nice-coils/paper/paper.tex

\documentclass[conference,compsoc]{IEEEtran}

\usepackage[T1]{fontenc}
\usepackage[
    backend=biber,
    style=numeric,
    natbib=true,
    url=false, 
    doi=true,
    eprint=false
    ]{biblatex}
\addbibresource{paper.bib}
\usepackage{amssymb,amsmath}
\usepackage{eurosym}
\usepackage{wasysym}

\usepackage[binary-units]{siunitx}
\DeclareSIUnit{\baud}{Bd}
\DeclareSIUnit{\year}{a}
\usepackage{commath}
\usepackage{graphicx,color}
\usepackage{subcaption}
\usepackage{array}
\usepackage{censor}
\usepackage{hyperref}
\usepackage{makecell}

\renewcommand{\floatpagefraction}{.8}
\newcommand{\degree}{\ensuremath{^\circ}}
\newcolumntype{P}[1]{>{\centering\arraybackslash}p{#1}}
\newcommand{\partnum}[1]{\texttt{#1}}
\newcommand{\todo}[1]{\textbf{TODO}\footnote{#1}}

\begin{document}

\date{}
\title{Wireless Power Transfer with a Twist:
Achieving Rotation-Invariant Coupling using Multi-Layer PCB Inductors}
\maketitle

\begin{abstract}
    % FIXME
\end{abstract}

\section{Introduction}

Inductive wireless power transfer (WPT) is a widely used technology supported by a large corpus of research literature.
% FIXME cite
While working on a novel application of Inductive wireless power transfer in a Inertial Hardware Security Module (IHSM)
as proposed by Götte and Scheuermann, % FIXME cite
we found ourselves presented with an unusual set of constraints around inductive wireless power transfer through a
rotating joint using a PCB inductor that does not yet seem to be addressed adequately in the existing literature on
inductive wireless power transfer.

Our application poses the challenge of transferring power between a stationary and a rotating part. To reduce
manufacturing cost of both parts, and to reduce weight, and thereby inertia as well as susceptibility to vibration in
the rotating part, we decided to use inductors that are directly patterned onto the IHSM's printed circuit boards.
The primary constraint that results from this choice is a highly constrained turn count that is limited by the PCB
manufacturing processes' pattern resolution and by ohmic heating.

We found that the limited turn count of PCB inductors results in a \emph{slightly} asymmetric field, which means that
the coupling coefficient of two such inductors oscillates at one oscillation per revolution when the inductors are
rotated on-axis, even if both inductors are perfectly coaxially aligned.

In other inductive wireless power transfer systems, this oscillation is mitigated by one of several factors: First, for
this effect to matter in the first place, the two coils have to be rotating with respect to one another. In ferrite or
iron-cored inductors, the core shapes the magnetic field and evens out any such imperfection. In wire-wound inductors,
the (much) higher turn count and circular aspect ratio of the wires reduces this effect to almost nothing. Finally, the
output ripple caused by this oscillation can be filtered through a voltage regulator or by using a large decoupling
capacitor on the secondary side.

While there exist a number of prior works focusing on efficient power transfer between two coils whose position relative
to one another cannot be precisely controlled as is the case in wireless phone charging systems, it is generally assumed
that the two coils remain (almost) stationary with respect to one another throughout the charging process. % FIXME cite

There exists a small body of work on inductive power transfer through rotating joints, % FIXME cite
but here the focus lies on higher power budgets than our application requires, which often requires ferrite or iron-core
inductors.

Our application is unique in that it requires power transfer through a joint that is constantly rotating at high speed,
while we simultaneously want to avoid heavy components on the (rotating) receiver side. (Liquid) electrolytic capacitors
cannot be used due to the large centrifugal acceleration that the rotating part experiences, and other heavy components
such as large ceramic or polymer electrolytic capacitors or ferrite-core power inductors are inadvisable since they will
exert large stresses onto the assembly due to the same centrifugal acceleration, and any imbalance caused by tolerances
in the placement of heavy components will quickly cause a strong vibration.

\subsection{Twisted inductors}

Applying a principle inspired by rectangular or octagonal RFIC inductor design as well as by the polygonal basket-woven
air coils used in early radio sets, we propose a novel way of laying out circular PCB inductors that twists the
inductor's windings around one another using a ring of vias each on the inside and outside of the inductor's windings.
Applying some math, we show that we can layout a twisted inductor for any number of twists that is co-prime to the
inductor's turn count.

We observe that in high-frequency applications, a moderate number of twists increases the spacing between the beginning
and end of the inductor's conductor, where the majority of the inductor's AC current flows. This decreases the parasitic
capacitance of the inductor and raises its self-resonant frequency, raising its maximum possible operating frequency and
improving its efficiency at lower operating frequencies. This is the same effect that is exploited in basket-woven
air core inductors that were commonly used in old radio sets.

\section{Related Work}

% TODO cite \cite{fanSimultaneousWirelessPower2024} (rotating coupling)
% TODO cite \cite{mullenEffectMisalignmentInductive} (misaligned coils)

\subsection{A Short Historical Diversion on Basket-Woven Air Coils}

Since the early days of radio engineering, the parasitic capacitance of inductors has been a point of
concern\cite{nesperHandbuchDrahtlosenTelegraphie1921,flemingPrinciplesElectricWave1910}. Going back to the early days of
wireless telegraphy after the turn of the twentieth century, coils with high inductance were needed for the construction
of both transmitters and receivers, but the ferrites that would later permit their compact construction were still being
developed. The ferromagnetic core material of choice back then was laminated iron, which was only useful at low
frequencies due to eddy current losses. As a result, the inductors in radio circuits of the era were constructed as
air-core coils. While air core inductors are immune to core saturation, the poor magnetic permeability of air leads
necessitates many large turns of wire for practical inductance values, which for reasons of practicality or leakage
inductance often could not be wound in a single layer. Winding an inductor with many turns on multiple layers improves
compactness and leakage inductance, but in turn gives rise to increased distributed capacitance.

Back then, a number of ways were devised to decrease distributed capacitance in multilayer inductors. These methods can
be divided into two general categories: Optimizing the connecting order of turns to minimize the voltage differential
between adjacent turns---a technique that is still used to this day\cite{lopeFirstSelfResonant2021}, and optimizing the
winding schema to increase the separation between turns. The main technique in the first category concerns winding the
turns of a cylindrical multilayer inductor not layer by layer, but instead layering them diagonally, effectively
connecting adjacent turns in a diagonal zigzag pattern. Then as now, wound inductors applying this technique were not
feasible to manufacture reliably by machine, but the technique can be closely replicated in PCB inductors as shown in
\textcite{leePrintedSpiralWinding2011a}. The main limiting factors in a PCB implementation are the requirement for a
large number of vias inside the inductor's turns limiting the achievable turn count\footnote{In PCBs, as opposed to
ICs, vias limit the achievablie turn count when they need to be placed in-line inside the turns as opposed to on the
inside or outside because a PCB's minimum trace/space widths are usually much smaller than the smallest feasible via,
consisting of a minimum-size drill surrounded by a minimum-size annular ring.} and increasing ESR through the thin trace
sections that are necessary to accomodate the via structure, as well as the layer pairing limitations when blind vias
are used in multilayer PCBs.

\begin{figure}
    \begin{center}
        \subcaptionbox{\raggedright A honeycomb coil in \textcite{saackeRadiotechnikIIIEmpfanger1926}}{
            \includegraphics[width=0.25\linewidth]{figures/saacke-radiotechnik-3-ledionspule.jpg}}
        \subcaptionbox{\raggedright A basket-woven coil in \textcite{kleinSpulenUndSchwingungskreise1941}}{
            \includegraphics[width=0.25\linewidth]{figures/klein-spulen-schwingkreise-korbspule.jpg}}
    \end{center}
    \caption{Illustrations of honeycomb and basket-woven coils from the early days of wireless radio.}
    \textbf{TODO}: Not final graphics. Get proper scans for camera-ready version
    \label{fig_illust_honeycomb_basket}
\end{figure}

This lack of a way to wind high frequency inductors with a machine led to the creation of a number of related winding
schemes that including honeycomb and basket woven coil
\cite{eppenAnforderungenEinzelteileRundfunkempfanger1927,
filbigLehrbuchHochfrequenztechnik1942,
kleinSpulenUndSchwingungskreise1941,
meinkeTaschenbuchHochfrequenztechnik1956,
nottebrockSpulen1950,
struttVerstarkerUndEmpfanger1951,
wiggeRundfunktechnischesHandbuch1930,
zicknerSpulen1927}.
Examples of both from contemporary literature are shown in Figure\ \ref{fig_illust_honeycomb_basket}. In a honeycomb
coil, subsequent winding layers are wound at a criss-cross pattern similar to how in sewing, a spool of thread is wound.
The characteristic feature of honeycomb coils is that the winding machine is adjusted to produce large air gaps between
adjacent windings on the same layer. When multiple layers like this are stacked, a rhomboid pattern results that is
vaguely reminiscent of a honeycomb's structure.

In basket-woven coils, a mandrel consisting of an odd number of sticks pointing either radially or axially is used, and
the wire is fed woven between adjacent sticks in an alternating direction. While visually similar to honeycomb coils,
this winding technique is more suited to homebrew construction and less amenable to mass production by machine. In
axially basket-woven coils, the mandrel can be pulled out after the coil is finished. Like honeycomb coils, the
resulting structure can be made mechanically stable with some lacquer, with the turns carrying the layers where they
cross.

Both construction techniques apply similar principles to those leading to the improved high-frequency behavior of
twisted inductors. Interestingly, both honeycomb and basket-woven coils are also governed by the same coprimality
condition between the number of turns and the number of inversions within each turn that we describe for our twisted
inductors below, although in contemporary literature, this condition is never explicitly stated
\cite{eppenAnforderungenEinzelteileRundfunkempfanger1927, kleinSpulenUndSchwingungskreise1941,
wiggeRundfunktechnischesHandbuch1930}.
% TODO cite \cite{querfurthCoilWindingDescription1954}

\subsection{PCB inductor design for wireless power transfer}

For wireless power transfer, air-core inductors with or without ferrite magnetic shielding are the standard solution.
Since in most applications, an air gap of several millimeters between the sending and receiving assemblies is expected,
adding a ferrite core does not result in a large improvement in coupling. Meanwhile, in many WPT applications,
especially for charging portable devices or medical implants, some misalignment between the sending and receiving coils
is expected. Using the available space with an air-core inductor that has a large cross-sectional area reduces the
impact of this misalignment.

Looking at such WPT inductors, they tend to be mostly planar coils with only a few layers, so implementing them in a PCB
process seems natural. Using a PCB for the inductor has the potential to reduce implementation cost since PCBs are
cheap, and they can also serve as structural support.

Implementing inductors in PCBs has a number of disadvantages. First, due to the limited layer count of common PCB
processes, and due to structure size limitations, the number of windings that can be fit into a given volume is much
lower than in wire-wound inductors. Second, due to a PCB's copper layers being thin compared to its dielectric
substrate, PCB inductors tend to have poor DC resistance. A PCBs' thin but wide trace cross-section helps with
skin effect compared to a solid conductor. However, PCBs can still not approach the performance of litz wire used in
high-frequency WPT coils, which commonly use wire diameters in the tens of micrometer
range\cite{zhaoDesignOptimizationLitzWire2023}. \textcite{lopeFrequencyDependentResistancePlanar2014} and
\textcite{nomotoSplittingConductorsCoils2024} propose a mitigation that aims to emulate a litz wire's structure in
large, high-current PCB inductors, but their mitigation is heavily limited by the structure size achievable in common
PCB manufacturing processes\cite{nguyenReviewComparisonSolid2020}.

A further factor that limits the high-frequency performance of PCB inductors is distributed capacitance. Not only do
large air coils exhibit more parasitic capacitance than much smaller ferrite-core inductors simply due to their size,
when implemented in a PCB process a large fraction of the electrical fields responsible for this capacitance pass
through the PCB's substrate, not air. The relative permittivity $\epsilon_r$ of common PCB substrates typically lies in
the range of $4$ to $5$\cite{mumbyDielectricPropertiesFR41989}, which increases the distributed capacitance compared to
a pure air-core inductor.

\subsection{Twisted Inductors in RFIC Design}

The simplest twisted inductor as shown below with $k=1$ inversion corresponds to the counterwound scheme that is
commonly used for two-layer planar
inductors\cite{lopeFirstSelfResonant2021,sproHighVoltageInsulationDesign2021,leePrintedSpiralWinding2011a}, and
which has been used to stack planar coils for more than a century\cite{flemingPrinciplesElectricWave1910}.

Another, more recent design interleaves two counter-wound planar spiral inductors on the same layer with the help of
some jumper connections on a second layer, as shown in \cite{daneshDifferentiallyDrivenSymmetric2002}. The use of this
design in RFIC design is mainly focused on its electrical symmetry, so that the two input ports can be fed with a fully
differential signal, with both driver outputs being loaded equally across the inductor's frequency range.

% They note that the main point behind the design is electrical symmetry of the two ports to make driving the thing
% differentially cleaner. We should adopt this observation for our inductors, which likewise are electrically symmetric
% when compared to a single-layer spiral inductor.

\subsection{Inductive Wireless Power Transfer in Practice}

Inductive WPT has been proposed in a large number of scenarios, each of which comese with a set of
unique constraints. When WPT is used to charge an electric toothbrush, the implementation cost of the system is
critical, while efficiency and total power output are of little concern. Mechanically, in an electric toothbrush's
charging system, the position and spacing of the transmitter and receiver coils can easily be controlled down to
millimeter precision.

In contrast to this, wireless smartphone charging is a much more demanding application. Here, the total cost of the
system is only secondary, but the receiver's form factor is critical, and total power output as well as efficiency
become major objectives. At the same time, in wireless smartphone charging, position tolerances are very coarse, and the
two coils in the charging base and in the phone may be positioned more than a centimeter off-axis, with a gap of several
millimeters and potentially not even in parallel planes.

Power transfer across large distances is even more of a concern in implantable medical devices. Where a wireless phone
charger must be able to bridge distances of a few millimeters, an implantable medical device might be situated
underneath several centimeter of tissue and bones. At the same time, cost is of (almost) no concern in this medical
application, which enables the use of complex manufacturing techniques, customized electronic components and exotic
materials.

While all of the aforementioned applications transfer somewhere between a few hundred milliwatts and several watts of
power, at the other end of the spectrum there is a large body of research suggesting the use of inductive wireless power
transfer for the charging of electric vehicles (EVs). In this application, the wireless power transfer system replaces
the conventional wired charging connector, which improves the systems' user experience given the strong force required
to seat or unseat these rather large connectors, as well as the heft of the required water-cooled cables. In this
application, size is of (almost) no concern, but at several kilowatt up to dozens or even a hundred kilowatt, the
transferred power is enormous and consequentially efficiency becomes of utmost importance. When charging an EV at a
rate of 30 kW, an efficiency improvement of just $0.1\%$ corresponds to a reduction in power dissipation of 30 W.
Besides the monetary cost of the power lost this way, each small improvement enables a reduction in size of heat sinks
and other cooling components, which directly translates to a decrease in cost.

\subsection{Air-Core Inductors for Inductive Power Transfer}

In inductive wireless power transfer, air-core inductors are often used since in most applications, an air gap of
several millimeters or more is expected, and adding a ferrite core would not change the system's performance by much in
these circumstances. A common way to use ferrites in WPT applications is magnetically shielding the inductor's back side
with a ferrite plate such that the field does not extend beyond the coil's back side, and to reduce eddy current losses
when the WPT coils are placed near metal
objects\cite{batraEffectFerriteAddition2015,leeSimpleWirelessPower2017,muehlmannMutualCouplingModeling2012}.


\section{Twisted Inductor Design} 

We can approach twisted inductors by construction. Let us first consider a simple, planar, circular spiral coil with a
fixed pitch. We will ignore trace width for now, and consider the trace a thin wire. We will assume the inductor's ports
are both located on the positive $x$-Axis. We can rotate it so its first port aligns with the $x$-Axis. To
minimize the loop area of the inductor's connections, inductors are usually designed with both ports close to one
another, so we can also assume its second port aligns with the $x$-Axis.

The trace trajectory of a standard planar spiral inductor can be parameterized in polar coordinates $r, \varphi$ based
on an Archimedean spiral: \todo{For the lulz, cite Archimedes here}

\begin{equation}
    r = a\cdot\varphi
    \label{eqn_arch_spi_basic}
\end{equation}

An Archimedean spiral defined this way always starts at the origin, and it continues to infinity. Let us re-parameterize
this spiral to a curve parameter $t$ with range $\left[0,1\right]$, such that $t=0$ corresponds to the start of the
inductor and $t=1$ corresponds to its end. As is customary in PCB inductors, we place the inductor's start on its outer
radius. To make handling of this easier, we introduce a variable $r' \in \left[0,1\right]$ representing the radius
normalized to the spiral's width. Let $n$ be the turn count of our inductor. The resulting parametrization is:

\begin{align}
    \varphi &= 2\pi n t\\
    r' &= 1 - t \\
    r &= r_1 + r' \left(r_2 - r_1\right)
    \label{eqn_simple_spiral_ind}
\end{align}

The resulting spiral trace starts at radius $r_2$ on the positive $x$ axis, and spirals inward until it meets $r_1$. In
its PCB realization, at $r_1$, a via would be placed to connect the end of the spiral trace to a jumper trace on another
layer of the PCB leading back to the start.

To improve layer utilization, a common technique in PCB inductor design is to use both layers of the PCB for the
inductor's spiral trace, instead of only using the bottom layer for a straight jumper trace. Using both layers this way
allows for wider traces, which lowers resistive losses. We can accomodate this optimization in our definition by
re-defining our normalized radius to allow both positive and negative values, defining negative values to designate
traces on the PCB's bottom layer as follows. Figure\ \ref{fig_nk_interleave_illust} shows both a simple and a two-layer
spiral inductor. 

\begin{align}
    \varphi &= 2\pi n t\\
    r' &= 1 - 2 t \\
    r &= r_1 + \left|r'\right| \left(r_2 - r_1\right)
    \label{eqn_twolayer_spiral}
\end{align}

\subsection{From Spiral to Twisted Inductor}

Extending the above parametrization of a spiral inductor's layout, we propose planar \emph{twisted inductors} based on
two core observations:

\begin{itemize}
    \item When using an archimedean spiral, multiple such spirals using the same pitch can be interleaved by spreading
        out their start and end points at regular angular intervals.
    \item In a two-layer spiral inductor as shown in Figure\ \ref{fig_nk_interleave_illust}, we can adjust the turn
        count of the pair of traces to move the end point of the bottom layer trace anywhere on the inductor's outer
        radius.
\end{itemize}

Combining these two observations, we find that by choosing a number $k$ of inversions that is coprime to the number of
total turns of the inductor $n$, we achieve a layout where when we connect all $k$ pairs of top and bottom-layer traces
in series, the resulting spirals on the top and bottom layers interleave cleanly. Figure\ \ref{fig_nk_interleave_illust}
shows a layout with $n=3$ turns with both a single inversion ($k=1$) as in a conventional two-layer inductor, and with
$k=2$ inversions, creating two interleaved spirals on both the top and the bottom layer of the PCB. Figure\
\ref{fig_nk_complex_illust} in Appendix\ \ref{sec_appendix_layout_examples} shows additional layout examples for larger
values of $n$ and $k$.

\begin{figure}
    \begin{center}
        \includegraphics[width=\linewidth]{figures/nk_interleave_illust.pdf}
    \end{center}
    \caption{single-layer spiral inductor's layout (left), a conventional two-layer planar inductor's layout (middle),
    and a twisted inductor with two inversions (right). All three inductors have $n=3$ turns. Traces on the PCB top
    side are shown in red, traces on the bottom side in blue. In the twisted inductor, each layer contains two
    archimedean spirals that interleave at a regular spacing. The four spirals of the inductor are connected in series
    such that they form three total turns.}
    \label{fig_nk_interleave_illust}
\end{figure}

Figure\ \ref{fig_nk_chinese_remainder_illust} illustrates how we arrive at the coprimality requirement. If we plot the
spiral in polar coordinates on a cartesian plot we observe that for a $n$-turn coil with $k$ inversions, the trace
crosses the $\varphi$ axis once for each inversion, wrapping around $r$. Likewise, it crosses the $r$ axis once for
each turn of the inductor, wrapping around $\varphi$. Based on this, we can re-label the angular axis in steps from $0$
to $k$, and re-label the radial axis in steps from $0$ to $n$. Labelling the new angular axis $i$ and the new radial
axis $j$, in the resulting integer lattice, the trace has slope $1$. We can state the trace's trajectory as a function
of a curve parameter $t \in [0, nk]$ as $f(t) = (i, j) = (t \mod n, t \mod k)$. To produce a valid inductor, the trace
must not intersect anywhere. Thus, the system of congruences

\begin{align}
    t &\equiv i \mod n\\
    t &\equiv j \mod k
\end{align}

must have a unique solution $t \in [0, nk]$ for all $(i, j)$. This statement corresponds exactly to the Chinese
Remainder Theorem, which states that this solution is unique when $k$ and $n$ are coprime.

\begin{figure}
    \begin{center}
        \includegraphics[width=0.8\linewidth]{figures/nk_chinese_remainder_illust.pdf}
    \end{center}
    \caption{Illustration of the winding pattern of two twisted inductors. The upper plots show the inductor's actual
    layout with the traces on each side of the substrate colored in red (top) and blue (bottom), respectively. The lower
    plots show the same traces, but \emph{unwrap} the annulus by plotting the traces' polar coordinates on cartesian
    axes. The left axis labels show the normalized radius, with $0$ being at the inductor's inner diameter, $1$ being at
    its outer diameter on the top layer, and $-1$ being at its outer diameter on the bottom layer. The top and right
    axes labels show the axis scaled to match indices $i\in\left[0, n\right]$ and $j\in\left[0, k\right]$,
    respectively.}
    \label{fig_nk_chinese_remainder_illust}
\end{figure}

\subsubsection{Ohmic Resistance}

The arc length $l$ of a spiral can be calculated from its turn count $n$ and the average of its inner and outer diameter
$\frac{2 r_1 + 2 r_2}{2}=r_1+r_2$ as $l = n\pi\left(r_1 + r_2\right)$. Since going from a standard inductor to a twisted
inductor does not change its turn count or dimensions, the combined arc length of all traces of the twisted inductor
does not change. Twisted inductors require two additional vias per inversion, which will increase DC resistance
slightly, but the contribution of these vias will remain small in practical applications since the overall number of
vias is still no more than a couple per turn, and since each via only bridges the short distance between the inductor's
layers.

As a general expression, for a standard or twisted inductor with turn count $n$ and twist count $k$ ($k=0$ for a
single-layer spiral inductor, and $k=1$ for a standard two-layer spiral inductor), given via resistance $R_\text{via}$
we derive a first order approximation of the inductor's DC resistance as follows.

\begin{equation}
    R_L = n\pi\frac{r_1 + r_2}{2} + \left(2k-1\right)R_\text{via}
\end{equation}

\subsubsection{Inductance}

Even for geometrically simple inductors, analytically calculating their inductance is a surprisingly hard problem whose
complexity quickly escalates with the inductor's geometric complexity, with realistic wire shapes as opposed to
approximations assuming an infinitely thin wire, or when taking into account differing magnetic permeabilities of
air or dielectrics and core materials. Instead, a number of approximations are commonly used. A commonly referenced
approximation for the inductance of planar spiral inductors is given by \textcite{mohanSimpleAccurateExpressions1999},
whose current-sheet approximation for circular planar spiral inductors we will use here to estimate our inductor's
inductance. The current-sheet approximation from \textcite{mohanSimpleAccurateExpressions1999} reads:

\begin{equation}
    \label{eqn_mohan_approx}
    L = \frac{\mu n^2 d_\text{avg} c_1}{2}\left(\ln\left(c_2/\rho\right)+c_3\rho+c_4\rho^2\right)
\end{equation}

In this equation, $c_{1-4}$ denote four empirically determined coeficcients that together describe the coil's shape. The
values for circular coils are $c_{1-4}=(1.00, 2.46, 0.00, 0.20)$. $\mu$ is the magnetic permeability of air (for an
air-core inductor), $n$ is the number of turns, $d_\text{avg}$ is the \emph{average} turn radius, i.e. $d_\text{avg} =
2\frac{r_1 + r_2}{2} = r_1 + r_2$. $\rho = \frac{r_2-r_1}{r_2+r_1}$ is the planar spiral inductor's \emph{fill ratio}.
The fill ratio encodes the fact that the inductor's turns have less flux linkage the closer to the inductor's center we
get. While turns close to the outside have good flux linkage due to their inner area overlapping well with that of other
turns, turns close to the center not only have a loop area that is only a fraction of that of turns further outwards,
the closer we get to the center, the larger is also the fraction of the field lines returning as leakage flux on the
outside of the inner turn that pass through the inner part of turns further outwards, flipping the sign and contributing
\emph{negative} mutual inductance.

As Equation\ \ref{eqn_mohan_approx} approximates the inductor's whole set of windings as a single, uniform current
sheet, the turn count only appears as a single factor of $n^2$ in the equation, with $\rho$ and $c_{1-4}$ correcting for
the inductor's geometry. To account for twisted inductors, we can separate the inductor into a set of $2k$ simple planar
spiral inductor \emph{branches} that are connected in series by the twisted inductor's vias. Compared to a simple spiral
inductor, for each branch, the inductance according to Equation\ \ref{eqn_mohan_approx} stays the same except that the
factor $n^2$ drops to $\left(\frac{n}{2k}\right)^2$ because the $n$ windings are evenly distributed across the $2k$
branches. Let us now make two assumptions. First, we will assume that the flux linkage between both sides of the
inductor is approximately one. This assumption is grounded in the fact that for practical designs, the substrate
thickness will be small compared to the inductor's diameter. Second, we will for now ignore the spiral inductor's field
asymmetry and assume that the flux linkage between two intertwined branches on the same side of the substrate is
approximately one. In our measurements below we show that for simple spiral inductors this asymmetry, while problematic
in our application, is small in absolute terms, and grows smaller with increasing turn count.

Based on these two assumptions, we can model the twisted inductor as a set of $2k$ series-connected spiral inductors
that are perfectly coupled, with full flux linkage. This results in the total series inductance gaining back the factor
$\frac{1}{2k^2}$ that each branch lost, resulting in identical inductances for a simple planar spiral inductor and a
twisted inductor with the same size and turn count according to Equation\ \ref{eqn_mohan_approx}. This approximation
introduces an error due to the imperfect flux linkage between the two sides of the substrate, and between two spiral
branches located at an angular offset from each other. In our experiments, we found that for our test inductors,
compared to inductances measured with an LCR meter, this error is below \qty{10}{\percent} for $n=5$ turns or more, and
for our test samples matches the performance of Equation\ \ref{eqn_mohan_approx} for the simple planar spiral inductor
case.

\subsection{CAD Integration}

To allow for easy design with twisted inductors, and to speed up the laboratory prototyping we performed for this paper,
we created a tool that generates arbitrary twisted inductor layouts, and that is able to output these layouts as PCB
footprint files for the open source KiCad EDA CAD tool. We integrated the ESR and ESL approximations as derived above
with our tool, so that it provides immediate design feedback when generating inductors. In order to minimize ESR and
maximize PCB area utilization, we made the tool automatically calculate the largest possible trace width when given a
minimum clearance specification.

To handle outputting PCB geometry in a format that can be read from KiCad, we utilized the open source EDA file format
library \emph{gerbonara}\todo{Cite gerbonara}. To support the FEM simulations that are described in the next section
below, our tool contains functionality to map gerbonara's geometry representation into that of gmsh\todo{Cite gmsh}, the
FEM mesher that we chose to interface with Elmer FEM\todo{Cite Elmer}.

Our inductor design tool is available in this paper's supplementary material as well as at the git repository linked at
the end of this paper.

\section{FEM Simulation}

To validate our analytical approximations, we performed a series of FEM simulations in Elmer FEM. For a number of
inductor layouts, we performed simulations to determine ohmic resistance and inductance. Due to limitations in our
gmsh/Elmer toolchain, we were unable to run simulations for parasitic capacitance and self-resonance, or for coupling
behavior of coil pairs. We found that for these cases which require larger, more complex meshes, gmsh would frequently
crash during meshing, and where we were able to produce meshes, Elmer would only converge for some of them. While these
are problems that can be solved through either a more skillful description of the problem in gmsh and Elmer, or by using
more robust software such as Simulia CST, we decided to instead experimentally measure these quantities instead
(Section\ \ref{sec_experiments}).

\paragraph{Ohmic Resistance}
Determining ohmic resistance by FEM is reasonably easy. In Elmer FEM, we can use the built-in joint static current and
joule heating solver to determine the ohmic resistance at a given current.

\paragraph{Inductance}
We let Elmer determine inductance by first using its coil solver to determine the volumetric current density in our mesh
given a test current, then applying its magnetodynamics solver to solve the electromagnetic field. Elmer provides
routines to derive the total magnetic field energy $U_\text{mag}$ from an EM field solution. Since we have only our
inductor under test inside the simulation volume, with test current $I_\text{test}$, we can then derive the inductor's
inductance according to the well-known relation\todo{Find decent source}:

\begin{equation}
    L = \frac{2\cdot U_\text{mag}}{I_\text{test}^2}
\end{equation}

\section{Experimental Validation}
\label{sec_experiments}

To experimentally validate our design with real-world inductors, we produced test coupons with a number of variations of
twisted inductors with winding count $n$ between $1$ and $25$, and twist count ranging from $k=0$ (simple single-sided
spiral inductor) to $k=37$. All test inductors had an inner diameter of \qty{15}{\milli\meter} and an outer diameter of
\qty{35}{\milli\meter}.

\subsection{Inductance, Q-factor and DC resistance}

We measured inductance and Q-factor of each test coupon using a Keysight U1733C LCR meter at \qty{100}{\kilo\hertz}. We
measured DC resistance using a Keysight 34465A multimeter in four-wire resistance mode. We further determined the
self-resonant frequency of each inductor using a LiteVNA64 handheld vector network analyzer. The results of our
measurements are shown in Table\ \ref{tab_inductor_params}.

We found our inductance approximation to be accurate within \qty{10}{\percent} and our ESR approximation to be accurate
within \qty{20}{\percent} for inductors with three turns or more. For lower turn-count inductors, inductance
measurements are difficult because the small absolute inductances involved are easily disturbed by stray inductances,
and ESR measurements are affected by contact and trace resistance even when measurements are taken in four-wire mode.

In accordance with our design intuition, we found that for high turn count inductors, the doubled trace width that is
afforded by splitting a simple spiral inductor across two PCB layers in any two-layer configuration improves ESR by
approximately a factor of two. Going from a simple single-layer spiral inductor to a simple two-layer spiral inductor
($k=1$), we observe that the resulting inductance decreases by up to \qty{15}{\percent}. We suspect that the main factor
leading to this decrease is radial magnetic flux leakage through the PCB material between the inductor's layers.
Comparing simple two-layer inductors with $k=1$ to the twisted inductors with larger $k$ values that we propose in this
paper, we observe almost identical performance for $k>1$ with decreases of less than \qty{0.5}{\percent} going from
$k=1$ to $k=3$ irrespective of turn count. From these measurements we can conclude that the flux linkage of twisted
inductors almost perfectly matches that of simple two-layer inductors.

Finally, while not particularly relevant for our application, we decided to evaluate the high-frequency performance of
twisted inductors. We found that going from a single-layer spiral inductor to a two-layer spiral inductor decreases the
self-resonant frequency, this effect being more pronounced with higher turn count. Intuitively, this makes sense if we
consider the mechanics of inductor self-resonance. The primary contributor to self resonance, particularly in higher
turn count inductors, is capacitive coupling between the inductor's windings. In a single-layer spiral inductor, this
effect gets partially mitigated since the strongest coupling exists between adjacent windings.
the SRF have a small voltage differential as only a fraction of the inductor's total voltage appears across each
winding. Compared to this, when the inductor is constructed as a simple two-layer inductor with $k=1$, now the start and
end windings of the inductor, which have the highest voltage differential, are located right on top of each other with
the substrate in between. Making things worse, common PCB substrates have a relative permittivity much larger than air
(usually around $4$).

Interestingly, we observe that this decrease in high-frequency performance is counteracted by larger inversion count
$k$. While our test samples focused on smaller turn counts, we observe an increase from an SRF of
\qty{8.9}{\mega\hertz} for a standard $n=25,k=1$ inductor to \qty{10.6}{\mega\hertz} for $n=25,k=13$.

In conclusion, we observe that twisted inductors \emph{improve} high-frequency performance compared to simple two-layer
inductors while closely matching them in ESR and inductance. While they peform worse than simple single-layer inductors
in high-frequency performance, the increased trace width that two-layer inductors allow for lowers resistive losses by
approximately a factor of four. In applications where resistive losses lead to the choice of a two-layer inductor,
twisted inductors provide improved high-frequency performance at no additional cost and without compromising other
performance parameters.

\begin{table*}
    \begin{tabular}{cc|ccc|}
        Turn Count $n$&
        Inversion count $k$&
        Inductance $L \left[\mu H\right]$&
        Q-factor&
        DC resistance $R_\text{ESR} \left[\Omega\right]$\\
    \end{tabular}
    \caption{Inductor sample design parameters and measured characteristics. All inductors have outer diameter
    \qty{35}{\milli\meter} and inner diameter \qty{15}{\milli\meter}.}
\end{table*}

\subsection{Coupling and its Sensitivity to Radial Offset}

The key performance criterion in our application is the voltage ripple that appears on the secondary side of a WPT link
when one of the inductors is rotating. To experimentally evaluate the magnitude of this ripple in a realistic scenario
across a large set of rotations and relative displacements, we created a test setup consisting of a 3D gantry built from
an old 3D printer, with a fourth rotation axis provided by a small servo that allows us to position two inductor test
coupons at arbitrary offsets and angles to one another while measuring their coupling.

\todo{pics of 3d printer test setup}

\begin{figure}
    \begin{center}
        \includegraphics[width=.85\linewidth]{figures/test_schematic.pdf}
    \end{center}
    \caption{The test schematic used in all measurements. For direct coupling factor measurements, the load resistor was
    disconnected. We measure voltage at the output of the function generator to account for drop in its internal output
    resistance.}
    \label{fig_test_schematic}
\end{figure}

To evaluate a realistic scenario, we loaded the secondary inductor with a resistive load of \qty{10}{\ohm}, while
providing a signal at a \qty{300}{\kilo\hertz} carrier frequency to the primary inductor from a Siglent SDG6022X
function generator as shown in Figure\ \ref{fig_test_schematic}. We measured both the input and output voltages
of the coupled inductor pair using Keysight 34465A multimeters in AC RMS mode. The results of these measurements, with
the voltage ratio between the coupled inductors' input and output voltages graphed across one revolution in Figure\
\ref{fig_symmetry_3turn_n_twist} for a set of three-turn inductors with multiple inversion amounts $k$. A plot for a set
of 10-turn inductors is shown in Figure\ \ref{fig_symmetry_10turn_n_twist} in the Appendix. A key observation here is
that while the asymmetry in the inductor's field is small, the ripple induced by rotation is considerable. Figure\
\ref{fig_field_plot_3d} shows a 3D plot of an inductor's coupling. While in the plot the field looks perfectly
rotationally symmetric, the sharp dropoff with radial offset, equivalent to a large gradient, ``amplifies'' any small
asymmetry and leads to the ripple voltages we observed, amounting up to several percent of total RMS output voltage.

\begin{figure}
    \begin{center}
        \includegraphics[width=\linewidth]{figures/symmetry_3turn_n_twist.pdf}
    \end{center}
    \caption{RMS output voltage of the test circuit from Figure\ \ref{symmetry_test_circuit} for three pairs of matching
    inductors with one inductor rotating w.r.t.\ the other. The inductors have $n=3$ turns each and $k=0$, $k=1$, and
    $k=3$, respectively. For each $k$, voltage curves are plotted for a number of different radial offsets
    between the two inductor's centers.}
    \label{fig_symmetry_3turn_n_twist}
\end{figure}

From the ripple plots in Figures\ \ref{fig_symmetry_3turn_n_twist} and \ref{fig_symmetry_10turn_n_twist} we observe
slightly lower coupling for $k>0$ compared to a single-layer spiral inductor, which is in line with our previous
inductance measurements. Across one revolution, we find that single-layer spiral inductors exhibit the worst voltage
ripple, with simple two-layer inductors with $k=1$ already improving ripple by a large margin. Increasing $k$ above $1$
does not decrease the amplitude of this ripple further, but it does shift the ripple into higher frequencies that are
easier to passively filter, as we originally intended.

\subsection{Total Coupling Variation}

To analyze the behavior of our test inductors under offset and rotation, we had our measurement setup sweep
through the full range of rotation of each of the two inductors when placed at a fixed height of \qty{1}{\milli\meter}
and radial offset of \qty{4}{\milli\meter}. The resulting plots show the variation in RMS output voltage compared to its
mean across all rotations as a percentage plotted against both angular dimensions. Figure\ \ref{fig_rms_ripple_n3} shows
the resulting coupling plot for a set of three-turn inductors, and Figure\ \ref{fig_rms_ripple_n5} for a set of
five-turn inductors. Measurements for 10- and for 25-turn inductors are shown in Figures \ref{fig_rms_ripple_n10} and
\ref{fig_rms_ripple_n25} in the Appendix.

From these plots, we can draw a number of conclusions. First, our primary objective of reducing coupling variation
across rotations works, with twisted inductors ($k>1$) showing a further improvement over simple two-layer inductors,
which prove to be better than simple single-layer spiral inductors. As one would expect, this gain is greatest for
inductors with low turn count, as their turns deviate the furthest from a set of ideal, concentric circles. For the
our test inductor with an inner diameter of \qty{15}{\milli\meter} and an outer diameter of \qty{35}{\milli\meter},
$k=3$ inversions pairs already provided an improvement over standard configurations, with still better performance observed
for $k=7$ inversions.

\todo{concrete coupling factor measurements}

\begin{figure}
    \begin{center}
        \includegraphics[width=.6\linewidth]{figures/field_plot_3d_n5_k0.pdf}
    \end{center}
    \caption{The coupling between a pair of identical coils (here two simple spiral inductors with $n=5$ and $k=0$)
    visualized in three dimensions. The $x$ and $y$ axis show in-plane displacement, and the $z$ axis shows output
    amplitude in arbitrary units. Height and rotation are fixed to \qty{1}{\milli\meter} and \qty{0}{\degree},
    respectively. The most prominent aspects of this plot are that coupling falls off steeply with distance, and that
    the rotation-dependent variation is small in comparison. The circular valley around the central peak is the region
    where one inductor is mostly outside the other inductors, and intersects the field lines returning from the other
    inductor's back, leading to a negative coupling coefficient.}
    \label{fig_field_plot_3d}
\end{figure}

\begin{figure}
    \begin{center}
        \includegraphics[width=.75\linewidth]{figures/rms_ripple_double_rotation_n3_r4.pdf}
    \end{center}
    \caption{RMS ripple magnitude as a percentage of mean RMS output voltage, plotted against the rotation of each of
    the two inductors. The two coils were kept at a constant \qty{4}{\milli\meter} radial offset, and the output coil
    was loaded with a \qty{10}{\ohm} load. All RMS ripple plots in this paper share the same color scale to allow for
    visual comparison. This figure shows four variants of 3-turn coils, plots for $n=5$ can be found in Figure\
    \ref{fig_rms_ripple_n5} and plots for $n=\{10,25\}$ in Figures \ref{fig_rms_ripple_n10} and \ref{fig_rms_ripple_n25}
    in the Appendix.}
    \label{fig_rms_ripple_n3}
\end{figure}

\begin{figure}
    \begin{center}
        \includegraphics[width=.75\linewidth]{figures/rms_ripple_double_rotation_n5_r4.pdf}
    \end{center}
    \caption{RMS ripple magnitude as shown in Figure\ \ref{fig_rms_ripple_n3} for four different 5-turn coils.}
    \label{fig_rms_ripple_n5}
\end{figure}

\section{Future Work}

As part of our inductor design tool, we extended the EDA file format library gerbonara with code to automatically map
gerbonara's geometry description to the gmsh FEM mesher. This code may be of independent interest since it allows for
the extraction of FEM meshes from PCBs in any file format supported by gerbonara such as KiCad's native file format, as
well as the Gerber file format supported by the majority of EDA tools.

In the measurements we performed on our set of test inductors, we observed that while at the dimensions we chose, a
twisted inductor has slightly lower inductance by \qty{2.0}{\percent} for $n=2$, or \qty{0.11}{\percent} for $n=25$,
when compared to a simple two-layer planar spiral inductor, its inductance \emph{increases} with increasing inversion
count $k$. In one of our test coupons with $(n, k)=(25, 37)$, we even measured \emph{higher} inductance compared to a
simple two-layer planar spiral inductor. We suspect that this increase in inductance is due to the twists of our twisted
inductor effectively forming the structure of a planar toroidal inductor, with twisted inductors with $k\gg n$
approximating planar toroidal inductors. In particular, except for the slight curvature of our twisted inductor's
traces, a twisted inductor with $(n, k)=(1, n')$ \emph{is} effectively a planar toroidal inductor with turn count $n'$.

We suspect that for some choices of parameters, this effect might lead to an appreciable increase in useful inductance
as well as potentially interesting high-frequency behavior, and we aim at producing additional simulations and new
measurements for some of these choices of parameters in a future paper.

\section{Conclusion}

In this paper, we introduced a novel layout approach for planar, multi-layer inductors inspired by classic basket-wound
inductors used in the early days of radio. Our \emph{twisted} inductors produce field distributions that have better
rotational symmetry along the inductor's main axis compared to either simple single-layer spiral inductors or
counter-wound two-layer spiral inductors. Furthermore, we found that our sample twisted inductors have slightly higher
self-resonant frequency compared to both traditional layouts. We base this evaluation on laboratory measurements on a
set of 24 test inductors, which include an automated, four-dimensional mapping of the coupling between a pair of
identical inductors. We provide both an analytical description of twisted inductor construction as well as a set of
Open-Source tools for their design.

\section*{Availability}
This is version \texttt{\input{version.tex}\unskip} of this paper, generated on \today.

The git repository with the LaTeX source for this paper, the data analysis code underlying our measurements as well the
set of tools for the generation of twisted inductor layouts that we wrote can be found at:

\todo{link here}
% \center{\url{https://git.jaseg.de/nice-coils.git}}

\printbibliography[heading=bibintoc]

\clearpage
\appendix
\section{Layout examples}
\label{sec_appendix_layout_examples}

\begin{figure*}
    \begin{center}
        \includegraphics[width=.75\textwidth]{figures/nk_complex_illust.pdf}
    \end{center}
    \caption{Layout examples for a number of combinations of turn count $n$ and inversion count $k$. Note that in this
    illustration we chose values for $n$ and $k$ such that all pairs are coprime.}
    \label{fig_nk_complex_illust}
\end{figure*}

\section{Supplemental plots}

\begin{figure}
    \begin{center}
        \includegraphics[width=\linewidth]{figures/symmetry_10turn_n_twist.pdf}
    \end{center}
    \caption{Coupled RMS output voltage of three pairs of matching inductors with $n=10$ turns each and $k=0$, $k=1$,
    and $k=3$, respectively, shown as in Figure\ \ref{fig_symmetry_3turn_n_twist}}
    \label{fig_symmetry_10turn_n_twist}
\end{figure}

\begin{figure}
    \begin{center}
        \includegraphics[width=.75\linewidth]{figures/rms_ripple_double_rotation_n10_r4.pdf}
    \end{center}
    \caption{RMS ripple magnitude as shown in Figure\ \ref{fig_rms_ripple_n3} for four different 10-turn coils.}
    \label{fig_rms_ripple_n10}
\end{figure}

\begin{figure}
    \begin{center}
        \includegraphics[width=.75\linewidth]{figures/rms_ripple_double_rotation_n25_r4.pdf}
    \end{center}
    \caption{RMS ripple magnitude as shown in Figure\ \ref{fig_rms_ripple_n3} for four different 25-turn coils.}
    \label{fig_rms_ripple_n25}
\end{figure}

\end{document}