Sunteți pe pagina 1din 59

Transverse Phase Gradient Coil

Development for Low-Field TRASE MRI

Sashika S. Kumaragamage

Undergraduate Honours Thesis


Advisor: Dr. Chris Bidinosti

University of Winnipeg
Winnipeg, MB Canada
August 9, 2016
Abstract

Magnetic Resonance Imaging (MRI) has become a ubiquitous tool for diagnostic

medicine, and is known for delivering soft tissue contrast with minimal risk to patients.

In this project, the target field method is used to design radiofrequency phase gradi-

ent coils for use in TRansmit Array Spatial Encoding (TRASE), a recently developed

method for MRI [1]. Although most MRI is done at high magnetic fields, there are

also advantages for exploring the low field regime, especially due to the simplicity and

adaptability of experiments.

The target field method was used and the stream function was calculated using a

discrete wire approximation to determine the wire winding pattern in Mathematica.

The simulations were tested using the superposition of the Biot-Savart fields from many

small wire segments in free space. The design included numerous free parameters; it

was found that the winding pattern was heavily dependent on both the argument of

the trigonometric functions s (field “twist”) in the target field and the strength of the

apodization function.

After optimization of all parameters, a coil was designed which produced a phase

gradient transverse to its axis. A prototype of the coil was built on a Lucite acrylic

former and the resulting magnetic field was measured with a field sensor. Although

the measured phase gradient (6.6 ± 0.1◦ /cm) was not in agreement with the simula-

tion (5.93 ± 0.10◦ /cm) the experiment did show that the target field approach can be

used to design radiofrequency coils to close approximations. The measurements likely

suffered from possible alignment errors, causing the discrepancy in phase gradient mea-

surements. The field magnitude was similar between simulation and measurement, and

was uniform to within 10% within the region of interest. The phase gradient coil de-

signed and prototyped in this project can be used for a x phase gradient or a y phase

gradient, which is sufficient for 1D imaging. 2D imaging is possible with the inclusion

of a standard uniform coil.

i
Contents

1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.2 Optimize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.3 Build . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.4 Field Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 TRASE 7

3 Method for Coil Design 12


3.1 Target Field Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.2 Stream Function for Transverse Phase Gradient . . . . . . . . . . . . . . . . 14
3.3 Application to y Phase Gradient . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.3.1 Orders in m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.4 Biot-Savart Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

4 Design and Simulation Results 19


4.1 Effects of Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.1.1 Twist parameter s . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.1.2 Apodization strength N . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.1.3 Field distance d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.1.4 Number of wires n . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.1.5 Radius of target field c . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.1.6 Weighting of m0 term . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.2 ”Best” Coil Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

ii
5 Construction and Mapping of Prototype Coil 34
5.1 Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.2 Initial Field Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.3 Improved Field Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.4 Comparing Simulation to Measurement . . . . . . . . . . . . . . . . . . . . . 36

6 Discussion 39
6.1 Phase Gradients and Maxwell’s Equations . . . . . . . . . . . . . . . . . . . 39
6.2 Experimental Data and Error . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6.2.1 Alignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6.2.2 Background Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.3 Wire Placement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

7 Conclusion 44

A Summary of Stream Function from Target Field 48

iii
List of Figures

1 Phase gradient along y axis . . . . . . . . . . . . . . . . . . . . . . . . . . . 5


2 Phase gradient vector field . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3 One-Dimensional TRASE Example . . . . . . . . . . . . . . . . . . . . . . . 12
4 Twisted saddle for z phase gradient . . . . . . . . . . . . . . . . . . . . . . . 14
5 Apodization function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
6 Target field designs for varied s . . . . . . . . . . . . . . . . . . . . . . . . . 22
7 Target field designs for varied N . . . . . . . . . . . . . . . . . . . . . . . . . 23
8 Target field designs for varied d . . . . . . . . . . . . . . . . . . . . . . . . . 25
9 Target field designs for varied n . . . . . . . . . . . . . . . . . . . . . . . . . 27
10 Target field designs for varied m0 . . . . . . . . . . . . . . . . . . . . . . . . 30
11 Winding pattern for phase gradient coil . . . . . . . . . . . . . . . . . . . . . 32
12 Field maps in z=0 plane: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
13 The probe used for field measurement . . . . . . . . . . . . . . . . . . . . . . 35
14 Full setup for measuring magnetic field produced by phase gradient coil . . 36
15 The digital sensor used for field measurement . . . . . . . . . . . . . . . . . 37
16 Phase along y axis for simulation and prototype coil . . . . . . . . . . . . . . 38
17 Field magnitude along y axis for simulation and prototype coil . . . . . . . . 38
18 Adding phase offset in phase plot . . . . . . . . . . . . . . . . . . . . . . . . 43
19 Background field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

iv
1 Introduction

Magnetic resonance is a phenomena that is exploited to image the human body (magnetic
resonance imaging, or MRI) and chemical samples (nuclear magnetic resonance, or NMR).
MRI is typically performed within the bore of a powerful magnet, where the strong magnetic
field (denoted B0 ), conventionally along the ẑ direction, causes nuclei of hydrogen atoms in
a sample to align their spins parallel or antiparallel to B0 , producing a net magnetization
~ 0 = M0 ẑ [2]. In the classical description of NMR, a second magnetic field (B1 field) can then
M
be applied using a radiofrequency (RF) coil [3], which ultimately rotates the magnetization
away from B0 , and the component transverse to B0 will precess at the Larmor frequency ω0
related to the magnetic field strength in its local environment. This relationship is given by
the Larmor equation: [4]

ω0 = γB0 . (1)

For many methods in magnetic resonance, the first RF burst causes all magnetization to fall
to the transverse plane (the RF pulse is known as a 90◦ pulse, since it rotates the magneti-
zation vector by 90◦ ). The precessing transverse magnetization can be written as

~ =M
M ~ 0 e−iω0 t . (2)

As the spins precess, the magnetization vector rotates, creating a changing magnetic field.
By Faraday’s law of induction, the changing magnetic field would induce a voltage signal
which can be measured in a separate receiving coil. When the RF field is shut off, the mag-
netization returns to equilibrium along B0 and the signal in the pickup coil will fall to zero;
the time taken for this to occur characterizes the bonding environment of the nucleus.

If the magnetic field is made to spatially and linearly vary (conventionally by the use of

1
~ · ~r, the precession frequency ω
B0 gradient coils) so that the effective field is Beff = B0 + G
of nuclei will differ depending on position from the center of the coil ~r.

ω = γBeff (3)
~ · ~r)
= γ(B0 + G (4)

For simplicity, assume that the gradient is in the y direction, so that

ω = γ(B0 + Gy y) (5)

While the gradient is applied, some magnetization vectors would precess faster than others.
As time goes on, the nuclei precessing faster would be “ahead” of the nuclei which are
precessing slower. Mathematically speaking, nuclei would be gaining or losing phase φ with
respect to other nuclei at different locations:

φ = ωt

= γ(B0 + Gy y)t (6)

and the phase of an affected nucleus would differ from a nucleus unaffected by the gradient
by

∆φ = γGy yt. (7)

The magnetization vector precesses with time dependence and a spatially varying phase.

~ (y, t) = M
M ~ 0 e−i(ω0 t+∆φ(y)) (8)

2
If the detection is done at ω0 , then the detectable magnetization is [2]

~ (y) = M
M ~ 0 e−iγGy yt . (9)

The phase therefore contains information about position: this is called phase encoding.
(Some sources may refer to this as frequency encoding instead since the gradient was never
turned off, but the phase is really the quantity that changes over time, so this is the terminol-
ogy used here.) Working backwards from the phase of the signal picked up in the receiving
coil, the origin location of each signal could be determined if the magnetic field gradient
strengths and time are known. However, breakthroughs in MRI show that this conventional
method is not the only way to make use of magnetic resonance.

1.1 Motivation

TRansmit Array Spatial Encoding (TRASE) is a method recently developed by Sharp


and King for performing NMR and MRI [2], where the B0 gradient coils (which are costly
to maintain, power-hungry, loud, and space consuming) [5] are replaced with more compact
phase gradient coils. As opposed to varying the magnitude of the B0 field, TRASE uses
phase gradients to alter the phase of the RF field, which is an unconventional and novel
way of phase encoding. However, TRASE is only a different method for performing MRI;
therefore, many of the technologies developed for conventional MRI can still be utilized with
little to no modifications. The method has potential applications for magnetic resonance
where the quieter, simpler, and cheaper equipment can prove particularly advantageous;
some examples include a more accessible and affordable alternative for clinical-use MRI or
use in fundamental physics research on nuclear spin. This paper focuses on the design of
a radiofrequency phase gradient coil for use in TRASE, so that spatial information can be
encoded into phase data.

3
In this study we will use low fields, where RF waves oscillate at the KHz level. It is
known that low fields reduce signal strength and thus image clarity. However, these set-
backs are counterbalanced by the simplicity and convenience which come from the low field
approach. An entire MRI system can be built and modified easily and inexpensively, while
fitting comfortably on a lab bench.

Low-field experimentation of TRASE is desirable to explore physics principles governing


magnetic resonance, especially as the phase gradient coils can be designed based on direct
current (DC) approximations. This allows inverse methods of the Biot-Savart law to be
used to design the coil. At low field, the Biot-Savart law can be used to simulate the field
produced by the currents as the fields are near DC; quasi-static conditions were assumed.
Although multiple inverse methods exist, this paper will employ the relatively simple target
field approach. Many projects have focused on developing the theory for low frequency
coil design [3, 4, 6, 7, 8, 9], but there have been few publications on the optimization and
construction of an RF coil producing a phase gradient transverse to its axis (see Figure 1),
which is the goal of this project. Much of the design method depends on the stream function
[10], which provides insight on the required wire winding pattern for the coil.

4
Figure 1: This is the transverse phase gradient which will be studied in this paper. The
arrows represent the strength and direction (phase) of the magnetic field along the y-axis.
Note that the length of the arrows (magnitude of the vector) is constant across the cylinder.
The coil is shown with both the Cartesian (x, y, z) and cylindrical (ρ, φ, z) coordinate axes.
Figure from [3]
.

1.2 Outline

The majority of the theoretical groundwork for the design of a transverse phase gradient
coil has already been laid out in previous projects, and the method for developing a coil to
produce a required field is known. The primary goal of this work is to test the model, and
follows a few steps, shown below.

1.2.1 Design

The objective for the coil design was a phase gradient of approximately 5◦ /cm and a trans-
verse field magnitude difference of no more than 10% within a region of interest (ROI) in
the coil (dimensions of ±2 cm in x and y, and ±10 cm in z). A Mathematica script was

5
written to numerically evaluate the stream function. A contour plot was then produced, with
equally spaced contour lines clearly marked. Wrapping this contour plot around a cylinder
of radius a allowed visualization of the winding pattern required to produce the transverse
phase gradient field. The target field method is presented in section 3.1, with the stream
function discussed in section 3.2 and applied to a y phase gradient in section 3.3.

1.2.2 Optimize

There exist many parameters in the design of the coil, so it was necessary to have a method
for optimization. By treating the stream function contour lines like many miniature straight
wires carrying current, the magnetic field from each segment was calculated using the Biot-
Savart law (section 3.4). This simulation was independent of the design method, therefore
it served to check the validity of representing a surface current using a finite number of
discreet current paths. Maps of magnetic field phase and strength were used to identify
any potential issues with linearity of the phase gradient or magnitude inhomogeneity of the
transverse field. The design parameters were manually adjusted until a suitable field was
produced. The parameters and their effects will be discussed in section 4.1. The maps of
the finalized coil design are presented and explained in section 4.2.

1.2.3 Build

Using the optimized version of the design, the stream function contour plot was used to
build a prototype phase gradient coil. The aim of the project was only to ensure that the
design produced a reliable phase gradient; as such, the contour plot was simply printed on
paper which was wrapped around an acrylic former (cylinder around which wire is wound
to produce a coil). Section 5.1 discusses some flaws with this method, and some ideas for
future prototypes.

6
1.2.4 Field Map

In order to compare the simulated magnetic field and that which was measured, a map of
the magnetic field was produced by measuring the field inside the coil. This was done using
a field sensor capable of measuring the two transverse components of the field (Bx and By )
and gathering data at many points in the x = 0 plane. The measurement apparatus is
presented in section 5.2 and 5.3. These measurements were used to determine the transverse
field magnitude and phase. The field is compared to simulation in section 5.4.

2 TRASE

In 1979, Hoult [11] theorized an alternative method for magnetic resonance which would
utilize magnitude gradients in B1 instead of B0 , which could theoretically increase signal
to noise ratios (SNR). This method, known as rotating frame zeugmatography, performed
spatial encoding by utilizing spatially dependent flip angles rather than spatially dependent
static fields. However, rapid imaging required high RF field strengths, resulting in unsafe
levels of power deposition into the sample being imaged, making the method better suited
for spectroscopy than clinical imaging.

This paper will focus on a particularly promising recent method utilizing a gradient in
B1 phase, rather than magnitude, known as TRansmit Array Spatial Encoding (TRASE).
An array of phase gradient RF coils allows a novel method of k-space traversal, which is
required for image formation. Among the chief advantages of the method are acoustically
quiet gradients and reduced expense for electronic equipment [5].

The field produced by the special RF coils must be such that the phase φ (arbitrarily

7
assigned to be zero at the center of the coil) varies linearly, such that [1]

~ r) = Be
B(~ ~ iφ(~r) , (10)

where

~s
φ(~r) = · ~r (11)
a

~s
The phase gradient is written here as where ~s indicates the gradient strength and a is
a
the coil radius, and the reason for this notation will become apparent shortly. Imaging data
is acquired in the Fourier domain, and thus requires the traversal of k-space; it is useful to
consider the phase in k coordinates:

φ(~r) = 2π~kA · ~r. (12)

Equation (12) defines the phase gradient as a point in k-space known as the k-space origin
~kA ; any two coils with different phase gradients will have different k-space origins, and a

uniform coil (any coil that does not produce a phase gradient) will have a k-space origin
at zero (kAx = 0, kAy = 0). For a coil with its axis along z, a phase gradient in the same
direction can be produced from a fairly intuitive coil design, as the solution is a standard
uniform coil (saddle coil) which is twisted clockwise or conterclockwise along its axis. It is
of interest to consider a B1 phase gradient transverse to the axis of the coil: the required
coil geometry is not so evident for this configuration. Either the x or y direction can be
chosen for the phase gradient direction, but a 90◦ rotation interchanges between these coils,
so a successful design can be used for both orthogonal phase gradients. For this paper, the

8
y direction was arbitrarily chosen, so the phase could be simplified to

s
φA (y) = y = 2πkA y (13)
a

where the subscript is used to refer to the phase and k values corresponding to the y phase
gradient coil. It can now be seen that s is the linear proportionality constant between phase
φ and the spatial displacement y; it can be considered a measure of how much the magnetic
field twists along the y direction. Written this way, this “twist” parameter shows that the
gradient exists from −a to a. Then, Equation (10) can be written (using Euler’s formula) as

~ = x̂B cos s y + ŷB sin s y ,


   
B (14)
a a

and this ideal field can be visualized in Figure 2. The magnitude of B is constant for all y,
so all spatial information is completely contained within the phase. If the initial longitudinal
~ 0 = M0 ẑ, then a 90◦ pulse from the y phase gradient results
magnetization of the sample is M
in a detectable transverse magnetization of [2]

~ transverse = M
M ~ 0 eiφA . (15)

note the similarity to conventional phase encoding in Equation (9), since spatial information
is again contained within the phase, but here the phase is constant (in the conventional case,
it changes with t). This can equivalently be written using the k-space definitions of the
phase gradient coils:

~ 0 ei2πkA y .
=M (16)

Now consider the effect of a 180◦ pulse on the transverse magnetization; with any coil

9
Figure 2: The phase gradient in Equation (14) as a vector field, with B = 1. From the length
of the vectors, it is apparent that the magnitude is conserved across the entire region, with
only the phase changing. The x and y dimensions are scaled to the radius of the coil a.

the resulting magnetization vector Mf is

Mf = (Mtransverse )∗ e2iφB (17)

where the asterisk (∗) represents the complex conjugation operator. With normal coils, φB
would be zero as no phase gradient would exist, leading to the familiar flip across the z
axis. If a 180◦ pulse is applied using a coil with a phase gradient along x and its own phase
φB = 2π k~B · ~r and k-space origin kB , the new transverse magnetization Mf would be [1]

Mf = (Mtransverse )∗ e2iφ2 (18)

= M0 ei(−φ1 +2φ2 ) (19)

= M0 ei2π(−k1 +2k2 )·~r (20)

10
The behaviour in k-space can be found from the arguments of the exponential functions; the
general position of a set of nuclei would change according to the TRASE equation:

kf = −ki + 2kcoil (21)

where the i (f ), subscripts refer to the initial (final) k-space “locations”, and the k-space
origin of the coil in use. The arguments of the exponential functions represent spatially
dependent phase increments, or shifts in k-space. In theory, this single pulse is sufficient
to phase-encode all spatial data. However, in a manner somewhat reminiscent of Hoult’s
zeugmatography method, this approach is flawed since the phase gradients required for high
resolution imaging are unachievable for a volume coil [2]. Instead, successive 180◦ pulses are
used to gradually encode spatial data. With each pulse, the k-space vector shifts to a new
position, and the pulses are continued until all spatial information is encoded into the NMR
signal phase. An immediate issue with this method comes up in the second pulse, where
kf = −(−ki + 2kcoil ) + 2kcoil = ki : there was no traversal of k-space! In order to encode
information, TRASE therefore requires that the successive RF pulses be produced by coils
with different k-space origins (different gradient strengths or orientations).

By changing coils between each 180◦ pulse, a unique point in k-space is obtained with
every acquisition, and so all spatial points in the region of interest (ROI) can be encoded
into the phase. This can be seen in the one-dimensional example in Figure 3. Initially, an
“excitation” pulse (90◦ ) is used to tip the magnetization vector into the transverse plane.
After the first rephasing 180◦ pulse of RF, the k-space position would be at the k-space
origin of the first coil (coil A in the example). Following Equation (21), a second 180◦ pulse
from a second coil (B, which is a uniform coil in the example) would first cause a reflection
about the origin of the second coil (−ki ) and the addition of twice the k-space origin of the
coil (2kcoil ), which happens to be zero for the uniform coil. This process continues until all

11
spatial information is encoded into k-space (which is related to phase by Equation (12)).
While the conventional B0 gradient method utilizes a constantly evolving phase, TRASE
uses 180◦ pulses to evoke a shift in phase with every pulse.

Figure 3: a shows the pulse sequence used to phase encode in TRASE, while b shows
the point encoded in k-space during each numbered observation period. The X markings
represent the k-space origin of each coil. Since the B coil has an origin at k = 0, it is a
conventional uniform coil, while the A coil has an origin displaced from k = 0, and thus
must be a phase gradient coil. Figure from Sharp and King [2]

3 Method for Coil Design

The field required for a y phase gradient is known (Equation (14)), but the current pat-
tern required to produce this field is not intuitive. One method for field design is to use
superposition principles to sum known fields produced by well-understood coils (Helmholtz,
saddle, Maxwell. etc.), or to modify these coils to produce the required fields. Examples
include spiral birdcage coils for z phase gradients or perpendicularly aligned Maxwell and
Helmholtz coils for transverse gradients [4]. However, these resonator designs (including
capacitors in the wire paths to build resonant circuits) are not suitable at low field strengths

12
as the required capacitors are larger; making simple wire-wound coils much more convenient.

For the z direction, twisted saddle coils have been shown to produce a satisfactorily
linear phase gradient at low field (see Figure 4), but no simple modification of a common
coil is known to produce a transverse phase gradient. If the problem were reversed (where
the field is unknown but the current distribution is known), the field can easily be found
using the Biot-Savart law or its variations. Fortunately, inverse methods of the law do exist,
such as the time-harmonic inverse method [12] and boundary element method [13], which
are powerful iterative methods of computing the current densities from the magnetic field
[3]. The target field method used in this study, however, was chosen for its simplicity in
computation while delivering accurate results for the current density at low frequencies.

3.1 Target Field Method

The target field approach was originally proposed by Turner [14] as a fairly straightforward
technique for calculating currents from magnetic fields without using iterative computations.
It utilizes a relationship between the discrete Fourier components of the magnetic vector po-
tential and the components of a current density constrained on a cylindrical surface of radius
a and infinite length to yield powerful results: If the Fourier components of a desired field
are known at some radius c, then the Fourier components of the surface current needed to
generate that field can be solved. The target field radius c should be close to the the radius
at the ROI, but this is not always possible (see section 4.1).

The method has been successful for the manufacture of high quality uniform coils and
shields at DC [9, 14, 15]. Cobus [7] has explored some examples, presented an RF coil design
using the method, and also identified useful apodization functions (discussed in section 3.2)
to cut off the field at finite distances. Liu [3] expanded on these ideas to produce a number
of different RF phase gradient coil designs for TRASE; his results for a y phase gradients

13
were used at high field by Bellec [4]. The same results are optimized for low fields in this
paper. A brief summary of the derivation is provided in the appendix and a more complete
derivation can be found in [7], but only the main results from [3] will be presented here.

Figure 4: Although it may seem unrecognizable now, this coil is a standard uniform saddle
coil, which has been twisted along the z-axis to produce the same uniform field, but with a
z-dependent phase. Figure from Liu [3].

3.2 Stream Function for Transverse Phase Gradient

The initial paper on the target field method by Turner [9] specified a uniform field in the z
direction; Bidinosti et al [6] then expanded its usability by deriving expressions relating the
Fourier components of a general surface current on a cylindrical surface to the components
of its corresponding magnetic field. Although the target field method has many strengths,
the surface currents it predicts are not always simple to build; therefore, much of the utility
of the method comes from the use of the stream function: Brideson [10] determined that
the stream function, commonly used in fluid dynamics to visualize the flow of a velocity

14
vector field, could be used to represent current flow. Because the surface current F must
~ · F~ = 0) just as a fluid would, and is also constrained to
obey the continuity condition (∇
the non-radial directions, the current density can be described by the curl of a purely radial
function ψ [6]:

F~ = ∇
~ × (ψ ρ̂). (22)

ψ is known as the stream function. The target field method allows us to write the Fourier
components of F using the Fourier components of the desired magnetic field B. Summing
all these field terms, integrating, and performing an inverse Fourier transform yields the
stream function. The continuity condition constrains the components of F , so that only one
component (Fφ or Fz ) is required. The critical importance of ψ comes from the fact that its
contour lines will always follow the direction of F ; this is used to visualize current patterns.
Equally spaced contours of the stream function will show lines of equal current required to
produce the required B. This implies that any magnetic field specified on a cylinder of radius
c can be used to compute the winding pattern along a larger cylinder of radius a required to
produce that field.

3.3 Application to y Phase Gradient

The magnetic field required from the phase gradient was given in Equation (14). However,
this would require an infinitely long coil since the field exists at all z. This impracticality can
be avoided by modifying the field with an apodization (also known as tapering) function.
The purpose of the function is to restrict the spreading of the field along z by having a
functional value which quickly falls off to zero for large |z|. The modified field would be

~ z) = x̂ B s  B s 
B(y, cos y + ŷ sin y . (23)
1 + ( dz )2N a 1 + ( dz )2N a

15
1
where the prefactor is the apodization function determined by Cobus to be ap-
1 + ( dz )2N
propriate for use in target field coil design. The reasoning here was that it balances the
need to sharply cut off the magnitude of B and J for large z, and the need for the target
field Fourier transform integrals to remain convergent [7]. A graph of the functional value is
shown in Figure 5.

Figure 5: The apodization function 1+(Bz )2N plotted against z for the parameters N = 3 and
d
d = 1. When z ≈ 0, the functional value is 1, but this drops off rapidly when |z| > d. If d
were increased, the plateau would persist for greater z, and if N were increased, the fall-off
slopes would become steeper.

The parameter N controls the rate of fall-off for large z. In the final coil design, this
parameter was crucial as the length of the coil was limited by the dimensions of the available
former. In order to ensure that all wire windings were within the allowable coil dimensions,
N was increased a number of times. The field permeation parameter d controls the region
permeated by the field. Referring to Figure 5, it can be considered to determine the span
of the plateau in the center of the function. For coil design, the changes in field near the
edges of the coil alters the corresponding current paths, so this parameter was useful for ad-
justing wire placement without significant disruption of the target field. The effects of these

16
apodization parameters on coil design will be presented and discussed further in section 4.1.

Following the target field method (see Appendix A for derivation) Liu’s final result for
the transverse phase gradient coil was the stream function given in Equation (24) [3].

( ∞ ∞
−Bd X h  sc   sc i Z cos(kz)
ψ(φ, z) = cos(mφ) J1−m + J1+m dk 0 (kc)K 0 (ka)
µ0 aN m=1
a a 0 k 2 Im m
N  
X (2n−1)π (2n − 1) (2n − 1)π
· ekd sin 2N sin + kd cos
n=1
2N 2N

 sc  Z cos(kz) − 1
+ J1 dk
a 0 k 2 I00 (kc)K00 (ka)
N  
X
−kd sin
(2n−1)π(2n − 1)π (2n − 1)π
· e sin
2N + kbl cos
n=1
2N 2N
∞  sc i Z ∞
X cos(mφ) h  sc  cos(kz)
+c J1−m − J1+m + dk 0 (ka)
m=1
m a a 0 kIm (kc)Km
N  )
X (2n−1)π (2n − 1)π (2n − 1)π
· e−kd sin 2N sin + kbl cos (24)
n=1
2N 2N

Jm represents the m’th order Bessel function of the first kind, while Im and Km represent
the m’th order modified Bessel functions of the first and second kind, respectively. There
are three terms in Equation (24): The first comes from the ρ component of B, and the third
is derived from the φ component. These would normally be solved individually; Bellec [4]
showed that only one of the terms is necessary under ordinary conditions, so only the Bρ
term was used in this paper (Bφ was set to zero). The second term describes the 0th order
term in m, which has to be treated in a slightly different manner in the derivation. Each
order in m corresponds to a different kind of field (see section 3.3.1), though the higher
order terms play a minimal role for a linear phase gradient [4], allowing the higher order
terms to be omitted for calculations. Because the stream function is perfectly tangential to
the current and is equivalent to the integrated surface currents around the cylinder, equally
spaced contour lines of ψ reveal the surface current paths. Winding wires on these contour

17
lines approximates the necessary surface current to produce the initially specified target field
[6, 10].

3.3.1 Orders in m

In the treatment of the target field, m has served as an important index for the Fourier
components (notably for the order of the modified Bessel functions), and initially arose from
a Green’s function expansion of the magnetic vector potential. Different orders yield distinct
patterns in the field and coil shape. For example, m = 0 yields a pair of opposed infinite
solenoids whose field in the z direction cancel completely at the center, but have fringe fields
which contribute to a perfectly radial field in central axial slice of the coil. The m = 1 term
corresponds to a saddle coil, and produces a uniform field in a transverse direction. The
m = 2 term corresponds to a transverse magnitude gradient; though this may appear to be
of little use in a phase gradient, it can in fact cancel out any magnitude gradients resulting
from the other orders in m. The lower order terms contribute to the field more than higher
order terms; studies have found that all terms of m > 2 were negligible for phase gradient
coil design [4].

3.4 Biot-Savart Simulation

As mentioned, the Biot-Savart law can be used to calculate the magnetic field from any
straight portion of wire carrying a steady current. For curved current paths, the field can
be computed by splitting the wire into miniature discrete segments, such that the current
is approximately straight at the local level. The field from a such a filament is known [16],
~ as the number of segments is increased.
so the superposition of all fields approaches the true B

The main use for the Biot-Savart law in this project was for magnetic field simulations;
after the target field method was used to determine a winding pattern, the field produced
by the windings could be used to check for agreement with the desired field. The advan-

18
tage of this simulation method was that it was entirely independent of the coil design; the
simulations could thus offer a way to check the validity of the approximations made when
extracting finite contours of the stream function (sometimes known as discretization error).
The simulations were done using thousands of wire segments at DC. The calculated result
would not be an accurate representation of the field behaviour at high frequencies (> 100
MHz), as magnetostatics is inherently assumed in the calculation. The simulations were
used to determine each (Cartesian) component of the magnetic field at various points in
space, which in turn were used to calculate the phase and magnitude of the field. Using this
simulation, the design parameters could be adjusted to optimize the expected field without
having to build a prototype for each promising design.

4 Design and Simulation Results

The objective for the coil design is restated here as a phase gradient of approximately
5◦ /cm and a transverse field magnitude difference of no more than 10% within a 4×4×10cm
ROI. This gradient strength was suggested by King, who first published the method; rough
homogeneity is required because a uniform flip angle is necessary for imaging (only the
phase should vary with position) [1]; the ROI was chosen to support water sample sizes
which would likely be used for first attempts at imaging. The design was to be wound on
a coil with an outer radius of 4.75 inches (6.2 cm) and a length of 11.5 (30 cm) inches in
order to be utilized in the same apparatus as z phase gradient coils developed in the past.
The coil was designed using a target field numerical calculation and the winding pattern
was determined using equally spaced contours of the stream function, both done using a
program written in Mathematica. There were many adjustable parameters which were used
to alter and improve the design and field characteristics. It was found that the parameters
were not entirely independent, so many iterations of the target field were done using various
combinations of parameter values. This section will highlight the effects of each of these

19
parameters on coil design, and then present the combination of parameters which produced
the results which most closely matched ideal coil requirements.

4.1 Effects of Parameters

Unlike many mathematical problems, the transverse phase gradient solutions are not
unique. As such, many possible designs exist to produce very similar phase gradients, so an
“ideal” coil is difficult to pinpoint. This is made especially difficult because perfect phase
gradients and uniformity are physically not achievable (see section 6.1). A design was con-
sidered satisfactory if it met the gradient linearity and magnitude homogeneity requirements
outlined above. Due to the interdependence of the parameters, the process could not easily
be automated; the parameters were narrowed down through manual iterations. The purpose
of this section is to highlight the effect of each of the parameters on the final design. The
different parameters used and brief definitions are shown in Table 1 for reference.

Table 1: Design parameters and their definitions.

s Strength of phase gradient


N Strength of apodization
d Distance to which field permeates
n Number of wires
c Radius at which the field is specified
m0 Weighting of m = 0 term

4.1.1 Twist parameter s

The twist parameter ultimately determines the strength of the phase gradient. From Equa-
tion (23),

~ z) = x̂ B s B s
B(y, z 2N cos( y) + ŷ z 2N sin( y), (25)
1 + (d) a 1 + (d) a

20
it can be seen for example that s = π/2 would result in a complete 180◦ change in phase
from −a to a. For s = 0, one would expect no phase gradient at all, ie: a uniform field.
This is indeed the case, as the target field method yields precisely a sin φ saddle coil. As s
increases, the design differs from the saddle by having wires concentrate more on one side
of the coil. Beyond this, solenoidal rings begin to form at the end to strengthen the phase
gradient. The s parameter had the most impact in coil design, though gradient strength
could also be altered by restricting the region encompassed by the field (d) and changing
the expression of the m = 0 term. The effect of changing s can be seen in Figure 6. It can
be seen that there is seemingly no limit to how twisted the field can be made to be; the
target field method continues to allow the design of coils even for extreme phase gradients.
This provokes the question of why single excitation TRASE was deemed impractical. The
solution can be found when attempting to optimize a highly twisted field for homogeneity.
As the phase gradient increases, the magnitude uniformity of the field necessarily decreases
(see section 6.1). Biot-Savart simulation revealed that coils producing large phase gradients
were extremely inhomogeneous in magnitude. This makes smaller phase gradients and more
180◦ pulses a better option for TRASE applications.

4.1.2 Apodization strength N

The apodization function used in this target field was the G3 function from [7]. As mentioned
in section 3.2, the purpose of the function is to restrict the spreading of the field along z
by having a functional value which quickly falls off to zero (see Figure 5). The function is
controlled by two parameters, N and d. N controls the fall-off rate; a high value more closely
approximates the ideal boxcar function, but as discussed in [3] and [7], it can also lead to
divergent integrals due to the step function-like behaviour. Adjusting this parameter can
help contain the wire windings to within the coil specifications (total length < 11.5 inches).
The effect is more easily observed in the contour plot of the stream function than on the
coil, and can be seen in Figure 7.

21
(a) s=0 (b) s=π/16

(c) s = π/8 (d) s = π/7

(e) s = π/4

Figure 6: Target field designs for various s parameters. The lines are the contours of ψ, and
represent the surface current patterns required to produce the required magnetic field. All
other parameters are from Table 2.

22
(a) N = 1 (b) N = 2

(c) N = 3 (d) N = 4

Figure 7: Target field designs for various N parameters. The lines are the contours of ψ, and
represent surface current patterns. In 7a and 7b, the current paths exceed the z dimension
requirements for the coil, so the field had to be apodized more harshly in order to reduce
extension along the axis, and can be seen in 7c and 7d. Note that the inner windings, which
are closer to the ROI, remain almost unchanged. This suggests that the field at the center
changes very little when N is changed. All other parameters are from Table 2.

23
4.1.3 Field distance d

Unlike N , which altered the rate of apodization, the field permeation parameter d determined
how far the field could reach before being apodized. Although the parameter initially seems
to be of little consequence, it significantly impacted wire placement, and was especially useful
in eliminating small loops of wire (which are necessary to correct the magnetic field at a very
local level, but are very unwieldy in reality since magnet wire is not easily manipulated into
such fine shapes). Expanding the reach of the field under these circumstances allowed for
more constructible contour shapes, as the wires would be further apart.Such an effect can
be seen in Figure 8. Unlike N , this parameter can have impacts on the inner windings as
well a the outer windings. This makes it more versatile, but also more difficult to predict.

24
(b) d = 1.0 coil
(a) d = 1.0 contour

(c) d=1.2 contour (d) d = 1.2 coil

(e) d = 1.6 contour (f) d = 1.6 coil

Figure 8: Target field contours and coil shapes as the d parameter is changed. d has to be
sufficiently small for the apodization function to have any effect, but if it is too small as in
8b and 8a small current loops begin to form as a way of performing local field corrections.
Increasing d can eliminate these effects. All other parameters are from Table 2.

25
4.1.4 Number of wires n

For the current density to be perfectly simulated on the cylinder, an infinite number of
wire traces would be required. To make the modeling (and construction) practical, a finite
number were used instead, with equal spacings of the contour lines in ψ so that the same
current can be sent through each wire. The maximum (ψmax ) and minimum (ψmin ) values
of the stream function were found and the spacing was chosen according to [4]

(ψmax − ψmin )
ψj = ψmin + (j + 1) (26)
n

where j is an index denoting the contour. This is the only parameter which does not depend
particularly on the target field, but rather from the way information is extracted from the
stream function. Note that each pair of end rings will have the same value for the stream
function so specifying n sets the number of independent contours and not the total number
of wires. Increasing the number of wires would make the target field more accurate, but the
number of wires was restricted to n = 4 to simplify construction. The effect of changing n
can be seen in Figure 9.

4.1.5 Radius of target field c

For the target field calculation, a radius needs to be specified at which the field should
apply. This radius c should clearly be less than a, the radius of the coil, since the region of
interest exists within the coil. Beyond that, there are no mathematical restrictions for the
target radius. Bellec [4] had found that the best results come from the smallest value for c.
Practically, the target field should be specified close to the borders of the region of interest
(radius of 2 cm). However, it was found that numerical integration did not converge unless
c was sufficiently large. Cobus [7] had previously identified this condition to be

π
c > a − d sin . (27)
2N

26
(b) n = 3 coil
(a) n = 3 contour

(d) n = 4 coil
(c) n = 4.0 contour

(f) n = 5 coil
(e) n = 5 contour

Figure 9: Target field contours and coil shapes as the n parameter is changed. The number
of contours (current lines) increases with n. Adding more wires increases the accuracy of the
target field, but also makes winding the wire more difficult. All other parameters are from
Table 2.
27
This serves as direct evidence for the interrelationship between the program parameters.
Depending on the other parameters, this limit could be reached, but for the particular coil
parameters in use, c = 0.6 a was the lowest attainable value without requiring complicated
integration methods.

4.1.6 Weighting of m0 term

In the first theoretical treatment of transverse phase gradients, the stream function was
obtained by integrating the current density Fφ (obtained by using the magnetic field com-
ponents in (48) followed by inverse Fourier transform).

Z
ψ(φ, z) = Fφ (φ, z)dz (28)
∞  sc i Z ∞
−Bd X h  sc  cos(kz)
= cos(mφ) J1−m + J1+m dk 2 0 0 (ka)
µ0 aN m=0 a a 0 k Im (kc)Km
N  
X (2n−1)π
kd sin 2N (2n − 1)π (2n − 1)π
· e sin + kd cos +S
n=1
2N 2N

where S is an integration constant (with respect to z). The value of the constant is unknown
unless some boundary condition is found. To prevent a singularity in the m = 0 term, the
constant (see Appendix A) was chosen to be

−Bbl  sc  Z ∞ −1
S= 2J1 dk 2 0
µ0 aN a 0 k I0 (kc)K00 0(ka)
N  
X (2n−1)π
kd sin 2N (2n − 1)π (2n − 1)π
· e sin + kd cos . (29)
n=1
2N 2N

However, when re-deriving these formulas, an erroneous factor of two was found in the co-
efficient of this constant in [3]. Also, in the numerical calculation, a programming error was
found which omitted a factor of 4 when using the recurrence identities for the derivatives of
the modified Bessel functions; overall there was a factor of 2 missing from the calculation.
Somehow, the results of the simulation were more favourable (more linear phase gradient

28
and greater magnitude uniformity) when the mistake was left in place. This led to the un-
derstanding that the integration constant acts acts as a weighting factor for the m = 0 term,
and could be modified to suit the coil design.

Note that the integration constant does not directly affect this coefficient, but does change
the expression of the m = 0 characteristics in the final design. Modifying the coefficient was
the approach taken in this project to take into account the arbitrary value of the constant; it
is unknown whether this is the most effective method. The 0th order term in this expansion
represents the radial contributions to the field in the total stream function i.e the solenoidal
aspects of the design. To prevent an overall field along z, the solenoidal rings are arranged in
an anti-Helmholtz fashion, and the fringe fields from the finite solenoids is the source of the
radial field. As the m = 0 term is emphasized, the number of rings in the design increases,
and the phase gradient also strengthens within the region of interest. Curiously, the effects
of weighting of the m = 0 term resemble the effect of changing s. Since much of these results
are yet unexplained, this was only used as a method to slightly fine-tune the design. The
effect of the m0 weighting factor is shown in Figure 10. Note the immense similarity to
Figure 6, where s was being changed. This serves to confirm that the main contribution to
the target field comes from the m = 0 term.

29
(a) m0 = 0.5 contour (b) m0 = 0.8 coil

(c) m0 = 1.0 contour (d) m0 = 1.5 coil

(e) m0 = 2.0 contour (f) m0 = 4.0 coil

Figure 10: Target field contours and coil shapes as the m0 parameter is changed. Since the
term adds to the solenoidal component of the current, the appearance of end-ring structures
are heavily affected. All other parameters are from Table 2.

30
4.2 ”Best” Coil Design

With each coil design shown in section 4.1, the magnetic field was simulated using the
summed Biot-Savart fields from many small wire filaments. Using the values of the transverse
components of the field (Bx and By ), the magnitude of the transverse field (Bx2 + By2 ) and the
By
phase (arctan( Bx
)) were calculated, and were used to check the phase gradient linearity and
magnitude homogeneity. Eventually, a coil design with sufficient linearity and homogeneity
was found. The parameters used for this design are summarized in Table 2 and the coil
design is shown in Figure 11. The field maps of phase and magnitude can be found in Figure
12. In the Biot-Savart simulation, the phase gradient was found to be 5.93 ± 0.06◦ /cm along
y, and the magnitude was uniform to within 8% in a region ±2cm in y and ±5cm in z.

Table 2: Parameters used to produce coil with the most suitable phase gradient.

Parameter Definition Value


s Phase gradient strength π/7
n Number of wires 5
c Radius for target field 0.6a
N Apodization strength 3
d Field spread distance 1.6a
m0 Weighting of m0 term 1.15

31
Figure 11: Winding pattern required to produce the target field. Many possibilities are
available, but this was chosen due to high linearity in target field and high homogeneity
within region of interest. (left) shows the contour plot of the stream function, with the
cylindrical coordinates φ (degrees) plotted against z (m) with the contour lines showing the
wire paths. (right) shows this contour plot wrapped around a cylinder of radius a = 6.2cm

32
Magnitude of B in z=0 Plane Phase of B in z=0 Plane
0.03 0.03

0.02 0.02

0.01 0.01
x (m)

x (m)

0.00 0.00

-0.01 -0.01

-0.02 -0.02

-0.03 -0.03
-0.03 -0.02 -0.01 0.00 0.01 0.02 0.03 -0.03 -0.02 -0.01 0.00 0.01 0.02 0.03
y (m) y (m)

Figure 12: Contour lines show 1% changes in the magnitude of the transverse field as a
function of position (left).Within the region of interest (±2cm) there is less than 8% change
in magnitude). Contour lines show 10◦ changes in the phase (right). The phase gradient
cannot be determined quantitatively from this map, but the linearity is quite evident from
the straightness of the contour lines within the ROI.

33
5 Construction and Mapping of Prototype Coil

After winding copper wire along the contours shown in Figure 11 on paper, the contour was
wrapped around an acrylic cylinder (with the same radius a that the coil was designed for).
The magnetic field was measured using a field sensor, but initial measurements had a high
degree of error associated with them, so finer measurements were done using a more precise
apparatus.

5.1 Construction

The four quadrants of the stream function contour plot were printed on four separate sheets
of paper, and were taped together to fit around the former. Magnet wire was manually
wound along the lines and held in place with tape. However, there were still some sharp
corners which the wire needed to turn, especially when crossing from one contour line to
another. This could not be done perfectly, so some small current loops were accidentally
created which were not part of the original design. It was in anticipation of these difficulties
that the number of contours n was made to be small. The coil could only be effectively
wound if the contour lines were sufficiently spaced apart, which is why n = 4 was used as
opposed to larger values. The currents going into the coil and the currents coming out were
wound in a twisted pair when possible to prevent outside interference in field measurements,
or they were simply placed beside one another. In future designs, grooves can be cut into
the cylinder, allowing many more wires to be used to approximate the current density; this
will lead directly to more accurate simulated and experimental results.

5.2 Initial Field Measurements

A field probe was made using square loops of wire (Figure 13). After calibrating the probe,
the field from the phase gradient coil was mapped near the region of interest using the grid
shown in Figure 14. The probe rested on a piece of wood such that the center of the probe was

34
at the center of the coil, and a grid was drawn so that the probe could accurately be moved
to pre-determined positions within the coil. The coil was powered with 20 kHz RF using a
lock-in amplifier, and the receive probe was connected to the same amplifier to obtain a low-
noise measurement. The strength of the phase gradient was measured to be 6.6 ± 0.6◦ /cm,
which agrees with the expected value from the simulation. The parabolic variation in field
magnitude which was noticed in the simulation was not observed in experimental data, as any
such fine detail was obscured by the error bars. The data and associated error is discussed
in section 6.2. Since the uncertainty in measurement appeared quite large, some sources of
error were tackled, and another attempt was made to achieve finer measurements.

Figure 13: Field measurement probe with orthogonal square coils visible (left). One set will
measure Bx and the other will measure By . This grid was drawn on wood and was used to
accurately position and align the probe (right). The marked x was used to aid the eye with
positioning.

5.3 Improved Field Measurements

Due to the large error bars in the initial measurement, the experiment was redone. Some
major sources of error in the experiment were avoided, allowing more precise results. A
notable difference was the use of a digital field sensor instead of a home-built one. The
HMC5883L 3-axis magnetic field sensor chip was capable of measuring with a resolution of
200 nT with a high degree of orthogonality. It was powered by an Arduino microcontroller
and the coil was run using a DC power supply as the chip was not designed for RF field

35
Figure 14: Full setup for measuring magnetic field produced by phase gradient coil

measurements. The mounting setup is shown in Figure 15. The rest of the apparatus was not
changed significantly, but extra materials were added to prevent movement of the apparatus
while measurements were being taken and a block of wood was used to prevent the probe
from tipping while measuring values near the edges of the ROI. In addition, the apparatus
was nailed together to improve potential misalignments which were common in the previous
iteration of the apparatus. The chip was calibrated in a uniform field to convert its ADC
measurements into µT, and to check the orthogonality of its sensors. Data was collected for
the entire ROI, but only the field along the y-axis is presented in this paper.

5.4 Comparing Simulation to Measurement

The field from the phase gradient design was simulated using Biot-Savart methods, and
the field from the coil was measured using field sensors. Phase was calculated with φ =
By p
arctan and the magnitude with |B| = Bx2 + By2 . For a quantitative measure of the
Bx
phase gradient, the phase can be plotted against y, and a linear fit can be used to determine

36
Figure 15: The field sensor chip was mounted onto the same acrylic piece used for initial
measurements, so that the relative heights of the coil and grid need not be adjusted.

the gradient. The gradient strengths were found to be similar within the ROI, but only the
case of x = 0, z = 0 was analyzed for this report. The comparison between phase gradient
simulation and measurement is shown in Figure 16 and a similar comparison of magnitude
is shown in Figure 17. The phase gradient from the target field model simulation was found
to be 5.93 ± 0.06◦ /cm and the error was found using the standard deviation of gradient
strengths for different x values, added in quadrature with the fit error. Measurements shown
a phase gradient strength of 6.6 ± 0.1 ◦ /cm, which is not in agreement with the simulation.
The measurements used were the “improved” measurements from the HMC5883L sensor,
and the discrepancy between expected and measured gradient strength is discussed further
in section 6.2.

37
Phase of B along y-axis
30

Magnetic Field Phase (Degrees)


20

10

-10

-20

-30
-0.03 -0.02 -0.01 0.00 0.01 0.02 0.03
y (m)

Figure 16: Plot showing the phase gradient determined from the simulation (red line) and
field measurements (points with error bars). A global offset was added to the phase of the
measured values (this is discussed in section 6.2.

Magnitude of B along y-axis

1.04
Normalized Field Magnitude

1.02

1.00

0.98

0.96

-0.03 -0.02 -0.01 0.00 0.01 0.02 0.03


y (m)

Figure 17: Plot of transverse field magnitude along y-axis for both the simulated coil (red
line) and built prototype (points with error bars).

38
6 Discussion

The results of the simulation show promise for the target field method in designing coils
producing reliable phase gradients, and the data collected from the prototype demonstrate
that the these theoretical results are obtainable in reality. However, there are some areas for
improvement in this project, which may allow for more effective designs, and more reliable
field measurements.

6.1 Phase Gradients and Maxwell’s Equations

In order for the phase gradient design to be ideal, it must produce a magnetic field with a
perfectly linear phase gradient while also being perfectly uniform. If we consider the phase
gradient from Equation (14), then we see that the magnitude is

r
q sy sy
Bx + By = B sin2 ( ) + cos2 ( ) = B,
2 2 (30)
a a

and the phase is

 
By sy sy
arctan = arctan(tan )= . (31)
Bx a a

This satisfies both requirements. Unfortunately, such an ideal field is not only difficult to
produce, but in fact impossible. In the low frequency regime, electromagnetism obeys the
~ ×B
quasi-static conditions: Maxwell’s equations state that ∇ ~ = 0 and as always, ∇
~ ·B
~ = 0.

Writing these out explicitly,

∂Bx ∂By ∂Bz


+ + =0 (32)
∂x ∂y ∂z
(33)

39
and

∂Bx ∂By
= (34)
∂y ∂x
∂Bz ∂Bx
= (35)
∂x ∂z
∂By ∂Bz
= (36)
∂z ∂y

However, if we consider the apodized target field

~ z) = x̂ B s  B s 
B(y, cos y + ŷ sin y (37)
1 + ( dz )2N a 1 + ( dz )2N a

we see that none of these conditions are satisfied. Clearly, this field is simply not an
allowable solution to Maxwell’s equations, so the field must be modified to satisfy these
conditions. The only way to satisfy Equations (35) and (36) is if there existed some Bz which
was a function of x and y: Bz = Bz (x, y). This already shows that total field homogeneity
is lost if a phase gradient exists. The only way to satisfy both Equation (31) and Equation
(34) is if

*0


s  sy  ∂B  sy
  sy  ∂B
−B sin + cos
 = sin (38)
a a ∂y
 a a ∂x


the cancellation occurs due to the orthogonality of trigonometric functions

s ∂B
B = (39)
a ∂x

which has the solution

 sx 
B = A exp + C(y). (40)
a

40
Then our target field equation becomes

sx sx
 
~ A exp a
+ C(y). s  A exp a
+ C(y). s 
B(x, y, z) = x̂ cos y + ŷ sin y + Bz (x, y)ẑ
1 + ( dz )2N a 1 + ( dz )2N a
(41)

This shows that the magnitude will not remain constant in the transverse plane either.
Maxwell’s equations force a compromise between gradient linearity and magnitude unifor-
mity. This is the inviolable limit for phase gradient coil design.

6.2 Experimental Data and Error

The error bars in the initial field measurements were fairly large, in comparison to the spread
in data. As the measurements were only the results of the first transverse phase gradient of
this kind, little effort was put into minimizing this error; as long as the phase gradient and
magnitude uniformity was apparent, the measurements fulfill their purpose. Nevertheless,
this section will highlight the main sources of uncertainty which were eliminated and the
ones which remained, so that the precision of any follow-up attempts at field-mapping may
be improved.

6.2.1 Alignment

Alignment was a large cause for uncertainty in initial measurements. This was primarily
due to the experimental setup, where none of the individual components (coil, grid, probe)
were locked in place with respect to one another. It was vital that the probe be oriented
such that the transverse field sensors were in the plane of x and y respectively. This only
affects phase, since any offset in one component is made up for by the other when taking the
magnitude Bx2 + By2 . For future mapping, a method needs to be devised to ensure that the
coil is oriented with its y axis pointing directly upwards, and the grid must be perfectly along
the axis of the coil. In the experiment, two planks of wood were used to level the grid, but a

41
balance indicated a small mismatch in plank height. Also, it must be ensured that the probe
points directly along z, to avoid any Bz (from section 6.1) creeping into the transverse field
measurement. This was done in the experiment by lining up the sides of the probe along the
grid lines, so any error in the drawing of the grid lines carries into the measurement this way
as well. In the second iteration of the experiment, the only parts free to move was the entire
wooden apparatus (instead of its individual pieces, and the coil (rotation or translation was
possible). These were all adjusted by eye during the measurement.

Misalignment was almost certainly the reason for the offset in phase measurements (the
gradient does not pass through the origin as expected). The measurements were corrected
by adding a global phase of 3.5◦ (both shown in Figure 18). It should be noted that the
phase offset does not make any difference for imaging purposes, as section 2 shows that it
the phase gradient which impacts encoding, not the phase itself. However, this offset is a
good tool for error-checking. It is possible that the grid was spatially offset from the center
(in y), or that the coil was rotated slightly so that the area being mapped was not in fact
the x = 0 plane but a in a plane tilted about z; however, this is unlikely since this could
only reduce the measured gradient, because the largest phase gradient should be along y.
The more likely issue would be some horizontal misalignment between the grid and the coil,
causing the introduction of Bz components to be measured. Another strong candidate for
the source of the large phase gradient would be the extrapolation used to subtract away
background field.

An estimate of the error from misalignment was obtained by purposefully slightly mis-
aligning the probe and measuring the field values; the difference with the original measure-
ment was taken to be the error due to misalignment. This was added in quadrature with
other estimates of uncertainty to determine the total uncertainty.

42
Phase of B along y-axis Phase of B along y-axis
30 30

Magnetic Field Phase (Degrees)

Magnetic Field Phase (Degrees)


20 20

10 10

0 0

-10 -10

-20 -20

-30 -30
-0.03 -0.02 -0.01 0.00 0.01 0.02 0.03 -0.03 -0.02 -0.01 0.00 0.01 0.02 0.03
y (m) y (m)

Figure 18: Plots of phase along the y direction for the prototype coil (dots with error bars)
compared with simulation (line). (left) shows the original measurements, while (right) shows
the same measurement with a global offset of 3.5◦ .

6.2.2 Background Field

It is quite well known that the rooms in the second floor of Centennial Hall at the University
of Winnipeg are exposed to many changes in magnetic field over the course of a work-day,
so mapping was done at nighttime on a Saturday. Although the stray background field was
not accurately mapped, a somewhat rough measure was obtained using the x, y, z = 0 point
and 25 points around it in the x = 0 plane. The measurements of Bx and By can be seen in
Figure 19. This data was extrapolated to roughly represent linear functions, and subtracted
away from the field measurements to determine the field produced by the coil alone. The
current through the coil was maximized, which minimizes the effect of stray fields, to further
prevent any effects from the background field. Measurements of Bz were not used. For future
field-mapping experiments, all mapping should be done by taking separate measurements
with the current on and off, obtaining a background field measurement at every point at
the same time when experimental data is being collected. This eliminates any extrapolation
error and prevents any effects of time dependent background fields.

43
X component of background B along y-axis (m) Y component of background B along y-axis (m)
-49.1
-49.2 21.5
-49.3
-49.4 21.0
Bx (μT)

By (μT)
-49.5
-49.6 20.5
-49.7
-49.8 20.0
-49.9
-0.02 -0.01 0.00 0.01 0.02 -0.02 -0.01 0.00 0.01 0.02
Position along y-axis (m) Position along y-axis (m)

Figure 19: The background field was measured in the ROI, although the spacing was more
coarse to save time. The Bx component (left) and By component both showed essentially
constant fields with small linear dependence on y. This was subtracted from the data col-
lected when the phase gradient was running so that the true field from the coil could be
determined.

6.3 Wire Placement

Although a significant effort was made to ensure that the wire windings were at the correct
places when building the prototype, the windings were still done on paper, which could have
been misaligned with respect to the coil. For a first protoype, this method was fast and
effective, but later renditions should use a more precise method for wire placements around
the cylinder. A proposed idea is to laser-cut grooves into the cylinder, and to simply insert
wire into the grooves. The number of wires n would no longer be a limiting parameter, and
could be increased to produce a more favourable field as well.

7 Conclusion

A phase gradient coil was designed to produce a transverse phase gradient, using the tar-
get field method and the stream function approximation. In order to test each design,
Biot-Savart (magnetostatics assumption was made here) methods were used to simulate the
magnetic field inside the region of interest, and the transverse phase gradient and field mag-
nitude were calculated. It was found that the design and field relied heavily on the field twist

44
(s) and apodization parameters (N, d), and partially relied on four other parameters as well;
these were iteratively altered until a coil with a satisfactory phase gradient strength and
linearity, and relatively uniform magnitude was found. A prototype was built on a cylinder
and tested at DC. The measured phase gradient (6.6 ± 0.1◦ /cm) fairly similar to the simula-
tion (5.93 ± 0.10◦ /cm), and expressed the same parabolic characteristics in magnitude. The
uncertainty in the gradient was likely due to misalignment between the coil and the field
sensor, so a more permanent measurement apparatus should be constructed for more precise
results. However, the work presented here does confirm that the target field approach and
stream function approximation are entirely valid for use in phase gradient design.

45
References

[1] J. C. Sharp et al., “High-resolution mri encoding using radiofrequency phase gradients,”
NMR in Biomedicine, 2013, doi: 10.1002/nbm.3023.

[2] J. C. Sharp and S. B. King, “Mri using radiofrequency magnetic field phase gradients,”
Magnetic Resonance in Medicine, vol. 63, pp. 151–161, 2010.

[3] C.-Y. Liu, “The target field approach for phase gradient radio frequency coil design,”
Undergraduate Honours Thesis, University of Winnipeg, 2010.

[4] J. Bellec, “A target field based design of a phase gradient,” Master’s thesis, University
of Manitoba, 2015.

[5] J. C. Sharp and S. B. King, “A cheaper, quieter, mri machine,” Nature, vol. 501, pp.
138–139, 2013, doi: 10.1038/501138e.

[6] C. P. Bidinosti et al., “Active shielding of cylindrical saddle-shaped coils: Application


to wire-wound rf coils for very low field nmr and mri,” Journal of Magnetic Resonance,
vol. 177, pp. 31–43, 2005.

[7] L. Cobus, “The target field approach for radio frequency coil design,” Undergraduate
Honours Thesis, University of Winnipeg, 2008.

[8] Q. Deng et al., “B1 transmit phase gradient coil for single-axis trase rf encoding,”
Magnetic Resonance Imaging, vol. 31, pp. 891–899, 2013.

[9] R. Turner, “A target field approach to optimal coil design,” Journal of Physics D.,
vol. 19, pp. L147–L151, 1986.

[10] M. Brideson and L. Forbes, “Determining complicated winding patterns for shim coils
using stream functions and the target-field method,” Concepts in Magnetic Resonance,
vol. 14, pp. 9–18, 2002.

46
[11] D. I. Hoult, “Rotating frame zeugmatography,” Journal of Magnetic Resonance, vol. 33,
pp. 183–197, 1979.

[12] B. Lawrence et al., “A time-harmonic inverse method- ology for the design of rf coils in
mri,” IEEE Transactions in Biomedical Engineering, vol. 49, pp. 64–71, 2002.

[13] M. Poole and R. Bowtell, “Novel gradient coils designed using a boundary element
method,” Concepts in Magnetic Resonance, vol. 31, pp. 162–175, 2007.

[14] R. Turner and R. M. Bowley, “Passive screening of switched magnetic field gradients,”
Journal of Physics E., vol. 19, 1986.

[15] F. Qi et al., “A new target field method for optimizing longitudinal gradient coils’
property,” Piers Online, vol. 3, pp. 865–869, 2007.

[16] J. D. Hanson and S. P. Hirshman, “Compact expressions for the biot-savart fields of a
filamentary segment,” Physics of Plasmas, vol. 9, no. 10, pp. 4410–4412, 2002.

[17] W. R. Byers et al., “Correction of concomitant gradient artifacts in experimental mi-


crotesla mri,” Journal of Magnetic Resonance, vol. 177, pp. 274–284, 2005.

[18] R. Ullah and M. S. Anwar, “Undesired gradients in low-field magnetic resonance imag-
ing,” Concepts in Magnetic Resonance, vol. 34, pp. 173–190, 2009.

[19] Y. Zheng et al., “Very-low-field mri of laser polarized xenon-129,”


Journal of Magnetic Resonance, vol. 249, pp. 108–117, 2014, url:
http://dx.doi.org/10.1016/j.jmr.2014.09.024.

[20] C. Bidinosti et al., “In-vivo nmr of hyperpolarized 3he in the human lung at very low
magnetic fields,” Journal of Magnetic Resonance, vol. 162, pp. 122–132, 2003, preprint.

47
A Summary of Stream Function from Target Field

The general form for the calculation of the stream function will be presented very briefly
here. Refer to [10], [4], or [7] for a more complete derivation. Cylindrical coordinates will
be used since it is more natural for this problem. For any magnetic field, there exists a
~ such that ∇
vector potential A ~ ×A
~ = B.
~ Since A
~ typically points in the direction of the
~ and the current must be constrained to the cylinder, it is only necessary
current density J,
to consider the φ and z components:

Jφ (~r) sin(φ − φ0 )
Z
µ0
Aφ = dτ 0 (42)
4π |~r − ~r 0 |
Z
µ0 Jz (~r)
Az = dτ 0 (43)
4π |~r − ~r 0 |

where the dτ 0 is an incremental volume in the cylinder, the primed coordinates refer to
points on the cylinder, and Jz , Jφ refer to the components of current density along the z
and φ directions. The key step is in rewriting the denominator using a Green’s function
expansion, such that


1 X im(φ−φ0 ) ∞
Z
1 0

0
= e dkeik(z−z ) Im (kρ< )Km (kρ> ), (44)
|~r − ~r | π m=−∞ −∞

where Im and Km represent the m’th order terms of the modified Bessel functions of the 1st
and 2nd kinds, and the <, > superscripts denote the smaller or larger between ρ, ρ0 . Also,
since the current is constrained onto a cylinder of radius a, the current density exists as
a surface current F existent only at a, and can be written using a Dirac δ function. The
current density components can then be written as

Jφ (~r 0 ) = δ(ρ0 − a)Fφ (φ0 , z 0 ) (45)

Jz (~r 0 ) = δ(ρ0 − a)Fz (φ0 , z 0 ) (46)

48
After these simplifications and discrete Fourier decomposition, it can be shown that Equa-
tions (42) and (43) can directly relate the Fourier components of the surface current density
with the Fourier components of the magnetic field B specified at the radius c [3].

im
Fzm = 0 (kc)K 0 (ka)
Bρm (k)
µ0 a2 k 2 Im m
c
= 0 (ka)
Bφm (k) (47)
µ0 a2 kIm (kc)Km
im
= 0 (ka)
Bzm (k)
µ0 a2 k 2 Im (kc)Km
−i
Fφm = 0 0
Bρm (k)
µ0 akIm (kc)Km (ka)
−c
= 0
Bφm (k) (48)
µ0 amIm (kc)Km (ka)
−i
= 0
Bzm (k)
µ0 akIm (kc)Km (ka)

where the primes on the Bessel functions indicate derivatives with respect to their arguments.

The stream function could be used to represent current flow. Because the surface current
~ · F~ = 0) and is also constrained to the non-radial
must obey the continuity condition (∇
directions, the current density can be described by the curl of a purely radial function ψ:

F~ = ∇
~ × (ψ ρ̂). (49)

ψ is known as the stream function. This yields

∂φ
Fz = − (50)
a∂ψ
∂ψ
Fφ = , (51)
∂z

The utility of ψ comes from the fact that its contour lines will always follow the direction of
F ; equally spaced contour lines will show current flow. From Equations (47) and (48), the

49
Fourier components of F can be written in terms of the Fourier components of B. Summing
all the terms and integrating, then, can directly yield the stream function. Applying all this
to our target field

~ z) = x̂ B s B s
B(y, z 2N cos( y) + ŷ z 2N sin( y) (52)
1 + (d) a 1 + (d) a

or in cylindrical coordinates

~ B sρ B sρ
B(φ, z) = ρ̂ z 2N cos(φ − sin φ) + φ̂ z 2N sin(φ − sin φ) (53)
1 + (d) a 1 + (d) a

The Fourier transform of this field at ρ = c is

Z π Z ∞
1 −imφ
Bρm (k) = dφ e dz e−ikz Bρ (φ, z)
2π −π −∞
Z π Z ∞
1 −imφ B sc
= dφ e dz e−ikz z 2N cos(φ − sin φ)
2π −π −∞ 1 + (d) a
Z π Z ∞
B sc cos(kz)
= 2 dφ cos(mφ) cos(φ − sin φ)2 dz
2π 0 a 0 1 + ( dz )2N
Z π h
B sc sc i
= dφ cos((1 − m)φ − sin φ) + cos((1 + m)φ − sin φ)
2π 0 a a
Z ∞
cos(|k|z)
· 2(d)2N dz 2N .
0 (d) + z 2N

For the y phase gradient specified at a radius c, the relationship between the current
density and magnetic field were derived by Liu: [3]

N  
Bblπ h  sc   sc i X (2n−1)π
−|k|d sin 2N (2n − 1)π (2n − 1)π
Bρm (k) = J1−m + J1+m e sin + |k|d cos
2N a a n=1
2N 2N

N  
iBblπ h  sc   sc i X (2n−1)π
−|k|d sin 2N (2n − 1)π (2n − 1)π
Bφm (k) = J1−m + J1+m e sin + |k|d cos
2N a a n=1
2N 2N

50
Solving for Fz or Fφ will allow the stream function to be determined. Using the Bρm (k)
version of Equation (48), it results in the current density Fourier components:

−iBblπ h  sc   sc i
Fφm (k) = 0 (kc)K 0 (ka)
J1−m + J1+m
2N µ0 akIm m a a
N  
X (2n−1)π
−|k|d sin 2N (2n − 1)π (2n − 1)π
· e sin + |k|d cos
n=1
2N 2N

Such that the surface current density is


1 X imφ ∞
Z
Fφ (φ, z) = e dk eikz Fφm (k)
2π m=−∞ −∞

1 X imφ ∞ −iBdπ
Z h  sc   sc i
= e dk eikz 0 (kc)K 0 (ka)
J1−m + J 1+m
2π m=−∞ −∞ 2N µ0 akIm m a a
N  
X (2n−1)π
−|k|d sin 2N (2n − 1)π (2n − 1)π
· e sin + |k|d cos
n=1
2N 2N
∞  sc i ∞
−iBd X imφ h eikz
 sc  Z
= e J1−m + J1+m dk 0 0 (ka)
2N µ0 a m=1 a a −∞ kIm (kc)Km
N  
X (2n−1)π
−|k|d sin 2N (2n − 1)π (2n − 1)π
· e sin + |k|d cos
n=1
2N 2N
Z ∞
−iBd h  sc i eikz
+ 2J1 dk 0 0 (ka)
4N µ0 a a −∞ kIm (kc)Km
N  
X (2n−1)π
−|k|d sin 2N (2n − 1)π (2n − 1)π
· e sin + |k|d cos
n=1
2N 2N

Note: the second term differs from the derivation in [3] by a factor of 2. Using identities

51
of the modified Bessel functions, this can be rewritten as:

∞  sc i Z ∞
Bd X h  sc  i sin(kz)
Fφ (φ, z) = cos(mφ) J1−m + J1+m dk 0 0 (ka)
µ0 aN m=1 a a 0 kIm (kc)Km
N  
X (2n−1)π
−kd sin 2N (2n − 1)π (2n − 1)π
· e sin + kd cos
n=1
2N 2N
Z ∞
Bd h  sc i i sin(kz)
+ J1 dk 0 0 (ka)
µ0 aN a 0 kIm (kc)Km
N  
X (2n−1)π
−kd sin 2N (2n − 1)π (2n − 1)π
· e sin + kd cos .
n=1
2N 2N

(Note: the imaginary part in Kn0 (−x) can be ignored [6].)


The stream function is a integrated current density with S as an integration constant.

Z ∞
−Bd sc −1
S= J1 ( ) dk 2 0
µ0 aN a 0 k I0 (kc)K00 (ka)
N  
X (2n−1)π
−kd sin 2N (2n − 1)π (2n − 1)π
· e sin + kd cos .
n=1
2N 2N

The integration constant was chosen such that the integrand


eikz
Z
dk 0 (kc)K 0 (ka)
−∞ kIm m

remains finite at k = 0. However, this is not the only possibility, so this may lead to an
erroneous result. For example,

Z ∞
−Bd sc k−1
S= J1 ( ) dk 2 0
µ0 aN a 0 k I0 (kc)K00 (ka)
N  
X (2n−1)π
−kd sin 2N (2n − 1)π (2n − 1)π
· e sin + kd cos .
n=1
2N 2N

is also valid for preventing the integral from blowing up. The constant used by Liu was
maintained for this project, but other integration constants may be worth considering in the

52
future. The total stream function comes to
Z
ψ(φ, z) = Fφ (φ, z)dz
∞  sc i Z ∞
−Bd X h  sc  cos(kz)
= cos(mφ) J1−m + J1+m dk 2 0 0 (ka)
µ0 aN m=1 a a 0 k Im (kc)Km
N  
X (2n−1)π
−kd sin 2N (2n − 1)π (2n − 1)π
· e sin + kd cos
n=1
2N 2N
Z ∞
−Bd h  sc i cos(kz)
+ J1 dk 2 0 0 (ka)
µ0 aN a 0 k Im (kc)Km
N  
X (2n−1)π
−kd sin 2N (2n − 1)π (2n − 1)π
· e sin + kd cos +S
n=1
2N 2N
∞  sc i Z ∞
−Bd X h  sc  cos(kz)
= cos(mφ) J1−m + J1+m dk 2 0 0 (ka)
µ0 aN m=1 a a 0 k Im (kc)Km
N  
X (2n−1)π
−kd sin 2N (2n − 1)π (2n − 1)π
· e sin + kd cos
n=1
2N 2N
Z ∞
−Bd h  sc i cos(kz)
+ J1 dk 2 0 0 (ka)
µ0 aN a 0 k Im (kc)Km
N  
X (2n−1)π
−kd sin 2N (2n − 1)π (2n − 1)π
· e sin + kd cos
n=1
2N 2N
Z ∞
−Bd  sc  −1
+ J1 dk 2 0
µ0 aN a 0 k I0 (kc)K00 (ka)
N  
X (2n−1)π
−kd sin 2N (2n − 1)π (2n − 1)π
· e sin + kd cos
n=1
2N 2N

The final result for the transverse phase gradient coil was the stream function given in

53
Equation (24).

( ∞ ∞
−Bd X h  sc   sc i Z cos(kz)
ψ(φ, z) = cos(mφ) J1−m + J1+m dk 0 (kc)K 0 (ka)
µ0 aN m=1
a a 0 k 2 Im m
N  
X (2n−1)π (2n − 1) (2n − 1)π
· ekd sin 2N sin + kd cos
n=1
2N 2N

 sc  Z cos(kz) − 1
+ J1 dk
a 0 k 2 I00 (kc)K00 (ka)
N  
X
−kd sin
(2n−1)π(2n − 1)π (2n − 1)π
· e sin
2N + kd cos
n=1
2N 2N
∞  sc i Z ∞
X cos(mφ) h  sc  cos(kz)
+c J1−m − J1+m + dk 0 (ka)
m=1
m a a 0 kIm (kc)Km
N  )
X (2n−1)π (2n − 1)π (2n − 1)π
· e−kd sin 2N sin + kd cos (54)
n=1
2N 2N

54

S-ar putea să vă placă și