networks. For details, see here: http://mosaic-5g.io/flexran/ Scenarios setup: Raw datasets are recorded for one eNB and a single mobile User Equipment (UE) in five different mobility scenarios by following different motions and distance patterns relative to the eNB . All raw data have been recorded without including Tx power amplification on the RF frontend (0 dBm transmit power), which implies an approximately 10m maximum range of coverage. Future versions of the datasets will refer t o multiple UEs monitoring and an eNB with Tx power amplification. How to use: The contributed raw datasets can be processed and used for training intelligent 4G/5G models. The processed datasets that are also contributed here follow a proposed stepwise paradigm procedure. Therefore, you are advised to customize and adapt the proposed processing steps to match your own needs. Dataset DOI: 10.15783/c7-s58c-qn61 / https://doi.org/10.15783/c7-s58c-qn61 datasets: ftp://ftp.eurecom.fr/incoming/01-RawDatasets.zip and ftp://ftp.eurecom.fr/incoming/02-PreprocessedDatasets.zip

last modified

2019-08-28

nickname

elasticmon5G2019

institution

eurecom

reason for most recent change

the initial version

release date

2019-08-28

date/time of measurement start

2019-12-11

date/time of measurement end

2019-12-11

website

www.crawdad.org/eurecom/elasticmon5G2019

network type

None of the above

collection environment

The 10 datasets come in two different grouped versions, namely:
01-RawDatasets: raw statistics containing MAC, RRC and PDCP metric values provided by the FlexRAN controller; 02-PreprocessedDatasets: processed monitoring data by adding timestamps and cleaning out (i) corrupt/inaccurate metric values and (ii) static values; Each grouped version is composed of five comma-separated files: -1- moving-away.csv: the UE moves away from the eNB to a maximum distance of 10 meters. -2- movingcloserfarcloser.csv: the UE moves back and forth relative to the eNB, from a 0 distance up to approximately 10 meters. -3- stableshortdistance.csv: the UE stands still in a short distance (approx., 0-1m) away from the eNB. -4- stablemiddistance.csv: the UE stands still in a mid distance (approx., 1-5m) away from the eNB. -5- stablelongdistance.csv: the UE stands still in a long distance (approx., 5-10m) away from the eNB. In what follows, we describe our processing process as a paradigm. Prospective users are advised to customize this process in order to match their needs. Step 1 - Raw datasets: Medium Access Control (MAC), Radio Resource Control (RRC), Packet Data Convergence Protocol (PDCP) data provided by the FlexRAN controller recorded for 1 UE in a JSON format. Each JSON measurement contains more than a 100 metrics. A detailed description of mea surement metrics available by the FlexRAN controller is available here: http://mosaic-5g.io/apidocs/flexran/#api-Stats. Step 2 - processed Datasets: Pre-processing takes place to give a proper structure to raw recordings and to reduce the number of metrics per measurement from over a 100 to 42. Pre-processing is necessary for a series of reasons: - Adding a timestamp: Exact dates in raw measurements do not give useful information. It is necessary to add timestamps inside the recorded JSON tree of each measurement. This is needed for computing the time elapsed between consecutive measurements. - Cleaning out static values: Omitting specific metric fields that do not change over time. Such metrics maintain in a constant value across measurements regardless of the UE being in motion or not. Therefore, they offer no valuable information for prediction. Note that the rem aining 'dynamic' metrics after this step drops to 42. - Adjusting corrupt/inaccurate metric values: There are measurements such as 'macStats_phr', with corrupt/inaccurate values due to integer overflow. The problem is addressed based on the type of metric and number of consecutive corrupt/inaccurate values by replacing evidently corrupt/inaccurate values with either (a) the median value of their neighboring rows or (b) the mean value over a period of time (e.g., past 100ms) out of a series of neighboring rows (resp., 100 rows). Option (a) was used in cases where consequent values created a trend that was not matched by the identified as corrupt/inaccurate; value, while option (b) was preferred for particular types of metrics such as macStats_phr.

network configuration

One user using an Android v8.0 (Oreo) Nexus 6P phone connected to an eNB (carrier band 7). 5G network based on FlexRAN v2.0 and OpenAirInterface snap packages as follows: oai-ran rev. 16 openairinterface5g tag 2018.w42 (see https://snapcraft.io/oai-ran) and oai-cn rev. 26 (see https://snapcraft.io/oai-cn).

data collection methodology

(1) Collection environment: Raw datasets are collected using our prototype version of ElasticMon v0.1. ElasticMon is a novel elastic monitoring 5G framework for OAI-RAN (see https://snapcraft.io/oai-ran) and OAI-CN (see https://snapcraft.io/oai-cn) built over the F lexRAN v2.0 programmable SD-RAN platform (see: http://mosaic-5g.io/flexran/) (2) Data collection: A single mobile user engaged into different mobility scenarios by following different motion patterns for moving further away or closer to the eNB, or by remaining in a static distance relative to the eNB. The adopted measurement frequency was 50ms.

sanitization

 The datasets contain no sensitive information that could raise any privacy concerns. Therefore, the datasets are not sanitized

disruptions to data collection

(1) Part of the recorded measurements of 'movingcloserfarcloser.csv' were not included in the first place in the raw dataset, as data recording failed (NULL/erroneous values) due to the very long distance travelled by the mobile user. (2) Although neither disrupted or erroneous, some static values can be found in the raw datasets over time regardless of the UE being in motion or not. Such metric values offer no valuable information for most prediction models

error

There are erroneous'wbCqi' values at parts of both the raw and processed datasets. These values stand out easily in comparison to their neighboring row values and can be replaced with either the median or the mean wbCQI of their neighboring row values.

This dataset contains the following 2 tracesets:

When writing a paper that uses CRAWDAD datasets, we would appreciate it if you could cite both the authors of the dataset and CRAWDAD itself, and identify the exact dataset using the appropriate version number. For this dataset, this citation would look like:

Berkay Koksal, Robert Schmidt, Xenofon Vasilakos, Navid Nikaien, CRAWDAD dataset eurecom/elasticmon5G2019 (v. 2019‑08‑28), downloaded from https://crawdad.org/eurecom/elasticmon5G2019/20190828, Aug 2019.

We also provide bibliographic information in common citation formats below:

If you do not use the provided citation formats, please include a reference with the same information, as described in the CRAWDAD FAQ.

The eurecom/elasticmon5G2019 dataset (v. 2019-08-28)

This dataset contains the following 2 tracesets: