Find file Copy path
457 lines (353 sloc) 20.6 KB

Signal Metadata Format Specification v0.0.2


The Signal Metadata Format (SigMF) specifies a way to describe sets of recorded digital signal samples with metadata written in JSON. SigMF can be used to describe general information about a collection of samples, the characteristics of the system that generated the samples, and features of the signal itself.

Status of this Document

This document is currently under active development and is not stable. We encourage anyone and everyone interested in this effort to participate. We are using the Issue Tracker for the repository as the medium for discussion, and changes are submitted as Pull Requests.

This effort was kicked off at the DARPA Brussels Hackfest in early February 2017, and first announced at FOSDEM'17.

Copyright Notice

This document is Copyright of The GNU Radio Foundation, Inc.

This document is available under the CC-BY-SA License.

Creative Commons License

Table of Contents

Table of Contents


Sharing sets of recorded signal data is an important part of science and engineering. It enables multiple parties to collaborate, is often a necessary part of reproducing scientific results (a requirement of scientific rigor), and enables sharing data with those who do not have direct access to the equipment required to capture it.

Unfortunately, these datasets have historically not been very portable, and there is not an agreed upon method of sharing metadata descriptions of the recorded data itself. This is the problem that SigMF solves.

By providing a standard way to describe data recordings, SigMF facilitates the sharing of data, prevents the "bitrot" of datasets wherein details of the capture are lost over time, and makes it possible for different tools to operate on the same dataset, thus enabling data portability between tools and workflows.

Conventions Used in this Document

The key words “MUST”, “MUST NOT”, “REQUIRED”, “SHALL”, “SHALL NOT”, “SHOULD”, “SHOULD NOT”, “RECOMMENDED”, “MAY”, and “OPTIONAL” in this document are to be interpreted as described in RFC 2119.

JSON keywords are used as defined in ECMA-404.

Augmented Backus-Naur form (ABNF) is used as defined by RFC 5234 and updated by RFC 7405.

Fields defined as "human-readable", a "string", or simply as "text" shall be treated as plaintext where whitespace is significant, unless otherwise specified.


The SigMF specification fundamentally describes two types of information: datasets, and metadata associated with those datasets. Taken together, a dataset with its SigMF metadata is a SigMF Recording.

Datasets, for purposes of this specification, are sets of digital measurements generically called samples in this document. The samples can represent any time-varying source of information. They may, for example, be digital samples created by digital synthesis or by an Analog-to-Digital Converter. They could also be geolocation coordinates from a GNSS receiver, temperature readings from a thermal sensor, or any other stored digital measurement information.

Metadata describes the dataset with which it is associated. The metadata includes information meant for the human users of the dataset, such as a title and description, and information meant for computer applications (tools) that operate on the dataset.


A SigMF Recording consists of two files: a SigMF metadata file and a dataset file. The dataset file is a binary file of digital samples, and the metadata file contains information that describes the dataset.

  1. The metadata and dataset MUST be in separate files.
  2. The metadata file MUST only describe one dataset file.
  3. The metadata file MUST be stored in UTF-8 encoding.
  4. The metadata file MUST have a .sigmf-meta filename extension.
  5. The dataset file MUST have a .sigmf-data filename extension.
  6. The names of the metadata and dataset files must be identical (excepting their extensions).

SigMF Archives

The metadata and dataset files that comprise a SigMF Recording may be combined into a file archive. A SigMF Archive may contain multiple SigMF Recordings.

  1. The archive MUST use the tar archive format, as specified by POSIX.1-2001.
  2. The archive file's filename extension MUST be .sigmf.
  3. The archive MUST contain the following files: for each contained recording with some name given here meta-syntactically as N, files named N (a directory), N/N.sigmf-meta, and N/N.sigmf-data.
  4. The archive MUST NOT contain any other files unless their pathnames begin with N/N, for some N which has .sigmf-meta and .sigmf-data files as described above.
  5. It is RECOMMENDED that if recordings in an archive represent a continuous dataset that has been split into separate recordings, that their filenames reflect the order of the series by appending a hyphenated zero-based index (e.g., N-0, N-1, N-2, etc.,).

Each recording in an archive, even if connected to others by being part of a larger dataset, MUST be evaluated independently for compliance to the SigMF standard, and thus all metadata MAY be different between the recordings.

Dataset Format

The samples in the dataset file must be in a SigMF-supported format. There are four orthogonal characteristics of sample data: complex or real, floating-point or integer, bit-width, and endianness. The following ABNF rules specify the dataset formats defined in the SigMF core namespace:

    dataset-format = (real / complex) ((type endianness) / byte)

    real = "r"
    complex = "c"

    type = floating-point / signed-integer / unsigned-integer
    floating-point = "f32"
    signed-integer = "i32" / "i16"
    unsigned-integer = "u32" / "u16"

    endianness = little-endian / big-endian
    little-endian = "_le"
    big-endian = "_be"

    byte = "i8" / "u8"

So, for example, the string "cf32_le" specifies complex 32-bit floating-point samples stored in little-endian, the string "ru16_be" specifies real unsigned 16-bit samples stored in big-endian, and the string "cu8" specifies complex unsigned byte.

Note that only IEEE-754 single-precision floating-point is supported by the SigMF core namespace.

The samples should be written to the dataset file without separation, and the dataset file MUST NOT contain any other characters (e.g., delimiters, whitespace, line-endings, EOF characters, etc.,).

Complex samples should be interleaved, with the in-phase component first (i.e., I[0] Q[0] I[1] Q[1] ... I[n] Q[n]).

Metadata Format

SigMF is written in JSON and takes the form of JSON name/value pairs which are contained within JSON objects. There are three types of top-level objects: global, captures, and annotations. The names of the name/value pairs must be namespaced.

The format of the name/value pairs is:

"namespace:name": value,
  1. The metadata MUST be written in JSON, as specified by ECMA-404.
  2. The entire contents of the metadata file MUST be contained within a single JSON object. This object is hereafter called the top-level object.
  3. The top-level object MUST contain three values named global, captures, and annotations.
  4. Metadata name/value pairs SHALL NOT be assumed to have carried over between capture or annotation segments. If a name/value pair applies to a particular segment, then it must appear in that segment, even if the value is unchanged relative to the previous segment.


The values in each name/value pair must be one of the following datatypes:

type long-form name description
int integer Signed 64-bit integer.
uint unsigned long Unsigned 64-bit integer.
double double-precision floating-point number A 64-bit float as defined by IEEE 754.
string string A string of characters, as defined by the JSON standard.
boolean boolean Either true or false, as defined by the JSON standard.
null null null, as defined by the JSON standard.
array JSON array an array of other values, as defined by the JSON standard.
object JSON object an object of other values, as defined by the JSON standard.


Namespaces provide a way to further classify name/value pairs within metadata objects. This specification defines the core namespace, which contains the foundational name/value pairs for describing signal data.

Some keys within the core namespace are optional, and others are required. The fields that are required are those that are minimally necessary to parse & process the dataset, or that have obvious defaults that are valid. Other fields are 'optional', even if they are highly encouraged.

Extension Namespaces

Fields not defined in the core namespace may be defined in extension namespaces. The SigMF specification defines some extension namespaces to provide canonical definitions for commonly needed metadata fields that do not belong in core. These canonical extension namespaces can be found in the extensions/ directory of the official SigMF repository. Other extension namespaces may be defined by the user as needed.

  1. An extension namespace MUST be defined in a single file, named meta-syntactically as, whereN is the name of the extension.
  2. A file MUST be a Github-Flavored Markdown file stored in UTF-8 encoding.
  3. Extensions MUST have version numbers. It is RECOMMENDED that extensions use Semantic Versioning.
  4. An extension namespace MAY define new top-level SigMF objects, name/value pairs, new files, new dataset formats, or new datatypes.
  5. New name/value pairs defined by an extension namespace MUST be defined in the context of a specific SigMF top-level object (i.e., global, captures, annotations, or a new user-defined object).
  6. It is RECOMMENDED that an extension namespace file follow the structure of the canonical extension namespaces.
Canonical Extension Namespaces

This is a list of the canonical extension namespaces defined by SigMF:

  • antenna - Used to describe the antenna(s) used to for the recording.
  • modulation - Defines how to describe modulations used in wireless communications systems.
  • volatile - Allows for continously time-varying fields, such as a moving receiver or rotating antenna.

Global Object

The global object consists of name/value pairs that provide information applicable to the entire dataset. It contains the information that is minimally necessary to open and parse the dataset file, as well as general information about the recording itself.

The following names are specified in the core namespace and should be used in the global object:

name required type description
datatype true string The format of the stored samples in the dataset file. Its value must be a valid SigMF dataset format type string.
sample_rate false double The sample rate of the signal in samples per second.
version true string The version of the SigMF specification used to create the metadata file.
sha512 false string The SHA512 hash of the dataset file associated with the SigMF file.
offset false uint The index number of the first sample in the dataset. This value defaults to zero. Typically used when a recording is split over multiple files.
description false string A text description of the SigMF recording.
author false string The author's name (and optionally e-mail address).
meta_doi false string The registered DOI (ISO 26324) for a recording's metadata file.
data_doi false string The registered DOI (ISO 26324) for a recording's dataset file.
recorder false string The name of the software used to make this SigMF recording.
license false string A URL for the license document under which the recording is offered; when possible, use the canonical document provided by the license author, or, failing that, a well-known one.
hw false string A text description of the hardware used to make the recording.
extensions false object A list of extensions used by this recording.
The extensions Field

The core:extensions field in the global object is JSON array of name/value pairs describing SigMF Extension namespaces, where the name is the namespace provided by an extension and the value is a string that specifies whether the extension is optional or the version of the extension required to properly parse & process the SigMF Recording. In the example below, extension-01 is used, but not required, and version 1.2.3 of extension-02 is required.

  "global": {
    "core:extensions" : {
      "extension-01": "optional",
      "extension-02": "v1.2.3",

Captures Array

The captures value is an array of capture segment objects that describe the parameters of the signal capture. It MUST be sorted by the value of each capture segment's core:sample_start key, ascending.

Capture Segment Objects

Capture segment objects are composed of name/value pairs.

Each capture segment object must contain a core:sample_start name/value pair, which indicates the first index at which the rest of the segment's name/value pairs apply.

The following names are specified in the core namespace and should be used in capture segment objects:

name required type description
sample_start true uint The sample index in the dataset file at which this segment takes effect.
global_index false uint If the sample source provides a global sample count, this is the global index that maps to sample_start.
frequency false double The center frequency of the signal in Hz.
datetime false string An ISO-8601 string indicating the timestamp of the sample index specified by sample_start. More details, below.
The global_index Pair

Some hardware devices are capable of 'counting' samples, or assigning sample indices relative to the sample stream produced or consumed by the device. Note this is different from the sample index used to reference a sample in the SigMF dataset file.

These numbers are most commonly used to indicate that data was dropped by the hardware device. For example, if the hardware driver provides a packet of data, labeled with samples 0 to 1000, and the following packet labels its first sample as number 1500, that indicates that 500 samples were dropped between those two packets. This field allows you to indicate such a discontinuity in the recorded sample stream as seen by the application (e.g., a SigMF writer or reader).

The datetime Pair

This name/value pair must be an ISO-8601 string, as defined by RFC 3339, where the only allowed time-offset is Z, indicating the UTC/Zulu timezone. The ABNF description is:

   date-fullyear   = 4DIGIT
   date-month      = 2DIGIT  ; 01-12
   date-mday       = 2DIGIT  ; 01-28, 01-29, 01-30, 01-31 based on
                             ; month/year
   time-hour       = 2DIGIT  ; 00-23
   time-minute     = 2DIGIT  ; 00-59
   time-second     = 2DIGIT  ; 00-58, 00-59, 00-60 based on leap second
                             ; rules
   time-secfrac    = "." 1*DIGIT
   time-offset     = "Z"

   partial-time    = time-hour ":" time-minute ":" time-second [time-secfrac]
   full-date       = date-fullyear "-" date-month "-" date-mday
   full-time       = partial-time time-offset

   date-time       = full-date "T" full-time

Thus, timestamps take the form of YYYY-MM-DDTHH:MM:SS.SSSZ, where any number of digits for fractional seconds is permitted.

Annotations Array

The annotations value is an array of annotation segment objects that describe anything regarding the signal data not part of the captures and global objects. It MUST be sorted by the value of each annotation segment's core:sample_start key, ascending.

Annotation Segment Objects

Annotation segment objects contain name/value pairs.

Each annotation segment object must contain a core:sample_start name/value pair, which indicates the first index at which the rest of the segment's name/value pairs apply.

The following names are specified in the core namespace and should be used in annotation segment objects:

name required type description
sample_start true uint The sample index at which this segment takes effect.
sample_count true uint The number of samples that this segment applies to.
generator false string Human-readable name of the entity that created this annotation.
comment false string A human-readable comment.
freq_lower_edge false double The frequency (Hz) of the lower edge of the feature described by this annotation.
freq_upper_edge false double The frequency (Hz) of the upper edge of the feature described by this annotation.
latitude false
longitude false

There is no limit to the number of annotations that can apply to the same group of samples. If two annotations have the same sample_start, there is no defined ordering between them.

The freq_lower_edge and freq_upper_edge fields should be at RF if the feature is at a known RF frequency. If there is no known center frequency (as defined by the frequency field in the relevant capture segment object), or the center frequency is at baseband, the freq_lower_edge and freq_upper_edge fields may be relative to baseband.

Dataset Licensing

You may specify any license of your choosing. Recommended licenses for SigMF recordings are:

SigMF Compliance by Applications

In order to be SigMF Compliant, an application must meet the following requirements:

  1. Adheres to and supports the file rules, dataset formats, objects, namespaces, and names specified by this document.
  2. Must be able to ignore any object or namespace not specified by this document and still function normally.
  3. Capture segments referring to non-existent samples should be ignored.
  4. Must treat consecutive capture segments whose metadata is equivalent for purposes of that application (i.e., it may be different in values ignored by the application such as optional values or unknown extensions) as it would a single segment.
  5. Supports all fields in the core namespace.


[TODO] Provide an example of metadata file contents.

Citing SigMF

To cite the SigMF specification, we recommend the following format:

The Signal Metadata Format (SigMF), <release>, <date of release>,


This specification originated at the DARPA Brussels Hackfest 2017.