Introduction to Magnetometer Calibration

Rishav
December 2020

1 Introduction
Magnetometers are used in spacecrafts to measure the geomagnetic field and deduce the orientation of
the spacecraft by utilizing the reference model of Earth’s magnetic field. Magnetometers are important
sensor used in attitude determination of microsatellites with small orbits as the magnetic field is strong
closer to Earth.

2 Magnetometer noise modeling

The magnetometers are subjected to various sources of error which can be either due to the fabrication
of the magnetometer or due to the environment on which it used. Due to these sourses of errors, the
readings of magnetometer is unsuitable to be used to compute the attitude of spacecraft. Hence, it
is important to identify those errors and compensate them in software to get rid of errors. Renaudin
classifies the magnetometer errors in two broad categories,
1. Instrumentation error
2. Magnetic deviation

2.1 Instrumentation error

Instrumentation error are the constant error source that is due to the device itself. It is related to the
way the magnetometer is fabricated.

Scale factor Scale factor is the proportionality constants related to each axes of the magnetometer by
which the true value is scaled.
ĥ = Shm (1)
Here,  
sx 0 0
S =0 sy 0
0 0 sz
is the scale matrix and sx , sy and sz are the scaling constants associated with each axis of the three-axis
magnetometer.

Bias Bias bo is the offset of the magnetic field measurements by the magnetometer from the true value
which it was supposed to read. In three-axis magnetometer there exists some bias associated with each
coordinate axes i.e. bi = [ bix , biy , biz ]⊺ .
ĥ = hm + bi (2)

Non-orthogonality Non-orthogonality represents the misalignment of the three-axis magnetometer

sensor from the coordinate frame that is regarded as the sensor frame. This misalignment can be modeled
using orthogonal rotation matrix M that maps the actual sensor readings in misaligned frame to the
sensor frame.
ĥ = M hm = [εx , εy , εz ]−1 (3)
where, εx , εy and εz ∈ R3 are the directions of the sensor’s x, y and z axes respectively in the sensor
frame.

1
2.2 Magnetic deviation
Magnetic deviation represents the error introduced due to the magnetic field on the surrounding of the
sensor.

Hard-iron Hard-iron error refers to the offset resulting due to the permanently magnetized ferromag-
netic components on the setup on which magnetometer is attached to. Since magnetometer and hard-iron
sources rotates together, it can me mathematically modeled as a bias vector bh = [ bhx bhy bhz ]⊺ whose
elements corresponds to the offset of magnetometer readings in each coordinate axes.

ĥ = hm + bh (4)

Soft iron Soft-iron effect is defined as the error introduced when external geomagnetic field induces
interfering magnetic field by interacting with ferromagnetic materials present around magnetometer.
Soft-iron effect can change the magnitude as well as the direction of the ĥ. Soft-iron effect is modeled
using a matrix As ∈ R3×3 such that,

ĥ = As hm (5)
Using Eq. (2.1), Eq. (2.1), Eq. (2.1), Eq. (2.2), and Eq. (2.2) representing each of the error sources
discussed above we can now perform the complete error modeling of the magnetometer.

ĥ = SM (As hm + bh ) + bi + ε (6)
Here, ε is Gaussian whiteband noise ∼ N (0, σε2 ),
and considered to be zero for simplification.
Eqn. (6) is simplified to
ĥ = Ahm + b (7)
3×3
where, A ∈ R = SM As is the matrix representating the combined effect of scale factors, mis-
alignments, and soft iron disturbances. b = SM bh + bi is the combined bias.
We can extract the true magnetometer reading ĥ from hm if the matrix A and vector b is determined
by some algorithm. Doing this is known as magnetometer calibration. It is the calibrated magnetometer
sensor readings that is used to determine orientation of spacecraft.

3 Magnetometer Calibration
Ideally, the norm of sensor readings from magnetometer should be equal to the magnitude of the geo-
magnetic field at that place. so, if three-axis magnetometer were rotated freely in three dimensions, then
the locus described by the sensor readings should describe a sphere with radius equal to the magnitude of
local Earth’s magnetic field. Following equation constraints the measurements of perfect magnetometer
in a perturbation free environment.
2
Hm − ||hm ||2 = Hm
2
− h⊺m hm = 0 (8)
where, Hm is the norm of geomagnetic field intensity at the place where measurement hm is done
determined using geomagnetic model like IGRF model. Above equation implies that the expected norm
and measurements are same so the magnetometer can be considered ideal. However from Eq. (7) we
know that sensor measurements can be modeled as follows.

hm = A−1 (ĥ − b) (9)

From Eq. (8) and Eq. (9) we get,
2
(ĥ − b)⊺ Q (ĥ − b) − Hm =0 (10)
where, Q = (A−1 )⊺ A−1 . The equation can be expanded to obtain quadratic equation.

ĥ⊺ Qĥ + u⊺ ĥ + k = 0 (11)

⊺ ⊺ 2
where, u = −2Q h, and k = b Qb − Hm .
Eq. (11) represents the general equation of plane of second
order so it might represent a hyperboloid, a cone, or an ellipsoid. The equation is ellipsoid if following
condition holds:

2
 u⊺ Q−1 u > 4k (12)
This inequality in the equation form is,

u⊺ Q−1 u − 4k = Hm
2
(13)

We know that the magnitude of the magnetic field of earth is strictly positive and due to this reason
˙
Eq.(12) holds. It implies that Eq. (11) is general equation of ellipsoid.
This connection suggests that locus described by the readings of uncalibrated three-axis magnetometer
is ellipsoid in nature. If we could determine parameters in Eq. (11), A and b can be determined which
are used to estimate ĥ using Eq. (11).

3.1 Ellipsoid fitting algorithm

Least Squares Ellipsoid Specific Fitting is the ellipsoid fitting algorithm that we used to estimate the
parameters of Eq. (11) from the magnetometer readings. This paper develops a sufficient condition for
a quadratic surface to be ellipsoid and develops closed form solution for ellipsoid fitting based on this
constraint. This algorithm is claimed to be stable, fast and robust to noise in the data. If the algorithm
were treated as black box, it generates the coefficients of the general equation of ellipsoid for given n sets
of three dimensional points. In our case, those points will be the readings from magnetometer.
The general equation of the second degree in three variables is

ax2 + by 2 + cz 2 + 2f yz + 2gxz + 2hxy + 2px + 2qy + 2rz + d = 0 (14)

Let,

I =a+b+c
J = ab + bc + ac − f 2 − g 2 − h2
a h g (15)
K= h b f
g f c

Eq. (14) with the constraint of 4J − I 2 > 0 represents the general equation of ellipsoid. Hence the
least squares fitting problem is formulated as follows:

min||Dv|| subject to 4J − I 2 = 1 (16)

where D ∈ R10×n is the design matrix defined as

D = [ w1 , w2 , ... , wn ]. (17)

The solution to this least squares problem appears as the eigenvalue problem
−1 ⊺
C1−1 (S11 − S12 S22 S12 ) v̄ = λ v̄ (18)

where C1 ∈ R6×6 , S11 ∈ R6×6 , S12 ∈ R6×4 and S22 ∈ R4×4 are defined as
 
−1 1 1 0 0 0
 1 −1 1 0 0 0
   
S S12 1 1 −1 0 0 0
DD ⊺ = 11

⊺ and C 1 =  . (19)
S12 S22 0
 0 0 −4 0 0
0 0 0 0 −4 0 
0 0 0 0 0 −4
−1 ⊺
Let v1 ∈ R6 be the eigenvector associated with largest eigenvalue of C1−1 (S11 − S12 S22 S12 ) ∈ R6×6 .
The ellipsoid vector v can now be evaluated as

v = [ v1⊺ , v2⊺ ]⊺ (20)

−1 ⊺
where v2 ∈ R4 = S22 S12 v1 . v in Eq. (20) is solution to Eq. (16). Algorithm (3.1) shows the
implementation procedure to implement ellipsoid fitting.

3
Algorithm 1 Least Squares Ellipsoid Fitting
1: Inputs:
x = [ x1 , x2 ... , xn ], y = [ y1 , y2 ... , yn ], and z = [ z1 , z2 ... , zn ]
2: Initialize design matrix D and C1 using Eq. (17) and Eq. (19) respectively.
3: Find S11 , S12 and S22 using Eq. (19).
−1 −1 ⊺
4: Find the eigenvector v1 associated with largest eigenvalue of C1 (S11 − S12 S22 S12 )
−1 ⊺
5: Compute v2 = S22 S12 v1 .
⊺ ⊺
6: Return ellipsoid vector v = [ v1 , v2 ]⊺