Supported file types

Overview

Plugins are under active development. Stable plugins are documented below with their available options. Plugins under beta testing are available upon request.

File format	File extensions	Status
CSV	.csv, .txt, .zip, ...	Stable
MDF	.mdf, .mf4	Stable
MATLAB	.mat	Stable
HDF5	.h5	Stable
TDMS	.tdms	Stable
Ulog (PX4)	.ulg	Stable
MAVLink	.bin	Stable
Apache Avro	.avro	Stable
Parquet	.parquet, .pqt	Stable
Rosbag	.bag	Beta testing

Custom plugins

Do you have files in a different format? Our technical team can create a custom plugin for your data format.

Get in touch with [email protected] to discuss your use case.

CSV

Plugin name: CSV

The CSV plugin has a lot of options for configuration (see below). Key points:

• The time column can be configured (but defaults to time)

• Time can be a number or date string format. (preferably YYYY/MM/DD hh:mm:ss)

• CSV zips are supported through --unzip

• Two main shapes are supported: tab (example 1) and tsv (example 2)

Example 1: Tabular (tab)

time,altitude,speed,engine_temp,fuel_level,battery_voltage
s,m,km/h,°C,%,V
0.0,0.0,0.0,90.0,100.0,12.6
1.0,10.0,20.0,91.0,99.5,12.6
2.0,25.0,40.0,92.0,99.0,12.5
3.0,50.0,60.0,93.0,98.4,12.5
4.0,80.0,75.0,94.0,97.8,12.4
5.0,120.0,85.0,95.0,97.2,12.4
6.0,160.0,90.0,96.0,96.5,12.3
7.0,200.0,88.0,96.5,95.7,12.3
8.0,240.0,70.0,97.0,94.9,12.2
9.0,260.0,40.0,96.0,94.3,12.2

Download the example file:

1MB

example.csv

Open

Example 2: Time Signal Value (tsv)

time,signal,value
0.0,altitude,0.0
0.0,speed,0.0
0.0,battery_voltage,12.6
1.0,altitude,10.0
1.0,speed,20.0
1.0,battery_voltage,12.6
2.0,altitude,25.0
2.0,speed,40.0
2.0,battery_voltage,12.5
3.0,altitude,50.0
3.0,speed,60.0

Available options

Option	Default value	Details
-t --time-col	"time"	Name of the time column(s), the first matching item in the list will be used (e.g. -t 'time' -t 'datetime')
-T --combined-time-col		Combine multiple column names using space to form 1 date column (e.g. -T 'date' -T 'time' -T 'UTC Offset')
--shape	tab	If the data is organised tabular (tab), or tsv with for example columns (time, signal, value)
--delimiter	,	Character that splits the columns. To specify a tab delimited file, use --delimiter "<paste tab here>"
--decimal	.	Character that show decimal point
--thousands	None	Character that splits 1000s
--quotechar	"	Character indicating strings
--encoding	latin-1	See list of available encodings here
--chunksize	1 million (1e6)	Number of rows consumed at once
--skiprows	0	Rows to be ignored at the top of the file
--dayfirst --no-dayfirst	--no-dayfirst	Parse dates with the day first (e.g. 23/08/2025)
--time-factor	1 billion (1e9)	Factor to convert time to nanoseconds
--time-offset	0	Offset [ns] to add to time. Time [ns] = [time-col] * time-factor + time-offset
--use-index --no-use-index	--no-use-index	Use row index as time column
-h --header		Names for the columns, instead of using the first row (e.g. -h 'time' -h 'col0' -h 'col1')
--units --no-units	--no-units	Read units from the second line (tab only)
--unzip --no-unzip	--no-unzip	The input file is a .zip file which contains .csv files
--group-by-filename --no-group-by-filename	--no-group-by-filename	Add the file name as a signal group
--signal-col	signal	Column to use for signal names, in case of multiple, they will be combined using . (tsv only)
--value-col	value	Column to use for values (tsv only)
--unit-col	None	Column to use for units (tsv only)
-M --metadata-col		Columns to use for metadata (tsv only)

A general example of how these options can be correctly used in Marple DB:

MDF

Plugin name: MDF/ MF4

Can import files generated according to the ASAM Measurement Data Format version 4 (MF4) standard. Binary MDF files cannot be parsed but needs to converted using a .dbc file first.

Available options

Option	Default value	Details
--chunksize	10 million (1e7)	Number of datapoints in each processed chunk. Large values may result in too much use of RAM and crashing
--time-factor	1 billion (1e9)	Factor to convert time to nanoseconds
--time-offset	0	Offset [ns] to add to time. Time [ns] = [time-col] * time-factor + time-offset
--use-header-start-time --no-use-header-start-time	--use-header-start-time	Use the start time from the header of the file.
-s / --select		Channel names to include. If not provided, all channels are included.
--skip-empty-signals --no--skip-empty-signals	--no--skip-empty-signals	Skip empty channels during ingestion.

MATLAB

Plugin name: MATLAB

MAT files are used a lot when using MATLAB or certain loggers such as dSpace, Speedgoat, .. MAT files are treated similarly to HDF5 files (as described above), so the same rules apply.

If you have a MAT file of version 7.3 or higher, this will be treated as a HDF5 file. MAT files of a lower version are significantly less efficient and will require more effort to be parsed by Marple.

Download the example.mat file below:

939B

example.mat

Open

Available options

Option	Default value	Details
-t --time-col	-t "time" -t "Time"	Name of the time column(s), the first matching item in the list will be used. (e.g. -t 'time' -t 'datetime')
--use-index --no-use-index	--no-use-index	Use row index as time column
--time-factor	1 billion (1e9)	Factor to convert time to nanoseconds
--time-offset	0	Offset [ns] to add to time. Time [ns] = [time-col] * time-factor + time-offset
--include-groups --no-include-groups	--include-groups	Include group names in signal names.
--structure	flat	How to interpret the data structure in the file. [flat/matrix/signal]
--names-key	names	Key name for signal names when using 'matrix' structure.
--units-key	units	Key name for signal units when using 'matrix' structure.
--descriptions-key	descriptions	Key name for signal descriptions when using 'matrix' structure.

HDF5

Plugin name: HDF5

HDF5 files are widely known for their excellent handling of large data sets.. and we agree! We love HDF5 data sets!

Unfortunately, HDF5 files come in different shapes and formats, so we need to agree to some structure.

Marple supports 3 structures for HDF5: flat, matrix (row-oriented) and signal (nested groups).

Flat (default, signal-based)

Groups act like dictionaries: each signal name is a key, and the value is a Dataset (array). There is typically one time signal per group, shared by all signals in that group.

file (h5.File)
└── group_1 (h5.Group)
    ├── time (h5.Dataset)
    ├── signal_1 (h5.Dataset)
    └── signal_2 (h5.Dataset)

We also allow for signal grouping, as can be seen in the example below:

Units can be stored as an attribute on each signal Dataset.

Example file:

10KB

example_flat.hdf5

Open

SIGNAL (nested groups)

Each signal is a Group; inside it are Datasets for time and values (e.g. time, value, value_text). No shared time column—each signal carries its own.

file (h5.File)
└── group_1 (h5.Group)
    └── signal_1 (h5.Group)
        ├── time (h5.Dataset) (Can be omitted when using '--use-index')
        └── value (h5.Dataset) (Can be text or numeric data)
        └── value_text (h5.Dataset) (Can only be text data, in that case 'value' can only be numeric)

Groups can be nested further; any Group that contains both a time-like Dataset and a value Dataset is treated as one signal.

Example file:

53MB

example_signal.h5

Open

MATRIX (row-oriented)

A matrix structure is a row-oriented construct.

A group holds a single data matrix plus separate keys for signal names, units, and descriptions. Suited to row-by-row processing.

Key names are configurable (--data-key, --names-key, --units-key, --descriptions-key).

file (h5.File)
└── group_1 (h5.Group)
    ├── data (h5.Dataset)     # matrix: rows = time, columns = signals
    ├── names (h5.Dataset)    # signal names
    ├── units (h5.Dataset)    # optional
    └── descriptions (h5.Dataset)  # optional

Example file:

511KB

example_hdf5_matrix.h5

Open

Available options

Option	Default value	Details
-t --time-col	-t "time" -t "Time"	Name of the time column(s), the first matching item in the list will be used. (e.g. -t 'time' -t 'datetime')
--time-factor	1 billion (1e9)	Factor to convert time to nanoseconds
--time-offset	0	Offset [ns] to add to time. Time [ns] = [time-col] * time-factor + time-offset
--use-index --no-use-index	--no-use-index	Use row index as time column
--include-groups --no-include-groups	--include-groups	Include group names in signal names.
--structure	flat	How to interpret the data structure in the file. [flat/matrix]
--names-key	names	Key name for signal names when using 'matrix' structure.
--units-key	units	Key name for signal units when using 'matrix' structure.
--descriptions-key	descriptions	Key name for signal descriptions when using 'matrix' structure.

TDMS

Plugin name: TDMS

Available options

Option	Default value	Details
--chunksize	10 million (1e7)	Number of datapoints in each processed chunk. Large values may result in too much use of RAM and crashing
--time-factor	1 billion (1e9)	Factor to convert time to nanoseconds
--time-offset	0	Offset [ns] to add to time. Time [ns] = [time-col] * time-factor + time-offset
-t --time-col	-t "time"	Name of the time column(s), the first matching item in the list will be used. (e.g. -t 'time' -t 'datetime')
--use-index --no-use-index	--no-use-index	Use row index as time column
--unit	unit_string	Name of the channel property which contains the unit.

ULOG

Plugin name: ULog

The ULOG format is often used when dealing with systems that operate using the PX4 software or ecosystem. We think it's a great format because it is standardized!

As long as you adhere to the ULOG structure, .ulg formats are plug & play for Marple. No configuration needed!

Download the example.ulg file below:

5MB

example.ulg

Open

MAVLINK

Plugin name: MAVLink

MAVLink is a very lightweight messaging protocol for communicating with drones. The messages can be stored in a file, often with a .bin extension and can be easily imported by Marple.

Available options

Option	Default value	Details
--include-types		Message types to include (if empty, include all types). Multiple options allowed.
--exclude-types		Message types to exclude. Multiple options allowed.
--no-bad-data	False	Ignore BAD_DATA messages
--parameters	metadata	How to add the values from the PARM message type. Valid options are: skip : Ignore PARM messages metadata : Add parameters as metadata to the dataset (use first value if more than 1 occur for same parameter) signal : Add each parameter as a new signal to the dataset both : Add parameters as metadata & signals

Apache Avro

Avro is a row-oriented serialization framework that is used in Apache Hadoop. Avro files store data in binary format but it’s syntax definition resembles a JSON file. Here is an example of what a .avro file JSON structure should look like to be supported by Marple:

[
    {
        "Timestamp": "2024-06-01T12:00:00Z",
        "Name": "Measurement",
        "Value": "0.108"
    },
    {
        "Timestamp": "2024-06-01T12:00:00Z",
        "Name": "Measurement1",
        "Value": "48.6"
    },
    {
        "Timestamp": "2024-06-01T12:01:00Z",
        "Name": "Measurement",
        "Value": "0.235"
    },
    ...
]

Dataset metadata and signal metadata can be embedded in the Avro file’s metadata section using the following structure:

• Dataset metadata: A dictionary containing key–value pairs, where each key represents a metadata field and each value stores the corresponding metadata value.

• Signal metadata: A dictionary where each key is a signal name, and the value is another dictionary containing all metadata fields associated with that specific signal.

Example of metadata in Avro file:

{
    "type": "record",
    "name": "marpledb.example",
    "fields": [...],
    "datasetInfo": {
        "Location": "Antwerp",
        "Date": "11/12/2025",
        "Description": "Temporary configuration for testing purposes"
    },
    "signalInfo": {
        "signal1": {
            "unit": "m/s",
            "description": "Signal 1 description"
        },
        "signal2": {
            "unit": "kg",
            "description": "Signal 2 description"
        }
    }
}

Available options

Option	Default value	Details
--chunksize	100k (1e5)	Number of records in each processed chunk. Large values may result in too much use of RAM and crashing
--time-factor	1 billion (1e9)	Factor to convert time to nanoseconds (only in case timestamps are not stored as dates).
--time-offset	0	Offset [ns] to add to time. Time [ns] = [time-col] * time-factor + time-offset
--time-col	Timestamp	Name of the time column.
--signal-col	Name	Name of the signal column.
--value-col	Value	Name of the value column.
--signal-metadata-key		Key name for signal metadata dictionary in file metadata (if any). See example above for more info.
--dataset-metadata-key		Key name for dataset metadata dictionary in file metadata (if any). See example above for more info.
--unit-key	Unit	Key name for signal unit in signal metadata (if any).
--dayfirst --no-dayfirst	--dayfirst	Parse dates with the day first (e.g. 23/08/2025)

Parquet

The .parquet plugin allows 2 file shapes: long and wide.

Wide format

Use this format if your data is already structured as a time-series table where each signal is its own column and one column contains the time values.

time	temperature	pressure	rpm
0.00	23.1	101.3	1200
0.01	23.4	101.5	1205

Example file:

137KB

example_wide.parquet

Open

Long format

One row per measurement, with a fixed time, signal, and value column.

time	signal	value
0.00	temperature	23.1
0.00	pressure	101.3
0.01	temperature	23.4

Example file:

143KB

example_long.parquet

Open

Available options

Option	Default value	Details
--shape	wide	Shape of the dataset.
--time-col		Name of the time column. If not set, "time" is used if present.
--time-factor	1	Factor to convert time to nanoseconds (only in case timestamps are not stored as dates).
--time-offset	0	Offset [ns] to add to time. Time [ns] = [time-col] * time-factor + time-offset
--chunksize	100k (1e5)	Number of records in each processed chunk. Large values may result in too much use of RAM and crashing
-s --signal-col	["signal"]	Columns to use for signal names. (long only) If multiple values are provided, the names in these columns will be combined to make the signal name.
--value-col	value	Column to use for values. (long only)
--ordered-by-signal --not-ordered-by-signal	--not-ordered-by-signal	Assume rows are ordered by signal, enabling per-signal sequential processing (improves performance for large datasets). (long only)
--use-index --no-use-index	--no-use-index	Use the row index as relative time instead of a time column.