MathWorks News&Notes

The Magazine for the MATLAB and Simulink Community

Power Electronics for

a More Electric Aircraft

ALSO IN THIS ISSUE

Deep Learning for

Computer Vision

Generating C Code
from MATLAB

Sensor Fusion for

Scania AEBS

Internet of Things

Cleves Laboratory

MATLAB SPEAKS

WIRELESS
DESIGN

MATLAB CENTRAL
TURNS FIFTEEN!
Tap into the knowledge and experience of thousands of community members.
Find or contribute downloadable code, technical tips, and answers.

2000

You can simulate, prototype, and

verify wireless systems right in
MATLAB. Learn how todays MATLAB
supports RF, LTE, WLAN, and 5G
development and SDR hardware.
mathworks.com/wireless

Problem sets and

games on Cody

25,000
Downloadable files
on File Exchange

120,000
Questions answered
on MATLAB Answers

3000
Posts on MATLAB
and Simulink blogs

mathworks.com/matlabcentral

10

14

18

22

MathWorks News&Notes
featu res

6 Real-Time Simulation and Testing of Power Electronics on a More Electric Aircraft

Model-Based Design enables Microsemi to develop power control modules that can perform reliably for 50,000 to
150,000 hours of flight and under a wide range of failure conditions.

10 Developing and Verifying Sensor Fusion Methods for Advanced Emergency

Braking Systems on Scania Trucks and Buses
Scanias AEBS alerts the driver of an imminent collision and automatically applies the brake if the driver doesnt
respond.

14 HS Bochum Students Design and Build a Motor Controller for an E-Longboard
with Model-Based Design
When the company learned of Kevins success with the e-longboard, they hired him on the spot.

30

18 Creating Computer Vision and Machine Learning Algorithms That Can Analyze
Works of Art
Using MATLAB and a database of thousands of paintings, Rutgers University researchers discovered that an
algorithm could classify a painting as easily as a non-expert human.

22 Deep Learning for Computer Vision with MATLAB

An object detection and recognition example shows how easy it is to use MATLAB for deep learning, even if youre
new to computer vision algorithms and neural networks.

abo ut th e cov e r
The cover shows a rendering of a commercial aircraft, highlighting the THSA (trimmable
horizontal stabiliser actuator). On a more electric aircraft, mechanical actuators are replaced with
all-electric actuators, reducing weight, improving reliability, and minimizing environmental
impact. The article on page 6 describes how engineers in Microsemis Aviation Center of Excellence
developed a power core module (PCM) for a more electric aircraft. The PCM controls the electric
motors used in primary flight control actuation and landing gear systems.

26 The Joy of Generating C Code with MATLAB

Generate efficient, readable C code with this three-step workflowand explore successful applications of
generated code by companies such as Delphi and Baker Hughes.

de partm e nts
4 MATLAB and Simulink in the World: Physical

34 Tech Spotlight: Smart Devices and Analytics Spur

Modeling

Innovation in the Internet of Things

Managing Editor
Linda Webb

Production Staff
L. Bai, J. Mespelli, A. Pollack

Editor
Rosemary Oxenford

Editorial Board
T. Andraczek, S. Gage, C. Hayhurst,
M.Hirsch, S. Lehman, D. Lluch,
M.Maher, A. May, C. Moler,
M. Mulligan, L.Shure, J. Tung

Art Director
Robert Davison

30 Cleves Corner: Introducing Cleves Laboratory

36 Third-Party Products: Solutions for Image

Acquisition and Computer Vision

Graphic Designer
Chris Roth
Technical Writer
Jack Wilber
Production Editor
Julie Cornell

2016 The MathWorks, Inc.

MATLAB and Simulink are registered trademarks of The MathWorks, Inc. See mathworks.com/trademarks
for a list of additional trademarks. Other product or brand names may be trademarks or registered trademarks of their respective holders.

Printed on 30% postconsumer waste materials

Made in the U.S.

Printer
DS Graphics

Contributors and Reviewers

J. Abraham, J. Andersson, A.Bergmann,
G. Bourdon, M. Carone, B. Chou,
K.Cohan, S. DeLand, D. Doherty,
A.Elgammal, T. Erkkinen, J. Friedman,
T.Gerke, A. Himmeldorf, K. Hyman,
W.Jin, T. Jones, J. Kelly, T. Kush, J. Lerche,
V. Leung, C. Lin, K. Lorenc, R. Mawrey,
T.McKay, S. Miller, S. Myschik,
A.Nehemiah, S. ODonnell, B. Patel,
P.Pilotte, S. Prasanna, R. Purser, G. Reith,
R. Rovner, G. Sandmann, H. Scharler,
F.Smith, B. Tannenbaum, A. Th,
G.Thomas, S. Velilla, M. Vetsch, E.Wetjen

Subscribe
mathworks.com/subscribe
Comments
mathworks.com/contact
Find Us Online

MATLAB AND SIMULINK IN THE WORLD

Physical Modeling
Engineers in the automotive, aerospace, manufacturing, and other industries use Simulink and Simscape to assemble
system-level models that include the control system and span mechanical, electrical, and other physical domains. By
simulating the plant model and the controller in a single environment, they optimize system-level performance and test their
designs under scenarios that would be difficult, expensive, or unsafe to test on physical prototypes.

enabled them to test the system under conditions that would be difficult or unsafe to test at sea.
mathworks.com/dcns

AIRBUS
Developing a fuel management system for the
A380 aircraft
The Airbus A380 is the largest commercial aircraft currently in
operation. Its 11 fuel tanks enable nonstop flights of more than
8000 miles. Airbus engineers modeled the control logic for the
A380s fuel management system in Simulink and Stateflow. This
model defines modes of operation on the ground (including refuel,
defuel, and ground transfer) and in flight (including center of gravity
control, load alleviation, and fuel jettison). The team developed
a parameterized plant model of the tanks, pumps, and valves, and
incorporated relays and other elements of the electrical power system
with Simscape. Engineers can reconfigure this model to represent
fuel systems for any Airbus aircraft.
mathworks.com/airbus

DCNS
Modeling and simulating a helicopter handling system
The SAMAHE handling system can transfer a 10-ton helicopter between a hangar and a navy ships flight deck in less than two minutes.
It operates safely in waves up to 6 meters high, and can be configured
for use on a variety of ships, from corvettes to frigates. Working in
Simulink and Simscape, DCNS built a three-dimensional mechanical model of the system model comprising the handling system, helicopter, and ship. DCNS ran more than 1200 simulations for various
helicopter mass configurations, center of gravity locations, and positions, as well as ship motion and wind conditions. The simulations

MathWorks News&Notes | 2 0 1 6 2 0 1 7

The pressure at the nipthe line of contact between rollsmust be

within 0.2 bar of its set value. Metso developed a digital hydraulic
system that consumes 98% less energy and is more reliable than the
proportional hydraulic systems traditionally used for nip control. The
team used Simulink and Simscape to model the control system, digital valves, the mechanical frame supporting the calender, and other
elements of the equipment. By simulating the hydraulic systems and
controls with the model, they could quickly explore fault conditions,
valve configurations, and other design options.
mathworks.com/metso

SANDIA NATIONAL LABORATORIES

Simulating microgrid and photovoltaic systems
As part of Hawaiis Clean Energy Initiative, a 1.2 megawatt photovoltaic solar farm was installed on Lanai, an island served by Maui
Electric Company. Maui Electric partnered with Sandia National
Laboratories to evaluate the battery capacity and control systems required for reliable operation. Sandia developed a model of the Lanai
microgrid using Simulink and Simscape Power Systems, and conducted simulations to assess various configuration and control options. Initial estimates for the Lanai system included a 700 kilowatthour battery. The Simulink simulations demonstrated that a battery
about half that size would be sufficient.
mathworks.com/sandia
GasTOPS
Developing propulsion control algorithms for the
USS Makin Island
The USS Makin Island is an 850-foot-long, twin-shaft amphibious assault ship in active service with the U.S. Navy. It uses a hybrid-electric
propulsion system with gas turbines for high-speed travel and electric motors for low-speed operation. Using Simulink and Simscape,
GasTOPS engineers developed models of the propulsion system, including submodels for the hull, propellers, shafting, gearboxes, motors,
generators, and gas turbines. For the electrical plant, they modeled six
diesel generators, eight transformers, numerous smart breakers, and
the distribution system, as well as electric motors and other loads. They
performed simulations to evaluate the system response to short circuits, generator failures, and various fault conditions.
mathworks.com/gastops

METSO
Developing a controller for an energy-saving digital
hydraulic system for papermaking equipment
In industrial papermaking equipment, the calender rolls must be precisely controlled to ensure smoothness and glossiness in the paper.

VINTECC
PLC system development for a multi-axle harvesting
machine
With a 780hp engine driving three independent rear axles and two
independent wheels on the front axle, the JPS Mega Star harvester
can collect and haul 100 tons of produce in a single load. Vintecc developed the harvesters complex control system. Using Simscape they
modeled tire and vehicle body elements; hydraulic pumps, motors,
and cylinders; powertrain components; and mechanical linkages.

They developed algorithms for each controller using Stateflow charts

to manage execution modes and Simulink PID Controller blocks to
control the harvesters hydraulic and mechanical systems. To verify
the traction control, axle alignment, cruise control, auto-reverse, and
other functions, they ran model-in-the-loop simulations of the controller and plant models.
mathworks.com/vintecc

VOLVO CONSTRUCTION EQUIPMENT

Streamlining product development with a real-time,
human-in-the-loop simulator
Volvo Construction Equipments Virtual Machine Simulator
(VMS) gives construction machine operators realistic visual, auditory, and motion feedback during simulation, enabling engineers to
evaluate new designs before a prototype is built. The Volvo CE team
used Simscape to model the physical system, with 3D mechanical
models of the boom, arm, and bucket connected to hydraulic circuits containing control and relief valves, a swing motor, and other
components. Integrating this model with models of the engine and
control system in Simulink resulted in a multidomain model of the
complete machine that supported control design tasks and realtime simulations in the VMS.
mathworks.com/volvo

LEARN MORE
Physical Modeling Solutions

mathworks.com/physical-modeling

User Stories

mathworks.com/user-stories

MathWorks News&Notes | 2 0 1 6 2 0 1 7

IN TODAYS AIRCRAFT, HYDRAULIC AND PNEUMATIC ACTUATION

systems are increasingly being replaced by electrical systems. Actuators for primary flight control surfaces,
as well as actuators in landing gear, braking systems, and fuel delivery systems, are now driven by power
electronics. The electric motors that drive these actuators need to be small, light, and inexpensive. They also
need to perform reliably for 50,000150,000 hours of normal flight operation and under a wide range of
failure conditions.

o meet these requirements, the Microsemi

Aviation Center of Excellence is developing a
line of Intelligent Power Solutions (IPS) based
on a power core module (PCM) designed
and tested with MATLAB and Simulink.
Model-Based Design has enabled us to push
our design to the limits because we can simulate failures, optimize performance, and lower
risk by conducting real-time reliability tests
of motor drive hardware and control software
early in the development process.

Modeling the PCM and Running

Closed-Loop Simulations
A complete power electrical control unit
consists of functions for pulse width modulation (PWM) control, data conversion, and
communications; filtering and protection; a
three-phase permanent magnet synchronous
motor (PMSM) drive; a control module; and
a monitoring module (Figure 1). The motor
current, motor speed, and actuator position
are fed into the monitoring module, and the

control module uses this information to direct the PCM to speed the motor up or slow it
down. Because this was a new design, we had
to develop the PCM without having working
versions of the monitoring module or control
module available to test it.
We modeled the PCM in Simulink, using Simscape Power Systems and Simscape
Electronics to model the three-phase PMSM
drive and electronic components and the control and monitoring modules. We then ran

Real-Time Simulation and Testing of Power

Electronics on a More Electric Aircraft
By Shane ODonnell, Microsemi
FIGURE 1. Architectural diagram of the power core module within the larger power electrical control unit.

MathWorks News&Notes | 2 0 1 6 2 0 1 7

MathWorks News&Notes | 2 0 1 6 2 0 1 7

What We Learned
Through our extensive modeling and simulations, we established that units equipped with
motor drives based on silicon carbide (SiC)
MOSFETs operate at a temperature approximately 40 Celsius lower than similar units
with IGBTs.
Because active cooling is not possible with
todays smaller and lighter hardware designs,
managing the temperature of the device while
in operation is vital to ensuring that it will func-

tion reliably for 150,000 flight hours. Simulations also showed that power dissipation with
IGBTs is considerably higher than with SiC
MOSFETs (Figure 4). These insights informed
our design decisions for the PCM and point to
SiC MOSFETS as an enabling technology as the
industry moves towards increased fly-by-wire
controls in the more electric aircraft (MEA).
Simulink, Simulink Real-Time, and
Speedgoat target hardware have enabled us to
demonstrate the application-specific reliabil-

ity of our early designs without installing the

units on an actual aircraft. With Model-Based
Design, we can do continuous validation and
verification without waiting until all aspects of
the power electrical control unit are developed.
The feedback weve received from our customers has been very positive. With our realtime simulation results, we are confident that
we can meet the PCMs reliability targets as
we continue to reduce the units size, weight,
and cost.

LEARN MORE
Real-Time Simulation and Testing with
Simulink Real-Time 39:57

mathworks.com/video-100956

Concurrent Execution with Simulink Real-Time

and Multicore Target Hardware
mathworks.com/concurrent-execution

FIGURE 2. The Speedgoat setup with prototype PCM hardware.

both the PCM controller and the other modules were run on the target hardware for realtime tests. In a different setup, we deployed our
controller to a production ProASIC3 FPGA on
the PCM and ran hardware-in-the-loop tests
with the target hardware system performing
the functions of the control and monitoring
modules. We tested normal operation using
both test setups. We also tested the controllers
response to several fault conditions to perform
failure mode, effects, and criticality analysis.
FIGURE 3. A plot of motor current for a typical
flight mission of a single-aisle aircraft.

closed-loop simulations to characterize the

systems electrical and mechanical behavior.
Next, we deployed the three models to a
Spartan-6 FPGA in the Speedgoat target system using Simulink Coder and Simulink
Real-Time (Figure 2). The modules communicate through a low-voltage differential
signaling (LVDS) interface. In one test setup,

MathWorks News&Notes | 2 0 1 6 2 0 1 7

Testing Real-World Flight Profiles

Under Real-World Conditions
To demonstrate the PCM under realistic
flight profiles, we developed Simulink and
Stateflow models that translate flight characteristics into electrical and mechanical requirements for an actuation system. As the
aircraft proceeds through the typical phases
of a flighttaxiing, taking off, climbing,
cruising, descending, approaching, and land-

ingthe motor current demands for an aileron actuator, for example, vary significantly.
Simulations that we ran using our Simulink
and Stateflow mission and flight profile models enabled us to accurately estimate motor
current demands for ailerons and other components on specific aircraft (Figure 3).
For our reliability tests, we generated aircraft-specific motor current demands based
on the flight profile simulation results. We use
environmental chambers that vary the pressure and temperature. For example, the ambient temperature in Boston is much lower than
that of Dubai in summer, and our tests must
take that into account. With the environmental chambers, we can expose the systems to
temperatures of -55 Celsius and pressures of
less than 0.2 bar. Long-term reliability tests
representative of 150,000 flight hours require
careful monitoring and thorough analysis of
the results. We conduct this monitoring and
data analysis in MATLAB.

FIGURE 4. Plots showing power dissipation over time for an IGBT 3-phase bridge (top) and
SiC MOSFET 3-phase bridge (bottom).

MathWorks News&Notes | 2 0 1 6 2 0 1 7

REAR-END COLLISIONS ARE THE MOST COMMON TYPE OF ACCIDENT

for freight-carrying trucks and other heavy vehicles. To reduce the risk of rear-end collisions, in 2015 the EU
mandated advanced emergency braking systems (AEBS) for all new vehicles.

Developing and Verifying Sensor Fusion

Methods for Advanced Emergency Braking
Systems on Scania Trucks and Buses
By Jonny Andersson, Scania

10

MathWorks News&Notes | 2 0 1 6 2 0 1 7

ike other advanced driver assistance systems (ADAS), an AEBS uses input from sensors to screen the environment. When a collision is imminent, the system warns the driver
with an audio alarm. If the driver does not respond, it applies a warning brake. If the driver
still does not respond, the system applies the
brakes fully to avoid the collision (Figures 1a
and 1b). The AEBS also provides brake assist: When the driver brakes, but with insufficient force to avoid a collision, the system
calculates and then applies the required extra
braking force.
AEBS uses both radar and camera sensors mounted on the front of the vehicle to
scan for objects in the area ahead. The system leverages the particular strengths of each
sensor to gain a more precise environment
model. Radar sensors excel at determining
an objects range, relative velocity, and solidity but are less able to determine its shape or
lateral position. A system using radar alone
would find it difficult to distinguish a car
parked at the side of the road from one in the
drivers lane. Cameras, on the other hand, can
pinpoint an objects size and lateral position
but do not detect range well and are unable
to assess density (a dense cloud may be perceived as a solid object).
My colleagues and I built a sensor fusion
system that matches and merges data from
both sensors into a single object. The system
uses four weighted propertieslongitudinal
speed and position and lateral speed and positionto calculate the probability that both
sensors have detected the same object. Once
the sensor fusion system has identified an object in the host vehicles path, it passes the ob-

FIGURE 1A. AEBS overview.

FIGURE 1B. A typical AEBS scenario: truck with AEBS installed approaching a slow-moving vehicle.

jects position and the vehicles projected path

to the AEBS, which determines when to alert
the driver or engage the brakes.
Our group had previously used ModelBased Design to develop an adaptive cruise
control system using radar technology, but
we had never before developed a sensor fusion system. Because it was a new design, we
knew we would need a readable, understandable architecture to visualize signal flow. We
also anticipated many design iterations, so we

wanted an easy way to visualize results and

debug our designs. In addition, we wanted
to save time by generating code, but the code
had to be efficient, as the CPU load on our
electronics control unit (ECU) was already
about 60% when we started the sensor fusion project. Lastly, we needed to thoroughly
verify our designour plan was to run simulations based on more than 1.5 million kilometers worth of sensor data. Model-Based
Design met all these requirements.

MathWorks News&Notes | 2 0 1 6 2 0 1 7

11

tected by any sensor and the unified object

perceived by the sensor fusion system. These
MATLAB objects enabled us to quickly step
forward and backward in time as we visualized the data.
We used the same tool during road tests to
visualize live data coming in from the vehicle
network (Figure 4).

FIGURE 2. Simulink model of the sensor fusion system showing independent functional blocks.

FIGURE 3. Sensor visualization tool developed in MATLAB.

Building the Sensor Fusion System

We began by partitioning the system design into functional units, such as object
matching and projected path placement,
and building a separate Simulink block
for each unit. The result was a clear software architecture with well-defined interfaces (Figure 2). We wrote MATLAB
code for the track association, to compute
variances, calculate weighted probabilities,
and perform other tasks that are easier to
implement with a script than with blocks,
and incorporated this code into our Simulink
model with MATLAB Function blocks. These
algorithm blocks made it easy for team mem-

12

MathWorks News&Notes | 2 0 1 6 2 0 1 7

bers to merge their algorithms and integrate

them with the control system.
To debug and refine our initial design, we
ran simulations using recorded radar sensor data, corresponding camera images, and
other vehicle sensor data. During debugging we found it useful to visualize the sensor data alongside a camera view from the
front of the vehicle. We built a visualization
tool in MATLAB that displays sensor fusion
data synchronized with a web camera view
of the surrounding traffic (Figure 3). Taking
advantage of the object-oriented programming capabilities of MATLAB, the tool uses
a MATLAB class to represent each object de-

Implementing the System and

Optimizing Performance
To deploy the sensor fusion system to the ECU,
we generated C code from our Simulink model
with Embedded Coder. With code generation, we were able to get to an implementation
quickly, as well as avoid coding errors. Most of
the ECU processors resources were allocated
to maintenance functionsmonitoring dashboard alerts, physical estimations, data gateway, adaptive cruise control, and so on. As a
result, we needed to optimize our initial design
to increase its efficiency.
In order to get the most performance
out of the generated code, we worked with
the MathWorks pilot team, who helped us
optimize the code generated from MATLAB
Coder. To further reduce the processing load
we divided the model into separate parts that
were executed on alternating cycles. For example, instead of running calculations for
stationary and moving objects on every cycle,
we ran them on alternating cycles. We realized that the processor was bogged down by
the trigonometric functions our system was
calling. To alleviate this problem, we wrote
trigonometric approximation functions in C
and called them from a MATLAB Function
block. These modifications not only increased
the efficiency of the sensor fusion code, they
also enabled the AEBS software to react faster,
which is vital when vehicles are traveling at
highway speeds and every millisecond counts.
Verifying and Refining the Design
We tested the design in-vehicle on a closed
course, but we needed to know how the system would react in real-world driving scenarios, such as different weather conditions, traf-

FIGURE 4. A controlled road test of the

AEBS software. The trapezoidal object
between the two vehicles is a soft target,
designed to resemble a vehicle, that is used
to fool the radar and the camera.

fic patterns, and driver behaviors. It would be

impractical as well as unsafe to test the AEBS
directly under these conditions. Instead, we
used a simulation-based workflow. We began
by gathering data from a fleet of trucks. We
decided to collect all data available on the
ECUnot just data from the radar and camera used for sensor fusionas well as images
from a separate reference camera.
Using this fleet test data we ran simulations to identify interesting driving scenariosscenarios in which the AEBS intervened
to warn the driver or engage the brakes, and
scenarios in which the system could have intervened but did notfor example, when the
driver pressed the horn and braked simultaneously, swerved, or braked sharply. Focusing on these scenarios, we then analyzed the
performance of the AEBS to identify areas in
which we could improve the design.
We needed to resimulate every time we
updated the AEBS software. However, with
more than 80 terabytes of real traffic data
logged over more than 1.5 million of kilometers of driving, it took several days to run
a single simulation.
To accelerate the simulations, we built
an emulator using code generated from our
Simulink models with Embedded Coder. The
emulator reads and writes the same MATfiles as our Simulink model but runs simulations 150 times faster. To further speed up
simulations, we wrote MATLAB scripts that
run simulations on multiple computers in

FIGURE 5. The Situation Classification Assistant Module, a MATLAB based tool for processing
logged ECU data and automatically identifying situations relevant to emergency braking.

our department as well as on dedicated multiprocessor servers, where we ran up to 300

simulations in parallel. With this setup, we
cut the time needed to simulate all 1.5 million
kilometers to just 12 hours. When we identified a new interesting scenario in the emulator, we reran the simulation in Simulink
to analyze it in depth.
Identifying and classifying potentially interesting scenarios in terabytes of data was
a tedious and time-consuming task, so we
developed the Situation Classification Assistant Module, a MATLAB based tool that
automates that part of the process (Figure
5). The tool generated a list of events from
the simulations, such as collision warnings,
warning brakes, and full brakes initiated by
the system, as well as hard brakes and sharp
turns initiated by the driver. We could then
compare these lists for any two versions of
our software.
The ability to perform extensive simulations enhanced the robustness and safety of
the AEBS function and production code implementation for the ECU. It also enabled us
to make changes more quickly. We had confidence in those changes because we were using all the available data in our simulations
to test thousands of scenarios.

Deploying the Generated Code in

Production ADAS
Most Scania trucks and buses are now
equipped with AEBS running production code
generated from Simulink models and verified
via extensive simulations. We have reused our
sensor fusion system design in Scanias adaptive cruise control system, and there are now
more than 100,000 units on the road.

LEARN MORE
Radar System Modeling and Simulation for
Automotive Advanced Driver Assistance
Systems 26:00
mathworks.com/video-107121

Cost and Benefit of Model-Based

Development of Automotive Software
Results of a Global Study 30:30

mathworks.com/video-92771

Model-Based Approach to Resource-Efficient

Object Fusion for an Autonomous Braking
System 21:21
mathworks.com/video-108103

MathWorks News&Notes | 2 0 1 6 2 0 1 7

13

WHEN I SAW MY GRADUATE STUDENTS COMPETING TO SEE WHO

could maintain the highest average speed on the electric motor-powered skateboard they had built, I knew
that I had achieved my two most important goals for their project. Not only had they gained a deep understanding of Model-Based Design, they also had a great deal of fun doing so.

evin Leiffels and Raphael-David

Volmering designed and built the e-longboard as a final project for my course on
electric drives and field-oriented control
(vector control). Powered by two independent brushless DC motors (BLDCs), the
longboard can carry a rider up to 25 km
(15.5 miles) with a top speed of more than
40 km/hour (25 mph).

HS Bochum Students Design and Build

a Motor Controller for an E-Longboard
with Model-Based Design
By Dr. Arno Bergmann, Bochum University of Applied Sciences (HS Bochum)

14

MathWorks News&Notes | 2 0 1 6 2 0 1 7

The Value of Hands-On Projects

with Model-Based Design
Students acquire a much deeper understanding of engineering concepts by completing
hands-on projects than by listening to a lecture. Even students who get excellent grades
on tests dont truly understand concepts such
as field-oriented control for electric motors
until they have applied those concepts in the
real world.
Model-Based Design enables students to
tackle meaningful projects in the limited time
available. For example, in a single semester
Raphael completed the e-longboards printed
circuit board design while Kevin designed,
implemented, and tested the controller. Kevin
generated more than 15,000 lines of code
much more code than he could have written
by hand in one semester.
HS Bochums acquisition of a Total Academic Headcount (TAH) license was a major milestone for the university and a boon
to my course. The students are free to use
MATLAB and Simulink on assignments
both inside and outside the lab. This flex-

FIGURE 1. The underside of the e-longboard showing two BLDC motors on the left.

ibility is highly motivating. MATLAB and

Simulink are industry standard tools, and the
students know that to develop the skills required in industry they need more practice
with the tools than they can get just by working in the lab. The TAH license also makes
my job easier because I no longer have to
keep track of individual licenses.
Establishing Project Requirements
For all student projects I establish a set of basic
requirements and then let the students come
up with their own ideas. The e-longboard project had to include a power device, two separate
motors that are not mechanically connected,
and a DSP on which the field-oriented control
is implemented (Figure 1). The control part
of the project must be challenging but simple
enough for a student to complete in a single
semester. Each student must use Model-Based

Design. In my view, Model-Based Design is the

state of the art for control system development
because it enables early and thorough verification of the design, low implementation costs,
portability to multiple hardware platforms,
and short development times.
Before Model-Based Design became a
requirement, the students did not learn as
much about how real engineering projects
are conducted. For example, they would
often start with a prepackaged third-party
motor controller, hack a few lines of C code
together to get the motor spinning, and then
move directly to constructing the rest of the
system for trial-and-error testing. When
they were done they would find out that the
system did not meet their power demands
or satisfy all the real-time requirements.
Because they had not verified their designs
via simulation, they only discovered these

MathWorks News&Notes | 2 0 1 6 2 0 1 7

15

boards range. It wasnt long before the students were competing to see who could
deplete the batteries fastest by maintaining
the top average speed. Following test runs,
Kevin post-processed metrics captured during the runs. To visualize the e-longboards
speed response, for example, he created a
combined graph of desired speed and actual
speed in MATLAB (Figure 4).
Next Steps for the Student and
the Course
When Kevin completed his studies at HS
Bochum he began working for an engineering company that uses Model-Based

Design. When the company learned of

Kevins success with the e-longboard they
hired him on the spot.
One of the most valuable lessons Kevin
learned is the importance of verifying requirements as thoroughly as possible via modeling
and simulation before the actual implementation. The company that he now works for will
be supporting our upcoming use of Simulink
Verification and Validation because they,
too, are seeking to verify requirements at earlier stages in development.
I am planning a few changes for the
next group of students taking the fieldoriented controls course. I will still re-

quire hands-on projects with Model-Based

Design, and I will be encouraging more
students to build an enhanced version
of the e-longboard. Next years version
of the course will place a stronger emphasis on requirements, logical modes,
and physical modeling using Simulink
Verification and Validation, Stateflow, and
Simscape Power Systems, respectively.
I plan to use the board to inspire next years
students. I will bring the board to class and
let the students take a test drive outside to stir
their interest before they head to the lab to
begin their own projects using Model-Based
Design and field-oriented control.

LEARN MORE
Simulation of FOC Using PMSM Model
mathworks.com/foc-pmsm

MATLAB and Simulink in the World:

The MATLAB Enabled Campus
mathworks.com/miw-tah

FIGURE 2. Simulink model for simulating field-oriented control with a permanent magnet synchronous machine.

problems when it was too late to do anything

about them.
Supplementing the basic project requirements, the students included several requirements specific to the e-longboard. In addition
to specifying a minimum range for a single
battery charge, these requirements defined a
maximum braking distance and a minimum
hill grade that the e-longboard would need to
be able to climb.
Designing and Implementing
the Controller
A principal design challenge in field-oriented
control is maintaining a 90 angle between
the rotor and stator field in the motor. In addition to minimizing changes of the magnetic
flux to enable fast transient responses, maintaining this angle maximizes motor torque
for a given current. Kevin used Hall sensors

16

MathWorks News&Notes | 2 0 1 6 2 0 1 7

to measure rotor position, which is a key input to both the field-oriented control and the
boards speed control loop.
Kevin based his controller design on an
example field-oriented control project from a
MathWorks webinar. The example included a
Simulink model for controlling the speed and
torque of a three-phase permanent magnet
synchronous machine (PMSM), which was
modeled using Simscape Power Systems
(Figure 2). We found the example to be a good
implementation of field-oriented control.
After downloading the example project
from mathworks.com, Kevin modified the
parameters for the e-longboard, removed
unneeded parts, and added features. After
running simulations in Simulink, Kevin used
Embedded Coder to generate C code for
the boards TI F28069 microcontroller. At
that point, he began evaluating the real-time

response of the system to see if it met the realtime requirements he had established.
Kevin and Raphael worked largely on
their own, meeting me once a week so that I
could monitor their progress. On this project, Kevin applied and expanded the basic
knowledge of Model-Based Design with
MATLAB and Simulink that he had acquired in earlier control design electives at HS
Bochum. He relied on technical support from
MathWorks to resolve any technical issues that
he encountered, and that enabled him to work
with surprisingly little assistance from me.
Once the printed circuit board was ready
(Figure 3) and the rest of the e-longboard had
been constructed, Kevin and his classmates
began testing the board in and around the HS
Bochum campus.
Kevin rode around a nearby lakea
distance of more than 25 kmto test the

FIGURE 3. The e-longboard printed circuit.

FIGURE 4. Plot of desired speed (pink) and actual speed (yellow) showing how closely the
two matched during a test run.

MathWorks News&Notes | 2 0 1 6 2 0 1 7

17

WHEN YOU STUDY A PAINTING, CHANCES ARE THAT YOU CAN

make several inferences about it. In addition to understanding the subject matter, for example, you may
be able to classify it by period, style, and artist. Could a computer algorithm understand a painting well
enough to perform these classification tasks as easily as a human being?

y colleagues and I at the Art and

Artificial Intelligence Laboratory at Rutgers
University explored this question using
MATLAB, Statistics and Machine Learning
Toolbox, and a database of thousands of
paintings from the past six centuries. We also
addressed two other intriguing questions about
the capabilities and limitations of AI algorithms: whether they can identify which paintings have had the greatest influence on later artists, and whether they can measure a paintings
creativity using only its visual features.

Creating Computer Vision and

Machine Learning Algorithms That
Can Analyze Works of Art
By Ahmed Elgammal, Rutgers University

18

MathWorks News&Notes | 2 0 1 6 2 0 1 7

Extracting Visual Features for

Classifying Paintings
We wanted to develop algorithms capable of
classifying large groups of paintings by style
(for example, as Cubist, Impressionist, Abstract Expressionist, or Baroque), genre (for
example, landscape, portrait, or still life), and
artist. One requirement for this classification
is the ability to recognize color, composition,
texture, perspective, subject matter, and other
visual features. A second is the ability to select
those visual features that best indicate similarities between paintings.
Working with MATLAB and Image
Processing Toolbox, we developed algorithms to extract the visual features of a painting. The feature extraction algorithm is fairly
common in computer vision, and straightforward to implement. The more challenging
task was finding the best machine learning

FIGURE 1. Left: Diego Velzquezs Portrait of Pope Innocent X. Right: Francis Bacons
Study After Velzquezs Portrait of Pope Innocent X.

techniques. We began by testing support vector machines (SVMs) and other classification
algorithms in Statistics and Machine Learning
Toolbox to identify visual features that are useful in style classification. In MATLAB, we then
applied distance metric learning techniques to
weight the features and thereby improve the
algorithms ability to classify paintings.
The algorithms we developed classified the
styles of paintings in our database with 60%
accuracy, where chance performance would
have been about 2%. While art historians can
perform this task with much more than 60%

accuracy, the algorithm outperforms typical

non-expert humans.
Using Machine Learning to Uncover
Artistic Influences
Once we had algorithms that could reliably
identify similarities between pairs of paintings,
we were ready to tackle our next challenge: using machine learning to reveal artistic influences. Our hypothesis was that visual features
useful for style classification (a supervised
learning problem) could also be used to determine influences (an unsupervised problem).

MathWorks News&Notes | 2 0 1 6 2 0 1 7

19

FIGURE 2. Left: Frederic Bazilles Bazilles Studio; 9 rue de la Condamine. Right: Norman Rockwells Shuffletons Barbershop. Yellow circles
indicate similar objects, red lines indicate similar composition, and the blue rectangle indicates a similar structural element.

Art historians develop theories of artistic

influence based on how the artists worked,
traveled, or trained with contemporaries. Our
MATLAB based machine learning algorithms
used only visual elements and dates of composition. We hypothesized that an algorithm
that took into account objects and symbols
in the painting would be more effective than
one that relied on low-level features such as
color and texture. With this in mind, we used
classification algorithms that were trained on
Google images to identify specific objects.
We tested the algorithms on more than
1700 paintings from 66 different artists working over a span of 550 years. The algorithm
readily identified the influence of Diego
Velazquezs Portrait of Pope Innocent X on
Francis Bacons Study After Velazquezs Portrait of Pope Innocent X (Figure 1).
The similarities in composition and subject
matter between these two paintings are easy
even for a layman to spot, but the algorithm
also produced results that surprised the art
historians we worked with. For example, our
algorithm identified Bazilles Studio; 9 rue
de la Condamine, painted by French Impres-

20

MathWorks News&Notes | 2 0 1 6 2 0 1 7

sionist Frederic Bazille in 1870, as a possible

influence on Norman Rockwells Shuffletons
Barbershop, completed 80 years later (Figure
2). Although the paintings might not look similar at first glance, a closer examination reveals
similarities in composition and subject matter,
including the heaters in the lower right of each
work, the group of three men in the center, and
the chairs and triangular spaces in the lower left.
In our data set, the algorithms correctly
identified 60% of the 55 influences recognized by art historians, suggesting that visual similarity alone provides sufficient information for algorithms (and possibly for
humans) to determine many influences.
Measuring Creativity by Solving a
Network Centrality Problem
Recently, our research has focused on developing algorithms to measure creativity in art. We
based this project on a widely used definition
that identifies an object as creative if it is both
novel and influential. In these terms, a creative
painting will be unlike the paintings that came
before it (novel), but similar to those that came
after it (influential).

In addressing this problem, we once again

saw an opportunity to apply our MATLAB algorithms for identifying similarities between
paintings. In MATLAB we created a network
in which the vertices are paintings and each
edge represents the similarity between the
two paintings at its vertices. Through a series of transformations on this network we
saw that making inferences about creativity
from such a graph is a network centrality
problem, which can be solved efficiently using MATLAB.
We tested our creativity algorithms on
two data sets containing more than 62,000
paintings. The algorithm gave high scores to
several works recognized by art historians as
both novel and influential, including some of
the works shown in Figure 3. Ranking even
higher than Pablo Picassos Young Ladies
of Avignon (1907) in the same period were
several paintings by Kazimir Malevich. This
result initially surprised me, as I knew little
about Malevichs work. I have since learned
that he was the founder of the Suprematism
movement, one of the earliest developments
in abstract art.

FIGURE 3. Computed creativity scores (y-axis) for paintings from 1400 to 2000 (x-axis), showing selected highest scoring paintings for
individual periods.

To perform a basic validation of our algorithm, we changed the date on specific

works of art, effectively shifting them backwards or forwards in time. In these time
machine experiments, we saw significant
creativity score increases for Impressionist
art moved back to the 1600s and significant
reductions for Baroque paintings moved
forward to the 1900s. The algorithms correctly perceived that what was creative 300
years ago is not creative today, and that
something that is creative now would have
been much more creative if introduced far
in the past.
A Scalable and Extensible
Framework for Arts Research
Humans have the innate perceptual skills to
classify art, and they excel at identifying simi-

larities in pairs of paintings, but they lack the

time and patience to apply these skills objectively to thousands or millions of paintings.
Handling tasks at this scale is where computers come into their own. By developing machine learning algorithms that have perceptual capabilities similar to humans, our goal is
to provide art historians with tools to navigate
vast databases of images.
The framework we developed in MATLAB
for identifying similarities and measuring
creativity is not confined to art. It could be
applied to literature, music, or virtually any
other creative domain, as long as the individual works can be encoded in a way that is
accessible to the algorithms.
For now, however, our focus remains on the
visual arts. We are interested not only in ensuring that machine learning algorithms pro-

duce good results but also in how they arrive

at those results. In this area, too, MATLAB
is a tremendous advantage because it provides
many ways to quickly and easily visualize results. These visualizations enable us to understand the results and use them to inform
ongoing AI research.

LEARN MORE
Machine Learning with MATLAB 41:26

mathworks.com/video-81984

Woods Hole Oceanographic Institution

Improves Underwater Imaging
mathworks.com/woods-hole

Applying Modern PDE Techniques to Digital

Image Restoration
mathworks.com/image-restoration

MathWorks News&Notes | 2 0 1 6 2 0 1 7

21

Deep Learning for Computer

Vision with MATLAB
By Avinash Nehemiah and Valerie Leung, MathWorks
FIGURE 1. Workflow for using a pre-trained CNN to extract features for a new task.

We initialize an imageDatastore to access the images in MATLAB.

COMPUTER VISION ENGINEERS HAVE USED MACHINE LEARNING

%% Set up image data

dataFolder = ' \data\PetImages';

techniques for decades to detect objects of interest in images and to classify or identify categories of objects.

categories = {'Cat', 'Dog'};

They extract features representing points, regions, or objects of interest and then use those features to train a

imds = imageDatastore(fullfile(dataFolder, ...



function Iout = readAndPreprocessImage(filename)


I = imread(filename);

categories), 'LabelSource', 'foldernames');

% Some images may be grayscale. Replicate the image


model to classify or learn patterns in the image data.

We then select a subset of the data that gives us an equal number of

dog and cat images.

% 3 times to create an RGB image.

if

tbl = countEachLabel(imds)

ismatrix(I)
I = cat(3,I,I,I);

end

n traditional machine learning, feature selection is a time-consuming manual process. Feature extraction usually involves processing each image with one or more image processing operations, such
as calculating gradient to extract the discriminative information
from each image.
Enter deep learning. Deep learning algorithms can learn features,
representations, and tasks directly from images, text, and sound, eliminating the need for manual feature selection.
Using a simple object detection and recognition example, this article illustrates how easy it is to use MATLAB for deep learning, even
without extensive knowledge of advanced computer vision algorithms
or neural networks.
Getting Started
The goal in this example is to train an algorithm to detect a pet in a
video and correctly label the pet as a cat or a dog. Well be using a
convolutional neural network (CNN), a specific type of deep learning algorithm that can both perform classification and extract features
from raw images.
To build the object detection and recognition algorithm in MATLAB,
all we need is a pre-trained CNN and some dog and cat images. Well use
the CNN to extract discriminative features from the images, and then
use a MATLAB app to train a machine learning algorithm to discriminate between cats and dogs.

22

We use the readAndPreprocessImage function to resize the images to 227x227 pixels.

MathWorks News&Notes | 2 0 1 6 2 0 1 7

Importing a CNN Classifier

We begin by downloading a CNN classifier pretrained on ImageNet, a
database containing over 1.2 million labeled high-resolution images in
1000 categories. In this example well be using the AlexNet architecture.

%% Use the smallest overlap set

minSetCount = min(tbl{:,2});

Iout = imresize(I, [227 227]);

% Use splitEachLabel method to trim the set.

imds = splitEachLabel(imds, minSetCount, ...


websave('\networks\imagenet-caffe-alex.mat',...


imagenet-caffe-alex.mat');

% Notice that each set now has exactly the same


% number of images.

We import the network into MATLAB as a SeriesNetwork using

Neural Network Toolbox, and display the architecture of the CNN. The
SeriesNetwork object represents the CNN.
% Load MatConvNet network into a SeriesNetwork

convnet = helperImportMatConvNet(cnnFullMatFile);


end

'randomize');

'http://www.vlfeat.org/matconvnet/models/beta16/

% Resize the image as required for the CNN.



Performing Feature Extraction

We want to use this new dataset with the pretrained AlexNet CNN.
CNNs can learn to extract generic features that can be used to train a

countEachLabel(imds)

Since the AlexNet network was trained on 227x227-pixel images, we

have to resize all our training images to the same resolution. The following
code allows us to read and process images from the imageDatastore
at the same time.

% View the CNN architecture



%% Pre-process Images For CNN

convnet.Layers

% Set the ImageDatastore ReadFcn

imds.ReadFcn = @(filename)readAndPreprocessImage...


Weve stored the images in separate cat and dog folders under a parent
called pet_images. The advantage of using this folder structure is that
the MATLAB imageDatastore we create will be able to automatically
read and manage image locations and class labels. (imageDatastore
is a repository for collections of data that are too large to fit in memory.)

(filename);

%% Divide data into training and testing sets


[trainingSet, testSet] = splitEachLabel(imds, ...



0.3, 'randomize');

FIGURE 2. Visualization of first layer filter weights.

MathWorks News&Notes | 2 0 1 6 2 0 1 7

23

w1 = convnet.Layers(2).Weights;
% Scale and resize the weights for visualization

w1 = mat2gray(w1);
w1 = imresize(w1,5);
% Display a montage of network weights. There are 96

% individual sets of weights in the first layer.
figure
montage(w1)
title('First convolutional layer weights')


FIGURE 3. Classification Learner app.

Notice that the first layer of the network has learned filters for capturing blob and edge features. These primitive features are then processed by deeper network layers, which combine the early features to
form higher-level image features. These higher-level features are better
suited for recognition tasks because they combine all the primitive features into a richer image representation. You can easily extract features
from one of the deeper layers using the activations method.
The layer right before the classification layer fc7 is a good place to
start. We extract training features using that layer.
featureLayer = 'fc7';
trainingFeatures = activations(convnet, ...


FIGURE 4. Result of using the trained pet classifier on an image of a cat.

new classifier to solve a different problemin our case, classifying cats

and dogs (Figure 1).
We pass the training data through the CNN and use the activations method to extract features at a particular layer in the network.
Like other neural networks, CNNs are formed using interconnected layers of nonlinear processing elements, or neurons. Input and output layers connect to input and output signals, and hidden layers provide nonlinear complexity that gives a neural network its computational capacity.
While each layer of a CNN produces a response to an input image,
only a few layers are suitable for image feature extraction. There is no
exact formula for identifying these layers. The best approach is to simply try a few different layers and see how well they work.
The layers at the beginning of the network capture basic image features, such as edges and blobs. To see this, we visualize the network
filter weights from the first convolutional layer (Figure 2).

trainingSet, featureLayer, '

MiniBatchSize', ...

32, 'OutputAs', 'columns');

Training an SVM Classifier Using the Extracted

Features
Were now ready to train a shallow classifier with the features extracted in the previous step. Note that the original network was trained to
classify 1000 object categories. The shallow classifier will be trained
to solve the specific dogs vs. cats problem.

24

MathWorks News&Notes | 2 0 1 6 2 0 1 7

FIGURE 7. Accurately classified cats and dogs.

The Classification Learner app in Statistics and Machine Learning Toolbox lets us train and compare multiple models interactively (Figure 3).
Alternatively, we could train the classifier in our MATLAB script.
We split the data into two sets, one for training and one for
testing. Next, we train a support vector machine (SVM) classifier
using the extracted features by calling the fitcsvm function using
trainingFeatures as the input or predictors and trainingLabels
as the output or response values. We will cross-validate the classifier on
the test data to determine its validation accuracy, an unbiased estimate
of how the classifier would perform on new data.

LEARN MORE
Deep Learning for Computer Vision with MATLAB 36:50
mathworks.com/video-116080

Object Detection Example Code (download)

mathworks.com/fx-57116

%% Train a classifier using extracted features


trainingLabels = trainingSet.Labels;

% Here I train a linear support vector machine

% (SVM) classifier.
svmmdl = fitcsvm(trainingFeatures ,trainingLabels);

% Perform cross-validation and check accuracy

cvmdl = crossval(svmmdl,'KFold',10);

fprintf('kFold CV accuracy: %2.2f\n',...



% Get the network weights for the second


% convolutional layer

FIGURE 6. Image Region Analyzer app.

Performing Object Detection

In most images and video frames, there is a lot going on. For example, in
addition to a dog, there could be a tree, or a flock of pigeons, or a raccoon
chasing the dog. Even a reliable image classifier will only work well if we
can locate the object of interest, crop the object, and then feed it to the
classifierin other words, if we can perform object detection.
For object detection we will use a technique called optical flow,
which uses the motion of pixels in a video from frame to frame. Figure
5 shows a single frame of video with the motion vectors overlaid.
The next step in the detection process is to separate out pixels that
are moving and then use the Image Region Analyzer app to analyze the
connected components in the binary image to filter out noise caused
by the motion of the camera. The output of the app is a MATLAB
function that can locate the pet in the field of view (Figure 6).
We now have all the pieces we need to build a pet detection and
recognition system (Figure 7). The system can:
Detect the location of the pet in new images using optical flow
Crop the pet from the image and extract features using a pretrained
CNN
Classify the features using the SVM classifier we trained to determine if the pet is a cat or a dog
In this article we used an existing deep learning network to solve
a different task. You can use the same techniques to solve your own
image classification problemfor example, classifying types of cars in
videos for traffic flow analysis, identifying tumors in mass spectrometry data for cancer research, or identifying individuals by their facial
features for security systems.

FIGURE 5. A single frame of video showing the motion vectors overlaid.

1-cvmdl.kfoldLoss)

We can now use the svmmdl classifier to classify an image as a cat or

a dog (Figure 4).

MathWorks News&Notes | 2 0 1 6 2 0 1 7

25

The Joy of Generating C Code

from MATLAB
By Bill Chou, MathWorks

FIGURE 1. Three-step iterative workflow for generating code.

ENGINEERS HAVE TRANSLATED LOW-LEVEL LANGUAGES LIKE C

into machine code for decades using compilers. But is it possible to translate a high-level language like
MATLAB to C using coders? Most engineers would agree that its possible in theorybut does it work
in practice? Is the generated code readable or spaghetti? Efficient or bloated? Fast or slow? And does it
support industrial workflows, or just R&D?

his article addresses these concerns

head-on. It provides tips and best practices for
working with MATLAB Coder, as well as industry examples of successful applications of
generated code by companies such as Delphi,
Baker Hughes, iSonea, and dorsaVi.
Comparing MATLAB and C Code:
A Multiplication Example
The simple MATLAB function below multiplies two inputs.

double myMult(double a, double b)

c[i0 + 3 * i1] = 0.0;


return a * b;

f
or (i2 = 0; i2 < 4;

i2++) {

c[i0 + 3 * i1] +=

As you can see, the generated code maps

clearly back to the MATLAB code.
The same piece of MATLAB code, when
given two matrix inputs, generates three nested for-loops in C:

a[i0 + 3 * i2] * b[i2

+ (i1 << 2)];
}
}
}
}

#include "myMult.h"
function c = myMult(a, b)
void myMult(const double a[12],
% Multiply two inputs

const double b[20], double c[15])

c = a * b;

{
int i0;

Given scalar inputs, MATLAB Coder generates the following C code:

int i1;
int i2;
for (i0 = 0; i0 < 3; i0++) {

#include "myMult.h"

26

MathWorks News&Notes | 2 0 1 6 2 0 1 7

FIGURE 2. Left: Automated checks for features and functions not supported for code generation. Right: Automated analysis and proposal for input data
type and sizes.

f
or (i1 = 0; i1 < 5; i1++)

Recommended Three-Step Iterative

Workflow
The simple function shown above can be
implemented in a single step. But for more
substantial projects, we recommend a structured approach using a three-step iterative
workflow (Figure 1).
1. Prepare your algorithm for code generation. Examine and modify the MATLAB

code to introduce implementation considerations needed for low-level C code, and

use the MATLAB language and functions
that support code generation.
2. Test the MATLAB codes readiness for code
generation using default settings. Check for
run-time errors by generating and executing a MEX file. If successful, move to the
next step. If not, repeat step 1 until you can
generate a MEX function.
3. Generate C code or keep the MEX function
from step 2. You can iterate on the MATLAB
code to optimize either the generated C code
(for look and feel, memory, and speed) or
the MEX function (for performance).
The MATLAB Coder app guides you

through this iterative process while enabling

you to stay within the MATLAB environment. It analyzes your MATLAB code to
propose data types and sizes for your inputs.
It tests whether your MATLAB code is ready
for code generation by generating a MEX
function, then executes the MEX function to
check for run-time errors (Figure 2). Equivalent command-line functions provide the
same functionality so you can generate code
as part of a script or function.
Implementation Constraints
As you prepare your MATLAB algorithm for
code generation, you need to take account of
implementation constraints resulting from

the differences between MATLAB and C

code. These include:
Memory allocation. In MATLAB, memory
allocation is automatic. In C code, memory
allocation is manualit is either allocated
statically (using static), dynamically (using malloc), or on the stack (using local
variables).
Array-based language. MATLAB provides a
rich set of array operations that allow concise coding of numerical algorithms. C code
requires explicit for-loops to express the
same algorithms.

Dynamic typing. MATLAB automatically determines the data types and sizes
as your code runs. C requires explicit type

MathWorks News&Notes | 2 0 1 6 2 0 1 7

27

Industry Success Stories

dorsaVi generated C++ code from motion analysis algorithms and

FIGURE 3. Polymorphism example.

compiled it into a DLL, which was then integrated into their C# ap-

declarations on all variables and functions.


Polymorphism. MATLAB functions can
support many different input types, while C
requires fixed type declarations. At the top
level, you must specify the intended C function declaration.
Lets take a closer look at polymorphism.
Polymorphism can give a single line of
MATLAB code different meanings depending on your inputs. For example, the function shown in Figure 3 could mean scalar
multiplication, dot product, or matrix multiplication. In addition, your inputs could
be of different data types (logical, integer,
floating-point, fixed-point), and they could
be real or complex numbers.
MATLAB is a powerful algorithm
development
environment
precisely
because you dont need to worry about
implementation details as you create
algorithms. However, for the equivalent C
code, you have to specify what operations
mean. For example, the line of MATLAB
code shown above could be translated into
this single line of C code that returns B*C:
double foo(double b, double c)
{
return b * c;
}

Or, it could be translated into 11 lines of C

code with three for-loops that multiply two
matrices:

28

MathWorks News&Notes | 2 0 1 6 2 0 1 7

void myMult(const double a[12],


const double b[20], double c[15])
{

tions of your MATLAB code by generating a MEX function that calls the compiled C/C++ code.

int i2;
for (i0 = 0; i0 < 3; i0++) {
for (i1 = 0; i1 < 5; i1++) {
c[i0 + 3 * i1] = 0.0;
for (i2 = 0; i2 < 4; i2++) {
c[i0 + 3 * i1] += a[i0 +

3 * i2] * b[i2 + (i1 <<
2)];
}
}

Multicore-Capable Code Generation

and Other Optimization Methods
In MATLAB, for-loops whose iterations are
independent of each other can be run in parallel simply by replacing for with parfor.
MATLAB Coder uses the Open Multiprocessing (OpenMP) application interface to
support shared-memory, multicore code
generation from parfor-loops. OpenMP is
supported by many C compilers (for example,
Microsoft Visual Studio Professional).

}
}

diagnose injury.

VivaQuant generated fixed-point C code from heart rhythm monitoring algorithms and compiled it for an ARM Cortex-M processor.

sequence prediction algorithms and integrated it into surface de-

Respiri generated C code from acoustic respiratory monitoring al-

coding software running on a PC that enables downhole data to be

gorithms and compiled it for an iPhone app, an Android app,

decoded quickly and reliably during drilling operations.

and cloud-based server software.

f
or (i3 = 0; i3 < loop_ub;
i3++) {

planeHist_data[i3] =
originalHist_data[plane +
originalHist_size[0] * i3];
}

for (i1 = 0; i1 < 3; i1++) {

sz[i1] = (unsigned int)


Working with the Generated Code:

Four Use Cases
Once you have generated readable and portable C/C++ code from MATLAB algorithms
using MATLAB Coder, you have several options for using it. For example:

Integrate your MATLAB algorithms as
source code or libraries into a larger software
project such as custom simulators or software
packages running on PCs and servers.
Implement and verify your MATLAB algorithms on embedded processors such as
ARM processors and mobile devices.
Prototype your MATLAB algorithms as a
standalone executable on PCs.
Accelerate computationally intensive por-

algorithm and compiled it for an ARM10 processor.

plication running on a PC that analyzes the athletes movements to

Baker Hughes Dynamics & Telemetry group generated a DLL from

int i0;
int i1;

Delphi generated C code for an automotive radar sensor alignment

originalImage->size[i1];
}
N = sz[0];
M = sz[1];
normalizer = (L - 1.0) /

((double)sz[0] * (double)sz[1]);
#pragma omp parallel for \
num_threads(omp_get_max_threads()\
) private(s,r,planeHist_data,\

loop_ub,i3,y,x,j,d0,u0)
for (plane = 0; plane < 3;

plane++) {
loop_ub = originalHist_size[1];

You can use MATLAB Coder with

Embedded Coder to further optimize
code efficiency and customize the generated code. Embedded Coder provides optimizations for fine-grained control of the
generated codes functions, files, and data.
For example, you can use storage classes to
control the declaration and definition of a
global variable in the generated code, and
use code generation templates to customize banners and comments in the generated
code. Embedded Coder also improves code
efficiency by using code replacement libraries, which replace certain operators and
functions with implementations optimized
for popular processors like ARM Cortex-A
and ARM Cortex-M.

Testing the Generated Code

As you develop your MATLAB algorithm, you
can create unit tests to verify that the algorithm
produces the results you expect. Tests written
using the MATLAB unit testing framework
can be reused to verify that the generated
code behaves the same way as your MATLAB
algorithm. With Embedded Coder you can
reuse the unit tests in combination with
software-in-the-loop (SIL) and processorin-the-loop (PIL) tests on the generated
standalone code or library.
An Automated Workflow
MATLAB Coder enables an automated workflow for translating MATLAB algorithms into
C code. With this workflow you spend less
time writing and debugging low-level C code
and more time developing, testing, and tuning designs. By maintaining one golden reference in MATLAB, including the algorithm
and test benches, you can propagate algorithmic changes to your C code more quickly. Automated tools like the MATLAB unit testing

framework and the Embedded Coder SIL and

PIL testing framework let you test both the
MATLAB code and the C code thoroughly
and systematically. Whether you are implementing designs running on traditional PCs,
web servers, mobile devices, or embedded
processors, MATLAB Coder will help you
get from MATLAB to C code faster and with
fewer manual translation errors.

LEARN MORE
Generating C Code from MATLAB
(download)
mathworks.com/fx-56270

MATLAB and C/C++ Resources

mathworks.com/matlab-c

Supported Language and Functions for

MATLAB Coder
mathworks.com/coder-language

MathWorks News&Notes | 2 0 1 6 2 0 1 7

29

CLEVES CORNER

Introducing Cleves Laboratory

By Cleve Moler, MathWorks

n this Cleves Corner I want to give you a sneak preview of the

experiments in Cleves Laboratory. I launched the laboratory to collect much of the work I have done over the last several years in one
place. The experiments come from my two ebooks, my blog, Cleves
Corner columns in MathWorks News & Notes, and new work.
Each experiment centers on an interactive app that allows experimenters to try out the ideas for themselves. Several of the ex-

periments, including hello_world, klock, and biorhythms, are

designed to introduce newcomers to MATLAB.
The figure below shows snapshots of the graphical entry pages.
Even these are live pages driven by MATLAB. The wave in the
first icon moves every time you open the page. The clock reads
the correct time. The Sudoku puzzle changes. The double pendulum swings.

logo_wave. The MathWorks logo is the solution to the wave equation, a foundation of mathematical physics. MathWorks is the only
company in the world whose logo is the solution to a partial differential equation. This program demonstrates the vibration of a membrane stretched over an L-shaped region.

lifex. This is a version of John Conway's Game of Life. "Life" is

a cellular automaton that involves life and death in an infinite rectangular, 2D, cellular universe. The lifex program accesses the Life
Lexicon, a historical collection of nearly 500 starting populations
available online. It uses sparse matrix operations and an elegant, oneline implementation of Conways rules for evolution in this universe.
The icon shows Bill Gospers glider gun, which emits a continuous
stream of 5-element agents that move across the space.

pdeapp. This program demonstrates finite difference methods for

solving model problems for four partial differential equations involving
Laplaces operator: the Poisson equation, the heat equation, the wave
equation, and an eigenvalue equation. The regions are a square, an Lshape, an H-shape, a disc, an annulus, and a pair of isospectral drums.

fern. The fern is a self-similar fractal. It was invented by Michael

Barnsley, and is described in his book Fractals Everywhere. Each leaf
is similar in structure to the larger fern and contains a miniature copy
of yet another leaf. The fern program keeps running until the stop
button is toggled.

T puzzle image courtesy of Shop New Zealand www.shopnewzealand.co.nz Shop New Zealand

30

MathWorks News&Notes | 2 0 1 6 2 0 1 7

fibonacci. This program is based on Fibonacci's rabbit pen. A

man puts a pair of rabbits in a place surrounded on all sides by a wall.
How many pairs of rabbits can be produced from that pair in a year if
it is supposed that every month each pair begets a new pair that from
the second month on becomes productive? Today, the solution to this
problem is known as the Fibonacci sequence, or Fibonacci numbers.

flame. This is an example of a stiff ordinary differential equation.

flame(r0) specifies the initial radius is r0. Default r0 = .02. A ball

of fire grows until its radius is just large enough for all the oxygen
available through the surface to be consumed by combustion in the
interior. The equation for the radius is rdot = r^2 - r^3. The problem
becomes stiff as the radius approaches its limiting value.

MathWorks News&Notes | 2 0 1 6 2 0 1 7

31

CLEVES CORNER

orbits. The orbits program solves Newtons equations governing

the gravitational attraction among several bodies. When the number
of bodies is equal to nine, this is the solar system with one sun and
eight planets.

eigsvdapp. With this program you can watch how MATLAB handles three different matrix eigenvalue problems: the eigenvalues of a
nonsymmetric matrix, the eigenvalues of a symmetric matrix, and
the singular values of any matrix. Watch the reduction to Hessenberg, tridiagonal, or bidiagonal form, then the QR iteration to obtain
Schur or diagonal form.

censusapp. This experiment is older than MATLABit started

as an exercise in Computer Methods for Mathematical Computations, by Forsythe, Malcolm and Moler, published in 1977. The data
comes from the decennial census of the U.S., 19002010. The task
is to extrapolate population data beyond 2010. Todays MATLAB
makes it easier to vary the parameters and see the results, but the
underlying mathematical principle is unchanged: Using polynomials of even modest degree to predict the future by extrapolating
data is a risky business.

durerperm. The icon shows Durer's magic square. To permute the

square, click on two different rows or columns. Is the result still a
magic square?

golden_spiral. In this program you can see a continuously expanding sequence of golden rectangles and inscribed quarter circles.

tictactoe. This program combines three games that initially appear to be unrelated: Pick15, TicTacToe, and Magic3. In Pick15, the
object is to generate a total of 15 using exactly three digits, where
each digit can be chosen only once. TicTacToe follows the traditional
game, but replaces Xs and Os with blue and green. The object is to
get three in a row, column, or diagonal. Magic3 superimposes a magic square of order three on TicTacToe to show that Pic15 is actually
the same game.

walker. This model, developed by Nikolaus Troje, is a five-term

Fourier series with vector-valued coefficients that are the principal
components for data obtained in motion-capture experiments involving subjects wearing reflective markers and walking on a treadmill. The components, also known as postures or eigenwalkers,
correspond to the subjects movements. The postures are also classified by gender.

t_puzzle. I first saw this wooden T puzzle at Puzzling World in

Wanaka, New Zealand. The underlying mathematics involves geometry, trigonometry, and arithmetic with complex numbers. The
t_puzzle program demonstrates some useful programming techniques. The four pieces all have the same width but different heights.
It turns out that they can be arranged to form a capital T as well as
an arrow and a rhombus.

mandelbrot. This program invites you to explore the Mandelbrot

fractal. You can use the mouse to select a region, zoom in on any region, increase the grid size, increase the iteration depth, and change
the color map.

waterwave. A 2D shallow water model with reflexive boundary

conditions. A random water drop initiates gravity waves. The surface
plot displays height colored by momentum. The solution is computed
by the Lax-Wendroff finite difference method. The plot title shows
simulated time and total variation.

32

MathWorks News&Notes | 2 0 1 6 2 0 1 7

waves. This program demonstrates the wave equation in 1D

and 2D space dimensions. Solutions are expressed as time-varying
weighted sums of the first four eigenfunctions. The 1D domain is an
interval. The 2D domains include a square, a disc, a three-quarter
circular sector, and the L-shaped union of three squares. With polar coordinates, the eigenfunctions of the disc and the sector involve
Bessel functions.

predprey. In this classic model, one species grows exponentially

while the other decays exponentially in the absence of the other.
The model is nonlinear, but the solutions are periodic. Dragging the
red dot changes the equilibrium point. Dragging the blue-green dot
changes the initial conditions.

tumbling_box. If you throw a rectangular box in the air with a

twist, you can make it tumble stably about its longest or shortest axis.
But if three sides of the box are of different lengths, you cannot make
it tumble about its middle-sized axis. The Euler differential equations
for the angular momenta about the three principal axes have two
stable critical points and one unstable critical point.

LEARN MORE
Cleve's Laboratory App
(download)
mathworks.com/cleves-lab

Cleves Corner Blog

blogs.mathworks.com/cleve

Cleves Corner Collection

mathworks.com/cleves-corner

MathWorks News&Notes | 2 0 1 6 2 0 1 7

33

TECH SPOTLIGHT

Smart Devices and Analytics Spur Innovation

in the Internet of Things
By Eric Wetjen, MathWorks
FIGURE 5. Categorizing live traffic data in ThingSpeak.

FIGURE 3. Simulink block diagram to find and count the cars.

he Internet of Things (IoT) is a rapidly evolving space in which

virtually any smart hardware devicea mobile phone, a pacemaker,
a wearable fitness sensor, even a refrigeratorcan be connected to
the internet to generate and receive data.
Combining internet-connected devices with cloud computing,
machine learning, and other data analytics approaches is enabling
products and solutions that are transforming the way we live and
work. Today, for example, thanks to the Internet of Things:
A doctor can remotely monitor how often a patients pacemaker fires.
Athletes can measure how many calories they burn during a run.
Farmers can optimize irrigation of crops.
Building managers can save electricity by optimizing controls for
HVAC equipment.
Asthma sufferers can manage their condition by using a mobile phone
app to monitor their wheezing levels.
Automakers are close to developing a driverless car capable of autonomously navigating through city streets.
Inside an IoT System
A typical IoT system works like this (Figure 1):

2. An analytic IoT platform processes and stores the sensor data. It
may integrate information from other sources, such as business
systems. It analyzes and takes action on the incoming data.
3. A systems engineer or data scientist accesses the historical data
from the cloud or the device and develops algorithms to preprocess and analyze it. These algorithms may involve machine learning techniques for predicting future values of a sensor quantity or
for classifying the sensor data.
4. The algorithm is deployed in the cloud or on a smart device, where
it operates on incoming live data.
Creating an IoT System with MATLAB and Simulink
MATLAB and Simulink support IoT systems by helping you develop and test smart connected devices, access and collect data in the
cloud, and analyze sensor data. Lets look at two examples.

Example 1. A Traffic Monitor

Tired of sitting in traffic every time they drove home from work, two
engineers decided to study traffic flow trends on the busy highway
outside their building. They installed a webcam in an office overlooking the highway and connected it to a Raspberry Pi board running
a computer vision algorithm (Figure 2).
The data was collected in the cloud using ThingSpeak, an analytic
IoT platform that can run MATLAB code.

FIGURE 1. A typical Internet of Things workflow.

1. A smart connected device produces sensor data and ultimately

sends data to the cloud. These devices are often smart enough to
run data-reduction algorithms on their embedded processors.

34

MathWorks News&Notes | 2 0 1 6 2 0 1 7

FIGURE 2. Webcam connected to a Raspberry Pi 2.

FIGURE 4. Simulink external mode detection of cars during algorithm

development phase.

They built a Simulink block diagram that included a Median Filter

block to clean up the image and a Blob Analysis block to identify cars
in the image (Figure 3). They then wrote a custom block to count the
cars, and sent the summary data to ThingSpeak.
Before deploying the algorithm to the Raspberry Pi, they verified it
using the video display in Simulink external mode (Figure 4).
Once the data was in ThingSpeak, they used the MATLAB app
integrated into ThingSpeak to create live visualizations of the traffic
density in each direction for the previous 48 hours. These include a
color-coded visualization (Figure 5) that categorizes the current state
of the traffic (light = green, moderate = yellow, and heavy = red).
Example 2. A Tide Gauge and Alert System
Getting your boat stuck in the mud is a distinct possibility without an
accurate measure of water depth. Because tide predictions and realtime water levels are not available for most bays and estuaries, one
boater and MATLAB user built a low-cost, real-time tide gauge. He
quickly learned that the timing and amplitude of tide levels is highly
dependent on location.
The key hardware components he used were an Arduino Mega
board, a SparkFun Electronics cellular shield, and an ultrasonic
range finder. The cellular shield provided the connection from the
Arduino to the internet. The Arduino was used to read the data from
the ultrasonic sensor.
ThingSpeak was used to collect and process the data in the cloud.
The ultrasonic sensor reports distance in mm. To convert from distance to water depth, the boater set up a ThingSpeak TimeControl
that runs MATLAB code to read the range data, convert it into water

FIGURE 6. ThingSpeak data collection and tide level display.

depth, and write the data to a new ThingSpeak channel. The result
was an internet-connected tide gauge that can be viewed on a mobile
phone or a web browser (Figure 6).
The boater set up additional MATLAB code to detect tidal thresholds. He then used the Twitter integration in ThingSpeak to send
alerts when particular conditions were met.
As these examples show, analytics is a vital ingredient of the innovations occurring in IoT. With MATLAB and Simulink you can
develop analytics that run on your smart devices or in the cloud.
With ThingSpeak, you can easily collect data from your devices in
the cloud. You can then use MATLAB to gain insight into the sensor
data you have collected.

LEARN MORE
Developing Internet of Things Systems with MATLAB
mathworks.com/iot

ThingSpeak

thingspeak.com

Traffic Monitor Data Visualizations

thingspeak.com/channels/38629

MathWorks News&Notes | 2 0 1 6 2 0 1 7

35

WHAT IF EVERYONE

THIRD-PARTY PRODUCTS

Solutions for Image Acquisition

and Computer Vision

ON CAMPUS HAD MATLAB?

MATLAB and Simulink provide a platform for engineers to explore images, develop computer vision algorithms,
and evaluate implementation tradeoffs. Third-party imaging hardware enables them to acquire visible, thermal, depth,
microscopy, and a range of other images. Engineers can analyze images, via live acquisition or by postprocessing
in MATLAB, and then deploy computer vision applications such as object detection, tracking, and recognition onto
embedded processors, FPGAs, multicore GPU systems, and other hardware.

More than 1 million students and 700 universities around the world
including the top 10 ranked universitieshave unlimited access to MATLAB and
Simulink with a Total Academic Headcount (TAH) license.

FLIR Infrared Cameras

FLIR thermal and visible-light imaging systems are used in a wide variety of thermal imaging, situational awareness, and security applications. Using Image Acquisition Toolbox,
engineers can configure features of FLIR infrared cameras and stream fully temperaturecalibrated data directly into MATLAB for
analysis, visualization, and modeling.
flir.com
3i SlideBook
SlideBook microscopy software enables scientists to acquire and analyze image data
across time, color, and specimen locations,
with customizable experiment protocols and
drivers for hundreds of research instruments.
SlideBook supports data import/export with
MATLAB, enabling users to call MATLAB
to set variables and perform advanced image
analysis, and extending SlideBooks built-in
image processing functions. MATLAB scripts
can drive live cell experiments by determining
regions for capture and photomanipulation.
intelligent-imaging.com

Allied Vision FireWire and GigE

Vision Cameras
Allied Vision offers high-performance IEEE1394 FireWire, GigE Vision, and USB3 Vision
compliant digital cameras for machine vision, computer vision, and other industrial or
medical applications. Cameras include CCD
and CMOS sensors ranging from VGA to 29
megapixels. Image Acquisition Toolbox, with
Allied Visions Vimba Software Development
Kit (SDK), enables direct access to the cameras
from MATLAB and Simulink.
alliedvision.com
Microsoft Kinect
Microsoft Kinect for Windows includes a 3D
depth sensor and integrated RGB camera for
applications such as robotics, kinesiology, 3D
scene reconstruction, point-cloud processing,
and skeletal tracking. Engineers can acquire
3D data from Microsoft Kinect into MATLAB
and Simulink using Image Acquisition Toolbox and Computer Vision System Toolbox.
developer.microsoft.com/kinect

Xilinx Zynq-7000
All-Programmable SoC
The Xilinx Zynq-7000 All-Programmable
SoC combines a dual-core ARM Cortex-A9
with Xilinx 7-series FPGA logic on a single
chip. As a result, users can integrate camera
control and image processing functions in a
single device while enabling hardware acceleration of video analytics. Evaluation kits include
hardware I/O, design tools, IP, and pre-verified
reference designs. MATLAB, Simulink, HDL
Coder, Embedded Coder, and Vision HDL
Toolbox support the design and simulation of
image processing applications and automated
deployment to the Zynq device.
xilinx.com/zynq

On multidisciplinary projects, students with

quite different educational backgrounds

42,000

can work together more easily because

they are using the same tools.

Faculty and students using MATLAB to program hardware

Professor Jakob Stoustrup, Aalborg University

JOB OPPORTUNITIES

82%

Hardware Support

mathworks.com/hardware

If you want to work at Google, make sure

you can use MATLAB.

Fortune 100 companies with a MATLAB license

Our teams are here to do world-class research,

and easy access to MATLAB enables them to be

LEARN MORE

HANDS-ON LEARNING

their most productive.

Shailesh Shenoy, Director of Research Computing,
Albert Einstein College of Medicine of Yeshiva University

Jonathan Rosenberg, Senior Vice President of

Products, Google

RESEARCH PRODUCTIVITY

1,970,000
Google Scholar results referencing MATLAB

Image Processing and Computer Vision

mathworks.com/solutions/image-video-processing

Third-Party Products and Services

mathworks.com/connections

mathworks.com/matlab-campus
36

MathWorks News&Notes | 2 0 1 6 2 0 1 7

MATLAB SPEAKS

ARDUINO
so you dont have to

You can design, build, test, and run a

systemon Arduino, Raspberry Pi, LEGO,
and morewithout writing traditional code.
Download free MATLAB and Simulink
hardware support packages at
hardware.mathworks.com

92984v00 11/16