generated: '2026-08-12'
method: searched
source: https://support.empatica.com/hc/en-us/articles/17727336158365-Raw-data-in-the-Empatica-Cloud
docs:
- https://support.empatica.com/hc/en-us/articles/17727336158365-Raw-data-in-the-Empatica-Cloud
- https://support.empatica.com/hc/en-us/articles/17405877853981-How-to-access-Avro-files-with-Python
- https://support.empatica.com/hc/en-us/articles/13879014347421-Accessing-Data-on-the-S3-Bucket
- https://www.empatica.com/rawdata/
name: Empatica raw-data model (Avro file schema)
description: >-
  There is no OpenAPI to derive an entity graph from, so this data model is read
  from the one place Empatica genuinely publishes a schema: the Apache Avro files
  it writes to the customer S3 bucket. The files are self-describing — the schema
  and per-sensor sampling frequency are embedded in each file — and Empatica's own
  Python example script names the field paths reproduced below. Every field here
  appears in provider-published documentation or in the provider's published
  example code. Field types are stated only where the documentation states them.
format: Apache Avro (1.11.x)
delivery: 'AWS S3, organization-scoped prefix: s3://bucket-name/version/dir-name/'
file_naming: '[participant_id]_[utc_timestamp_start].avro'
cadence: uploaded every 30 minutes; each file holds up to ~30 minutes of uninterrupted
  stream
timezone: all data and filename timestamps are UTC+00:00
entities:
- name: AvroFile
  root: true
  description: One self-contained Avro record covering one uninterrupted capture window
    for one participant.
  fields:
  - {name: schemaVersion, shape: '{major, minor, patch}', description: Avro schema version.
      Behaviour changes at the 6.5.0 boundary — see conversions below.}
  - {name: rawData, description: Container for every sensor stream in this window.}
  relationships:
  - {has_one: SchemaVersion, via: schemaVersion}
  - {has_one: RawData, via: rawData}
  - {belongs_to: Participant, via: participant_id in the filename}
  metadata_documented:
  - Participant Full ID
  - Device serial number
  - Participant's time offset from UTC+00:00
  - Sensors polling rate
  - ADC IMU parameters
- name: RawData
  description: The sensor stream container. Availability of individual streams depends
    on the study configuration, so a stream can legitimately be empty.
  relationships:
  - {has_one: Accelerometer, via: accelerometer}
  - {has_one: Gyroscope, via: gyroscope}
  - {has_one: BloodVolumePulse, via: bvp}
  - {has_one: ElectrodermalActivity, via: eda}
  - {has_one: Temperature, via: temperature}
  - {has_one: Steps, via: steps}
  - {has_one: SystolicPeaks, via: systolicPeaks}
  - {has_many: Tag, via: tags}
- name: Accelerometer
  path: rawData.accelerometer
  fields:
  - {name: timestampStart, description: Microsecond UTC timestamp of the first sample.}
  - {name: samplingFrequency, description: Samples per second; used to reconstruct per-sample
      timestamps.}
  - {name: x, description: Array of ADC counts on the X axis.}
  - {name: y, description: Array of ADC counts on the Y axis.}
  - {name: z, description: Array of ADC counts on the Z axis.}
  - {name: imuParams, shape: '{physicalMax, physicalMin, digitalMax, digitalMin, conversionFactor}',
    description: ADC IMU parameters used to convert counts to g.}
  conversion:
    before_6_5_0: 'g = value * (physicalMax - physicalMin) / (digitalMax - digitalMin)'
    from_6_5_0: g = value * conversionFactor
    note: >-
      A consumer MUST branch on schemaVersion to convert accelerometer counts
      correctly. This is a silent breaking change in the data contract with no
      version negotiation and no deprecation header — the only signal is the
      schemaVersion triple inside the file.
    source: https://support.empatica.com/hc/en-us/articles/17405877853981-How-to-access-Avro-files-with-Python
- name: ElectrodermalActivity
  path: rawData.eda
  fields:
  - {name: timestampStart}
  - {name: samplingFrequency}
  - {name: values, description: Array of EDA samples.}
- name: Temperature
  path: rawData.temperature
  fields:
  - {name: timestampStart}
  - {name: samplingFrequency}
  - {name: values, description: Array of skin temperature samples.}
- name: BloodVolumePulse
  path: rawData.bvp
  description: Blood volume pulse stream from the PPG sensor. Named in the raw-data
    stream list.
- name: Gyroscope
  path: rawData.gyroscope
  description: Gyroscope stream. Named in the raw-data stream list.
- name: Steps
  path: rawData.steps
  description: Step counts. Named in the raw-data stream list.
- name: SystolicPeaks
  path: rawData.systolicPeaks
  description: Detected systolic peaks. Named in the raw-data stream list.
- name: Tag
  path: rawData.tags
  description: Participant-marked events recorded on the device.
  source: https://support.empatica.com/hc/en-us/articles/17703560836381-How-tagged-events-data-look-in-the-RAW-data
- name: Participant
  description: >-
    The subject a file belongs to. Identified by participant_id in the filename and by
    Participant Full ID in the file metadata. Participants are created and terminated
    through the Care Portal or, for enterprise trials, through the Cloud API using the
    sponsor's own existing participant IDs.
  relationships:
  - {has_many: AvroFile, via: participant_id}
  - {has_one: Device, via: device serial number}
- name: Device
  description: EmbracePlus or EmbraceMini wearable. Identified by serial number carried
    in each Avro file's metadata.
sensors_documented: [PPG, EDA, Skin thermometer, Accelerometer, Gyroscope]
derived_products:
  digital_measures: '300+ digital measures / digital biomarkers, delivered as CSV. The
    per-measure catalog is not published publicly.'
  source: https://www.empatica.com/llms.txt
reader_support:
  official_avro_apis: [Python, Java, C, 'C#', C++]
  also_supported: [MATLAB]
  provider_example: 'Python script converting one Avro file into one CSV per sensor,
    published in the support centre (tested against Avro file schema v6).'
gaps:
- No JSON Schema, OpenAPI or published .avsc file — the schema exists only embedded
  inside delivered data, so it cannot be read before a study produces data.
- The full field-level reference ("Data Access Documentation") is reachable only from
  inside the Care Portal, at https://s3.amazonaws.com/box.empatica.com/ehmp/careportal/data_access/v2.4e/en.pdf,
  and is a PDF rather than a machine-readable schema.
- The digital-measure catalog (the 300+ biomarkers that are the actual product) has no
  public data dictionary.