Achieving High Navigation Accuracy

Using Inertial Navigation Systems

in Autonomous Underwater Vehicles
Robert Panish and Mikell Taylor
Bluefin Robotics Corporation
553 South Street
Quincy, MA 02169 USA

Abstract-This paper presents a summary of the current variable to meet the demands of the specific application – from
state-of-the-art in INS-based navigation systems in AUVs one to eight meters. The vehicles make use of free-flooded
manufactured by Bluefin Robotics Corporation. A detailed
description of the successful integrations of the Kearfott T-24 modularity to create a robust system that is field maintainable,
Ring Laser Gyro and the IXSEA PHINS III Fiber Optic Gyro easily expandable and customizable, and some are equipped
into recent Bluefin Robotics AUVs is presented. Both systems with field-swappable payload sections. Inside the free-
provide excellent navigation accuracy for high quality data flooding hull are a number of subsystems protected within their
acquisition. This paper provides a comprehensive assessment of own pressure vessels or oil-filled, pressure-tolerant assemblies.
the primary advantages and disadvantages of each INS, paying
particular attention to navigation accuracy, power draw, physical The most critical of these subsystems is the Main Electronics
size, and acoustic radiated noise. Additionally, a brief Housing (MEH), which contains the main vehicle computer,
presentation of a recently integrated Synthetic Aperture Sonar power distribution network, Doppler Velocity Log (DVL),
system will be used to highlight how critical a high-performance navigation system, and a variety of other subsystems.
INS is to hydrographic, mine countermeasures, and other SAS A wide array of payloads has been successfully integrated
applications.
into Bluefin AUVs, including side scan sonar (SSS), synthetic
Keywords—inertial navigation system, autonomous underwater aperture sonar (SAS), multibeam echosounders, sub-bottom
vehicles, AUVs, unmanned underwater vehicles, UUVs, Bluefin profilers, cameras, magnetometers, water sampling systems,
Robotics, Kearfott, IXSEA, ring laser gyro, RLG, fiber optic gyro,
fluorometers, conductivity and temperature sensors, and sound
FOG
velocity sensors, to name a few. Collection of high quality
I. INTRODUCTION data with these payloads requires a combination of high
accuracy navigation data and stable vehicle dynamics.
The Bluefin Robotics Autonomous Underwater Vehicle Stable vehicle dynamics are achieved through closed loop
(AUV) is an extremely versatile system that can be utilized for control of vehicle motion in trackline or trackcircle modes,
a wide range of missions. For shallow or deep survey while operating at any depth or altitude above the seafloor.
applications the Bluefin AUV is a highly capable, extensively Vehicle propulsion, as well as horizontal and vertical control,
configurable tool that can be used to obtain high quality data is achieved by an articulated, ducted thruster known as the
from a wide array of oceanographic sensors. An AUV is able tailcone, shown in Fig. 2. The system is designed to be
to maintain sensor positioning at an ideal height above the passively stable in roll, due to a proper separation of the center
seafloor during surveys, is unaffected by surface sea states, and of buoyancy and the center of gravity. The tailcone is designed
can follow a rough terrain to produce the best possible images.
Bluefin Robotics offers a wide range of products to meet the
needs of a variety of customers, including propeller-driven
torpedo form-factor AUVs, a hovering AUV, and a glider.
Bluefin’s torpedo form-factor AUVs come in three different
diameters – 9”, 12”, and 21”. A typical Bluefin 12” vehicle is
shown in Fig. 1. The length of the 12” and 21” vehicles is

Figure 1 - A 12" diameter Bluefin Robotics AUV. Figure 2 - A 12" Bluefin Tailcone.

978-1-61284-4577-0088-0/11/$26.00 ©2011 IEEE

Authorized licensed use limited to: China University of Petroleum. Downloaded on March 16,2021 at 15:16:39 UTC from IEEE Xplore. Restrictions apply.
to be torque-neutral over a wide range of speeds, and thus roll The differences between the T-24 and the PHINS III are
is minimally affected by propeller speed. much more than simply the brand and implementation details.
Navigation on a Bluefin AUV involves a fusion of data from Both units measure rotations using laser-based interferometric
several sensors. Selection of an appropriate navigation option techniques. However, the fundamental technology behind the
depends on the vehicle configuration, operating depth, gyroscopes is different between the two units. The T-24 uses
navigation accuracy specifications, and budget. Shallow water RLGs, which means the beam path is created by a set of
systems with moderate navigation accuracy requirements will mirrors redirecting the laser into a loop. To prevent a
often use a compass-based navigation solution. These systems phenomenon known as injection locking, where under slow
achieve a quality navigation solution through the integration of rotations the gyroscopes will not accurately measure rotation
a tactical-grade Inertial Measurement Unit (IMU), digital rates, the mirrors in the RLG are mechanically dithered. This
magnetic compass, DVL, GPS receiver, pressure/depth sensor, dithering in the T-24 produces an audible tone at the dithering
and a sound speed sensor. frequency. In a FOG-based INS, such as the PHINS III, the
Oftentimes a vehicle will require a higher degree of laser beam travels through a long optical fiber to create the
navigation accuracy to achieve its mission. In these cases, an beam path. One advantage of using optical fibers over dithered
Inertial Navigation System (INS) will be integrated with a mirrors is that there are no moving parts in the system, and it is
DVL, GPS receiver, pressure/depth sensor, and a sound speed therefore acoustically silent.
sensor. This type of navigation solution is most valuable in The standard measure for evaluating a system’s navigation
deep water applications, when surfacing for GPS is infeasible, accuracy is to state the position drift as a function of distance
or when the payload requires high navigational accuracy. SAS travelled since the last GPS fix. Both the T-24 [1] and the
processing, in particular, requires extremely accurate PHINS III [2] have a stated position drift of less than 0.1% of
navigation information. Another application that cannot be distance travelled, CEP. CEP refers to the Circular Error
dependent on a compass-based solution is operation in the Probability, or a circle about a mean value which includes 50%
polar-regions, where compasses are not useful for navigation. of the population. For instance, if a system with a drift of 0.1%
While INSs also suffer degraded performance at high latitudes, of distance travelled CEP travels five kilometers between GPS
an INS can still provide some level of navigation whereas a fixes, it is expected that 50% of the time the position drift will
compass-based solution may be rendered useless. be less than five meters.
This paper will present information on the integration and Partially due to the differences in the type of gyroscopes, the
performance of two different exceptional INS units into T-24 and the PHINS III have a number of differing properties
Bluefin AUVs - the T-24 Ring Laser Gyro (RLG) that provide advantages for the use of each unit. When
manufactured by Kearfott and the PHINS III Fiber Optic Gyro choosing an INS for an AUV there are a number of device
(FOG) manufactured by IXSEA. Data from hundreds of specifications that a system integrator may consider. The
kilometers of submerged survey will be presented to compare required navigation accuracy is certainly one of the most
the navigation accuracy of the two units in comparable vehicle important considerations. However, a proper design process
platforms. In addition to the navigation accuracy, a number of should also pay attention to the physical size and weight of the
other factors may influence the selection of an INS. These units, the power draw, and acoustic noise. In this comparison,
factors include physical size, power draw, and acoustic noise. the T-24 is significantly smaller than the PHINS III, whereas
This paper will also highlight how navigation accuracy is a the PHINS III draws less power and is acoustically silent.
key contributor to obtaining high quality sonar data with
favorable contact localization accuracy. High quality sonar A. DVL-INS Calibration
data will be presented that could only be obtained with a high When an INS is integrated into an AUV it must be calibrated
accuracy INS. Overall, this paper provides a summary of the with the DVL. This calibration is necessary to account for
state-of-the-art inertial navigation technologies for AUVs and mechanical misalignment in the installations of the INS and
their impact on data quality. Readers will find important DVL, as well as for potential errors in the velocity estimates of
information on the performance of INS technology and will the units. The calibration procedure need only be performed
take away a better understanding of the intricacies of INS once, unless the INS or the DVL is removed from the vehicle.
integration on AUVs. The calibration procedures are different for each INS, but the
basic procedure involves driving the vehicle in straight lines to
II. INSS ON BLUEFIN AUVS compare the INS and DVL estimated motion with GPS truth
data. To ensure a consistent calibration it is critical that the
Bluefin has integrated a number of different INSs in AUVs INS and DVL are both mounted to the same rigid structure and
over the years. The two systems most commonly used are the that the lever arms and orientation between the devices remain
Kearfott T-24 RLG and the IXSEA PHINS III FOG. Each of constant.
these systems makes use of DVL, pressure/depth sensor, and Calibration of the IXSEA PHINS III INS involves the use of
sound speed sensor inputs while submerged to achieve the the built-in calibration tool to put the INS into calibration mode.
desired navigation accuracy. In calibration mode the INS uses its inertial sensors and GPS to

Authorized licensed use limited to: China University of Petroleum. Downloaded on March 16,2021 at 15:16:39 UTC from IEEE Xplore. Restrictions apply.
determine its motion, but does not make use of the DVL vector are known, the North direction (and thus, heading) can
velocities. It records the DVL velocities and automatically be determined through vector algebra. [2] Alignment of the
determines the calibration parameters based on the difference INS may be performed on land, on the deck of a ship, or on the
between the known motion (from the INS and GPS) and the surface of the water. The alignment procedures are different
measured DVL motion. To perform the calibration, the AUV for each INS.
is sent on a single five kilometer trackline while remaining on Alignment of the IXSEA PHINS III INS requires GPS to
the surface to maintain continuous GPS contact. During this determine the initial position and altitude of the vehicle. It
procedure the various calibration parameters can be monitored does not use any inputs from the DVL or other sensors. The
to watch for convergence. These parameters are the roll, pitch, alignment proceeds in three phases – coarse alignment, fine
and heading misalignment angles, as well as the scale factor alignment, and aligned. The primary value that can be
for velocity magnitude adjustment. monitored to determine the status of the alignment is the
In practice on a Bluefin vehicle, where the DVL and the INS standard deviation of heading. During the coarse alignment
are rigidly mounted in the MEH, it has been observed that the phase, the INS should be kept fairly stationary. Minor motion
pitch and roll misalignment angles converge rapidly to a stable while on the surface of the water or on the deck of a ship is
value. The heading misalignment angle takes a bit longer to acceptable, but the ship should not be moving at more than
converge, as this direction of misalignment is slightly more about five knots. This phase lasts for about five minutes, at
difficult to measure. This value usually converges to a steady which point the PHINS III proceeds to the fine alignment
value within the first kilometer and becomes refined over the phase.
duration of the calibration. The scale factor is refined The fine alignment phase is marked by a decreasing heading
throughout the calibration and, for well functioning PHINS III standard deviation as the INS refines its heading estimate.
and Teledyne RDI DVL units, should converge to a value near During this phase the INS must be rotated through a few
zero. Once the calibration procedure for the PHINS III has heading changes of at least ninety degrees. IXSEA
been completed, the unit must be power cycled and realigned. recommends either a stair-step pattern or a box pattern. There
It is then ready for normal operations. is no restriction on the velocity of the system during this phase,
The INS-DVL calibration procedure for the T-24 INS is however, without heading changes the alignment will not
markedly different from that of the PHINS III. To calibrate the complete.
T-24, the INS is put into calibration mode and the vehicle is Once the heading standard deviation has fallen to an
sent on a submerged dive to determine the calibration acceptable level, the INS is considered aligned. At this point it
parameters. The dive consists of a pair of boxes with GPS is ready for operation. Further heading changes can refine the
surfacings at each corner. Each side of the box must be about alignment further to achieve even greater sensor performance.
fifteen minutes in duration, and should be completed with the While not particularly difficult, the PHINS III alignment
vehicle submerged. The vehicle should have DVL bottom lock procedure does require some operator attention to ensure that
for the entire duration of the calibration. During the calibration the heading changes are performed.
procedure the INS is using the GPS fixes at each corner to Alignment of the Kearfott T-24 INS also requires GPS to
determine its internal biases, scale factors, and misalignment determine the initial position of the vehicle, and does not use
angles. Once the vehicle has completed this dive, the INS is any other sensors during the alignment. The alignment
taken out of calibration mode, power cycled, and is ready for a proceeds in three phases – coarse alignment, fine alignment,
verification dive or normal operations. and aligned. The coarse alignment phase is very similar to that
Operationally speaking, the T-24 calibration procedure is of the PHINS III. As with the PHINS III, the vehicle must be
simpler to perform than the PHINS III’s. Although it takes kept at a low velocity. Coarse alignment lasts for about five
significantly longer to complete, the T-24 procedure involves minutes, at which point the INS proceeds to the fine alignment
the vehicle operating submerged. This is certainly preferable phase.
to operating the vehicle on the surface, where additional care During fine alignment, the INS further refines its heading
must be taken to ensure there is no boat traffic in the area and estimate. Unlike with the PHINS III, no heading changes are
that the sea state is low enough for stable surface operations. required to complete the alignment. When the T-24 reaches
the aligned state, it is ready for operation.
B. INS Alignment
Upon power-up, an INS must perform an alignment. The III. ASSESSING NAVIGATION ACCURACY
primary purpose of the alignment is to determine the initial
attitude of the INS. Roll and pitch are determined from a The navigation accuracy of an INS is an important measure
simple measurement of the direction of the gravity vector from of the quality of the system. The accuracy of a system is
the accelerometers, which must first be filtered to remove presented as the position drift as a function of distance
accelerations due to motion. Heading is determined by travelled since the last GPS fix. When evaluating navigation
tracking the time derivative of the gravity vector, which lies in position drift, it is important to consider over what submerged
the East direction. Once the gravity (down) vector and the East distances it can be discerned whether there is INS drift, relative

Authorized licensed use limited to: China University of Petroleum. Downloaded on March 16,2021 at 15:16:39 UTC from IEEE Xplore. Restrictions apply.
to GPS uncertainty. Standard, commercially-rated GPS has an
uncertainty of five meters. Since the published drift of the
T-24 and the PHINS III is 0.1% of distance travelled, a true
test of navigation accuracy should involve dives with
submerged distances of at least five kilometers. In practice,
shorter submerged distances may be used, but with the caveat
that GPS uncertainty may be corrupting the calculations.
The best type of mission to determine the navigation
accuracy is a long, straight trackline. The longer the trackline,
the more certainty there is that the measured position drift is
due to INS drift and not GPS uncertainty. And since the
performance specification is given as a statistical value (CEP),
it is important to have a statistically significant number of data
points for analysis. Since obtaining a statistically significant
number of sufficiently long tracklines is not always practical
Figure 3 – Position drift can occur in the along-track or cross-track direction.
within the concept of operations (CONOPS) of a vehicle, it is
often necessary to include different types of dives in the
analysis. This may include shorter tracklines or even circular in heading after alignment. Depending on the type of payload,
tracks. While these types of dives are not ideal for testing the different types of drift are often more acceptable than others.
navigation accuracy, they are useful because they provide a If the concern is with maintaining a constant range to a target,
measure of the navigation accuracy that a system will achieve for instance, then a low cross-track error is of greater
during its normal operations. importance.

A. Calculating Navigation Accuracy IV. COMPARISON OF T-24 AND PHINS III

Calculation of the navigation accuracy from a data set is NAVIGATION ACCURACIES
more complicated than simply comparing the INS position to
To present a comparison between the T-24 and the PHINS
the first GPS position fix. Since GPS can often be erroneous
III, data taken throughout 2010 on a pair of Bluefin vehicles
during the first few position fixes after surfacing, it is
has been analyzed. Each of these systems was delivered in
important to extrapolate the INS position forward in time based
2010 and completed an extensive set of missions during
on measured DVL velocities on the surface. Once the GPS has
Engineering Sea Trials in Boston and after delivery to the
reached an acceptable number of satellites and quality, the
customers. Both vehicles utilize a recent build of the Bluefin
reported positions are then compared.
Huxley operating system, a Teledyne RDI DVL, and a
The measure of distance travelled is not simply the distance
Valeport sound velocity sensor for navigation aiding.
between initial and final GPS positions. Since it is likely that
Although they have different payloads, both systems require
the AUV did not travel a perfectly straight path between the
high performance navigation to aid their SAS payloads.
two points, a simple GPS position difference would result in an
The two vehicles have different CONOPS, so the library of
underestimate of distance travelled. The AUV may have
dives available is not identical. The vehicle with a Kearfott
traversed various transits to and from the surface locations,
T-24 is most often used to conduct long, linear surveys. Thus,
which may not have been collinear with the trackline.
most of the data available is from dives where the vehicle
Additionally, normal dynamic variations about the nominal
conducted a ‘lawnmower’ pattern of a series of back and forth
trackline should be included in the calculation of distance
lines, acquiring GPS position information either after every leg
travelled. The distance travelled is the length of the trajectory
or after every pair of legs. The vehicle with an IXSEA
calculated by the INS when submerged plus the length of the
PHINS III is used for a combination of tracklines and circles.
trajectory extrapolated on the surface.
The standard missions for this vehicle could include one and a
half orbit circles with GPS at the start and end, single straight
B. Types of Drift
lines with GPS at each end, or a large box with GPS at
An INS will experience position drift while submerged, and
convenient locations. Although the two systems operate with
this drift may be decomposed into two types – along-track and
different types of dives, it is important to realize that the
cross-track error, as shown in Fig. 3. Along-track error is the
navigation accuracies presented represent the achieved
component of the drift in the direction parallel to the nominal
accuracy of each AUV system in its normal CONOPS.
AUV motion. Cross-track error is the component of drift in the
Fig. 4 shows the cumulative probability curves of the
direction perpendicular to the nominal motion. In general,
achieved navigation accuracies for the two systems, for
along-track error is the result of a poorly calibrated scale factor
submerged distances greater than five kilometers. The plot
or errors in sound speed, and cross-track error is the result of a
shows the position drift (percent of distance travelled) as the
poorly calibrated heading misalignment angle or residual errors

Authorized licensed use limited to: China University of Petroleum. Downloaded on March 16,2021 at 15:16:39 UTC from IEEE Xplore. Restrictions apply.
 V. CONSIDERATIONS FOR SELECTION OF AN INS
Section IV has demonstrated that the navigation accuracies
of both the Kearfott T-24 and the IXSEA PHINS III are superb
and both can be successfully integrated into a Bluefin AUV.
Selection of an INS between these two units should not be
based on navigation accuracy alone, as they are both excellent.
However, each unit has certain advantages over the other that
deserve to be presented and considered.
An important consideration with any underwater system is
power draw of components. Due to the differences in
architecture, the PHINS III draws significantly less power than
the T-24. The PHINS III draws only 15 Watts, whereas the
T-24 draws 30 Watts. While neither of these devices will be a
major tax on the system in comparison with the propulsion
systems, for low energy density or long endurance platforms
Figure 4 – Achieved navigation accuracies for submerged distances
greater than five kilometers. every Watt can impact the total endurance of the AUV.
The physical size of each system is also an important
consideration. At present, the PHINS III is a box with
independent variable on the horizontal axis, and the percent of
dimensions 180mm x 180mm x 160mm. The T-24, which is
the time the system achieves that level of drift or better
installed as an OEM system, consists of a cylinder of diameter
(percent occurrence) on the vertical axis.
115mm and length 155mm and a board set that is about
Recall that the specification is for each system to achieve a
135mm x 115mm x 65mm. The smaller size and flexible form
position drift of 0.1% of distance travelled (CEP), which
factor of the T-24 enables it to be fit more easily into a smaller
means that 50% of the runs should have a drift of less than
diameter vehicle.
0.1% of distance travelled. In analyzing the results for the two
Another differentiator that may be important to some
navigation systems, it is acceptable to present either the
customers is the acoustic noise generated by the devices. Since
achieved position drift CEP or to state the percentage of dives
the T-24 uses an RLG, which has mechanically dithered
that achieved the specification of 0.1% of distance travelled.
components, it makes noise. The FOG of the PHINS III has no
The data presented in Fig. 4 is summarized in each of these
moving parts, making it acoustically silent. Acoustic radiated
ways in Table I. It is clear from this table that both the Kearfott
noise may be an important consideration for certain types of
T-24 and the IXSEA PHINS III are superb navigation systems
sonar and should be considered in the selection of a navigation
that can be successfully integrated into a Bluefin vehicle. Both
system.
systems exceed the published drift specifications, achieving
high accuracy navigation in an AUV.
Another way of looking at the data, which shows the
importance of analyzing tracks longer than five kilometers, is
to examine a scatter plot of all data points as a function of
distance travelled. This type of plot, shown in Fig. 5,
highlights the greater distribution of results for shorter
submerged distances due to the effects of GPS uncertainty that
dominate the error budget.

TABLE I
POSITION DRIFT
Kearfott IXSEA
Specification
T-24 PHINS III

CEP Drift 0.1% dt 0.05% dt 0.07% dt

Percentage of Dives
50% 81% 83% Figure 5 – Scatter plot showing position drift as a function of trackline
with Drift < 0.1% dt
distance. Note that the drift appears much higher for shorter tracklines as
a result of GPS error being a non-negligible contribution.

Authorized licensed use limited to: China University of Petroleum. Downloaded on March 16,2021 at 15:16:39 UTC from IEEE Xplore. Restrictions apply.
 Figure 6 – Post-processed SAS image of a shipwreck, taken with an AST PROSAS Surveyor sonar on a Bluefin-12 AUV.
Note the constant resolution across the entire range.

system is presented in Fig. 6. It is clear that aided by enabling

VI. APPLICATIONS OF HIGH QUALITY NAVIGATION
technology such as high-performance tactical INS, SAS
High quality navigation is critical for many underwater systems integrated on Bluefin AUV platforms yield favorable
applications. In particular, the expanding application of SAS is improvements both in data quality and in the efficiency of data
dependent on high accuracy and high precision navigation. collection.
Side scan sonar is widely used throughout the AUV industry,
but SAS offers constant high-resolution imagery across a VII. CONCLUSIONS AND ONGOING WORK
greater range than a comparably sized SSS. In order to
generate this imagery, SAS payloads perform complex Integration of these navigation systems is an ongoing effort,
processing that requires a number of inputs from the platform with continuous improvements being made on both the Bluefin
with which the SAS is integrated – in this case, the AUV. side and the INS manufacturer side. The data presented in this
High-accuracy motion data from an INS – which has already paper for INS-based navigation accuracy was taken using
processed inputs from the DVL and other aiding sensors – is recently completed Bluefin AUVs. The data demonstrates that
fed directly to the SAS computer to allow processing to the navigation accuracies of both the Kearfott T-24 and the
compensate for vehicle motion. Low drift and high accuracy IXSEA PHINS III exceed the published specifications when
are critical to proper beamforming. integrated into a Bluefin AUV. Each of these systems provides
While high-accuracy motion data is needed for SAS exceptional navigation accuracy that can be used to collect
functionality, high-accuracy navigation is required for accurate high quality oceanographic data.
absolute positioning of sonar targets. In mine warfare Bluefin is continually expanding its library of dives that can
applications, for example, contact localization accuracy is the be processed to create plots like these. Additionally, Bluefin is
driving indicator of sensor and navigation performance for any refining its processes to facilitate faster and more accurate
minehunting technology. INS-DVL calibrations to produce even greater navigation
In 2010, Bluefin delivered a Bluefin-12 vehicle outfitted accuracy. In addition, as high-performance INS units shrink in
with a PROSAS Surveyor sonar, manufactured by Applied mechanical size and power consumption, cutting-edge
Signal Technology, Inc. aided by a Kearfott T-24 INS. This payloads such as SAS will become feasible for integration onto
vehicle regularly obtained seafloor imagery at constant smaller and lower-power AUV platforms.
1x1-inch resolution across 200 meter range on each side, with
an estimated contact localization accuracy of ten meters. This
was observed during both deep and shallow water surveys ACKNOWLEDGMENT
using long trackline missions. Overall this provided the The authors would like to acknowledge the entire Bluefin
operators with a six fold improvement in operational efficiency Robotics team for their work in integrating both of these
over a comparable SSS AUV. Sample SAS data from this

Authorized licensed use limited to: China University of Petroleum. Downloaded on March 16,2021 at 15:16:39 UTC from IEEE Xplore. Restrictions apply.
excellent navigation systems. Bluefin’s strong relationships
with Kearfott and IXSEA have enabled the creation of well-
integrated systems that are capable of meeting the customer’s
needs.

REFERENCES
[1] Kearfott Corporation, Seaborne Navigation System (SEANAV), KN-5050
Family. Data Sheet. 2010.
[2] IXSEA, PHINS User Guide. 2009.

ABOUT BLUEFIN ROBOTICS

Bluefin Robotics manufactures and develops Autonomous
Underwater Vehicle (AUV) systems and technology. Founded
in 1997, the company has grown to become a world leader in
AUV products designed for defense, commercial, and
scientific applications. Bluefin Robotics is a wholly-owned
subsidiary of Battelle. For more information, please visit
www.bluefinrobotics.com.

Authorized licensed use limited to: China University of Petroleum. Downloaded on March 16,2021 at 15:16:39 UTC from IEEE Xplore. Restrictions apply.