Fast Multi-Axis Tracking of Magnetically-Resonant Passive Tags:

Methods and Applications

By

Kai-yuh Hsiao

B.S., Electrical Engineering and Computer Science (1999)

Massachusetts Institute of Technology

Submitted to the Department of Electrical Engineering and Computer Science

In Partial Fulfillment of the Requirements for the Degree of
Master of Engineering in Computer Science and Electrical Engineering

At the

Massachusetts Institute of Technology

February 2001

 2001 Massachusetts Institute of Technology. All rights reserved

Signature of Author . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Department of Electrical Engineering and Computer Science
February 1, 2001

Certified by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Joseph A. Paradiso
Principal Research Scientist, Media Arts and Sciences
Thesis Supervisor

Accepted by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Arthur C. Smith
Chairman, Department Committee on Graduate Theses
2
 Fast Multi-Axis Tracking of Magnetically-Resonant Passive Tags: Methods and
Applications

By

Kai-yuh Hsiao

Submitted to the Department of Electrical Engineering and Computer Science

On February 6, 2001 in partial fulfillment of the requirements for the
Degree of Master of Engineering in Electrical Engineering and Computer Science

ABSTRACT

We have explored the design and development of magnetically-resonant tag readers for
application to tangible computer interfaces. To this end, we constructed a ringdown tag
reader and a swept-frequency tag reader, both capable of real-time continuous interaction
with multiple tagged objects. Although the ringdown reader worked well for smaller
numbers of tags, the swept-frequency tag reader proved more efficient for work with
twenty to thirty tags, and so we did further work to extend and apply it.

Graphical and musical applications were developed for the swept-frequency tag reader and
proved its usability for driving tangible desktop interfaces. Finally, a six-coil variant was
constructed in order to determine the three-dimensional position and orientation of tagged
objects. This thesis describes both reader systems, outlines our demonstration
applications, and gives first test results from the multiple-coil tracker.

Thesis Supervisor: Joseph A. Paradiso

Title: Principal Research Scientist, Media Arts and Sciences

3
Acknowledgements
First and foremost, I would like to thank Joe Paradiso, who is my advisor,
supervisor, colleague, and friend, and whose analog circuit wizardry provided the vast
majority of the basic concepts and the underlying electronics for my work on these
projects. Joe has given me nothing less than superhuman amounts of his time, attention,
and patience, and his singularly amazing spirit, supportiveness, and engineering abilities
are an inspiration to us all. Furthermore, I would like to thank the guys at British Telecom
and the rest of the Things That Think consortium for funding the past 1.5 years of this
endeavor, and also my parents for funding the several years preceding that. Thanks also to
the members of the ever-growing Responsive Environments group, especially Ari
Benbasat, whose occasional help and advice, along with his seemingly unwitting support
role for our entire group, often provided the extra kick necessary to just get things working.
And finally, thanks to the MIT Media Lab, and everyone who helps it to run smoothly, for
making every part of this experience even possible.

About the Author

Kai-yuh Hsiao’s primary interests in computers and music started at an early age.
He has been working at the Media Lab since his arrival at MIT as an undergraduate in
1995, and all of his projects have involved some combination of computers, electronics,
and music. After receiving his bachelor’s degree in 1999, he continued working under Joe
Paradiso in the Responsive Environments group in order to finish his master’s degree.

4
Table of Contents

1. Introduction 7

1.1. Types of wireless tags 8

1.2 Magnetically resonant tags as sensors 11
1.3 Tag readers for real-time CHI interaction 12

2. Ringdown tag readers 13

2.1 Design of a continuous real-time ringdown tag reader 14

2.2. Implementation of ringdown reader transmit circuitry 15
2.3. Selection of tags 17
2.4. Implementation of ringdown reader receive circuitry 18
2.5. Ringdown reader demo application and results 21
2.6. Ringdown reader conclusions 22

3. Swept-frequency tag readers 25

3.1. Design of a continuous real-time swept-frequency tag reader 26

3.2. Implementation of swept-frequency transmit circuitry 28
3.3. Selection of tags for swept-frequency tag reading 30
3.4. Implementation of swept-frequency receive circuitry 31
3.5. Swept-frequency tag reader results and refinements 34
3.6. Swept-frequency tag reader conclusions 40

4. Applications of the Swept-Frequency Tag Reader 41

4.1. Early applications 41

4.2. Musical Trinkets 43
4.3. Graphical extensions to the Musical Trinkets project 48

5
5. Swept-frequency tag readers with multiple coils 55

5.1. Limitations of a single-coil swept-frequency tag reader 55

5.2. Multiple-coil arrangements and Helmholtz coils 58
5.3. Two-coil Helmholtz arrangement 60
5.4. Six-coil Helmholtz arrangement 62
5.5. Results and application 66
5.6. Six-coil tag reader conclusions 70

6. Conclusion 73

6.1. The ringdown tag reader 73

6.2. The swept-frequency tag reader 74
6.3. Swept-frequency tag reader applications 75
6.4. The six-coil swept-frequency tag reader 76
6.5. Future directions and applications 77
6.6. Closing 80

References 81

Appendices 85

A.A. Schematics 85
A.B. Frequency drift compensation 89
A.C. Program code 93

6
Chapter 1
Introduction

An important direction that computer-human interface research has taken in recent

years focuses on enabling any object, anywhere, to interact with a computer. For instance,
the "tangible interface,"1 where the manipulation of physical, so-called “tangible” objects
is observed by a computer and used as input, is a topic of much exploration, in the hopes
that someday soon, interacting with a computer will be as natural as manipulating standard,
everyday items. The present-day graphical user interface (GUI) paradigm uses keyboards,
mice, and 2D displays as interface devices, but despite its versatility, the user is still fairly
remote from the data being manipulated. This stiff and somewhat artificial mode of
interaction will gradually be augmented by a physical dialog with real world objects as
they become interfaces to an omnipresent ubiquitous computing environment.2 Being able
to control a computer by simply manipulating objects on a physical desktop would allow a
seamless, intuitive interface well beyond what is commonly possible today on the virtual
desktop.
One approach to producing tangible objects is to embed each object with a small
wireless tag.3 These tags exhibit electromagnetic behavior that enables a suitable tag
reader device to detect their key properties, such as location and identification, remotely.
With a tag attached to the relevant objects, computers can observe their movements and
manipulations and use these data as input. In this paper, we describe the development of
several wireless tag readers well suited to these concepts and introduce several applications
of them as tangible interfaces.

7
 RF Coupled (RFID)
(2.4 GHz)
Inductively coupled
(RFID) (125 kHz)

LC Tag (EAS)
(13 MHz)

Electrostatically Coupled Magnetostrictor

(RFID) (125 kHz) (EAS) (59 kHz)

Figure 1. Various types of common tags.

1.1. Types of wireless tags

The term "wireless tag" describes any of a range of small, inexpensive electronic
circuits with electromagnetic properties that enable a tag reader system to detect the tags
and possibly read other properties of the tags as well. Several examples of wireless tags

8
are shown in Figure 1 and are described below. Wireless tagging is already a large, well-
established industry, and some wireless tagging technologies are already in very common
usage. Perhaps most popular are the ubiquitous “Electronic Article Surveillance” (EAS)
tagging systems found in many shops today; in order to prevent shoplifting, small wireless
tags are embedded on or in merchandise packaging of the items stocked in a store. When
the tags come within range of a reader device placed at the exit to the store, they are
detected, alarms ring and the shoplifter has been caught. When the item is purchased
properly, the tags are deactivated so they no longer respond and the item can be removed
without incident.
As prevalent as they are, the common types of EAS tags are a mere subset of the
types of tags that have been developed for a multitude of applications. First, tags fall into
two primary categories, active and passive. Active tags draw upon a source of power (e.g.
battery) attached directly to the tag. These tags typically operate in a low-power mode,
waiting for an appropriate signal to wake them up. Whenever they receive such a signal,
they draw upon their power sources to transmit their programmed information. Passive
tags, in contrast, have no attached power source, and instead they use the power
transmitted by the detection antenna.
Within the category of passive tags are RFID tags, as well as a whole assortment of
EAS tag types. RFID tags have an onboard microchip, which upon receiving power either
sends a signal directly back or manipulates the field from which they receive power, for
example by modulating the impedance of their antennae. This enables their stored
information to be read and/or written. EAS tags, on the other hand, are typically less
complex, contain no integrated circuits, and manipulate the reader’s fields in a simple and
detectable fashion. In this thesis we will concern ourselves with several types of EAS-
style tags, which are in turn a subcategory of passive tags.
RFID tags can be split into categories based on the frequency and type of field they
operate in. High-frequency tags derive their power from a rectified RF antenna
(“rectenna”) and commonly operate around 2.4 GHz. These tags typically transmit data
back to the reader via RF reflections of the interrogating signal, e.g. by modulating their

9
antenna impedance. Magnetically-coupled tags operate at lower frequencies (typically at
125 kHz or 13 MHz). Transmit coils at the reader produce a magnetic field which is
received by a coil wound around the chip. Power is inductively coupled into the chip tag’s
coil (analogously to a transformer), which is also used to transmit and receive data to and
from the reader. Finally, electrostatic tags 4 operate on a similar principle, but instead of a
coil they use two flat metallic plates, which extract power and communicate capacitively
via a changing electric field.
The other general class of passive tag, which the remainder of this thesis will be
working with, is the resonant EAS tag. This class of tags is very appropriate for simple
detection applications, most notably shoplifting detection. Compared to the more complex
RFID tags, these tags are often known as “1-bit tags”, because they traditionally provide
only one item of information: whether they are present in the detect field or not. The tags
used in these applications tend to be simple and cheap, and so they produce only simple
responses, thus providing a limited space of ID codes, as opposed to the more complex
dialog possible with chip-based RFID tags, which can store arbitrarily long ID bitstrings.
Currently, the most commonly found type of shoplifting tag is the magnetostrictor.5
This extremely inexpensive tag is simply a magnetically-coupled strip of metallic glass
with mechanical properties that create a high-Q resonance (meaning their resonance
response is very strong and responds to very specific frequencies) in the sub-100 KHz
range, so when it receives a magnetic pulse at its resonant frequency, it produces a
"ringdown" response which can be detected by the reader. Because the response of the tag
is dependent on a small strip magnet attached to the tag, these tags can be deactivated by
demagnetizing the strip magnet. This makes it convenient for stores to turn off these tags
by bringing the tags near an erase coil, allowing customers to leave with their purchases
without incident.
Another simple such resonant tag is the LC tag, which consists of an inductor (L) in
series with a capacitor (C). This, too, is magnetically coupled, through its inductor
component, and it is electromagnetically resonant, making it similarly detectable. This

10
type is usually cut out of a flat sheet of conductive foil and can thus be used for shoplifting
detection for items such as books or boxes.
Two other examples of shoplifting tags are the magnetic wire tag,6 which uses a
length of nonlinear magnetic wire to produce detectable high-order harmonics when
subjected to an AC magnetic field; and the microwave diode tag, 7 which uses a diode
whose bias varies when exposed to a capacitively-coupled 100 kHz electric field generated
near the store’s exit, thus producing detectable harmonics that distort an ambient
microwave signal reflected by the tag.

1.2 Magnetically resonant tags as sensors

The magnetostrictor and LC resonant tags have similar electromagnetic properties.

They both resonate and couple magnetically at relatively low frequencies, and thus both
types can be detected by similar tag readers. An appropriately designed tag reader would
be able to detect the presence and proximity of such tags to a coil generating an appropriate
AC magnetic field.
Typically these tags are used to provide their single bit of ID, i.e., whether they are
present or not. However, there have been several projects that examine more complex
sensing capabilities of such tags.8,9 These center around adjusting physically variable
parameters of the tags and detecting such changes using specialized readers. Specifically,
tags could be constructed such that their resonant frequencies or Q vary in proportion to
some other parameter, such as temperature, pressure, or local DC magnetic field.
Once such a shift in tag properties can be detected, these tags then enable yet
another dimension of manipulation aside from simple presence and position. For instance,
an LC tag could be embedded in the bottom of a milk carton, with an elastic material
between the capacitor plates.8 The presence and position of the milk carton can then be
identified, by virtue of the distinct resonant frequency of the attached tag and the
magnitude of its returned signal. Additionally, the quantity of milk remaining in the carton

11
could furthermore be measured, because the capacitor’s plate spacing, which determines
the capacitance and thus the tag’s resonant frequency, is shifted as a function of the weight
of the remaining milk.

1.3 Tag readers for real-time CHI interaction

Our purpose in working with tagging techniques was to design and explore systems
that could enhance real-time interaction with a computer, particularly to enable and
experiment with useful tangible interfaces. The systems available commonly on the
market tend to focus either only on presence detection of a small set of distinct tags over a
reasonably large range, which is the case with shoplifting-detection systems, or only on
detection of a large number of tags somewhat slowly and not so cheaply, which is the case
with chip tags, which have to use expensively manufactured onboard chips with time-
consuming anticollision protocols designed to prevent multiple tags from interfering with
each other during operation. Essentially no chip tags are optimized specifically for real-
time, continuous tabletop interaction between a user, an array of several distinct physical
objects, and a computer.
For the purposes of this paper, we have explored the use of magnetically resonant
(magnetostrictor and LC) tags in conjunction with two tag reading principles, pulsed
ringdown and continuous swept frequency, in order to develop an efficient and inexpensive
tag reader easily applicable to a wide variety of tangible interfaces.
The next two chapters introduce the two families of tag readers that we have
developed and the following two chapters present example applications and extensions.

12
Chapter 2
Ringdown tag readers

The first family of tag readers we examined was the “ringdown” tag reader. This
type of magnetically resonant tag reader works on a principle similar to that used in pulse-
induction metal detectors10 and nuclear magnetic resonance imaging systems.11 This
family of tag readers, already in fairly common usage in EAS systems, takes advantage of
the resonance properties of high-Q magnetostrictor and LC tags in order to detect them.
The basic operation of such tag readers can be described in three steps, as
diagrammed in Figure 2. First, a magnetic pulse is transmitted by the reader at the
resonant frequency of the desired tag. Next, the tag, upon exposure to this pulse, produces

Figure 2. Basic block diagram of the ringdown tag reader.

13
its own mechanical or electronic resonant response, which will continue for a short while
as the tag damps out after the transmitted pulse stops. Finally, because the tag produces
its own weak resonant signal, the magnetic drive at the reader can be turned off once the
tag rings up, and then the reader “listens” for this small but detectable response.
The response of the tag, which does not cut off immediately but rather decreases
exponentially over time, is described as a “ringdown,” and because these systems listen for
this ringdown response, they are called ringdown tag readers. Ringdown tag readers have
a number of advantages. First, they have a good range; the transmit pulse can be very
powerful and so can easily stimulate tags across a range of several meters. For example,
common EAS systems can detect tags up to three meters away. 12 Furthermore, the
configuration can be fairly flexible; one transmit antenna can broadcast the signal, and then
either the same coil can be used for detection, or other antennas can also be used to listen
for the response from various locations. Also, the design of these circuits is fairly
straightforward. Finally, the time required to detect one tag can be made relatively short,
with ring-up times within a millisecond or so, compared to the more complex chip-based
RFID systems, which typically require more time to power up CMOS chips and transmit
back ID’s. These advantages led us to design and examine ringdown tag readers.
Enabling a ringdown reader to have continuous real-time interaction with multiple tags
was the first step.

2.1 Design of a continuous real-time ringdown tag reader

The ringdown reader has two responsibilities; it must transmit a magnetic pulse
(“ping”), and then it must stop that pulse and listen for a response from a tag. However, in
order to satisfy our requirements for a tangible interface backend, several additional
specifications must be added to the list.
First, we would like to detect multiple tags. For each tag that we wish to detect,
this requires that we transmit a magnetic pulse at the tag’s resonant frequency. Second, we

14
must design the receive circuitry to detect a ringdown signal at the frequency of each of the
tags. Furthermore, in order to make the system continuous and real-time, the reader must
be able to distinguish the magnitude of the ringdown signal, and it must be able to perform
the transmit-receive cycle fairly quickly.

2.2. Implementation of ringdown reader transmit circuitry

The first goal was to be able to transmit a pulse for the detection of multiple tags at
various resonant frequencies. Two approaches were considered for this objective. First,
the transmit antenna could simply be wired to transmit a burst of broadband noise, which
has components at all frequencies of interest. This noise would stimulate all tags within
range and provoke ringdown responses from all the tags at once. Second, the transmit
antenna could cycle through a known list of resonant frequencies, pinging briefly at each
one and listening for a response.
Despite the relative simplicity of the broadband noise approach, we opted not to
use it because the transmission would require a large amount of power in order to provide
sufficient excitation at all relevant frequencies. Furthermore, the receiver would then be
somewhat complex; the simultaneous ringdown response of all the tags at once would
create the need for a real-time Fourier transform in order to determine which tags were
responding at the relevant frequencies.
Once it was determined that pinging sequentially on multiple known frequencies
would be more straightforward to implement, the issue became finding an efficient means
with which to do so. Generating and transmitting a relatively powerful signal at a given set
of frequencies was not prohibitively difficult. As seen in the schematics provided in the
appendix, we used a voltage-controlled oscillator to provide the signal, and the
amplification of the signal was done using a pair of power MOSFET’s that switched the
drive coils across the 40-volt supply rails.

15
 One of the difficult design issues was finding a way to drive a similarly large
amount of current through the coil setup at all the relevant frequencies, in order to produce
a reasonably large magnetic field. Because the transmit coil would have its own resonant
curve, frequencies further from the coil’s resonance would tend to produce weaker
transmitted signals. The apparent solution, then, was to change the resonant frequency of
the coil dynamically, tuning it to match the frequency of the ping being transmitted.
The resonant frequency of the coil assembly, like all inductor-capacitor resonant
1
circuits, would be given by f = , where L is the inductance of the coil and C is
2π LC
the series capacitance. The coil itself, which represented the inductive load of the
resonance, could not be changed, and thus the coil we used had a constant inductance
around 120 uH. However, the capacitors connected to the coil could be electrically
switched in and out, and this would enable real-time adjustment of the coil’s resonance.
The two possibilities were 1) to switch continuously and quickly between two capacitors,
allowing each one to have a partial impact on the total resonance of the system, as in a
switched-capacitor filter,13 or 2) to find some way to switch discretely between several
capacitors. The first approach would have the advantage of allowing any resonant
frequency to be attained between the values of the two attached capacitors, by adjusting the
switching duty cycle. However, it proved to be inefficient because of losses in signal due
to the continuous, high-voltage, high-frequency switching in the power MOSFET’s that we
used. This resulted in a low Q, meaning that the coil ended up transmitting too
inefficiently.
As a result, we implemented the second approach instead, using a binary-weighted
ladder of six sensitive-gate triacs and capacitors, which enabled us to switch the high-
power AC signals necessary for transmission using simple logic level signals. As seen in
the schematic given in Appendix 1, each triac was connected to capacitors of different
values, and thus the capacitance of the coil circuitry could be switched between various
values, enabling a discrete shift in resonance for pinging more efficiently, and thus more
powerfully, at different frequencies. Thus, when generating pings, the transmit system

16
first activated the proper triacs, and hence switched the proper capacitors into place, and
then the drive signal at the respective frequency was sent to the amplifier and coil. Each
ping was generated for about 1.5 milliseconds, and then the transmit circuit was rapidly
damped in order to silence it and listen for a ringdown response. A circuit board with this
transmit circuit, including the MOSFET drivers, the front-end receive amplifiers, the triacs
and all the capacitors, can be seen in Figure 3.
Having selected a transmit circuit design that was optimized to ping efficiently at a
set of discrete frequencies, we then had to select a set of tags that would respond well and
be reasonably optimized for detection by a receive circuit with our specifications.

2.3. Selection of tags

Once the transmit circuitry was designed, our next step was to decide on a type of
tag to use for detection. The choice was between magnetostrictor tags and LC tags, which
have differences in response power, resonance strength, and ease of tuning.
Magnetostrictor tags have a higher Q value than LC tags. This means that in
general magnetostrictors have lower losses and a sharper frequency response, hence they

Figure 3. Triac and capacitor ladder board with driver and front-end circuitry.

17
store more energy at their resonant frequency. This in turn produces a longer, more
powerful ringdown, which is easier to detect. LC tags, on the other hand, have a lower Q
value, meaning they are less sharply tuned and more lossy, so that they respond a little bit
more at frequencies near but not precisely at their resonance, but at resonance they are
weaker. The Q value in LC tags is mainly determined by the series resistance of the wire
in the circuit, which is not always easy to control.
Another factor is ease of tuning. LC tags have the advantage here because
inductors and capacitors for these tags can be selected in a multitude of values on any
electronics supply shelf. This means that LC tags can be created to resonate at any
arbitrary frequency. Magnetostrictors, in contrast, need to be tuned by cutting the metallic
strip to a different length, in order to affect its mechanically-based resonance. This cutting
process is fairly imprecise and leads to difficulties in tuning to specific frequencies as well
as a potentially significant loss in Q due to mechanical damping in a suboptimal package,
which nullifies the initial advantage of using magnetostrictors to begin with.
For the purposes of implementing our system, we opted to use magnetostrictors
because of their powerful responses. The ability to tune tags would play a bigger role in a
system with more tags, but for a simple test with few tags, we decided that
magnetostrictors would produce better responses, which would make the abilities of our
system clearer.
The next step, then, was to create a receive circuit for our reader that could detect
these tags after excitation.

2.4. Implementation of ringdown reader receive circuitry

Given that the transmit circuitry transmitted pings on a specific set of frequencies,
our receive system had to be designed to listen for a signal returned from the tags at their
respective resonant frequencies. Furthermore, in order to meet our additional

18
specifications, the system had to be able to determine the magnitude of a returned signal, in
order to produce proximity information for the tags.
Once the transmit system finished its ping, any tags in range of the coil would be
starting their ringdown responses. The entire receive section of the reader was disabled for
the duration of the transmit ping, and continued to stay off for between 100 and 350
microseconds from the end of the ping in order to allow the transmit system to quiet down.
After this, the receive circuit was activated and began processing the incoming signal from
the coil. Any signal detected by the coil had to be amplified so it could be used; we used a
differential amplifier across the coil for this task. As the driver circuit (with the resonator
capacitors) was effectively disconnected from the coil via a transmit/receive (T/R) diode
switch, the receiver front-end response was broadband. Accordingly, the frequency of the
receive signal had to be identified, to determine whether it was identical to the frequency
sent out in the ping. Finally, the total power of the response had to be determined in order
to infer the proximity of the tag to the coil.
As seen in the schematics in Appendix 1, amplification of the signal was
accomplished using a series of filters and operational amplifiers, producing an output that
could then be sent to a quadrature demodulator in order to bring the received signal
directly to baseband. This had the effect of determining the magnitude of the received
signal that was in-phase with the original ping signal, and also the magnitude of the
received signal that was ninety degrees out-of-phase with the original ping signal. This
was necessary because the ringdown responses of the tags could be subject to dynamic and
unpredictable phase shifts due to variations in the tags’ resonant peaks from small
mechanical differences between tag packages.
The demodulated signal was quite narrowband, i.e. it contained mainly the
component of the received signal that corresponded to the expected transmit frequency; out
of band noise was removed by simple low-pass filters following the demodulators. The
next step, finding the total power of the received signal, could be accomplished by taking
the square of each of the quadrature components and then adding them. This was all done
in hardware via an analog multiplier the squared the baseband signals, followed by a

19
 Figure 4. The completed functioning prototype ringdown reader setup.

summer that added them. A gated integrator accumulated the power signal for 10
milliseconds after the transmit ping, producing a single voltage whose magnitude
corresponded to the net magnetic coupling, and hence proximity and inclination, of a
single tag relative to the coil. Between the 1.5 millisecond ping time and this 10
millisecond accumulation time, along with the wait times in between states, the completed
system thus could take around 12 milliseconds in the best case to finish detecting one tag.
With the receive circuitry in place, we were able to finish off the system by taking
measurements, setting up a system to digitize the integrator’s output voltage, and
developing a simple demonstration application to test its abilities. A picture of the
completed system on our lab bench is shown in Figure 4. A series of voltage traces
showing the performance of the system is shown in Figure 5.

20
 Figure 5. Signals from completed ringdown reader system.

2.5. Ringdown reader demo application and results

In order to assess the capabilities of the system, a standard National Instruments

computer data-acquisition board was used to allow a computer to interface with the reader
and receive the finished signals coming from each tag detection. A simple Visual Basic
application was written, in which three tags were represented by bars on a constantly-
updating bar graph display. The height of each bar corresponded to the proximity of the
tag to the coil.
The reader was then configured to search at the resonant frequencies of three
magnetostrictor tags. By cycling sequentially between the three frequencies, the reader
provided constantly updating information about the proximity of each of the three tags.
The computer then received this information and updated the bar graphs.

21
 The end result was reasonably encouraging. The bar graphs corresponded
accurately to the proximity of the three tags. Each individual signal tended to be affected
by noise, but the time-average of each signal was in very good correspondence to
proximity, detecting magnetostrictor tags up to 1.5 feet away using a single 1-foot diameter
reader coil. However, the speed of the system was not quite fast enough to enable a real-
time interface; because our system performed one read in about a tenth of a second,
including delays for DAQ card commands, tuning the reader, and software overhead, it
took slightly less than a half second to update all three tags.

2.6. Ringdown reader conclusions

In the end, the ringdown tag reader could successfully detect high-Q magnetically-
coupled resonators such as magnetostrictor tags at different resonant frequencies and infer
the proximity of a tag to the detect coil, provided the tag was oriented correctly to couple
to the coil’s magnetic field. In addition, the power that can be transmitted through the
reader coil can be fairly large and thus the detect range can be relatively good. Because the
tuning of the coil can be changed dynamically, the reader is also effective at transmitting
efficiently over a range of frequencies, making it possible to detect multiple tags without
suffering from static, unchanging coil properties.
The timing in our system, while reasonable, presented several problems for real-
time interaction with multiple tags. Because of considerations such as the interval needed
for a single tag to be excited and ring down, detecting each tag took about 12 milliseconds.
However, in our system we took approximately three reads per tag in order to compensate
for the asynchronous nature of the connection between the data acquisition system on our
computer and the tag reader, as well as for the time necessary for each section of the circuit
to settle into a new frequency setting. Taking a time average measurement over multiple
reads also led to improved accuracy, but also required correspondingly longer intervals.

22
 Several disadvantages of this system became quickly evident as a result. First, the
system would be difficult to scale to larger numbers of tags. Supposing that we can
achieve the feeling of real-time interaction using about twenty detections per tag per
second, ten distinct tags would already require about two hundred reads per second. Even
using an optimized reader with a decreased detect time, the intrinsic timing considerations
inherent to this technique would make it impossible for it to update at this rate.
Furthermore, each tag’s frequency has to be preprogrammed into the hardware of
this system. The capacitors in the capacitor ladder have to be chosen to enable retuning of
the coil assembly to the proper frequency for each frequency range, and the hardware itself
must ping at the specific resonant frequency of each tag. This leads to more difficulties
because of the possibility of frequency drift and inaccuracies, both in the tags and in the
reader. Our device was especially prone to this because we used an open-loop analog
voltage-controlled oscillator.
The frequency limitations of magnetostrictor tags were another issue. Although
they can be manufactured by today’s vendors (e.g. Sensormatic Corp.) at frequencies
ranging from 55 to 65 kHz, they are commonly available only at 58 kHz (EAS tags) and 60
kHz (for industrial applications).14 One can manually cut the magnetostrictor strip inside a
tag package or change its magnetic backing to adjust its frequency, but these techniques
are imprecise and awkward at best, presenting an additional complication in coordinating
the various frequency-dependent sections of the system. The maintenance of such a setup,
given ten or more tags, could then be nearly untenable.
Therefore we concluded that although the ringdown reader has an advantageous
detect range and good power transmission abilities, we needed to examine other
approaches for our real-time, multiple-tag detection purposes before simply trying to
further optimize the ringdown reader for our tangible interface applications. Concepts of
the sort that we applied to our reader have been used in subsequent work by Fletcher et
al.15 However, for our purposes a system that could quickly read multiple tags without
prior knowledge of their precise resonant frequencies would be more suitable for our
purposes, and to this end we began examining tag reader designs that continuously

23
“sweep” through a given frequency range, instead of pinging on multiple frequencies
discretely.

24
Chapter 3
Swept-frequency tag readers

Because our prototype ringdown tag reader proved difficult to adapt to larger
numbers of tags, we began examining another approach to reading tags. The concept
behind swept-frequency tag reading involves passing through an entire range of
frequencies in one continuous sweep. This is in contrast to the ringdown reader, which
pings at separate, discrete frequencies. As a result, the swept-frequency tag reader uses a
different set of operational principles, places other constraints on the system architecture,
and as a result has numerous advantages and disadvantages over the ringdown reader.
When a resonant tag receives a signal at its resonant frequency, its circuit draws
upon the energy of the reader coil’s transient field to produce its sympathetic oscillations.
A ringdown reader would listen for the decreasing oscillation after excitation in order to
detect the tag. However, a swept-frequency tag reader uses a different principle: When a
tag enters the reader’s field and is exposed to a magnetically-coupled signal at its resonant
frequency, it pulls current from the reader coil. This energy drawn from the field causes a
slight, brief, but detectable change in the perceived properties (e.g. inductance) of the coil,
which manifests as a dip in the voltage or current being passed through the coil.
A coil transmitting at a constant frequency will only see a change in the coil
impedance as a tag with a resonance at that frequency is brought into the field; the further
away the tag, the smaller the tag’s response will be. Tags resonant at other frequencies
will have little to no impact on the coil impedance. By sweeping the frequency instead of
leaving it constant, however, any tags in the field produce sharp alterations to the coil
properties when the sweep reaches each of their resonances. These changes can be easily
emphasized and detected via simple analog signal processing.

25
 As a result, a tag can be clearly detected by sweeping through a range that includes
its frequency. This method of tag detection has also been applied to several commercial
EAS systems that use LC tags.16 Most of these systems take advantage of the range of the
sweep not to detect multiple tags, since this is currently irrelevant to EAS applications, but
rather to robustly detect tags within a small range while tolerating variance in the
resonance of manufactured tags. Unlike ringdown readers, the swept-frequency reader that
we’ve designed directly measures the properties of the transmit coil for its detection and so
cannot use separate coils for transmit and receive, although other designs can use separate
coils to monitor disturbances in the transmit field. Swept-frequency reader ranges also
tend to be less than those of ringdown readers, because the tag has to alter the read field
itself noticeably, and not just produce its own clear signal without background from the
reader. Likewise, since the frequency is continuously sweeping, the dwell time at any
particular tag’s frequency is limited.
These disadvantages can be easily outweighed by the increase in the frequency
tolerance of the system over ringdown readers, and so swept-frequency tag readers have
their uses. Specifically for the purposes of detecting multiple tags, an increase in the
sweep range to encompass a whole range of tags can bring about the scalability to
detecting larger numbers of tags in real-time, which our ringdown reader was unable to
produce.

3.1. Design of a continuous real-time swept-frequency tag

reader

In order to adapt the swept-frequency tag reading concept to the tangible interface,
several constraints must be met. The system must allow for multiple tags; it must be able
to determine the continuous proximity of the tag relative to the coil; and it must perform its
reading operations quickly enough to enable real-time interaction.

26
 First, the sweep range of the signal being transmitted must span the frequencies of
all the relevant tags. In order to detect multiple tags, the receive circuit then needs to be
able to measure and distinguish all of the frequencies at which sharp changes in the coil
impedance occur, as opposed to just measuring whether or not there is a change at one
point in the sweep range. Once this is done, the time during each sweep that a tag is
detected provides information about which tag is in the field, essentially converting
frequency properties of the tag into timing data.
Next, the system must be able to determine the proximity of the tag relative to the
detect coil. The effect of a tag on the coil’s measured impedance varies with how strongly
the tag couples to the field, as well as how quickly the sweep occurs and how much energy
the tag draws from the field. If the sweep rate is fast enough such that the tag appears to be
relatively unmoved during any given sweep, we can minimize the effect of the tag’s
motion on the magnitude of the impedance change. With a fixed sweep rate, the
magnitude of this impedance shift is determined by the amount of energy being coupled
into the tag to pump its resonance. This in turn is a function of the orientation of the tag
and the distance of the tag relative to the coil, both of which determine the coupling of the
tag’s inductive component to the field of the coil. Thus we can design the system such that
there is a relation between the proximity and orientation of a tag and the magnitude of the
change in the received signal.
Finally, the reader must be able to run the sweep quickly enough that interactions
appear to be real-time. This would involve reading all tags in the system at least ten times
per second or higher, depending on the application. Thanks to the nature of the swept-
frequency tag reader, each sweep should be able to detect all tags in the frequency range,
so increasing the number of tags in the system does not increase the time required to sense
those tags. Furthermore, single tags are now detected as soon as the sweep crosses the
tag’s frequency, since the time taken to listen for the tag’s ringdown is no longer
necessary.
The first step, then, is to implement a working transmit circuit that generates the
proper sweep.

27
 Figure 6. Block diagram of swept-frequency tag reader.

3.2. Implementation of swept-frequency transmit circuitry

In order to sweep through the necessary frequency range, the transmit circuit of the
tag reader must be able to produce an appropriate sweep signal. Magnetostrictor tags tend
to resonate in the range near 60 kHz, and so we planned our transmit circuitry to sweep
from slightly below this frequency to higher frequencies, which are attainable by trimming
the magnetostrictor tags to different lengths or by making simple LC tags with appropriate
off-the-shelf inductors and capacitors. Our planned frequency range eventually settled to a
low of near 50 kHz to a high of near 400 kHz.
In order to produce the sweep signal, we used a voltage-controlled sine wave
oscillator. This chip takes as input an analog voltage, which determines the frequency of a

28
sine wave that it outputs. The input to the oscillator came from a circuit that produces a
linear ramp signal. This signal simply varied linearly from one user-adjustable voltage to
another, and thirty times a second it would be reset to the original voltage. The ramp itself
was generated using a 555 timer chip and a single-transistor current source.17 The
essentials of this system can be seen in the upper-right section of Figure 6.
Two factors led to modifications to the input ramp over several revisions of the
system. First, it was eventually found that the ringdown of the magnetostrictor tags led to
extra artifacts in the detected signals for short periods of time after the frequency of the tag
had already been passed. We surmised that this was due to nonlinearities in the
magnetostrictor responses. By using LC tags, which have smaller ringdown responses, at
the higher frequencies, these false responses could be isolated to the lower end of our tag
frequency spectrum where the magnetostrictors resided. A fortuitous result from reversing
the sweep to start at a higher frequency and move to a lower frequency also led to these
false artifacts having less effect on the responses from other tags by allowing some of the
artifacts to take place after the end of the sweep.
Next, the responses of tags at lower frequencies tended to be weaker than the
responses of tags at higher frequencies. We hypothesized that this was because a linear
sweep spent the same amount of time at a given high frequency that it did at a given lower
frequency, but in any given time period there are fewer cycles of a low-frequency signal
than there are of a high-frequency one. The solution, then, was to switch to using an
exponential sweep, i.e. a sweep that varied less quickly at lower frequencies than at high
frequencies. By spending more time sweeping at lower frequencies, more cycles of those
low frequencies were produced, giving the tags a more adequate time to respond to the
field. The exponential voltage driving the voltage-controlled oscillator was attained by
charging a capacitor when the 555 went high and letting it decay as a normal RC circuit.
The end result in our final system was an exponential sweep from high to low
frequencies. The resulting sine wave was then passed through a pair of fast operational
amplifiers, buffered through push-pull power transistor drivers. These drove the transmit

29
coil differentially, at up to 40 volts peak-to-peak, producing the swept-frequency magnetic
field used to detect the tags.
Once the sweep was being transmitted properly, the tags to be used for this system
had to be selected.

3.3. Selection of tags for swept-frequency tag reading

As with the ringdown tag reader, the decision to choose one type of tag over the
other also came up with the swept-frequency tag reader. Because the two systems use
different principles for tag detection, the factors to consider were slightly different.
In a swept-frequency system, the Q value of each tag takes on a different
significance; the sharpness of the resonance of the tag now determines how much
frequency space in the sweep range the tag occupies, in addition to the strength of the
response. Also, ringdown can now become a detriment, because the time spent in one
tag’s ringdown can interfere slightly with the measurements of other tags that come soon
afterwards in the sweep, especially if that tag has nonlinear properties that can produce
multiple frequency peaks.
The magnetostrictor tags, with their sharp frequency responses and high Q value,
thus took on the advantage of having smaller footprints in the sweep range, which
produced the possibility of fitting more tags into any given sweep. However, their strong
ringdowns led to interference between tags, and the difficulty involved in tuning them
became a much greater issue now, given that the swept-frequency system would be able to
read far more tags in real-time than the ringdown reader.
LC tags, on the other hand, with their lower Q values, took up more space in the
frequency sweep, but their responses were still thoroughly detectable; in fact, by selecting
larger inductors, the tags could couple more effectively to the field of the coil, an effect
that helped significantly here, since our sensitivity was somewhat lower than for ringdown
tag -reading. The flexibility afforded by being able to select the components of each tag

30
also meant far easier tuning. The relative weakness of the ringdown furthermore decreased
interference levels between tags, although, with our analog signal processing, a strongly
interacting tag could still suppress the sensitivity slightly to tags at adjacent frequencies.
However, fewer LC tags could be fit into any given range due to the width of their
responses.
In the end, LC tags provided the most convenient benefits for the purposes of
swept-frequency tag reading. However, one or two magnetostrictor tags were also used in
the system, left at their original, relatively low frequencies, in order to avoid difficulties
associated with tuning them. As discussed in the previous section, adjusting the sweep to
hit lower frequencies at the end of the sweep rather than at the beginning allowed the
lingering ringdown responses of the magnetostrictor tags to be less of an issue. Having
magnetostrictor tags present in the system made it easy to demonstrate that they could be
sensed just like LC tags, only with slight complications.
Outside of the few magnetostrictor tags, the remainder of the twenty or so tags
eventually used in the system were LC tags, preferred for their ease of tuning as well as the
ability of larger inductors to receive more flux from the coil, increasing the response. Note
that at these relatively low frequencies, the LC tags were three-dimensional, bulky objects,
due to the large inductor coils, as opposed to the larger-area, flat, higher-frequency tags
shown in Figure 1.
Having specified these types of tags, we then looked to implementing the detect
circuitry for this system.

3.4. Implementation of swept-frequency receive circuitry

On the receive end of the swept-frequency tag reader, several things must be
accomplished. First, the slight shift in the coil properties caused by the presence of a tag at
a particular frequency must be detected. Next, this must be separated from the oscillations
of the original sine wave sent through the coil and processed into a signal that simply

31
indicates the presence or absence of a tag. Finally, this signal must be translated into
frequency and proximity information and sent to a computer for further processing.
As mentioned previously, when the resonance of a tag in the field matches the
frequency being transmitted through the coil, the tag will draw energy from the field. This
leads to a temporary change in the inductance of the coil, which can be translated into
voltage changes using an inductive bridge circuit. The bridge circuit is an assembly of
four matched inductors (see Figure 6), of which the sense coil is one. The purpose of a
bridge circuit is to provide the first step in measuring the difference in a given property of
some kind of sensor, relative to a reference value;18 the voltage across the bridge is
proportional to the difference in current flowing down its two legs. Thus, the other
inductors in this bridge circuit serve as references, relative to which the small perturbations
to the inductance of the read coil can be measured. When the sweep signal is sent through
the coil, changes in the inductance of the coil become changes in the current flowing
through that leg, which in turn become changes in the voltage across the bridge. We then
used a differential amplifier as the first stage to convert the voltage difference across the
bridge into an absolute voltage relative to ground. From here, the signal is passed through
a synchronous demodulator in order to bring the bridge signal down to DC (removing the
sinusoidal carrier) and filter out any noise components at frequencies other than the
transmit frequency. After tuning the sweep and nulling the bridge, the result was a smooth
signal with small discontinuities that reflected the presence and proximity of a tag. This
signal was then passed through a series of various high-pass and low-pass op-amp filters,
which enhanced the small bumps and reduced the slow baseline shift (high-pass), and
attenuated noise (low-pass). The resulting output came as close as possible to staying near
zero in the absence of a tag and rising towards five volts at points in the sweep at which
corresponding tags are introduced. The specific sequence of filters can be seen in the
schematics given in the Appendix.

32
 Figure 7. Sample oscilloscope plots of the swept-frequency tag reader signal after the
completion of analog processing. Names like “finger tags” and “goblin” refer to objects in
which tags have been embedded to demonstrate the system. The large static signals
framing the sweep interval signify the boundaries of the sweep in time.

This signal after the completion of filtering can be seen in the oscilloscope plots of
Figure 7. This is essentially a frequency analysis showing the response of all resonant
objects that approach the reader coil, and could be easily processed using an embedded
microprocessor. Thus, we added an onboard PIC microcontroller to measure the voltages

33
 Figure 8. An early prototype swept-frequency tag reader.

of the finished signal and send magnitudes and timing information after each sweep via a
serial link to a computer.
An early prototype reader with oscilloscope plots is shown in Figure 8. This
system used a coil underneath a Lego board and bricks embedded with tags to demonstrate
the basic abilities of a swept-frequency tag reader for applications like interactive games.
A later prototype, with improved signal conditioning, is shown with prototype reader
boards, coil, oscilloscope plots, and objects embedded with tags in Figure 9; this device
was used to develop and demonstrate our initial “Musical Trinkets” application discussed
in the next chapter. Figure 10 shows a more finalized version of the circuit board, together
with tags and scope traces, that we have used for later versions of the Musical Trinkets and
subsequent applications.

3.5. Swept-frequency tag reader results and refinements

As described, we have produced several versions of the swept-frequency tag reader

over the past few years, starting from an early prototype reader with very little signal
conditioning, up to our current printed circuit board, with selectable sweep types and many

34
 Figure 9. Second prototype of swept-frequency tag reader system.

adjustments. The key advantage common to these readers is that they all can run a sweep
30 times a second, detecting every tag present in the sweep’s frequency range and
measuring its parameters on each sweep. Because this rate is not affected by the number
of tags present in the field, the swept-frequency tag readers demonstrated their superiority
for the purposes of real-time multiple tag detection at even their earliest stages.
Depending on the size of the tags, the detection range of our swept-frequency tag
reader could extend to a little more than one foot with good reliability. Although not large,
this was comparable to the read range of our ringdown reader. Note, however, that the
ringdown reader could, in principle, easily be driven with even higher voltages, which
combined with its dynamically-tuned antenna would then lead to more improvements in
read distance. However, one foot is perfectly acceptable for the desktop-scale interaction
with tagged objects that we are exploring.

35
 Figure 10. A more mature tag reader. Electronics have been incorporated into a single
board, and the ring for the coil was made specifically for this purpose.

The jitter level of our system is quite good across our sensitive region. For
example, a typically strong tag was placed in the reader field and held steady for a
thousand samples. The PIC measured the integral of its signal and reported a mean of
1685 in 8-bit unscaled units, with a standard deviation of 2.13. This is an average jitter
level of around 0.126%, which is easily precise enough for our purposes. Typically,
additional background noise is caused by devices that release magnetic radiation in this
frequency range, which apparently include CRT monitors and other similar devices.
Keeping these noise sources at least six feet away allows the tag reader to retain its usual
precision. A commercial implementation of such a system would use higher frequencies,
ideally moved away from the dominant sources of environmental noise (again, primarily
CRT monitors). These measurements also compare favorably to the ringdown tag readers,
which required time-averaging on their proximity measurements because of phase-related
difficulties in reading tag responses.
One of the other concerns with the ringdown tag reader was the drifting of
frequencies, both within the reader and in the tags themselves. Because of the discrete
nature of the ringdown tag reader’s detection method, any significant drift in the
frequencies being read quickly made the system inoperable. In contrast, the swept-

36
 Figure 11. Plot of frequencies at ten sample points during each sweep for a period of 12
hours. Very little fluctuation occurred beyond rounding error. f1-f10 are variables referred to
in Appendix B for frequency drift compensation.

frequency tag reader, because it continuously sweeps through a frequency range, ensures
that frequency drift has a limited effect on tag detection. However, given enough drift it
can become difficult to distinguish one tag from another, as one tag’s perceived resonance
will overlap the expected resonance of a neighboring tag.
Measurements of voltage being input to the voltage-controlled oscillator, as well as
the swept-frequency signal coming out from the oscillator, were taken using the analog-to-
digital converter section and the external clock/counter of the PIC, respectively. Timing
was monitored using the crystal-locked clock of the microcontroller. The host computer
was set up to acquire these parameters periodically and to record them over a period of
time in order to determine the frequency drift of the tag reader. Apart from taking these
measurements in order to understand the sensitivity of the system, these data can also be
used in conjunction with algorithms running in the host computer, while the tag reader is in

37
 Figure 12. Frequencies sampled at ten points during each sweep while the tag reader was
subjected to a heat gun at two different settings, over a period of ten minutes. Notice that on
the right side the VCO has still not recovered to its original values on the left; several hours
are required for full recovery.

use, to dynamically compensate for such drifting effects and ultimately would allow the
system to run with great stability for an essentially unlimited time. The equations
necessary to implement such an algorithm are given in Appendix 2.
The measurements themselves provided good insight into the sensitivity of the tag
reader. At first glance, we had assumed that time and temperature were the primary
contributors to frequency drift in this system. However, monitoring the frequency range of
the reader over a twelve-hour period, with the temperature held nearly constant at 75
degrees Fahrenheit by the Media Lab’s central-air system, produced Figure 11, which
shows less than a 2-kHz variation from average frequency at any part of the sweep over the
course of an entire night. On the other hand, Figure 12 shows the same frequencies plotted
over a short period as the ambient temperature around the reader increases at a moderate
pace up to 150 degrees Fahrenheit, in order to determine its sensitivity to temperature
38
 Figure 13. Interpolated frequency curves during single tag-reader sweeps. Left: curves taken
during twelve hours of constant temperature, all plotted atop each other. Right: curves taken
during ten minutes of temperature change, plotted over each other.

under conditions similar to those used in the “Musical Trinkets” demo (described below).
The frequencies dropped dramatically, and the voltage input to the oscillator increased as
well. Further measurements determined that the voltage input to the voltage-controlled
oscillator returned to its initial value as the temperature returned to normal, but that the
frequency output of the oscillator took many hours to settle back towards its initial state.
In total, this leaves the system stability fairly sensitive to increases in temperature; Figure
13 shows the frequency curve for each sweep over twelve hours of exposure to steady
temperatures, compared to the curve for each sweep during exposure to heat. Finally,
some frequency drift was expected with the starting up of the system, but under stable
temperature conditions, frequencies decreased by less than 2 kHz between system start and
equilibrium, which while possibly significant, is small enough to fall into measurement
rounding error.
Even without the algorithms in place to correct for frequency drift, we expect that
when placed in an environment with relatively constant temperature, it will be stable
enough for most applications. With the algorithms to correct for the drift, the swept-
frequency tag reader will have little difficulty maintaining itself in any reasonable
demonstration environment for extended periods of time.

39
3.6. Swept-frequency tag reader conclusions

After numerous revisions and refinements of our swept-frequency tag reader

system, we have demonstrated that swept-frequency tag readers have sufficient stability
and robustness to drive real-time tangible interface applications. The drift and jitter levels
are not by any measure excessive for our intended purposes. Their superior ability to read
multiple tags without compromising their timing, the precision with which they can
measure tag proximity to the coil, and their stability all help to make this an ideal candidate
for further exploration. Because of our decision to use the swept-frequency tag reader in
our subsequent work, applications of the reader are described in a separate chapter.

40
Chapter 4
Applications of the Swept-Frequency Tag
Reader

Further work with the swept-frequency tag reader involved the creation of
applications to explore and demonstrate its capabilities. Music- and graphic-based systems
have generally proven to be a good way to demonstrate new sensing capabilities, as well as
to draw the attention of an audience to sensing projects in order to explore their utility at
public venues. As a result, our applications took this direction for the purposes of
demonstration. Several video files of these projects in action can be found linked off our
group’s web page (http://www.media.mit.edu/resenv). Also, the reader and applications
have been described in several papers published by our group.19,20,21

4.1. Early applications

The early prototype reader shown in Figure 7 provided a simple but effective
demonstration of this system. The coil of the reader was placed underneath the Lego board
shown in the picture, and several Lego bricks were embedded with both LC and
magnetostrictor tags. Although at this stage the setup was still a rough prototype and had
not been integrated with a computer and thus into a more complex system, the hardware
was able to generate tones when tags were detected, with the pitches corresponding to the
frequencies of the tags.
This simple output was effective in demonstrating the capability of the system to
detect and distinguish between tags, both LC and magnetostrictor. One particular object
had three tags placed within it, oriented along the three orthogonal axes, so that its

41
inclination could also be sensed relative to the coil; this was made possible because the
signal strength from a given tag always corresponds to both distance and orientation,
where increasing distance and rotation of the tag’s inductor out of the axis of the field lines
both decrease the detected signal. The other tag had a slider that continuously perturbed
the resonant frequency of a magnetostrictor by moving a small magnet placed behind it,
hence producing a glissando sound as it was pressed. Both concepts, that of using three
orthogonally-placed tags within the same object to detect the orientation of the object, and
that of adjusting a tag while it was in the field, were appealing for the purposes of
demonstration and were thus used in later applications as well.
Another application of a more mature revision of the swept-frequency tag reader
was produced in 1999 by Hiroshi Ishii’s Tangible Media Group, also at the MIT Media
Lab. Their musicalBottles project,22 in which three bottles on a table each represented a
different line of music, was an effective demonstration of the tag reader as a real-time
tangible interface backend, and has been widely exhibited. Figure 14 shows a picture of
the musicalBottles project in action. As each bottle was uncorked, a corresponding line of
music from Edouard Lalo’s “Piano Trio in C Minor” was played. The feeling produced by
this system was that a user was “releasing” the music from the bottle by opening it. The
mechanics behind the system were fairly straightforward; each bottle had a tag inside it,
and each cork contained a ferrite that altered the inductance of the tag, thus changing its
resonant frequency. The shift in resonance could easily be detected by the swept-
frequency tag reader, which reported this information to a computer for processing.

42
 Figure 14. The Tangible Media Group’s musicalBottles project. Opening
a cork produces a line of music.

The other application for the swept-frequency tag reader produced entirely by our
group (Responsive Environments), eventually came to be called “Musical Trinkets.” It has
a much longer and more involved history, summarized below.

4.2. Musical Trinkets

After numerous revisions, the swept-frequency tag reader evolved into a single
printed circuit-board, so it would be easier to reproduce and apply to various projects. In
order to demonstrate the abilities of this near-final version, we wrote a more elaborate
musical application involving a variety of some sixteen tagged objects.

43
 Figure 15. Musical Trinkets, with tag reader, read coil, and objects.

Figure 15 depicts the visual appearance of a fairly mature version of the music-only
project. The tag reader was used to identify and sense the proximity (coupled with
orientation) of the twenty-or-so tags. This information was gathered by the onboard PIC
processor and sent via serial stream to a laptop PC running the music application. The PC
sent music commands via MIDI to an Emu ClassicKeys synthesizer, which produced the
musical tones that comprised the music. Other MIDI commands were sent to a Lexicon
effects processor, which produced various distortions of the output sound.
The basic structure of the music was centered around a melody line and a harmonic
chord progression. Five tagged rings worn on the user’s fingers allowed the orientation of
five fingers relative to the coil to be detected. When the fingers were placed just above the

44
reader coil and rotated to point downwards, melodic notes would be triggered, with a
specific major-scale note corresponding to each ring. The effect produced was that of
playing a rudimentary keyboard in midair, using one’s fingers held somewhere over the
tabletop. The speed with which a ring’s signal increased, corresponding to the rapidity
with which the finger was turned towards the coil, adjusted the MIDI velocity (i.e. volume)
of a triggered note. This served to demonstrate the system’s ability to determine signal
strength, and the ability to play multiple notes at once by inserting multiple rings also
demonstrated the reader’s ability to read all the tags in real-time without any compromise
in timing.
The harmonic chords were controlled by three tagged objects, which happened to
be small goblins colored red, blue, and green (which, along with most of the other objects,
were all purchased at an after-Halloween sale). As each of these objects was moved closer
to the coil, a corresponding bass chord using a sustained instrument voice was faded up.
The ability to continuously control the volume in real-time by moving each goblin relative
to the coil further demonstrated the notion that the reader could determine the continuous
magnitude of the signal in real-time.
Each of the harmonic goblins represented one specific chord in the key of F major.
The red goblin represented the C chord (functional I chord), the blue one G (V), and the
green one F (IV). Inserting each of these goblins individually produced the corresponding
chord. Furthermore, inserting multiple goblins into the field at the same time produced
modulations and different chords according to a harmonic state machine, designed to
transition smoothly in a harmonic sense between the three primary chords. This state
machine is shown in Figure 16. Note that whenever only one goblin was present the state
machine always returned to the corresponding state 1, 2, or 3. The chord selected by the
goblin objects was used throughout the system; most of the notes produced by the other
objects were adjusted to harmonize on the selected chord. This ensured that whatever
users did, the resulting sound would not be too discordant.
The other tags in the music system served to either modify, embellish, or add
auxiliary layers of sounds to the melody and harmony. For instance, the pumpkin, when

45
 IF RED TAG SWITCHES, IF GREEN TAG SWITCH IF BLUE TAG SWITCH,
STATE KEY
GO TO STATE: GO TO STATE: GO TO STATE:
0 (NO TAG) C 1 2 3
1 (RED) C 0 5 14
2 (GREEN) F 9 0 10
3 (BLUE) G 15 10 0
4 B-FLAT 8 7 6
5 A 2 1 4
7 F 3 4 1
8 E-FLAT 16 3 2
9 G 2 1 4
10 C 11 3 2
11 E-FLAT 10 13 12
12 G 2 1 4
13 F 3 4 1
14 A 3 4 1
15 B 3 4 1
16 A-FLAT 19 18 17
17 B-FLAT 2 1 16
18 B-FLAT 3 16 1
19 D-FLAT 20 20 20
20 F-SHARP 21 21 21
21 B 22 22 22
22 E 23 23 23
23 A 24 24 24
24 D 25 25 25
25 G 26 26 26
26 C 27 27 27
27 F 28 28 28
28 B-FLAT 29 29 29
29 E-FLAT 16 16 16

Figure 16. State machine of harmonies in music system.

brought into the field, produced a high-pitched twinkling rush of random notes whose pitch
changed with its proximity. The dinosaur, which had a magnetostrictor attached to it,
served as a switch, which triggered an instrument voice change in the harmony and melody
lines. The proximity of the foot object caused a continuous pitch bend in the harmony line
moving smoothly down by up to an octave while transposing the other lines discretely

46
down by a fifth. The pig object added proportional amounts of vibrato to the melodic and
auxiliary lines as it approached the reader.
Several other embedded objects in the system demonstrated slightly more complex
interactions. The cube and eyeball each contained three tags, placed at right angles to each
other. Because of the correlation between coupling of the tag to field strength and the
orientation of the tag relative to the field direction, these right-angle arrangements could
yield information about both rotation of the object and distance from the reader, as in the
first application with the Lego blocks. The proximity of the cube was used to fade up a
low-pitched chord played with a “droning” voice, which then was pitch-bent up and down
depending on the orientation of the cube. The eyeball’s proximity and orientation was
used to adjust the magnitude of various distorting effects on the Lexicon unit.
Perhaps the most interesting user interaction came with the adjustable tag. As
discussed in Section 1.2, one of the possibilities with EAS tags is that their resonances can
be distorted by various factors, and this shift can be sensed by the tag reader as a change in
frequency response. In this system, one of the tags was designed such that the coil of the
inductor could be stretched out by the user. This in turn changed its resonant frequency,
and the distortion could be monitored by the system. Musically, inserting the tag faded up
a choral sound, and stretching the coil produced an additional orchestral string sound on
top of it, with the amount of pull on the coil proportionally adding in the second voice. In
the final version of this system, the user was able to pull a ferrite core up and down inside
a fixed coil, a much more robust interaction than pulling at the coil itself.
The music system was shown at numerous conferences and demonstrations both at
MIT and around the world and was well-received by attendees, who were usually
impressed by the unique interactions enabled by the system. The music provided intuitive
audio feedback, and users had little difficulty learning to use the system and enjoy it.
Eventually, the thought came up of proposing the system for exhibit at SIGGRAPH
2000,23 a very well-attended conference specializing in computer graphics and interaction
technologies, held in July of 2000. Because of the heavily graphical flavor of this

47
conference, the next step with this application was to develop a graphical layer of content
for the tag reader.

4.3. Graphical extensions to the Musical Trinkets project

For the purpose of adding graphics to the system to enrich the interaction, we
decided to go with simple colored shapes on the screen, on the grounds that adding
complexity to the graphics would distract from the interaction. Using OpenGL with
simple flat single-color polygons in three-dimensional space gave the requisite simplicity
while leaving plenty of parameters to adjust with the interaction.
The graphical display was designed to start with a flat gray disc, and other objects
would either alter the disc or move around the disc. Triggering the melodic rings produced
appropriately-colored falling balls that bounced off the disc. Adding the harmonic goblins
faded in red, green, and blue triangles which acted a little like spotlights in that they also
added their color to the central disc.
The pumpkin launched a swirling pattern of small diamonds rotating about the
central disc, whose radius varied with the tag proximity; the pig caused the melodic balls
to shimmer; the dinosaur caused the central disc to flip over; and the foot moved the
perspective further away from the central disc, effectively shrinking the view. The eyeball
added a floating, spinning eyeball image to the display that moved as the physical eyeball
was manipulated, and the cube rotated the entire point of view as it was turned.
For the purposes of demonstration, the stretchable tag was replaced by a Pez
dispenser, which had an internal spring that stretched when pulling the head of the
container in and out. This made for a very intuitive interaction, especially for people who
grew up familiar with Pez. Graphically, moving the Pez dispenser towards the coil added a
blue-colored disc behind the primary gray disc, and stretching out the head added still
another violet-colored disc.

48
 Finally, just before SIGGRAPH 2000, a new triangular-shaped tag was added,
which upon entering the field triggered a transient sound, switched between instrument
patches played by the stretching of the Pez dispenser, and also switched between the
simple graphical mode described thus far and a new graphical mode written by Marc
Downie from the Synthetic Characters group at the Media Lab. This mode, based on
Marc’s “MusicCreatures” program, 24 was a behavior-based system that listened to the
MIDI music commands coming from the computer and encouraged the motion of an
abstract animated set of “cloud-like” images on the screen in time with key musical events.
Examples of the first set of graphics are shown in Figure 17, where the visual
output of our system was redirected for display on a large screen at SIGGRAPH 2000. A
similar projection of the MusicCreatures graphics is shown in Figure 18, with attendees
using the Trinkets interface visible in the foreground.
The complete graphical display was set up in a specialized custom-built cabinet
that served as a narrow table. A projector at the bottom of the cabinet displayed an image
on a frosted glass screen on the top surface; this was used instead of other display devices
(e.g. CRT or LCD monitors) in order to minimize interference with the tag reader. The
reader coil was embedded just beneath the glass screen so users could move objects near
the table surface and receive immediate visual feedback from the graphics projected
directly beneath the area of interaction. Figure 19 shows the cabinet at SIGGRAPH with
users around it, and Figure 20 shows a closer view of the screen, with a user interacting
with the graphics. The system was also subsequently installed at the large SMAU
convention in Milan, Italy in October, 2000.25 Figure 21 shows several attendees
interacting with the system installed there.
The whole system made for a very intuitive interaction with aural and visual
feedback, which many people described as immersive, stimulating, and often relaxing to
use. The real-time capabilities allowed for immediate responses and the continuous values
yielded a very detailed level of control. The ability to use all the tags in the system
simultaneously also made for a great deal of flexibility in the interaction. On the whole,
the system was very well-received by the general public, many of whom were intrigued by

49
the novel mode of interaction. People were both fascinated by the Musical Trinkets
demonstration itself as well as the potential applications that could be enabled by such a
system.
Once the demonstration application was set up and running, and the capabilities of
the basic swept-frequency tag reader were clear, it became possible to consider extensions
of the system beyond its simple single-axis roots, to explore volumetric tracking
applications.

50
Figure 17. The first set of Musical Trinkets graphics, projected onto a large central
screen at SIGGRAPH 2000.

Figure 18. The graphics based on MusicCreatures, projected onto the large screen
with users using the Musical Trinkets in foreground.

51
Figure 19. A small audience gathers around the Musical Trinkets exhibit at
SIGGRAPH 2000.

Figure 20. A user interacts with the graphics screen.

52
Figure 21. Throngs of onlookers watch the Musical Trinkets project in action at SMAU in
Milan, Italy.

53
54
Chapter 5
Swept-frequency tag readers with multiple
coils

With a single read coil, the swept-frequency tag reader is capable of determining
signal strength from a tag in real-time. As mentioned, this signal strength is a function of
both proximity of the tag to the coil and the orientation of the tag’s inductive component
relative to the field lines emitted by the coil.
As discussed in the previous chapter, this in itself can provide a rich experience,
but we wondered if the system could be modified or extended in order to provide more
precision, more uniformity of interaction, or more information about each tag. Any of
these would enable a more interesting interaction for applications that might require
anything more than simple gross tag position in one direction relative to a tabletop. By
considering the limitations of the current system, modifications were considered and some
were implemented to extend the reader’s capabilities.

5.1. Limitations of a single-coil swept-frequency tag reader

A swept-frequency tag reader using a single read coil and simple, imprecise
tracking enabled a fairly rich interactive experience as demonstrated by the Musical
Trinkets application. In the interest of considering extensions to the system, we noted
several limitations, which mostly involved uniformity of tag reading and cross-axis
coupling with respect to position and orientation of the tag.

55
 Figure 22. Simple illustration of single-coil field and tag interaction. Tag A is turned parallel
to the field lines and so is detected strongly. Tag B is turned perpendicular to the field lines
and so is detected weakly. Tag D is turned in the same direction as Tag A, but the field lines
there run in a different direction. As a result, Tag D is detected less strongly than Tag C,
which is turned more parallel to the field lines there. This is an example of non-uniform
interaction, because of the differences in the detection of Tags A and D.

In a uniform interaction, changes in a tag’s orientation would have similar effects

on the magnitude of its signal, regardless of the location of the tag. One of the issues that
prevents uniformity of interaction with a single-coil tag reader is the nonparallel shape of
the field lines surrounding the coil. Right at the plane of the coil and on the central axis,
field lines run uniformly perpendicular to the plane, but anywhere off the plane and off-
axis, field lines begin to diverge, bending in elliptical shapes around the coil as the
magnetic flux follows its return path. Because of this shape, tags placed in the field
directly over the coil’s center are sensed with a magnitude corresponding to their
orientation relative to the axis of the coil, but tags not directly over the center have a
stronger magnitude when rotated slightly off-axis. Tags directly over the wires of the coil
itself are detected most strongly when turned completely perpendicular to the coil’s axis.
See Figure 22 for a simple illustration of this principle. If this problem could be solved,

56
the response of the system to the orientation of a tag would be more uniform, which would
increase the intuitiveness of the interaction.
Thus, the nonparallel field lines cause problems for the uniformity of the
interaction, introducing an ambiguity between angle and range depending on the tag’s
position. Another problem that arises is that the signal strength from a tag increases as a
tag approaches the wires of the coil, i.e. towards the outside of the ring. This is because
the field is really produced by the induced magnetic effect at the wires wound around the
ring, hence the magnetic flux is denser there and decreases with distance from those wires.
As a result, it is difficult to determine whether a decreasing signal strength is an indication
of a tag moving away from the tabletop or simply closer to the center of the coil area. This
is another issue that affects the uniformity of the interaction; in this case, reducing this
effect would allow for more precise determination of whether a tag was close to the plane
of the coil or not.
Even despite these uniformity difficulties, one more basic limitation was that with a
single read coil, only one piece of data about each tag could be generated. This single
piece of data, which corresponded to a function of both position and orientation, was in
itself already quite useful in producing a novel generic interaction, but moving a tag
horizontally around the tabletop, as many of our SIGGRAPH and SMAU attendees wanted
to do, produced no interesting effects as the motion could not be effectively detected. Nor
could a tag being rotated be differentiated from a tag being moved closer and further from
the reader. In the single-coil applications, this was circumvented by using three
differently-tuned orthogonal tags in a single object, which gave some idea of the
orientation of the object, but still produced no data about the horizontal position.
Because these limitations seemed to center around difficulties associated with
having only one reader coil, we turned our attention to possibilities for arrangements of
multiple coils.

57
5.2. Multiple-coil arrangements and Helmholtz coils

The extensions of the swept-frequency tag reader that we expected to produce the
most promising results, with respect to the limitations we noted, appeared to come from
adding more reader coils to the system. Given an appropriate arrangement, it seemed that
multiple coils could improve the uniformity of the system as well as provide more
information about the location and orientation of a tag. We realized early on that having
multiple coils did not necessarily mean running all the coils simultaneously, and in fact
cycling sequentially through a series of coils might lead to a slowing of the read time, but
may be necessary in order to prevent multiple coils from interfering with each other.
Despite this, for many applications, some degree of slowdown would be a reasonable
exchange to enable more precise varieties of interaction.
Several types of multi-coil arrangements were considered. First was the possibility
of having several coils placed in the same plane, either side by side or in a grid pattern. By
comparing the signal strengths for a given tag, it seemed like perhaps it would be possible
to tell which coil a tag was closest to, and from this the horizontal position of the tag could
be inferred. Likewise, by driving coils differentially, other field geometries could be
created that would provide additional information to resolve rotation, break ambiguities,
and determine a more precise location. At first glance, the math might work out similarly
to that in Josh Smith’s “School of Fish” project,26 which used an array of electric-field
sensing electrodes to accomplish a similar localization task for capacitively-coupled
objects. However, because of the nonuniformity of the magnetic fields coming from the
coils, and the vector coupling between fields and tags, the math could become unwieldy,
and no easily discernable reader coil configuration was derived for determining the precise
location of a tag. Thus, this solution promised to yield more information about the
whereabouts and state of a tag but without considerable calculation, would do nothing to
increase the uniformity of interaction.
Another arrangement that we considered was having a set of three orthogonal coils,
meeting at one corner and sequentially transmitting a field along its axis. This would

58
 Figure 23. Axial magnetic field of a typical Helmholtz coil, with coils spaced distance a apart,
where a is the radius of a coil.

provide capabilities similar to the single-coil system, except on three axes, giving some
degree of spatial knowledge to the measurement. Such an arrangement would give some
idea of three-dimensional position, but once again the field shapes were unchanged and the
uniformity of the interaction relative to any single coil would be compromised, as in Figure
22.
The dissatisfactions with both of these first two arrangements were the lack of
simple field shaping and uniformity. Our next concept, then, was to have two coils facing
each other, driven by the same signal simultaneously, in the hopes that this would take the
elliptical magnetic field lines of the single-coil system and pull them through two coils,
creating a space in-between where the field lines would be straighter and more uniform.
This arrangement is commonly known as the Helmholtz coil configuration.27
A little research yielded more information on Helmholtz coils. A single-coil

Nµ0 Ia 2
system has a magnetic field given by Bz ( z ) = , where N is the number of
2( z 2 + a 2 ) 3 / 2
wire turns in the coil, I is the current through the coil, a is the radius of the coil, µ0 is the

59
constant of magnetic permeability and z is the distance from the coil being measured. This
magnetic field decreases rapidly with increasing distance from the coil. Helmholtz coils,
on the other hand, have fields along their central axis given by:

Nµ 0Ia 2  1 1 
Bz ( z ) =  2 + 2 3/ 2 
, with 2b being the distance between
2 ( z + a ) 2 3/ 2
[( 2b − z ) + a ] 
2

the coils and other variables the same as above. While the equation looks messy, the field
it represents is much more uniform between the coils than in the single-coil system. A plot
of the Helmholtz field is shown in Figure 23. Notice that the field strength is
approximately constant between the coils when the distance separating the coils becomes
approximately equal to the radius of one of the coils.
Because of the improved uniformity of the field in the Helmholtz configuration, we
decided to try such a two-coil setup to observe its properties.

5.3. Two-coil Helmholtz arrangement

Using the boards already designed for the single-coil swept-frequency tag reader,
we were able to create a simple two-coil Helmholtz arrangement by wiring two reader
boards together such that the sweep generation circuitry from one board was used as the
input to the drive/transmit circuitry of both boards, with loading by the tag measured
separately for each coil as detailed in Chapter 3. A different coil was connected to each
board and they were placed parallel to each other approximately one coil diameter apart. A
simple block diagram of this setup is shown in Figure 24.
One of our uncertainties with using two simultaneously-transmitting coils was the
way in which the tag would couple to each of the coils. Would both coils receive equal
signals, would the signals between coils cancel, or would the signal received vary
proportionally to the distance between the tag and each coil? Our test showed that the tag
coupled with the two coils such that the signal received varied approximately linearly with
the proximity of the tag to one coil or the other.

60
 Figure 24. Two-coil Helmholtz arrangement block diagram.

Field shaping improved substantially as well. This two-coil configuration

apparently had the ability to shape the magnetic fields to be straighter and more evenly
spaced in a larger region between the two coils, relative to the one-coil setup. The result
was that the tag’s orientation decreased the signal strength in both coils proportionally
when rotated out-of-axis, provided that the tag was not too close to the edges of the coils.
Note that the optimal Helmholtz configuration involves spacing the coils one coil
radius apart from each other, but our systems used coils spaced one diameter apart, which
tended to lose some uniformity towards the fringes of the field and otherwise provided the
same benefits.

61
 From these observations, the results were promising. Inside a sizable region within
the coils, the field shaping of the Helmholtz configuration essentially decoupled the
orientation of the tag from the position of the tag along the axis between the two coils. As
summarized in Figure 24, the position of a tag could ideally be given by the difference in
signal strengths normalized by their sum, and the rotation of a tag out-of-axis was a
function of the combined signal strength in both coils.
Furthermore, the two coils were transmitting simultaneously, leading to no
slowdown in the response time, and also we were now receiving two pieces of information
about each tag, determining both position and orientation. The only desired feature
missing from this system was the ability to read the position of a tag along more than one
axis. Our next system used similar principles to do exactly that.

5.4. Six-coil Helmholtz arrangement

As seen in the two-coil Helmholtz arrangement, it was now possible to use two
coils to gather two pieces of information about a tag: position and orientation. Our next
hope was to extend this concept to gather six pieces of information about a tag, i.e. position
and orientation along three orthogonal axes.
The logical way to create such a system was to have three pairs of coils placed
along the outside of a space, each in different directions. The immediately apparent
disadvantage to such a system was that having coils totally surrounding a space would
limit a user’s ability to move completely flexibly in and out of the space. We decided that
using the six faces of a cubical frame as coils would be substantially better for user
mobility than having six attached circular rings. The edges of the cube were only needed
to support the wire of the coils, hence could be made very thin and unobtrusive.
A cube structure was built for this purpose, and wires were wound appropriately
around the six faces and each connected to a separate tag reader board. Our initial cube
was nine inches on a side, and we soon constructed a new cube out of Tinkertoys that was

62
 Figure 25. Block diagram of six-coil tag reader system.

twelve inches on each side, with ten windings of wire in each coil, making for an
inductance of between 105 and 115 µH, thus matching the inductors in the bridge circuit
on the reader board. The next step was to connect the reader boards such that the six coils
63
 Figure 26. The first completed six-coil cube frame with six tag reader boards.

transmitted sequentially, one opposing pair at a time. This pair-wise reading was
necessary in order to create Helmholtz fields between a pair of coils on one axis, while the
other four coils were turned off in order to prevent them from affecting the shape of the
magnetic field or the resulting signals. By cycling sequentially between three sets of coils,
we slowed our reading time by a factor of three over the single-coil case, but we anticipate
that many applications of the system would not suffer greatly from this effect. Otherwise,
the sweep could be correspondingly sped up with minor changes in the electronics,
keeping full updates of at least 30 Hz.
As depicted in Figure 25, the connections between the six reader boards were
closely related to the connections of the two-coil system. One board was designated the
master, and the sweep generator from this master board was fed through a multiplexer into
each of the drive circuits for all six boards. The multiplexer itself was controlled by the
PIC chip on the master board, so that every time a sweep was completed the PIC switched
the multiplexer to change connections to the next pair of boards. The drivers for all six
boards functioned normally and transmitted whenever they received a signal from the
multiplexer.

64
 Figure 27. Our final cube frame, along with three-axis tag and attachments for taking
measurements.

On the receive end, the signals coming back from all six coils went into their
respective boards, and after some analog processing were transferred back through a
multiplexer to the master board and its partner. The final filtering stages were performed
on these two boards, and the two PIC microcontrollers on these boards were used to
assemble the final information and transmit it via two serial cables to the host computer. A
picture of the completed system with all six coils around the cube is shown in Figure 26.
The newest six-coil cube, in the configuration used to take measurements for the next
section is shown in Figure 27.
Having completed the six-coil system, we observed its behavior and created a
simple demonstration application.

65
5.5. Results and application

Initial tests of the six-coil system yielded reasonable results. The behavior
observed earlier with the two-coil system was repeatable with all three pairs along the
different axes in this system. Differences in the response strengths of individual tags
created variations in the reconstruction necessary to determine position and orientation and
the precision achievable with each tag; some tags were weak and their signals faded out
towards the center of the cube, while other tags were strong and could saturate at the
cube’s edges. From this we inferred that the system must be carefully tuned in order to
achieve the best results. Increasing the analog conversion precision to ten bits of
information also improved the precision; this was added in the final version. A
dynamically-sampled baseline and smarter thresholding algorithms would further improve
the ability of the system to distinguish noise from signal, and has not yet been added; we
currently assume a static DC baseline across frequencies, but this is rarely the case (see
Figure 7 for example signals and baselines).
Noise levels were also higher for some of the reader boards than for others, which
we attributed to variations in calibration settings and soldering connections. Also, because
the bridge imbalance did not null as well across the entire decade of sweep used in our
other versions of the system, we swept across a narrower range of frequencies, over which
the baseline (and thus the null) was fairly flat. This improved the noise response as well.
Another potential contributor to noise level is the driver design; our coil driver circuitry
essentially holds non-driven coils effectively shorted, which may allow eddy currents to
build up and distort the magnetic fields near their edges. Subsequent work will investigate
this effect.
One of the great strengths of this system also turned out to be a disadvantage.
Because the field lines along each axis were much straighter than in a single-coil system,
tags rotated parallel to one axis and perpendicular to the other two were undetectable by
the coils along the perpendicular axes. This proves to be a problem for determining the

66
three-dimensional location of a tag, since some signal needs to be seen by all six coils to
precisely find the location and orientation.
The simple solution to this is to use a three-tag object, similar to the cube and
eyeball in the Musical Trinkets, in order to ensure that a signal is received for a given
object on all three axes. To conserve frequency usage it may be possible to use three tags
with identical resonant frequencies, as they would be representing the same object.
Likewise, orthogonal coils could be driven together to produce curved field geometries that
could resolve ambiguities (with significantly more math). For the purposes of using only
one tag, though, it would also be possible to increase the sensitivity of the system and
assume no change in the position of a tag along one axis when its signal drops out on that
axis, updating the position once the signal is reacquired. Under normal usage it seems
unlikely that a tag will spend any significant length of time completely undetectable along
any given axis.
Using a three-tag object, several series of measurements were taken for the system.
Holding the tags strictly parallel to their corresponding coil axes, the strength of the signal
for each tag from each coil along a 5 x 5 x 5 grid was recorded, normalized, and compared.
Because of the relatively high baselines, which caused tags not to be detected at all by
some coils, we simply took the differences between normalized signals, instead of dividing
the difference by the sum. Normalization involved dividing each signal by the highest
value recorded for that coil/tag combination at any position.
The differences between corresponding signals is plotted on a three-dimensional
graph in Figure 28, which shows that the raw information from the system is already not
too bad for three-dimensional position extrapolation. Magnetic field leakage, caused by
spacing the coils farther apart than the optimal Helmholtz distance, causes some warping
of the field near the edges; this is most apparent in the lowest plane plotted in the figure,
which happened to be slightly outside of the perimeter of the cube. However, because we
receive plenty of information from the tags and the positional data is uniform and not
multiply-valued, a simple fit algorithm could be designed to compensate for this and allow
for improved accuracy in position determination.

67
 Figure 28. Mesh visualization of raw data from tag readers along an evenly-spaced 5x5x5
grid. Although results were irregular, relative positions were fairly conserved and so a
simple fit algorithm could probably provide decent accuracy.

Noise data was also taken for three points within the cube. In the center of the
cube, the returned signals for a tag being held in place varied by up to 8% on average for
the noisiest coil, and towards the edge of the cube, the signals varied by up to 6% on
average for the noisiest relevant coil, where relevance was determined by which coil each
tag was on-axis with. This noise level, while not as low as in the single-coil version, is
sufficiently low to ensure that the precision for the system would be acceptably good for
simple interface applications given a properly-written fit algorithm.
Finally, a tag in the middle of the cube was rotated about the three axes to
demonstrate the correlation between the signals coming from the three sets of coils and the
orientation of the tag. These signals are plotted against time in Figure 29. The changes in
the response along one axis for one tag is accompanied by corresponding changes in the

68
 Figure 29. Normalized responses from three coils for one tag during tag rotation around coil
axes. As the tag signal decreases along one axis, it increases along another. Labels
indicate rotation represented by each graph peak.
other axes, making it likely that a good fit algorithm would be able to decouple the axes
without too much difficulty.
A simple graphical application was written to demonstrate the use of the system
with one tag, which, despite the sensitivity issues with perpendicular axes mentioned
above, produced a usable result. The application drew a dot on the screen when a tag was
detected, which moved around the screen based on the tag’s position in two axes and
became larger and smaller based on the tag’s position in the third axis. Coordinates were
determined simply by difference-over-sum, as described earlier. Furthermore, the disc
changed color depending on the orientation of the tag, with each axis of orientation
controlling one primary color component (red, green, blue).
This application successfully showed the basic feasibility of the six-coil tracking
system prototype. Other improvements to the system are left for future work.

69
5.6. Six-coil tag reader conclusions

Of the methods we envisioned as possibilities for extending the detection abilities

of the single-coil swept-frequency tag reader, we selected a concept based upon having six
coils wound around the faces of a cubical frame. This method was selected for its potential
to determine position in three dimensions, which in turn was motivated by our hopes to
extract more information for each tag, as well as the apparent expectations of people using
the one-coil system. Other advantages of using this configuration included its field-
shaping ability; because each pair of coils could shape field lines such that the area in
between the coils had straighter field lines than the area surrounding a single lone coil, tags
would produce more uniform signals based on their position along the coils’ axis
regardless of their position perpendicular to the axis.
By using multiplexers to alternate signals between the three pairs of coils, data
from each of the three dimensions could be gathered at the expense of a threefold decrease
in detection speed. Precision and accuracy of the data acquired were sufficient for
applications that we anticipate, and with the addition of an algorithm to fit the irregularities
of the data to the three-dimensional space, we expect that the results would allow a great
deal of control in three dimensions. The raw data already has enough regularity to be
useful in itself, and supports the potential for accurate 3-axis tracking after calibration.
Several immediate improvements could be made to the system at this point. First,
the fit algorithm could be added to the host computer, in order to map the raw data coming
from the coils to the actual position of the tags. Also, the baseline of each reader could be
taken at multiple points, instead of using a static baseline to determine presence of a tag in
the field. This would improve precision, as signals would be detected at a lower threshold.
Furthermore, the noise level of this system is higher than in the single-coil system, and
investigation into the nature of the noise, some of which seemed likely to have been caused
by loose connections and some of which seemed to arise from unexpected interactions
between reader boards or coils, would improve the precision substantially. Finally, with

70
appropriate field-shaping and design, the size of the cube could be expanded, which would
make the frame less of an obstruction to user interaction.
As it stands, the six-coil tag reader could already be useful in a number of
applications. Having a system that detects the motion of objects in real-time within a
three-dimensional space makes possible straightforward interactions, such as moving an
object within the field and controlling an onscreen counterpart. Given the ability of this
system to handle fairly large numbers of tags, they could be placed on all the fingers of a
hand, and tracked independently, creating an effect similar to a virtual reality glove, except
without wires, or an actual glove, or providing power to devices attached to the hand itself.
Similarly, the six-coil tag reader offers advantages over vision-based or sonar-based
interaction systems, in that multiple objects or fingers will not occlude each other so that
real-time interaction can continue uninterrupted regardless of position. Compared to
motion-capture technology like those offered by companies such as Polhemus, our system
might provide a simpler alternative for use with a smaller region.
In the next chapter we will discuss other possible uses for future versions of this
technology as well.

71
72
Chapter 6
Conclusion

In this thesis we explored the design and development of passive tag readers for
application in tangible-object desktop interfaces. Such interfaces would allow intuitive
interactions between a computer and a user not currently enabled by typical input devices.
Some of the capabilities that we deemed important in such a system were the ability to
handle multiple tags, the ability to sense continuous positions of the tags in real-time, and
an acceptable desktop read range, with reasonable stability, adjustability, precision, and
accuracy.

6.1. The ringdown tag reader

The first tag reader design we examined was the ringdown reader. The basic
concept behind this involved a coil through which we transmitted pings at the resonant
frequency of a predetermined tag, and then the system was made quiet in order to listen for
the ringdown responses of the tag. In order to detect multiple tags, we cycled our pings
between the resonant frequencies of multiple tags. For real-time, continuous interaction,
we performed the transmit-receive cycle in 1/30 seconds and used a gated integrator to sum
the strength of the demodulated received signal.
By adding a ladder of MOSFET’s and triacs, we were able to dynamically retune
the coil in order to optimize its resonance to match each successive tag being detected.
This enabled higher transmit power and better receive sensitivity. Although another design
possibility was to transmit broadband noise to excite all tags with a single ping, this had

73
implications limiting the transmit power. We used magnetostrictor tags for their high Q
values, hence strong response.
The system produced a usable result with the desired real-time and continuous
performance on multiple tags. Detection of one tag took about 12 milliseconds, which
gave essentially real-time performance with three tags; however, beyond about eight tags,
the speed of the interaction would suffer because each tag requires incrementally more
time for ringdown and detection. The other primary difficulty with the ringdown system
was its need to be preprogrammed for each resonant tag and the resulting vulnerability to
drift. Nonetheless, the ringdown reader has good power output and very capably detects a
small number of tags, making it a good choice for simple applications.

6.2. The swept-frequency tag reader

Next, we examined another approach, that of using a transmitted signal sweeping

continuously through the relevant frequency range. Because a continuous sweep cannot
stop to listen for ringdown responses, our reader used an inductive bridge circuit to detect
the energy being drawn dynamically from the magnetic field of the reader coil by a
resonant tag.
We decided to use LC tags for the ease of adjusting their frequencies. Our swept-
frequency reader swept a range of frequencies spanning the resonances of all tags. A
reverse exponential sweep ensured that lower frequencies received proportionally more
time for tag excitation. By running the sweep thirty times a second, we were able to
produce essentially real-time interaction, and by synchronously demodulating the output of
the differential amplifier of the inductive bridge, we were able to determine the magnitude
of a signal coming from a tag in the field.
Our finished reader had all the expected properties; it could sweep thirty times a
second and detect the continuous signal strength from all the tags in the frequency range.
Its receive range was slightly over one foot, which is acceptable for desktop-scale

74
interaction, and its jitter level was remarkably quiet. Frequency drift was not a severe
drawback except under relatively large temperature changes; even then, tag selection
allowed for some level of tolerance, and the addition of an algorithm to monitor and
compensate for frequency drift by comparing frequencies against the microprocessor’s
clock would remove this problem altogether. In many respects, the swept-frequency tag
reader made an ideal tangible interface backend.

6.3. Swept-frequency tag reader applications

The earliest swept-frequency tag reader was demonstrated using tags embedded
within Lego bricks, and produced simple tones when it detected the tags. Additional tag
interactions were produced by placing three orthogonal tags in a single object, thus
allowing for determination of orientation, as well as by perturbing the resonance of another
tag using a slider, thus allowing for an additional degree of control. These basic
interaction concepts have been used in subsequent applications as well.
Hiroshi Ishii’s Tangible Media Group also used the swept-frequency tag reader as a
backend for their musicalBottles project, which played various lines of music when bottles
on a tabletop were uncorked. Around the same time, our group began to produce the
Musical Trinkets system, which in its final incarnation produced music and graphics on a
small countertop with a frosted glass screen mounted on top. Twenty tags embedded in
sixteen toy “trinkets” each produced various interactions. Five rings worn on the fingers
produced melodic notes and graphical bouncing balls, interposed over a harmony and color
changes in the display controlled by three other tags. The remainder of the tags produced
various modifications and distortions of the basic music and graphics, including twinkling
sounds, vibrato, pitch bend, and several additional harmonizing voices, all accompanied by
viewpoint changes, swirling displays, and other additional graphics.
The Musical Trinkets project was displayed at several prominent conferences and
was received very well by the attendees as an interesting interface with notable aesthetic

75
qualities. With the success of the single-coil system, we turned our attention to extending
the abilities of our tag reader.

6.4. The six-coil swept-frequency tag reader

Using one read coil, it was possible to derive one piece of information about each
tag, which happened to be the proximity of the tag to the coil, scaled by the cosine of the
angle between the orientation of the tag and the orientation of the field at the tag’s location.
With some planning and experimentation, we found that using two parallel coils
transmitting simultaneously made it possible to deduce both position and orientation of a
tag along the axis of the coils, with the added bonus of shaping the magnetic field lines to
be more parallel to the axis and uniform in density. Known as Helmholtz coils, we
decided to design a fully three-dimensional tag reading system around this principle.
Using a cubical frame with coils wound around each face and six of our swept-
frequency tag reader boards, we multiplexed the signals in order to run each pair of facing
coils alternately. The uniformity of the interaction produced was encouraging, although
careful tuning was necessary to achieve the best results. Using a single object with three
orthogonal coils was generally the most reliable way to continuously receive data across all
six degrees of kinematic freedom. Data was taken and produced excellent results, even
without the addition of a calibration or fitting algorithm, and a simple graphical tag
tracking application was written for demonstration purposes.
Our six-coil tag reader is a work in progress, and numerous improvements to the
current system remain to be investigated. Beyond that, future improvements can come
from redesigns of the entire system and will be discussed in the next section.

76
6.5. Future directions and applications

All of our systems met basic expectations and demonstrated the feasibility of their
underlying concepts for application to tangible interfaces. However, many small
improvements can be made to the systems as they stand now, as well as more fundamental
redesigns which would enable applications of much greater scope than currently possible.
Both the ringdown tag reader and the swept-frequency tag reader have decent
transmit range and receive sensitivity now, but additional work in this area would involve
finding higher-quality components and tuning the resonances more carefully in order to
optimize the efficiency of the transmit and receive sections, and possibly exploring other
alternatives for tag detection beyond inductive bridges. With some effort, both systems
could potentially be adapted to much bigger coils and read areas than the 1-2 foot range we
have now, with the possibility of encompassing an entire physical desktop area or perhaps
even a larger, person-scale space. This would most benefit multiple-coil geometries like
the six-coil system, as larger coils would increase the size of the cube and make the frame
itself less of an obstruction for desktop interaction.
The swept-frequency tag reader would also benefit from the addition of a multi-
point dynamically-sampled baseline for the purposes of determining the threshold above
which a tag is detected and reported. The static baseline currently in use is necessarily
limited by frequencies in the range with higher noise susceptibility or distorted baselines
from dispersive bridge imbalances. Sampling each point of the baseline on system startup
would enable tags at less noisy frequencies and at suppressed baselines to be detected at
greater range.
Also for the swept-frequency reader, algorithms to compensate for temperature-
related frequency drift would be helpful in increasing the stability of the system. The host
computer would simply need to receive frequency measurements from the reader board
and adjust subsequently reported tag positions accordingly. Note however that tags near
the high and low extremes of the frequency range may drift out of range altogether, and

77
compensation algorithms cannot improve this without the addition of a feedback loop
directly between the microcomputer and the voltage controlling the oscillator.
Finally, in order to realize the full potential of the six-coil reader, a fit algorithm for
the data would allow the precision of the reader to yield excellent accuracy in three
dimensions. With the addition of such a fit algorithm, the reader could drive all sorts of
tangible interfaces, and the list of possible applications could even grow well beyond what
is possible just on a desktop.
On the simple end of tangible interfaces, the common everyday mouse as used
today could be adapted to a three-dimensional version, with clicks detectable (and in a
continuous manner as well) by tying the mouse “buttons” to changes in tag resonant
frequencies. More naturally, using any of the tag readers described, common desktop
objects could be moved around and monitored by the computer. In less computer-related
environments, for instance, a kitchen countertop with a series of tagged cooking utensils or
supplies could observe the location and presence of particular objects and furthermore use
resonant frequency distortions to sense the quantity of material present in a container (e.g.
as mentioned in Chapter 1).
The previous chapter also observed that the fully three-dimensional nature of the
six-coil system enables it to serve as not only a desktop-scale interface with the ability to
track an object being moved throughout space, but rings worn on the fingers of a hand
could be tracked in real time, enabling an interaction similar to that of virtual-reality
gloves. Even with a frame of the size we currently use, a sort of “virtual wetbox” can be
generated in this manner, in which the insertion and position of the user’s hand within the
cubical frame can be reflected in an onscreen representation. As a real wetbox involves
fixed gloves between a user and a hazardous manipulable environment in which the user
cannot be physically present, this virtual equivalent could provide similar interactions with
a computer-generated model.
Furthermore, just as the “wetbox” limits mobility, the cubical frame presents a
similar difficulty. However, the optimizations described above create the possibility that
the frame can be enlarged to the point that its presence is negligible from a usability

78
standpoint. Even with the frame in place, the ability to track fingers without either the
wires or the power sources of current active tracking systems, or the occlusion that
happens with current vision-based tracking, has substantial application-dependent
advantages, with the frame nicely defining the interaction volume.
Other applications for such a system include motion capture, which is an
increasingly important aspect of computer graphics, medicine, and other fields. This has
similar requirements (real-time, continuous, etc.) to those of tangible interfaces, except that
the data is typically streamed and recorded for later use and analysis instead of immediate
interaction. Placing tags on a subject’s hand, arm, or other body part that can be immersed
within the coil area would allow the tracking of that part in real-time, to be analyzed or
converted to other data by a computer. With a larger coil size, a larger part of a body could
be encompassed, and could thus provide a simpler alternative to the type of motion-capture
currently being offered by companies such as Polhemus,28 whose predominantly wired
active magnetic tracking systems have set a standard for motion capture.
Even further over the horizon, the principles of the six-coil reader, given sufficient
refinement, could be used to underlie medical technologies capable of tracking not only
body parts, but internal organs as well. This raises the possibility of using small tags,
injectable via biopsy needle, onto the surface of an internal organ in order to watch its
movement in real time. For instance, cancerous tumors on the surface of organs that move
around, such as lungs, force radiation therapy to use relatively diffuse beams for treatment,
in order to ensure that the tumor is struck, but this kills much more tissue than necessary. 29
The application of a tracking system such as an improved six-coil reader would enable the
beam of radiation to be more tightly focused on the exact tumor site, thus increasing
efficiency and decreasing danger to surrounding tissue.
Multiple improvements would have to be made to the multi-coil tag reader
technology, but in a relatively controlled RF environment, the needs of increased
sensitivity, larger coils, and smaller tags could be attained with further development.

79
6.6. Closing

We have investigated variations on two types of magnetic tag readers, and our
results show that applying these kinds of systems to tangible desktop interfaces can be
done cheaply and efficiently, and so common usage of such systems might not be that far
off. A plethora of applications is actually made possible by these principles, some as
simple and straightforward as three-dimensional computer mice, and others perhaps as far-
reaching as large-scale motion capture and cancer therapy.
Our primary hope, though, is simply that as computer technology matures further,
interfaces will transit from the direct and literal mouse-and-keyboard paradigm currently in
use to encompass a wider variety of interactions, potentially including everyday objects
being moved around intuitively. Incorporating these abilities could make computers easier
and more natural to use, and may lead to their seamless integration in many aspects of life
not currently computerized. The tag readers we described could be instrumental in
providing these capabilities, and so we encourage future research in interface development
to consider these projects as viable foundations for such work.

80
References

1. H. Ishii and B. Ullmer, “Tangible Bits: Towards Seamless Interfaces between

People, Bits and Atoms,” Proceedings of the Conference on Human Factors in Computing
Systems (CHI ’97), ACM, Atlanta, March 1997, pp. 234-241.

2. M. Weiser, “The Computer for the 21st Century,” Scientific American, 265(3),
September 1991, pp. 94-104.

3. K. Finkenzeller, RFID Handbook, John Wiley & Son, Ltd., New York (1998), p.
26.

4. “BiStatix Technology: A Whitepaper”, available at

http://www.motorola.com/GSS/SSTG/smartcard/white_papers/BiStatix_Whitepaper.pdf

5. See http://www.sensormatic.com/eas_www/easum.htm

6. K. Finkenzeller, RFID Handbook, John Wiley & Son, Ltd., New York (1998), p.
32.

7. K. Finkenzeller, RFID Handbook, John Wiley & Son, Ltd., New York (1998), p.
30.

8. R. Fletcher, J. A. Levitan, J. Rosenberg, and N. Gershenfeld, "Application of Smart

Materials to Wireless ID Tags and Remote Sensors," Materials for Smart Systems II,
Proceedings of the 1997 MRS Symposium, Boston, MA, Materials Research Society,
Pittsburgh, PA (1997), pp. 557-562.

9. W. B. Spillman, Jr., S. Durkee, and W. W. Kuhns, “Remotely Interrogated Sensor

Electronics (RISE) for Smart Structures Applications,” Proceedings of the SPIE Second
European Conference on Smart Structures and Materials, Volume 2361, Glasgow
(October, 1994), pp. 282-284.

10. J.A. Corbyn, “Pulse Induction Metal Detector,” Wireless World, March 1980, pp.
40-44.

81
11. E. Fukushima, S.B.W. Roeder, Experimental Pulse NMR: A Nuts and Bots
Approach. Addison-Wesley Publishing Company, New York (1981).

12. Hap Patterson, Sensormatic Corporation, personal communication, Spring 2000.

13. D. Lancaster, Active Filter Cookbook, Synergetics Press, Arizona (1995), p. 200.

14. Hap Patterson, Sensormatic Corporation, personal communication, Spring 2000.

15. R. Fletcher, O. Omojola, E. Boyden, and N. Gershenfeld, “Reconfigurable Agile

Tag Reader Technologies for Combined EAS and RFID Capability,” Proceedings of the
2nd IEEE Workshop on Automatic Identification Advanced Technologies, Summit, New
Jersey (1999).

16. K. Finkenzeller, RFID Handbook. John Wiley & Sons Ltd., England (1999), p. 27.

17. W.G. Jung, IC Timer Cookbook, Howard W. Sams & Co., Inc., Indianapolis
(1977), p. 107.

18. P. Horowitz and W. Hill, The Art of Electronics, Cambridge University Press, New
York (1989), p. 992.

19. J. Paradiso and K. Hsiao, “Swept Frequency, Magnetically-Coupled Resonant Tags

for Realtime, Continuous, Multiparameter Control,” Human Factors in Computer Systems;
CHI99 Extended Abstracts, ACM Press, New York (1999), pp. 212-213.

20. K. Hsiao and J. Paradiso, “A New Continuous Multimodal Musical Controller

Using Wireless Magnetic Tags,” Proceedings of the 1999 International Computer Music
Conference, International Computer Music Association, Sna Francisco (October 1999), pp.
24-27.

21. J. Paradiso, K. Hsiao, J. Strickon, J. Lifton, A. Adler, “Sensor systems for

interactive surfaces,” IBM Systems Journal, Vol. 39 No. 4 (2000), pp. 892-914.

22. H. Ishii, R. Fletcher, J. Lee, S. Choo, J. Berzowska, C. Wisneski, C. Cano, A.

Hernandez, C. Bulthaup, “musicBottles,” SIGGRAPH 1999 Conference Abstracts and
Applications, ACM Press, New York (1999).

23. J. Paradiso, K. Hsiao, A. Benbasat, “Musical Trinkets: New Pieces to Play,”

SIGGRAPH 2000 Conference Abstracts and
Applications, ACM Press, NY, July 2000, p. 90.

82
24. M. Downie, Behavior, Animation and Music: The Music and Movement of
Synthetic Characters. M.S. Thesis, MIT Media Lab (January 2001).

25. See http://www.smau.it

26. J. Smith, Electric Field Imaging. Ph.D. thesis, MIT Media Lab (February 1999),
page 49.

27. J.R. Reitz, F.J. Milford, Foundations of Electromagnetic Theory, Addison-Wesley

Publishing Company, Reading, Massachusetts (1967), p. 156.

28. See http://www.polhemus.com

29. Dr. Paul G. Seiler, Paul Scherrer Institut (PSI), Villigen, Switzerland, personal
correspondence, Spring 2000.

83
84
Appendix A

This appendix presents the schematics for the ringdown tag reader, the ringdown
tag reader front end driver, and the swept-frequency tag reader.

85
86
87
88
Appendix B

Here we present the information necessary to implement the frequency drift

compensation algorithms described in section 3.5.
On each normal sweep, the tag reader samples the output voltage of the tag
system several hundred times, and for each peak that exceeds the baseline threshold, it
outputs relevant information. For one sweep, the serial transmission provides the
following information:
1. The start time (where “time” is the counter of samples taken so far this
sweep) of each peak,
2. The stop time of each peak,
3. The sum of all samples under the peak (essentially an integral),
4. The number of samples taken in the course of the entire sweep.
By using frequency information and the number of samples to adjust the
measured start time and stop time, we can compensate for frequency drift.
On each frequency calibration sweep, the tag reader outputs samples of the
frequency at periodic intervals during the sweep. In a typical sweep, ten samples are
returned. When more samples are returned we can ignore the samples beyond the tenth;
although these are part of the sweep, tags detected beyond the tenth sample may not be
picked up under all common temperature conditions (one of the timing issues discovered
while taking heat measurements for section 3.5).
First, let us derive the compensation algorithm for the linear sweep. Let t start be
the start time for a tag, t stop be the stop time of the tag, and n be the number of samples
taken during an initial calibration sweep for that tag. Further, let us take frequency
samples so that f 1 and f 10 are the first and tenth frequency samples in that initial
calibration sweep. Then, our host computer’s program can store a value for the start and
stop frequencies of the tag that has been converted from the arbitrary counter timing of
the PIC code into something more closely resembling the actual frequencies of the tag.

89
 t
This can be done by scaling each time t such that f real = ( f1 − f 10) + f10 . After this, our
n
host computer can perform its computations by identifying tags based on their true
frequencies. Then, as time passes, temperatures change, and the frequency range drifts,
we can take new frequency samples periodically. Based on the new frequency
measurements, each subsequent incoming tag start and end time can be scaled also by the
equation just given, and tag identification can proceed based on the new scaled
frequencies, using the same lookup table as the one generated on startup.
The compensation algorithm for the exponential sweep works exactly the same
way, except that three frequencies are necessary to scale the tag time into an actual
frequency. By using f 1 , f 5 , and f 9 , the first, fifth, and ninth frequency samples, we can
scale a time given by the PIC into a real frequency once we can derive the exponential
function which would produce f 1 , f 5 , and f 9 . We use these particular values because they
are spaced equidistantly in time, which simplifies the math. By scaling the time axis so f 1
occurs at time 0, f 5 occurs at time 1, and f 9 occurs at time 2, we find we have to solve for
A, B, and k in a system of three equations
Ae −0k + B = f 1

Ae − k + B = f 5

Ae −2k + B = f 9
Taking logarithms of the second and third equation, it becomes possible to scale
both equations to be equal to k. Removing k and then substituting B using the first
equation, we obtain
( f5 − f1 ) 2
A=
2( f 9 − f 5 )
Thanks to the first equation, we thus know that
B = f1 − A
and finally we have
f5 − B
k = − ln
A

90
 Given the values of A, B, and k, we can derive a number analogous to the actual
frequency for each reported tag start and stop time t using the relation
f real = Ae −20kt / 9n + B

As with the linear case, by recording the values of f real for each t reported by the
tag reader, we can then use these values to distinguish between tags by their frequency in
the host computer, and when frequency measurements appear in periodic calibration
updates, it is possible to adjust the coefficients accordingly using the above relations,
which then continue to allow subsequent values to be scaled.
With a system like this implemented, it should then be possible to run the swept-
frequency tag reader for extended periods of time without the need to recalibrate for the
frequency drift of the reader. Frequency drift in the tags themselves is much harder to
compensate, but apart from extreme heat and physical distortion, the tags tend to be more
stable than the reader, particularly because the tags are in open air at room temperature,
while the reader board is often near other equipment (e.g. video projectors) which alters
its ambient temperature more often.

91
92
Appendix C

In this appendix we present the code used in the various systems described in this
thesis. The final PIC used with the PIC code was the Microchip PIC16F873. C code was
compiled using CCS PIC C (see http://www.ccsinfo.com) and written using MPLAB (see
http://www.microchip.com) using a PICSTART Plus chip programmer (also available
from Microchip).
Windows C++ code was written and compiled in Microsoft Visual Studio, with
the addition of Rogus McBogus, a MIDI library developed by the Brain Opera group at
the Media Lab, and with GLUT, the extension library for OpenGL (information available
at http://reality.sgi.com/opengl/glut3/glut3.html). The code used for the Musical Trinkets
is very long and will not be included here. Links and information are current as of
January 2001; contact the author for further information.

C.1. Swept-Frequency Master Board PIC

The first program is the one used to control the PIC on the master tag reader board
in the six-coil tag system, which can be used as is to control a single tag reader as well.

#include "tag10pic.h"

#define PIN_TRIGGER PIN_B1

#define PIN_CAL 1
#define PIN_INPUT 0
#define PIN_EXPORESET PIN_B0
#define PIN_TAGPRESENT PIN_B2
#define PIN_RESETSWEEP PIN_B3
#define PIN_OSCILLATE PIN_B4
#define PIN_EXPOTRIG PIN_B5
#define PIN_RUNCAL PIN_B6
#define PIN_HI555 PIN_B7
#define PIN_GATE PIN_C1
#define PIN_MUX0 PIN_C4

93
#define PIN_MUX1 PIN_C5

#fuses HS,NOWDT,NOPROTECT,PUT,NOBROWNOUT
#use rs232(baud=19200, xmit=PIN_C6, rcv=PIN_C7)
#use fast_io(B)
#use fast_io(C)

typedef union {
unsigned long two;
unsigned int ones[2];
} WORD;

unsigned int i;
WORD tmp;
unsigned int overflows=0;
unsigned long threshold=0;
unsigned int counter=0;
unsigned int last, start;
char axis=0; // which pair of boards are we running now? control the mux
int running=1;
char sweeping=1;
WORD sum;
unsigned int bufstart=0, bufend=0, buffer[80];

void addbuf(unsigned int val)

{
if (bufend==bufstart-1 || (bufend==79 && bufstart==0)) return;
buffer[bufend++]=val;
if (bufend>79) bufend=0;
}

unsigned int getbuf()

{
unsigned int val;
if (bufend==bufstart) return 255;
val=buffer[bufstart++];
if (bufstart>79) bufstart=0;
return val;
}

main()
{
#asm
bsf 0x9f, 7
#endasm
#use rs232(baud=19200, xmit=PIN_C6, rcv=PIN_C7)
setup_counters(RTCC_INTERNAL,RTCC_DIV_2);
setup_port_a(ALL_ANALOG);
setup_adc(ADC_CLOCK_INTERNAL);
set_tris_b(0b00000111);
set_tris_c(0b10000000);
printf("PIC starting up...\n");
output_high(PIN_RESETSWEEP);
output_high(PIN_OSCILLATE);
output_high(PIN_EXPOTRIG);
output_high(PIN_RUNCAL);
output_high(PIN_HI555);
output_low(PIN_MUX0);
output_low(PIN_MUX1);
while (!input(PIN_TRIGGER));
while (input(PIN_TRIGGER));

94
 for (i=0; i<127; i++)
{
set_adc_channel(PIN_INPUT);
delay_us(45);
while (!input(PIN_TRIGGER))
{
tmp.two=read_adc();
if (tmp.two>threshold)
threshold=tmp.two;
}
printf("%ld ", threshold);

while (input(PIN_TRIGGER));
}

setup_counters(RTCC_INTERNAL, RTCC_DIV_4);
enable_interrupts(RTCC_ZERO);
enable_interrupts(GLOBAL);

while(1)
{
set_adc_channel(PIN_INPUT);
while (!input(PIN_TRIGGER));
disable_interrupts(GLOBAL);
addbuf(overflows);
addbuf(counter);
addbuf(255);
addbuf(255-axis);
counter=0;
overflows=0;

axis++;
if (axis>2) axis=0;
if (axis==1) output_high(PIN_MUX0); else output_low(PIN_MUX0);
if (axis==2) output_high(PIN_MUX1); else output_low(PIN_MUX1);
while (input(PIN_TRIGGER))
{
if (bufstart!=bufend)
{
printf("%c", getbuf());
}
enable_interrupts(GLOBAL);

}
}

#INT_RTCC
void loop()
{
int start_o;
set_rtcc(100);

if (counter==255) {overflows++; counter=0;}

else counter++;

tmp.two=read_adc();
if (tmp.two>threshold)
{

95
 if (last==0)
{
start=counter;
// start_o=overflows;
}
last=1;
sum.two+=(tmp.two);
}
else
{
if (last==1)
{
// addbuf(start_o);
addbuf(start);
// addbuf(overflows);
addbuf(counter);
addbuf(sum.ones[1]);
addbuf(sum.ones[0]);
last=0;
sum.two=0;
}
}

if ((!(counter&0b00000011)) && (bufstart!=bufend)) // every 4 counts

{
printf("%c", getbuf());
}

96
C.2. Swept-Frequency Frequency-Drift Collector

This next program is similar to the first but its sole purpose is to return frequency
and voltage samples from the voltage-controller oscillator. Added to the code in C.1 and
combined with the algorithms for the host computer given in Appendix B, automatic
frequency drift compensation should be easily implementable.

#include "tag10cal.h"

#define PIN_TRIGGER PIN_B1

#define PIN_CAL 1
#define PIN_INPUT 0
#define PIN_EXPORESET PIN_B0
#define PIN_TAGPRESENT PIN_B2
#define PIN_RESETSWEEP PIN_B3
#define PIN_OSCILLATE PIN_B4
#define PIN_EXPOTRIG PIN_B5
#define PIN_RUNCAL PIN_B6
#define PIN_HI555 PIN_B7
#define PIN_GATE PIN_C3
#define PIN_MUX0 PIN_C4
#define PIN_MUX1 PIN_C5

#fuses HS,NOWDT,NOPROTECT,PUT,NOBROWNOUT
#use rs232(baud=19200, xmit=PIN_C6, rcv=PIN_C7)
#use fast_io(B)
#use fast_io(C)

typedef union {
unsigned long two;
unsigned int ones[2];
} WORD;

unsigned int i;
WORD tmp;
unsigned int overflows=0;
unsigned long threshold=0;
unsigned int counter=0;
unsigned int last, start;
char axis=0; // which pair of boards are we running now? control the mux
int running=1;
char sweeping=1;
WORD sum;

void exp_recal()
{
WORD vtmp1, vtmp2, vstmp1, vstmp2, vstmp3;
WORD ftmp1, ftmp2, ftmp3;

// get voltage and frequency 1st time

while (1) {

97
 while(!input(PIN_TRIGGER))
;
while(input(PIN_TRIGGER))
;
while(!input(PIN_TRIGGER))
;
while(input(PIN_TRIGGER))
;
while(!input(PIN_TRIGGER)) {

vstmp1.two = read_adc();
set_timer1(0);
delay_ms(1);
ftmp1.two=get_timer1();

if (input(PIN_TRIGGER)) break; // abort if cycle ended

printf("%c", vstmp1.ones[1]);
printf("%c", vstmp1.ones[0]);
printf("%c", ftmp1.ones[1]);
printf("%c", ftmp1.ones[0]);
}
printf("%c%c", 255, 255);

}
}

main()
{
#asm
bsf 0x9f, 7
#endasm
#use rs232(baud=19200, xmit=PIN_C6, rcv=PIN_C7)
setup_counters(RTCC_INTERNAL,RTCC_DIV_2);
setup_port_a(ALL_ANALOG);
setup_adc(ADC_CLOCK_INTERNAL);
set_tris_b(0b00000110);
set_tris_c(0b10000011);
printf("PIC starting up...\n");
output_high(PIN_RESETSWEEP);
output_high(PIN_OSCILLATE);
output_high(PIN_EXPOTRIG);
output_high(PIN_RUNCAL);
output_high(PIN_HI555);
output_low(PIN_MUX0);
output_low(PIN_MUX1);
set_adc_channel(PIN_CAL);

setup_timer_1(T1_EXTERNAL|T1_DIV_BY_1);
exp_recal();
}

98
C.3. PIC Header File

This header file is the one included in the programs in both C.1 and C.2.

//////// Standard Header file for the PIC16F873 device ////////

#device PIC16F873
#use delay(clock=20000000)
#nolist
/////////////////////////////// I/O definitions for INPUT() and OUTPUT_xxx()
#define PIN_A0 40
#define PIN_A1 41
#define PIN_A2 42
#define PIN_A3 43
#define PIN_A4 44
#define PIN_A5 45

#define PIN_B0 48
#define PIN_B1 49
#define PIN_B2 50
#define PIN_B3 51
#define PIN_B4 52
#define PIN_B5 53
#define PIN_B6 54
#define PIN_B7 55

#define PIN_C0 56
#define PIN_C1 57
#define PIN_C2 58
#define PIN_C3 59
#define PIN_C4 60
#define PIN_C5 61
#define PIN_C6 62
#define PIN_C7 63

/////////////////////////////// Useful defines

#define FALSE 0
#define TRUE 1

#define BYTE int

#define BOOLEAN short int

#define getc getch

#define getchar getch
#define puts(s) {printf(s); putchar(13); putchar(10);}
#define putc putchar

/////////////////////////////// Constants used for RESTART_CAUSE()

#define WDT_FROM_SLEEP 0
#define WDT_TIMEOUT 8
#define MCLR_FROM_SLEEP 16
#define NORMAL_POWER_UP 24
/////////////////////////////// Constants used for SETUP_COUNTERS()
#define RTCC_INTERNAL 0
#define RTCC_EXT_L_TO_H 32

99
#define RTCC_EXT_H_TO_L 48
#define RTCC_DIV_2 0
#define RTCC_DIV_4 1
#define RTCC_DIV_8 2
#define RTCC_DIV_16 3
#define RTCC_DIV_32 4
#define RTCC_DIV_64 5
#define RTCC_DIV_128 6
#define RTCC_DIV_256 7
#define WDT_18MS 8
#define WDT_36MS 9
#define WDT_72MS 10
#define WDT_144MS 11
#define WDT_288MS 12
#define WDT_576MS 13
#define WDT_1152MS 14
#define WDT_2304MS 15
#define L_TO_H 0x40
#define H_TO_L 0

#define RTCC_ZERO 0x0B20 // Used for ENABLE/DISABLE INTERRUPTS

#define INT_RTCC 0x0B20 // Used for ENABLE/DISABLE INTERRUPTS
#define RB_CHANGE 0x0B08 // Used for ENABLE/DISABLE INTERRUPTS
#define INT_RB 0x0B08 // Used for ENABLE/DISABLE INTERRUPTS
#define EXT_INT 0x0B10 // Used for ENABLE/DISABLE INTERRUPTS
#define INT_EXT 0x0B10 // Used for ENABLE/DISABLE INTERRUPTS

#define GLOBAL 0x0BC0 // Used for ENABLE/DISABLE INTERRUPTS

///////////////////////////////////// Constants used for Timer1 and Timer2
#define T1_DISABLED 0
#define T1_INTERNAL 5
#define T1_EXTERNAL 7
#define T1_EXTERNAL_SYNC 3
#define T1_CLK_OUT 8
#define T1_DIV_BY_1 0
#define T1_DIV_BY_2 0x10
#define T1_DIV_BY_4 0x20
#define T1_DIV_BY_8 0x30
#byte TIMER_1_LOW= 0x0e
#byte TIMER_1_HIGH= 0x0f
#define T2_DISABLED 0
#define T2_DIV_BY_1 4
#define T2_DIV_BY_4 5
#define T2_DIV_BY_16 6
#byte TIMER_2= 0x11

#define INT_TIMER1 0x8C01 // Used for ENABLE/DISABLE INTERRUPTS

#define INT_TIMER2 0x8C02 // Used for ENABLE/DISABLE INTERRUPTS

//////////////////////////////////// Constants used for SETUP_CCP1()

#define CCP_OFF 0
#define CCP_CAPTURE_FE 4
#define CCP_CAPTURE_RE 5
#define CCP_CAPTURE_DIV_4 6
#define CCP_CAPTURE_DIV_16 7
#define CCP_COMPARE_SET_ON_MATCH 8
#define CCP_COMPARE_CLR_ON_MATCH 9
#define CCP_COMPARE_INT 0xA
#define CCP_COMPARE_RESET_TIMER 0xB
#define CCP_PWM 0xC
#define CCP_PWM_PLUS_1 0x1c

100
#define CCP_PWM_PLUS_2 0x2c
#define CCP_PWM_PLUS_3 0x3c
long CCP_1;
#byte CCP_1 = 0x15
#byte CCP_1_LOW= 0x15
#byte CCP_1_HIGH= 0x16

#define INT_CCP1 0x8C04 // Used for ENABLE/DISABLE INTERRUPTS

//////////////////////////////////// Constants used for SETUP_CCP2()

long CCP_2;
#byte CCP_2 = 0x1B
#byte CCP_2_LOW= 0x1B
#byte CCP_2_HIGH= 0x1C

#define INT_CCP2 0x8D01 // Used for ENABLE/DISABLE INTERRUPTS

//////////////////////////////////// Constants used in SETUP_SSP()

#define SPI_MASTER 0x20
#define SPI_SLAVE 0x24
#define SPI_L_TO_H 0
#define SPI_H_TO_L 0x10
#define SPI_CLK_DIV_4 0
#define SPI_CLK_DIV_16 1
#define SPI_CLK_DIV_64 2
#define SPI_CLK_T2 3
#define SPI_SS_DISABLED 1
#define SPI_SAMPLE_AT_END 0x80 // Only for some parts
#define SPI_XMIT_L_TO_H 0x40 // Only for some parts
#define INT_SSP 0x8C08 // Used for ENABLE/DISABLE INTERRUPTS

#define INT_RDA 0x8C20 // Used for ENABLE/DISABLE INTERRUPTS

#define INT_TBE 0x8C10 // Used for ENABLE/DISABLE INTERRUPTS

///////////////////////////////////// Constants used for SETUP_ADC_PORTS()

#define ALL_ANALOG 0x80
#define ANALOG_RA3_REF 0x81
#define A_ANALOG 0x82
#define A_ANALOG_RA3_REF 0x83
#define RA0_RA1_RA3_ANALOG 0x84
#define RA0_RA1_ANALOG_RA3_REF 0x85
#define NO_ANALOGS 0x86

#define ANALOG_RA3_RA2_REF 0x88

#define ANALOG_NOT_RE1_RE2 0x89
#define ANALOG_NOT_RE1_RE2_REF_RA3 0x8A
#define ANALOG_NOT_RE1_RE2_REF_RA3_RA2 0x8B
#define A_ANALOG_RA3_RA2_REF 0x8C
#define RA0_RA1_ANALOG_RA3_RA2_REF 0x8D
#define RA0_ANALOG 0x8E
#define RA0_ANALOG_RA3_RA2_REF 0x8F
///////////////////////////////////// Constants used for SETUP_ADC()
#define ADC_OFF 0
#define ADC_CLOCK_DIV_2 1
#define ADC_CLOCK_DIV_8 0x41
#define ADC_CLOCK_DIV_32 0x81
#define ADC_CLOCK_INTERNAL 0xc1

#define ADC_DONE 0x8C40 // Used for ENABLE/DISABLE INTERRUPTS

101
#define INT_ADC 0x8C40 // Used for ENABLE/DISABLE INTERRUPTS

#list

102
C.4. Host computer calibration routine

Before running the Musical Trinkets program, the frequencies for each tag in the
system had to be recorded so the program could distinguish between tags. This program
was used for this calibration, and without being entirely too long, gives relevant
information on interfacing a Windows PC with the tag reader board.

#include <math.h>
#include <stdlib.h>
#include <stdio.h>
#include <iostream.h>
#include <fstream.h>
#include <conio.h>

const TOTALTAGS=20;
int maxes[TOTALTAGS];
int freqstart[TOTALTAGS], freqend[TOTALTAGS];
char tagnames[TOTALTAGS][30]={"red goblin","green goblin","blue goblin",
"white ring", "red ring","yellow ring","green ring","blue ring",
"distorting tag","dinosaur","porcupine","pikachu",
"box1","box2","box3","pig","eye1", "eye2", "eye3", "cone(switch)"};
static HANDLE hCom;
char serialPort[30] = "COM2";
int i, j, k;
int data[256];
int pos=0;
int totalcount=10000;
char filename[80]="c:\\tags.siggraph\\tagcal.txt";
void comSetup();
void calibtag(int *start, int *end, int *max);
void savetofile();
int* read();

void main()
{
// quick calibration-file-generating hack for the tags system

// first, read in the current calibration-file

ifstream infile(filename);
if (infile == NULL) {
cout << filename << " could not be loaded! dropping to defaults...\n";
}
else {
// load in current calibration
infile >> serialPort;
cout << "serial port " << serialPort << endl;
for (i=0; i<TOTALTAGS; i++) {

103
 infile >> freqstart[i] >> freqend[i] >> maxes[i];
cout << "tag " << i << ":" << freqstart[i] << " " << freqend[i] <<
“ " << maxes[i] << '\n';
}
}
comSetup();
infile.close();
cout << "Press key to continue...\n";
cout.flush();
_getch();

int laststart=0, lastend=0, lastmax=0, lasttag=-1;

int keeplooping=1;
char input=0, currtag=0;
while (keeplooping == 1) { // main menu loop
while (_kbhit()) _getch();
cout << "Current tag values (name, frequency range, amplitude):\n";
for (i=0; i<TOTALTAGS; i++) { // print menu
cout << (char)((i<10)?'0'+i:('a'+(i-10)))
<< ":" << tagnames[i] << " " << freqstart[i] << "-" <<
freqend[i] << " "
<< maxes[i] << endl;
}
cout << "Press number/lowercase letter to calibrate tag, or S:save current
calibration,\n"
<< "T:test tag (dump values), U:undo last tag calibrate, Q:quit\n";
cout.flush();
input=_getche();
cout<<endl;
switch(input) {
case 'Q':
keeplooping=0;
break;
case 'T':
while (!_kbhit()) {
cout.flush();
read();
for (i=0; i<pos; i++) {
cout << data[i] << " ";
}
cout << endl;
}
break;
case 'U':
if (lasttag!=-1) {
freqstart[lasttag]=laststart;
freqend[lasttag]=lastend;
maxes[lasttag]=lastmax;
lasttag=-1;
}
else cout << "Can't undo.\n";
break;
case 'S':
// calculate optimal frequencies and dump to file
// - sort by freqstarts, then average freqstarts.
// - if overlap, let lower tag take precedence
savetofile();
break;
default:
if (input >= '0' && input <= '9' && TOTALTAGS>input-'0') {
currtag=input-'0';

104
 }
else if (TOTALTAGS > 10 && (TOTALTAGS-10) > input-'a') {
currtag=input-'a'+10;
}
else {
cout << "invalid input.\n";
break;
}
lasttag=currtag;
laststart=freqstart[currtag];
lastend=freqend[currtag];
lastmax=maxes[currtag];
calibtag(&freqstart[currtag], &freqend[currtag],
&maxes[currtag]);
break;
}
}
}

void calibtag(int *start, int *end, int *max) {

int locounts[TOTALTAGS], hicounts[TOTALTAGS], sums[TOTALTAGS], middles[TOTALTAGS];
*start=10000;
*end=0;
*max=0;
while (!_kbhit()) {
read();
if ((pos-4)/4 <= 0) {
continue;
}
for (i=0; i<pos; i+=4)
{
locounts[i/4]=data[i];
hicounts[i/4]=data[i+1];
sums[i/4]=(data[i+2]<<8) + data[i+3];
middles[i/4]=(locounts[i/4]+hicounts[i/4])/2;
}
totalcount=(data[pos-4]<<8)+data[pos-3];
if (*start>locounts[0]) *start=locounts[0];
if (*end<hicounts[0]) *end=hicounts[0];
if (sums[0]>*max) *max=sums[0];
cout << (*start) << "-" << (*end) << " " << (*max) << endl;
}
}

void savetofile() {
int order[TOTALTAGS], realstart[TOTALTAGS], realend[TOTALTAGS];
int temp;
for (i=0; i<TOTALTAGS; i++)
order[i]=i;
for (j=0; j<TOTALTAGS; j++) {
for (i=0; i<TOTALTAGS-1; i++) {
if (freqstart[order[i]]>freqstart[order[i+1]]) {
temp=order[i];
order[i]=order[i+1];
order[i+1]=temp;
}
}
}
realstart[order[0]] = 0;
for (i=0; i<TOTALTAGS-1; i++) {
if (freqend[order[i]]>=freqstart[order[i+1]]) {

105
 cout << "warning! overlap between " << tagnames[order[i]] << " and
" <<
tagnames[order[i+1]] << endl;
realend[order[i]]=freqend[order[i]];
realstart[order[i+1]]=freqend[order[i]]+1;
}
else {
realend[order[i]]=(freqend[order[i]]+freqstart[order[i+1]])/2;
realstart[order[i+1]]=realend[order[i]]+1;
}
}
realend[order[TOTALTAGS-1]] = totalcount;

ofstream outfile(filename);
outfile << serialPort << endl;
for (i=0; i<TOTALTAGS; i++) {
outfile << realstart[i] << ' ' << realend[i] << ' ' << maxes[i] << endl;
}
outfile.close();
cout << "file " << filename << "written.";
_getche();
}

int* read() {
DWORD bytesread;
unsigned char buffer[256];

pos=0;
while (1)
{
// Sleep(1);
ReadFile(hCom, buffer, 1, &bytesread, NULL);
// cout << "read" << bytesread;
cout.flush();

if (bytesread)
{
// cout << "read";
// cout.flush();
for (int i=0; i<bytesread; i++)
{
data[pos++]=buffer[i];
if (pos>250) pos=0;
if (data[pos-1]==255 && data[pos-2]==255)
{
// cout << endl << pos << endl;
return data; // data in data, pos in pos (globals)
}
}
}

}
}

static
void
comSetup()
{
DCB dcb;
DWORD dwError;
BOOL fSuccess;

106
 hCom = CreateFile(serialPort,
GENERIC_READ | GENERIC_WRITE,
0, /* comm devices must be opened w/exclusive-access */
NULL, /* no security attrs */
OPEN_EXISTING, /* comm devices must use OPEN_EXISTING */
0, /* not overlapped I/O */
NULL /* hTemplate must be NULL for comm devices */
);

cout << "Opened serial port " << serialPort << " " << hCom;
if (hCom == INVALID_HANDLE_VALUE) {
dwError = GetLastError();

/* handle error */
// r->ew.w("Error opening %s",serialPort);
return;
}

/*
* Omit the call to SetupComm to use the default queue sizes.
* Get the current configuration.
*/

fSuccess = GetCommState(hCom, &dcb);

if (!fSuccess) {
/* Handle the error. */
// r->ew.w("Error getting the comm state");
return;
}

/* Fill in the DCB: baud=19200, 8 data bits, no parity, 1 stop bit. */

dcb.BaudRate = 19200;
dcb.ByteSize = 8;
dcb.Parity = NOPARITY;
dcb.StopBits = ONESTOPBIT;

fSuccess = SetCommState(hCom, &dcb);

if (!fSuccess) {
/* Handle the error. */
// r->ew.w("Error setting the comm state");
return;
}

COMMTIMEOUTS cto;
cto.ReadIntervalTimeout = 4;
cto.ReadTotalTimeoutMultiplier = 10;
cto.ReadTotalTimeoutConstant = 100;
cto.WriteTotalTimeoutMultiplier = 10;
cto.WriteTotalTimeoutConstant = 100;
if (!SetCommTimeouts(hCom, &cto)){
// r->ew.w("Unable to set proper timeouts");
}
}

107