Rotations and Coordinate Systems

One of DOLfYN’s primary advantages is that it contains simple tools for managing the coordinate system (a.k.a. the reference frame) of the vector data of data objects. This has taken considerable effort because the coordinate system definitions used by instrument manufacturers are not consistent, and the math/concepts of coordinate rotations can be described as somewhere between non-trivial and absolutely maddening. We hope that this page, and DOLfYN’s tools for managing coordinate system help to reduce the burden of tracking this information so that you – the researcher – can focus on what’s important.

Having said that, the coordinate-system/rotation tools provided in DOLfYN have been tested to varying degrees on different types of instruments and configurations. Instrument manufacturers use different conventions, and can change conventions with firmware updates. Therefore, we make no promises that these tools will work for any instrument type, but we do have higher confidence in some instruments and configurations than others. See the table at the bottom of this page for details on the degree of testing of DOLfYN’s rotations and coordinate-system tools that has occurred for several instrument types. With your help, we hope to improve our confidence in these tools for the wide-array of instruments and configurations that exist.

The values in the list dat.props['rotate_vars'] specifies the vectors that are rotated when changing between different coordinate systems. The first dimension of these vectors are their coordinate directions, which are defined by the following coordinate systems:

  • BEAM: this is the coordinate system of the ‘along-beam’ velocities. When the data object is in BEAM coordinates, the first dimension of the velocity vectors are: [beam1, beam2, … beamN]. This coordinate system is not ortho-normal, which means that the inverse rotation (inst to beam) cannot be computed using the transpose of the beam-to-inst rotation matrix. Instead, the inverse of the matrix must be computed explicitly, which is done internally in DOLfYN (in beam2inst()).

    When a data object is in this coordinate system, only the velocity data (i.e., the variables in dat.props['rotate_vars'] starting with 'vel') is in beam coordinates. Other vector variables listed in 'rotate_vars' are in the INST frame (e.g., dat.orient.AngRt). This is true for data read from binary files that is in beam coordinates, and also when rotating from other coordinate systems to beam coordinates.

  • INST: this is the ‘instrument’ coordinate system defined by the manufacturer. This coordinate system is orth-normal, but is not necessarily fixed. That is, if the instrument is rotating, then this coordinate system changes relative to the earth. When the data object is in INST coordinates, the first dimension of the vectors are: [X, Y, Z, …].

    Note: instruments with more than three beams will have more than three velocity components. |dlfn| does not yet handle these extra dimensions consistently.

  • EARTH: When the data object is in EARTH coordinates, the first dimension of vectors are: [East, North, Up, …]. This coordinate system is also sometimes denoted as “ENU”. If the declination is set the earth coordinate system is “True-East, True-North, Up” otherwise, East and North are magnetic. See the Declination Handling section for further details on setting declination.

    Note that the ENU definition used here is different from the ‘north, east, down’ coordinate system typically used by aircraft. Also note that the earth coordinate system is a ‘rotationally-fixed’ coordinate system: it does not rotate, but it is not necessarily inertial or stationary if the instrument slides around translationally (see the Motion Correction section for details on how to correct for translational motion).

  • PRINCIPAL: the principal coordinate system is a fixed coordinate system that has been rotated in the horizontal plane (around the Up axis) to align with the flow. In this coordinate system the first dimension of a vector is meant to be: [Stream-wise, Cross-stream, Up]. This coordinate system is defined by the variable dat.props['principal_heading'], which specifies the principal coordinate system’s \(+u\) direction. The \(v\) direction is then defined by the right-hand-rule (with \(w\) up). See the Principal Heading section for further details.

To rotate a data object into one of these coordinate systems, simply use the rotate2 method:

>>> dat_earth = dat.rotate2('earth')  # ("rotate to earth")
>>> dat_earth
<ADV data object>
  . 1.05 hours (started: Jun 12, 2012 12:00)
  . EARTH-frame
  . (120530 pings @ 32Hz)
  *------------
  | mpltime                  : <time_array; (120530,); float64>
  | vel                      : <array; (3, 120530); float32>
  + config                   : + DATA GROUP
  + env                      : + DATA GROUP
  + orient                   : + DATA GROUP
  + props                    : + DATA GROUP
  + signal                   : + DATA GROUP
  + sys                      : + DATA GROUP

Orientation Data

The instrument orientation data in DOLfYN data objects is contained in the orient data group. The orientmat data item in this group is the orientation matrix, \(R\), of the instrument in the earth reference frame. It is a 3x3xNt array, where each 3x3 array is the rotation matrix that rotates vectors in the earth frame, \(v_e\), into the instrument coordinate system, \(v_i\), at each timestep:

\[v_i = R \cdot v_e\]

The ENU definitions of coordinate systems means that the rows of \(R\) are the unit-vectors of the XYZ coordinate system in the ENU reference frame, and the columns are the unit vectors of the ENU coordinate system in the XYZ reference frame. That is, for this kind of simple rotation matrix between two orthogonal coordinate systems, the inverse rotation matrix is simply the transpose:

\[v_e = R^T \cdot v_i\]

Heading, Pitch, Roll

Some (most?) instruments do not calculate or output the orientation matrix by default. Instead, these instruments typically provide heading, pitch, and roll data (hereafter, h,p,r). For these instruments these variables are used to calculate an orientation matrix, \(R\), and the values of h,p,r are removed from the data object. These can be obtained from the orientation matrix using the dolfyn.orient2euler function (dolfyn.euler2orient provides the reverse functionality). This function computes these variables according to the following conventions:

  • a “ZYX” rotation order. That is, these variables are computed assuming that rotation from the earth -> instrument frame happens by rotating around the z-axis first (heading), then rotating around the y-axis (pitch), then rotating around the x-axis (roll).
  • heading is defined as the direction the x-axis points, positive clockwise from North (this is opposite the right-hand-rule around the Z-axis)
  • pitch is positive when the x-axis pitches up (this is opposite the right-hand-rule around the Y-axis)
  • roll is positive according to the right-hand-rule around the instrument’s x-axis

Considerable care has been taken to make sure that these definitions of h,p,r and orientmat are consistent within DOLfYN between instrument models. However, because the instrument manufacturer’s definitions of these variables are not consistent between instrument makes/models, this means that:

  • DOLfYN uses instrument-specific functions to calculate a consistent dat['orient']['orientmat'] from the inconsistent definitions of h,p,r
  • DOLfYN’s consistent definitions are different from the definitions provided by an instrument manufacturer (i.e., there is no consensus on these definitions, so DOLfYN uses the above definitions)

Instruments that provide an orientmat directly will have dat.props['has_imu'] = True. Otherwise, the orientmat was calculated from h,p,r. Note that an orientation matrix calculated from h,p,r can have larger error associated with it, partly because of the gimbal lock problem, and also because the accuracy of some h,p,r sensors decreases for large pitch or roll angles (e.g., >40 degrees).

So, while DOLfYN uses the instrument manufacturer’s definition of the instrument coordinate system (“XYZ”), the details of how this relates to the ‘earth’ coordinate system, and how h,p,r are computed are often distinct from the definitions specified by the instrument manufacturer (e.g., some manufacturers reference heading off of the y-axis rather than x, some use a ‘north-east-down’ earth reference frame, etc.). For practical purposes, when utilizing DOLfYN orientation data (for all instrument types), this all means that the user should:

  • Use the instrument manufacturers definitions of XYZ
  • Interpret h,p,r data according to the above definitions (ignore manufacturer definitions of these variables)
  • All rotations into the earth frame will yield vectors that are in a ENU coordinate system

It is also possible to obtain the raw h,p,r data as defined by the instrument manufacturer using the keep_orient_raw argument in the dolfyn.read() function. That data is contained in dat['orient']['raw']. Note that this data does not obey the above definitions, and instead obeys the instrument manufacturer’s definitions of these variables (i.e., it is exactly the data contained in the binary file). Also note that dat['orient']['raw']['heading'] is unaffected by setting declination as described in the next section.

Declination Handling

DOLfYN includes functionality for handling declination, but the value of the declination must be specified by the user. There are two ways to set a data-object’s declination:

  1. Set declination explicitly using the dat.set_declination method, for example:

    dat.set_declination(16.53)

  2. Set declination in the <data_filename>.userdata.json file (more details ), then read the binary data file (i.e., using dat = dolfyn.read(<data_filename>)).

Both of these approaches produce modify the dat as described in the documentation for set_declination() .

Principal Heading

As described above, the principal coordinate system is meant to be the flow-aligned coordinate system (Streamwise, Cross-stream, Up). DOLfYN includes the :func:<dolfyn.calc_principal_heading> function to aide in identifying/calculating the principal heading. Using this function to identify the principal heading, an ADV data object that is in the earth-frame can be rotated into the principal coordinate system like this:

dat.props['principal_heading'] = dolfyn.calc_principal_heading(dat.vel)
dat.rotate2('principal')

Note here that if dat is in a coordinate system other than EARTH, you will get unexpected results, because you will calculate a principal_heading in the coordinate system that the data is in.

It should also be noted that by setting dat.props['principal_heading'] the user can choose any horizontal coordinate system, and this might not be consistent with the streamwise, cross-stream, up definition described here. In those cases, the user should take care to clarify this point with collaborators to avoid confusion.

Degree of testing by instrument type

The table below details the degree of testing of the rotation, p,r,h, and coordinate-system tools contained in DOLfYN. The confidence column provides a general indication of the level of confidence that we have in these tools for each instrument.

If you encounter unexpected results that seem to be related to coordinate systems (especially for instruments and configurations that are listed as “low” or “medium” confidence), the best thing to do is file an issue.

Table 1: Instruments tested to be consistent with DOLfYN’s coordinate systems and rotation tools.
Make Model Config Confidence Notes
Nortek Vector modern (~2019) firmware Medium “Some direct “instrument on the desk” confirmation of orientation-matrix and p,r,h calcs.”
Nortek Vector with IMU, modern (2019) firmware High “Lots of direct ‘instrument on the desk’ confirmation of orientation-matrix and p,r,h calcs.”
Nortek Signature modern (~2019) firmware Medium “Some validation by reasonable results when working with data.”
Nortek Signature with IMU, modern (2019) firmware Medium “Some validation by reasonable results when working with data.”
Teledyne Workhorse modern (~2019-ish) firmware Medium “Some cross-validation with other sensors in post-processing, but minimal ‘instrument on the desk’ testing.”
ALL ALL External input orientation data NONE “There has been no testing of external heading, pitch, or roll inputs”
Nortek AWAC modern (~2019) firmware Low “This works, but there has been almost no testing to validate results”