Measurement 109 (2017) 297–303

Contents lists available at ScienceDirect

Measurement
journal homepage: www.elsevier.com/locate/measurement

Low-noise tunneling-magnetoresistance vector magnetometers with

flux chopping technique
Van Su Luong a,c, Jen-Tzong Jeng b,⇑, Chih-Cheng Lu a,⇑, Hua-Yi Hsu a
a
Graduate Institute of Mechatronic Engineering/Department of Mechanical Engineering, National Taipei University of Technology, Taipei 10608, Taiwan
b
Department of Mechanical Engineering, National Kaohsiung University of Applied Sciences, Kaohsiung 80778, Taiwan
c
International Training Institute for Materials Science, Hanoi University of Science and Technology, Hanoi 100000, Viet Nam

a r t i c l e i n f o a b s t r a c t

Article history: A concept for a low-noise three-axis magnetometer consisting of tunneling magnetoresistance (TMR)
Received 2 January 2017 sensors and a flux chopper was designed, implemented, and characterized in this work. The TMR sensors
Received in revised form 24 May 2017 used in this study were the model of TMR2102D from Multidimension Technology Inc. Three TMR sen-
Accepted 26 May 2017
sors were aligned orthogonally on a printed circuit board (PCB) and mounted inside a cylindrical flux
Available online 29 May 2017
chopper. The cylindrical flux chopper including a soft magnetic shielding tube and enameled copper
wires was 16 mm in length and 8 mm in diameter. The shielding tube was made of a cobalt-based soft
Keywords:
magnetic ribbon, Metglas-2714A, from Metglas Inc. The flux chopper modulated the external magnetic
3-axis magnetometers
Low-frequency noise
flux density using the fluxgate effect, which made the sensors respond to the quasi-static field at the
Magnetic noise chopping frequency. The demodulated output showed a reduction in low-frequency noise to the level
p
Magnetic shielding of 0.17 nT/ Hz@1 Hz. To demonstrate the technical feasibility for electronic compass application, the
demodulated output of the vector magnetometer prototype was recorded by rotating the device in
Earth’s magnetic fields about a fixed axis. The Cartesian components Bx, By, and Bz of the Earth’s fields
at various azimuth angles were retrieved by performing a calibration algorithm to correct the non-
orthogonality caused by the misalignment of TMR sensors and the chopper. The calibrated outputs were
linear and orthogonal to each other with an angle error less than 1° and nonlinearity of 0.7%, indicating
that the chopping technique is useful to realize an extremely low-noise three-dimensional magnetometer
for the geomagnetic application.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction
Three-dimensional (3D) magnetic sensors play important roles
in multiple applications, such as position sensing, automotive
application, and geomagnetic navigation [1]. In recent decades,
various kinds of vector magnetometers, including fluxgate sensor,
Hall sensor, anisotropic magnetoresistance (AMR) sensor, giant
magnetoresistance (GMR), and tunneling magnetoresistance
(TMR) sensor, were developed for applications in low-cost and
low-power geomagnetic measurements [2–13]. Among them,
TMR is the most promising technology to realize a micro 3D mag-
netometer with low-noise and low-power, which are crucial crite-
ria to the emerging portable and wearable electronic devices. To
build a 3D magnetometer, many techniques were introduced [2–
5]. The simple method is to use three magnetic sensors with their
sensing directions along the three coordinate axes, i.e. x-, y-, and
z- axes [2]. An alternative way is to use planar sensors and a mag-
⇑ Corresponding author.
E-mail addresses: jtjeng@kuas.edu.tw (J.-T. Jeng), cclu23@ntut.edu.tw (C.-C. Lu).
netic flux guide [3–5]. However, for all of the designs mentioned so
far, the experimental tolerances and environmental effects lead to
the non-orthogonality between three sensing directions. Notably,
this problem could be well solved using a calibration algorithm
[14]. The first TMR 3D magnetometer was developed in 2002 by
Tondra et al. [2,15] and the noise floor archived was around a
few nanotesla. But until recently, the typical noise level of a com-
mercial 3D TMR magnetic field sensor is still high. The noise level
p
is about 10 nT/ Hz at above 10 Hz [16]. As the 1/f noise behavior
exists, the noise level is even higher at a lower frequency region.
Therefore, a feasible solution to the 1/f noise problem is a crucial
study for a 3D TMR magnetometer intended for quasi-static low-
field applications. To deal with the noise problem, several extrinsic
methods have been reported, including TMR arrays [17], flux con-
centrator [18], and flux chopping techniques [19–21]. It was
reported that the 1/f noise in TMR is mainly caused by the fluctu-
ation of resistance [22,23]. The reduction in noise can be expected
with a series array of N TMR elements because the total output is
boosted by N times, while the weakly correlated fluctuation of
p
resistance is increased only by N. With a parallel array, the total

http://dx.doi.org/10.1016/j.measurement.2017.05.062
0263-2241/Ó 2017 Elsevier Ltd. All rights reserved.
298 V.S. Luong et al. / Measurement 109 (2017) 297–303
output is the same while the fluctuation of resistance is reduced by
p
N. For both cases, the field noise gives the same noise reduction
p
ratio of N. However, with the array method, the 1/f behavior
remains. In this way, the overall sensing area is quadratically
increased with the reduction ratio. The other methods, such as
the flux concentrator, can suppress the field noise by enhancing
the sensitivity. However, the magnetic hysteresis induced by mag-
netic thin film results in the drift in dc level of the sensor output.
Feasible methods to suppress the 1/f noise are an ac biasing [24]
and voltage chopping [25], which are similar to the chopping tech-
nique commonly applied in optical measurements. The two kinds
of electrical chopping methods can suppress the low-frequency
noise arising from the charge-induced fluctuation. However, for
TMR sensors, the 1/f noise originates from both electrical and mag-
netic fluctuations [26]. Thus, the noise due to the instability of
magnetization cannot be overcome by the electrical chopping
method. Notably, the magnetic flux chopping technique, which
was invented by Jander’s group [19], is a promising method to
reduce both the magnetic and electronic 1/f noise. The successful
implementations of flux chopper have been reported in recent
years [20,27]. The oscillating flux concentrator [27] and the modu-
lating magnetic shielding [20] are both viable methods to ‘‘chop”
the external magnetic flux, so as to reduce the low-frequency mag-
netic noise. Nevertheless, the unstable magnetization state of
mechanical flux chopper makes it ineffective for suppressing the
extremely low-frequency noise. In fact, the hysteresis could be
minimized by modulating the permeability of flux chopper [19].
It was shown that the shielding flux chopper was capable of reduc-
ing low-frequency noise by a factor above 20 at frequencies below
1-Hz [20]. To build a low noise three-axis magnetometer with a
shielding flux chopper, a challenge is that at least three sets of
flux-chopped TMR sensors must be used in a compact size. Hence
we designed and proposed a new concept for low-noise vector
magnetometers comprising three high-sensitivity TMR sensors
and a magnetic flux chopper in this study. The flux chopping tech-
nique was successfully applied to the TMR sensors and proven to
suppress the low-frequency noise using a single flux chopper,
which allows future integration of the system into a single pack-
age. The non-orthogonality problem was also well solved by using
a calibration algorithm. The noise and sensitivity characterizations
were performed and discussed, and the feasibility for geomagnetic
application was demonstrated.
2. Experimental details
2.1. Design of magnetometer and measurements
The design of a 3D magnetometer using TMR sensors and a
shielding chopper is shown in Fig. 1. The sensing elements were
Fig. 1. Design of the three-axis TMR magnetometer with a flux chopper: (a) side view, (b) quadrant cross section view, and (c) structure of the DFN8-TMR2102D sensor [28].
the TMR2102D sensors, from the Multidimension Technology Inc.
[28]. The TMR sensor is in a full Wheatstone bridge configuration
consisting of four tunnel-junction arrays, as depicted in Fig. 1(c).
The packaging size of the DFN8 type is 3 mm  3 mm  0.75 mm,
and the linear range is ±30 Oe with a nonlinearity of 1% at the full
scale. The intrinsic sensitivity is about 4.9 mV/V/Oe in the full-scale
operation range up to ±90 Oe of saturation field [28]. Three TMR
sensors were aligned orthogonally on a print circuit boards (PCB)
with the sensing directions along three-dimensional coordinates,
as shown in Fig. 1(b). These sensors were placed in the center of
a cylindrical chopper tube, which is 16 mm in length and 8 mm
in diameter. The core of the chopper is made of a cobalt-based
amorphous soft magnetic ribbon, Metglas-2714A, from Metglas
Inc. [29]. The magnetizing field for chopper operation was induced
by the enameled copper wire coil threading through and wrapping
the tube. The field strength to saturate the core of the chopper is
above 0.1 Oe. The chopping signal generator and phase sensitive
detection (PSD) were carried out digitally using a data acquisition
(DAQ) device, which is the multifunction device of the model USB-
6216 from National Instruments Inc. The outputs of three TMR sen-
sors were pre-amplified via a three-channel instrumentation
amplifier, INA111 from Texas Instruments Inc. A 1 kHz square
wave chopping signal was generated by the DAQ, and then it was
amplified via a power amplifier to induce a 200 mA chopping cur-
rent. The current switched the magnetization of the core between
the saturation and unsaturation, thus turning the magnetic flux ON
and OFF. The sensitivity of the device was determined through the
slope of a Volt/Tesla curve. To measure noise, the system was
placed inside a tri-layer magnetic shielding chamber. A dynamic
signal analyzer, model SR780 from the Stanford Research Systems,
was used to record the noise spectrum. To measure the azimuth
response, the device was fixed on a manual rotation stage. The
software-based PSD and device calibration were performed by a
commercial LabVIEW program.
2.2. Calibration algorithm
The 3D magnetometer aims to operate at a very low frequency
below 1 Hz. Therefore, the magnetometer has been tested as an
electronic compass with a calibration algorithm. Although the
sensing directions of three sensors were aligned as orthogonal as
possible to each other, the actual sensing directions were found
to be non-orthogonal, owing to the misalignment of the three sen-
sors on the separate PCBs as well as the flux bending effect of the
chopper. To overcome this problem, the conversion algorithm
employing voltage-to-field transfer matrix was performed. Sup-
pose that the Earth’s field is in an arbitrary direction relative to
the magnetometer, the Cartesian components Bx, By, and Bz are
 V.S. Luong et al. / Measurement 109 (2017) 297–303 299

determined using the response of three TMR sensors as following ~ ^ þ Bz^z

B ¼ Bx ^x þ By y ð8Þ
equations:
2 3 In general, the three sensing axes ^s1 , ^s2 ; and ^s3 deviate from the
Bx unit vectors ^ ^; and ^z of X, Y, and Z coordinates by angles of ai ; bi ;
x; y
6 7
B ¼ 4 By 5 ¼ A  V; ð1Þ and ci :
Bz 8
< ^s1 ¼ cos a1 ^x þ cos b1 y
> ^ þ cos c1 ^z
2 3 ^s2 ¼ cos a2 ^x þ cos b2 y
^ þ cos c2 ^z ð9Þ
ax1 ax2 ax3 >
:^
6 s3 ¼ cos a3 ^x þ cos b3 y
^ þ cos c3 ^z
A ¼ 4 ay1 ay2 ay3 7
5; and ð2Þ
az1 az2 az3 where i = 1, 2 or 3 represents the x, y or z- sensor, respectively. From
Eqs. (8) and (9), the components of ~
B detected by each sensors are
2 3
V1 given by the following equations,
6 7 8
V ¼ 4 V2 5 ð3Þ
< B1 ¼ Bx cos a1 þ By cos b1 þ Bz cos c1
>
V3 B2 ¼ Bx cos a2 þ By cos b2 þ Bz cos c2 ð10Þ
>
:
where V1, V2, and V3 are the direct outputs of x-, y-, and z- sensors, B3 ¼ Bx cos a3 þ By cos b3 þ Bz cos c3
respectively, and A is the voltage-to-field transfer matrix consists of
The output voltages V 1 , V 2 ; and V 3 of TMR sensors are respec-
the elements aij with i = x, y, or z, indicating the x, y, or z-direction,
tively proportional to the components of B1 ; B2 and B3 :
and j=1, 2, or 3 is the number of TMR sensor. The matrix A is diag- 8
onal when the actual sensing directions are certainly along the x-, < V 1 ¼ k1  B1
>
y-, and z- directions. In general, the off-diagonal elements are non- V 2 ¼ k2  B2 ð11Þ
zero and the nine elements of the matrix A must be determined via >
:
V 3 ¼ k3  B3
a calibration process.
To calculate the elements of matrix A, the device was fixed at where k1 ; k2 ; and k3 are the effective sensitivities of the three sen-
the center of a three-axis Helmholtz coils. The three-axis coils gen- sors inside the flux chopper. In terms of matrix notation, the rela-
erate the orthogonal reference magnetic fields, which are Bx-ref, By- tion between the output voltage and the components of ~ B is:
ref, and Bz-ref, to define the x-, y-, and z- axes, respectively. The cor- 2 3 2 3
responding direct outputs of three TMR sensors constituted a
k1 cos a1 k1 cos b1 k1 cos c1 Bx
6 7 6 7
matrix Vcal with the elements of Vji, where j=1, 2, or 3 indicates V ¼ 4 k2 cos a2 k2 cos b2 k2 cos c2 5  4 By 5 ð12Þ
the sensor number and i = x, y or z represents the x-, y-, and z- k3 cos a3 k3 cos b3 k3 cos c3 Bz
axis of the reference fields. Hence, the matrix A is determined by
According to Eq. (1), the output voltage can be calculated using
solving the relation between the reference fields and the matrix
Vcal as follows: the components of ~ B and the field-to-voltage transfer matrix is
shown by
Bref ¼ A  Vcal ð4Þ 2 31 2 3
ax1 ax2 ax3 Bx
in which 1 6 7 6 7
2 3 V¼A  B ¼ 4 ay1 ay2 ay3 5  4 By 5 ð13Þ
Bx-ref 0 0 az1 az2 az3 Bz
6 7
Bref ¼4 0 By-ref 0 5 ð5Þ
1
0 0 Bz-ref where A is the inverse matrix of the A. In comparison between
Eqs. (12) and (13), it is easily seen that
is a diagonal matrix consisting of the three reference magnetic 2 31 2 3
fields, and ax1 ax2 ax3 k1 cos a1 k1 cos b1 k1 cos c1
6 7 6 7
2 3 A 1
¼ 4 ay1 ay2 ay3 5 ¼ 4 k2 cos a2 k2 cos b2 k2 cos c2 5
V 1x V 1y V 1z
6 7 az1 az2 az3 k3 cos a3 k3 cos b3 k3 cos c3
Vcal ¼ 4 V 2x V 2y V 2z 5 ð6Þ
ð14Þ
V 3x V 3y V 3z

is the matrix of three TMR outputs. From Eqs. (4)–(6), it is straight-

forward to show that the voltage-to-field transfer matrix is

A ¼ Bref  V1
cal ð7Þ

where V1cal is the inverse matrix of Vcal . Thus, the Cartesian compo-
nents of an unknown magnetic field can be retrieved from the direct
output of three TMR sensors using Eqs. (1)–(3).

2.3. Actual sensing directions of TMR sensors

The orientations of the actual sensing directions were calcu-

lated in this section. Suppose that there is an external magnetic
field ~
B; which is detected by the x-, y-, and z- sensors with the
actual sensing directions along the unit vectors ^s1 ; ^s2 ; ^s3 ; respec-
tively, as shown in Fig. 2. In the coordinate frame ^ x, y ^, and ^z are
defined by the Bref . The vector ~B is represented by its Cartesian Fig. 2. Spatial distribution of unit vectors ^s1 , ^s2 ; ^s3 of the actual sensing directions in
components as the following expression the Cartesian coordinates defined through a calibration process.
300 V.S. Luong et al. / Measurement 109 (2017) 297–303

The elements of the matrix A1 are given by different sensitivities between x- and y-sensors were resulted from
2 3 2 3 the non-identical intrinsic sensitivity and the spatial change in the
A11 A12 A13 k1 cos a1 k1 cos b1 k1 cos c1 chopping efficiency. Anyhow, to construct a 3D magnetometer, the
6 7 6 7
A1 ¼ 4 A21 A22 A23 5 ¼ 4 k2 cos a2 k2 cos b2 k2 cos c2 5 sensitivities of three sensors should be similar. The identical sensi-
A31 A32 A33 k3 cos a3 k3 cos b3 k3 cos c3 tivities of 17.5 V/T were achieved by applying the bias voltage of
1.85, 1.65, and 5.0 V for x-, y-, and z- sensor, respectively.
ð15Þ
Using the identity of cos2 ai þ cos2 bi þ cos2 ci ¼ 1; it is straight-
forward to show that
3.2. Noise characterization
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ki ¼ A2i1 þ A2i2 þ A2i3 ð16Þ
The field noise was obtained by measuring the noise spectral
p
Therefore, the direction cosine angles ai ; bi ; and ci of the unit density in Volts/ Hz. It was used to determine the field noise in
p
vector ^si can be determined using the i-th row of A1 by the follow- Tesla/ Hz by dividing to the sensitivity in V/T. Fig. 3 shows the
ing equation field noise spectrum of each sensor with and without chopper to
8 evaluate the effectiveness of the flux chopping technique. The
>
>
> ai ¼ cos1 Ai1 intrinsic noise levels of x-, y- and z- bare sensors were 2.8, 2.4,
p
< ki
and 2.6 nT/ Hz@1 Hz, respectively, and the trend of the intrinsic
Ai2
bi ¼ cos1 ð17Þ
>
>
ki noise indicated the behavior of the low-frequency noise. With a
>
: ci ¼ cos1 Ai3
ki
chopper, the average noise levels of x-, y-, and z-sensor reduced
p
to 0.17 nT/ Hz@1 Hz. The noise spectrums were almost flat from
The angles h12 ; h23 ; h31 between the actual sensing directions of 10 Hz down to 0.1 Hz in the ‘‘white” region. The 1/f corner fre-
TMR sensors can be calculated by taking the scalar product of each quency shifted to less than 0.1 Hz. According to the theory of the
pair of ^s1 ; ^s2 ; and ^s3 : thermal noise or Johnson-Nyquist noise [26,30–32], the thermal
noise is a frequency independent, so that thermal noise is indepen-
h12 ¼ cos1 ð^s1 :^s2 Þ ð18Þ
dent on the chopper. The theoretical thermal noise is defined by
the following equation [32]
h23 ¼ cos1 ð^s2 :^s3 Þ ð19Þ

h31 ¼ cos1 ð^s3 :^s1 Þ ð20Þ S2V ¼ 4  R  kB  T ð21Þ

According to the analysis so far, it can be concluded that the

effective sensitivities of TMR sensors inside the flux chopper can
1
be calculated from A with Eqs. (15) and (16). The actual sensing
directions of TMR sensors are also well defined by A1 in Eq. (17).
Using the field-to-voltage transfer matrix for calibration, a three-
axis magnetometer can be formed using the linear sensors with
its actual sensing directions not necessarily to coincide with the
^ ^, and ^z coordinates.
x, y
3. Results and discussions
3.1. Sensitivities of three TMR sensors
The sensitivities of three TMR sensors with and without the flux
chopper were measured in 0.1 Hz sweeping field and determined
via the slopes of a V-B curve. The results are presented in Table 1.
The intrinsic sensitivities of x-, y- and z-sensors were 164, 174 and
170 V/T, respectively. With the flux chopper, sensitivities were
reduced due to the chopping efficiency. The x- and y-sensors were
operated with a perpendicular flux chopping. Meanwhile, the
z-sensor was operated with a parallel flux chopping. Herein, a per-
pendicular chopping is defined that the sensing direction of TMR
sensor is aligned perpendicular to the chopper axis. Whereas the
sensing direction parallel to the chopper axis is called a parallel
chopper [20]. The lower sensitivity of z-sensor is caused by the
poorer chopping efficiency of the parallel chopping. Besides, the
where kB ¼ 1:3806  1023 J K1 is the Boltzmann constant, T is the
absolute temperature of the sensing elements, and R is the resis-
tance of the sensing elements. Herein, R of the SOP-TMR is about
p
90 kX, so that the estimated Johnson noise is about 27 nV/ Hz
using Eq. (21). The intrinsic sensitivity of the TMR sensor is
170 V/T, as presented in Table 1. Hence, the estimated field noise
p
is about 0.15 nT/ Hz that is in agreement with the experimental
p
value of 0.17 nT/ Hz. The noise measurement was recorded
within the limited bandwidth, and the measurement range is from
0.1 Hz to 10 Hz owing to the demodulated and low-pass filtered
outputs of the TMR sensors. The cut-off frequency of the LPF is mea-
sured as 30 Hz in this work. The 1-Hz field noise levels were sup-
pressed by a factor more than 17 with the flux chopper. At the
lower frequency of 0.1 Hz region, the noise was substantially
reduced by a factor of above 30.

Table 1
Sensitivities of three TMR sensors with and without a flux chopper (V/T).

Sensor Without chopper With chopper

3.56 Vbias 3.56 Vbias Adjusted Vbias
x 164 33 17.5 (@1.85 Vbias)
y 174 37 17.5 (@1.65 Vbias)
z 170 13 17.5 (@5.00 Vbias)
Fig. 3. Noise spectrum of three TMR sensors with and without a flux chopper.
 V.S. Luong et al. / Measurement 109 (2017) 297–303 301

3.3. The azimuth response of magnetometer before and after a By, and Bz is estimated as 35 mT. The polar plot of BxBy was a circle,
calibration and centered at the origin, as shown in Fig. 5(b). The results corre-
sponded to the azimuth response of Bx and By with a negligible off-
A calibration processes involving a voltage-to-field transfer set, identical amplitude, and a 90° different phase to each other. It
matrix, as described in Eqs. (1)–(7), was employed to solve the reveals that the three sensing axes of the calibrated 3D magne-
non-orthogonal problem. The x-, y- and z-sensing directions of tometer were orthogonal to each other.
magnetometer were set by the three reference fields of Bx-
ref = 24.4 mT, By-ref = 25.4 mT, and Bz-ref = 27.4 mT, respectively. The 3.4. Actual sensing directions of TMR sensors
voltage-to-field transfer matrix A, calculated using Eq. (7), was
found to be The actual sensing directions of three flux-chopped TMR sen-
2 3 sors could be retrieved using the Eqs. (8)–(22), as follow. The
54:61 4:649 0:29
6 7 inverse matrix of the voltage-to-field matrix A in Eq. (22) was
A ¼ 4 6:441 57:57 10:29 5; ð22Þ found to be
5:89 18:55 63 2 3
0:01811 0:00152 0:00016
6 7
where the elements of matrix A are in units mT/mV. However, the dc A1 ¼ 4 0:00246 0:01813 0:00297 5 ð26Þ
offsets of chopped TMR sensors were different from each other, so 0:00242 0:0052 0:01673
that an additional matrix Vdc-level was applied. The elements of
Vdc-level were obtained by recording the dc output of the flux where the matrix elements are in units of mV/mT. The effective sen-
chopped sensors in a shielding chamber. The corrected output sitivities of TMR sensors using Eq. (16) were
matrix Vcorr is 8
< k1 ¼ 0:0181
>
Vcorr ¼ V  Vdc-level ; ð23Þ k2 ¼ 0:0185 ð27Þ
>
:
where k3 ¼ 0:0176
2 3
1:394 in units of mV/mT. In comparison to the sensitivities in Table 1, the
6 7 calculated effective sensitivities were slightly higher. The difference
Vdc-level ¼ 4 0:404 5 ð24Þ
can be explained by the fact that the actual sensing axes deviate
1:07
from the ^ ^, and ^z coordinate. The direction cosine angles of the
x, y
is in units of mV. Thus, Eq. (1) is described by actual sensing axes, using Eq. (17), were given by
8 8 8
< a1 ¼ 4:8 > < a2 ¼ 85:2 < a3 ¼ 90:5
  
B ¼ A  Vcorr ð25Þ > >
b1 ¼ 82:4 ; b2 ¼ 168:0 ; and b3 ¼ 80:8
 
ð28Þ
To investigate the performance of the device as an electronic >
: > >
compass, the device was rotated in the Earth’s magnetic field about c1 ¼ 82:1 : c2 ¼ 107:1 :
c3 ¼ 18:9
its z-axis vertically upward with an interval of 10°. The Cartesian The angles between the actual sensing directions calculated
components of the geomagnetic field were retrieved using Eqs. using Eqs. (18)–(20) were
(22)–(25). Fig. 4(a) shows the uncalibrated magnetometer 8 
response. The observed azimuth responses of x- and y-sensors < h12 ¼ 87:18
>
were found to be sinusoidal. The remaining dc levels and the vibra- h23 ¼ 62:78 ð29Þ
tion in the amplitude of the z-sensor were caused by the misalign-
>
:
h31 ¼ 84:07
ment, which leads to detecting the vertical component of the
geomagnetic field. The actual sensing directions and misalignment The results in Eq. (29) indicate that the actual sensing directions
angles will be discussed in the Section 3.4. The polar plot of x- and of TMR sensors aligned on the PCB boards are non-orthogonal to
y-sensor outputs yielded an elliptical shape which was distorted each other and deviate from the Cartesian axes. Hence, a calibra-
from a circle, and the center was shifted from the origin. The dc off- tion process is certainly necessary to achieve an accurate three-
set results in the shift, while the errors in phase and amplitude dimensional magnetometer.
may contribute to the distortion, as shown in Fig. 4(b). The azimuth
responses of the calibrated magnetometer are shown in Fig. 5, in 3.5. Accuracy of the magnetometer
which the z-component of the geomagnetic field Bz was almost
constant. The azimuth responses of horizontal components Bx To evaluate the accuracy of the 3D magnetometer, the errors of
and By were sinusoidal, as shown in Fig. 5(a). The magnitude of the reconstructed azimuth angles were analyzed. The relations in
the geomagnetic field B calculated from Cartesian components Bx, Fig. 6(a) are the reconstructed angles before and after a calibration.

Fig. 4. (a) The azimuth response of the uncalibrated magnetometer rotation about the z-axis, and (b) polar plot the outputs of x-y sensors.
302 V.S. Luong et al. / Measurement 109 (2017) 297–303
Fig. 5. (a) The azimuth responses of the calibrated magnetometer rotation about the z-axis, and (b) the polar plot of Bx and By.
Fig. 6. (a) The reconstructed angles before and after a calibration, and (b) the angle errors of the magnetometer.
It was found that the angle error was suppressed substantially with
the calibration process. The nonlinearity was improved from 3.6%
down to 0.7%. The angle error was reduced from ±5° down to less
than ±1°, as shown in Fig. 6(b). These favorable results revealed
that the proposed calibration algorithm was useful for the 3D mag-
netometer comprising the TMR sensors and a flux chopper. It is
expected that this algorithm will be applied to various kinds of
three-dimensional magnetometer using linear magnetic sensors.
4. Conclusion
The novel concept of a low noise 3D TMR magnetometer using
the magnetic chopping technique and the digital driving system
was demonstrated. The performance of the magnetometer, includ-
ing the sensitivity, the field noise spectrum characterization, the
actual sensing direction, and the accuracy in geomagnetic detec-
tion, were presented and analyzed. The noise level was improved
by a factor of 17 with the proposed chopping technique. The field
p [12] P. Ripka, M. Janosek, Advances in magnetic field sensors, IEEE Sens. J. 10 (2010)
noise was found to be 0.17 nT/ Hz@1 Hz. The angle error of less
than 1° was achieved by the calibration algorithm using the
voltage-to-field transfer matrix. Future work to apply the proposed
technique to monolithic magnetoresistance sensors is promising to
realize an accurately low-noise three-dimensional TMR magne-
tometer for geomagnetic detection application.
Acknowledgment
This work was supported in part by the Ministry of Economic
Affairs of Taiwan under Grant No. 103-EC-17-A-01-S1-219 and in
part by the Ministry of Science and Technology of Taiwan under
Grant No. MOST 104-2221-E-151-011 and Grant No. MOST 105-
2221-E-027-031.
References
[1] J.E. Lenz, A review of magnetic sensors, Proceedings of the IEEE 78 (1990) 973–
989.
[2] M. Tondra, A. Jander, C.A. Nordman, J. Anderson, Z. Qian, D. Wang, Three-axis
magnetometers using spin-dependent tunneling: reduced size and power, in:
SPIE Proceedings, 2003, pp. 208–213.
[3] C. Schott, R. Racz, A. Manco, N. Simonne, CMOS single-chip electronic compass
with microcontroller, Solid-State Circuits, IEEE J. 42 (2007) 2923–2933.
[4] P. Mather, J. Slaughter, N. Rizzo, Three axis magnetic field sensor, ed: US Patent
8390283B2, Everspin Technologies, Inc., 2013.
[5] J.-T. Jeng, C.-Y. Chiang, C.-H. Chang, C.-C. Lu, Vector Magnetometer with Dual-
Bridge GMR Sensors, Magnetics, IEEE Transactions on 50 (2014) 1–4.
[6] C. Roumenin, K. Dimitrov, A. Ivanov, Integrated vector sensor and magnetic
compass using a novel 3D Hall structure, Sensors Actuators A: Phys. 92 (2001)
119–122.
[7] J. Bretschneider, A. Wilde, P. Schneider, H.P. Hohe, U. Koehler, Design of multi-
dimensional magnetic position sensor systems based on HallinOneÒ
technology, in: Industrial Electronics (ISIE), 2010 IEEE International
Symposium on, 2010, pp. 422–427.
[8] F.C.S.d. Silva, S.T. Halloran, L. Yuan, D.P. Pappas, A z-component
magnetoresistive sensor, Appl. Phys. Lett. 92 (2008) 142502–142502.3.
[9] P. Ripka, Review of fluxgate sensors, Sens. Actuators A: Phys. 33 (1992) 129–
141.
[10] P. Ripka, Fluxgate sensors: magnetic sensors and magnetometers, Artech
House, Inc, 2001.
[11] P. Ripka, Advances in fluxgate sensors, Sens. Actuators A: Phys. 106 (2003) 8–
14.
1108–1116.
[13] P. Ripka, K. Záveta, Chapter Three Magnetic Sensors: Principles and
Applications, in: K.H.J. Buschow (Ed.), Handbook of Magnetic Materials,
Elsevier, 2009, pp. 347–420.
[14] C.-Y. Chiang, J.-T. Jeng, B.-L. Lai, V.S. Luong, C.-C. Lu, Tri-axis magnetometer
with in-plane giant magnetoresistance sensors for compass application, J.
Appl. Phys. 117 (2015) 17A321.
[15] M. Tondra, A. Jander, C.A. Nordman, Low-power 3-axis magnetometers using
spin dependent tunneling for UGS and security applications, in: SPIE
Proceedings, 2002, pp. 188–196.
[16] STJ-3D Micro Magnetics, Inc. Available: http://www.micromagnetics.com/
product_page_stj3d.html (Accessed date: December 24, 2016)
[17] R. Guerrero, M. Pannetier-Lecoeur, C. Fermon, S. Cardoso, R. Ferreira, P.P.
Freitas, Low frequency noise in arrays of magnetic tunnel junctions connected
in series and parallel, J. Appl. Phys. 105 (2009) 113922.
[18] J. Almeida, P. Freitas, Field detection in MgO magnetic tunnel junctions with
superparamagnetic free layer and magnetic flux concentrators, J. Appl. Phys.
105 (2009).
[19] A. Jander, C. Nordman, A. Pohm, J. Anderson, Chopping techniques for low-
frequency nanotesla spin-dependent tunneling field sensors, J. Appl. Phys. 93
(2003) 8382–8384.
[20] V.-S. Luong, C.-H. Chang, J.-T. Jeng, C.-C. Lu, J.-H. Hsu, C.-R. Chang, Reduction of
low-frequency noise in tunneling-magnetoresistance sensors with a
modulated magnetic shielding, Magnetics, IEEE Trans. 50 (2014) 1–4.
 V.S. Luong et al. / Measurement 109 (2017) 297–303
[21] V.S. Luong, J.T. Jeng, J.H. Hsu, C.R. Chang, C.C. Lu, Tunneling-magnetoresistance
vector magnetometer with deflection flux-chopper, IEEE Trans. Magnetics 52
(2016) 1–4.
[22] C. Ren, X. Liu, B.D. Schrag, G. Xiao, Low-frequency magnetic noise in magnetic
tunnel junctions, Phys. Rev. B 69 (2004) 104405.
[23] S.H. Liou, R. Zhang, S.E. Russek, L. Yuan, S.T. Halloran, D.P. Pappas, Dependence
of noise in magnetic tunnel junction sensors on annealing field and
temperature, J. Appl. Phys. 103 (2008) 07E920.
[24] R.H. Koch, J. Clarke, W.M. Goubau, J.M. Martinis, C.M. Pegrum, D.J. Harlingen,
Flicker (1/f) noise in tunnel junction dc SQUIDS, J. Low Temperature Phys., 51,
pp. 207–224.
[25] J.H. Huijsing, B. Shahi, Chopper chopper-stabilized instrumentation and
operational amplifiers, ed: U.S. Patent No. 7,132,883. Washington, DC: U.S.
Patent and Trademark Office, 2006.
[26] Z.Q. Lei, G.J. Li, W.F. Egelhoff, P.T. Lai, P.W.T. Pong, Review of noise sources in
magnetic tunnel junction sensors, IEEE Trans. Magn. 47 (2011) 602–612.
303
[27] J. Hu, M. Pan, W. Tian, D. Chen, J. Zhao, F. Luo, 1/f noise suppression of giant
magnetoresistive sensors with vertical motion flux modulation, Appl. Phys.
Lett. 100 (2012) 244102.
[28] TMR2102D - Multidimension Technologies Inc. Available http://
www.dowaytech.com (Accessed date: April 26, 2017)
[29] MetglasÒ 2714A Magnetic Alloy, Metglas Inc. Available: http://www.
metglas.com/products/magnetic_materials/2714a.asp (Accessed date: April
26, 2017)
[30] J.B. Johnson, Thermal agitation of electricity in conductors, Nature 119 (1927)
50–51.
[31] H. Nyquist, Thermal agitation of electric charge in conductors, Phys. Rev. 32
(1928) 110–113.
[32] C. Fermon, M. Pannetier-Lecoeur, Noise in GMR and TMR sensors, in: Giant
Magnetoresistance (GMR) Sensors: From Basis to State-of-the-Art
Applications, Springer Berlin Heidelberg, Berlin, Heidelberg, 2013, pp. 47–70.