RADIO INTERFEROMETRIC GEOLOCATION

yakzephyrΤεχνίτη Νοημοσύνη και Ρομποτική

24 Νοε 2013 (πριν από 3 χρόνια και 10 μήνες)

66 εμφανίσεις

RADIO INTERFEROMETRIC
GEOLOCATION

Will
Hedgecock

EECE 354

M.
Maroti
, B.
Kusy
, G..
Balogh
, P.
Volgyesi
, A.
Nadas
, K. Molnar, S. Dora, A.
Ledeczi
. "Radio Interferometric
Geolocation
". in Proc. ACM 3rd Conference on
Embedded Networked Sensor Systems (SenSys'05), November,
2005.

11/08/2010

Motivation


Many
WSN
(Wireless Sensor Network) applications
require knowledge of the location of individual nodes in
the system


Existing
localization
techniques have
limited range and
accuracy


Usually acoustic
-
based


Also true of
signal strength methods
with accuracies of up
to a few
meters


Tradeoff
between range and
accuracy


Difficult
to provide stealthy operation modes in traditional
approaches


Requires ultrasound


Most
existing techniques work only in 2D

Radio Interferometric Positioning System
(RIPS)

Background


Traditional radio interferometry is used
in
physics
,
geodesy, and astronomy to measure relative distances
to objects


Works
by measuring a single signal from two separate
directional antennae and performing cross
-
correlation


Resulting
signal
interference
can be used to determine
distance to an object, the precise relative location
of the
two
receiving antennae, or, if the
locations
of the two receivers are
known, the precise location of the
radio source


Standard
Radio Interferometric systems are quite
expensive and not conducive to WSN applications

General RIPS Overview


Directly
generate an interference signal using two
transmitters at high frequencies


If
the transmitters are
signaling
using slightly different carrier
frequencies, the resulting interference signal will have a
low
-
frequency envelope


Can
therefore be measured by cheap, low
-
precision
hardware


The
phase offset of the interference signal at the receivers
corresponds to the relative positions of the four nodes in
the system


Thus
, with at least 8 nodes, we can calculate the relative
location of all of the
nodes in 3D

General RIPS Overview

RIPS Key Points


The
phase offset of a low
-
frequency signal is measured
(cheaply), but it corresponds to the wavelength of a
high
-
frequency carrier
signal


This
allows low
-
precision techniques carried out on
resource
-
constrained
WSN nodes to produce highly
accurate results


Absolute

phase offset depends on several factors
including
exact
time instances when signal transmissions
were started,
BUT
relative

phases measured by the
receivers depends only the distances between the
transmitters and receivers and
the carrier
frequency


By
measuring the phase offsets at different carrier
frequencies, we can infer the relative positions of the nodes

Radio Interferometric Properties


This theorem forms the basis for using phase offsets from
a low
-
frequency envelope to obtain highly accurate results

Radio Interferometric Properties

Radio Interferometric Properties

Radio Interferometric Properties

Radio Interferometric Properties

Sources of Error


Carrier
Frequency Inaccuracy
:


Carrier
Drift and Phase Noise: All previous theorems operate under
the assumption
of completely stable signals


Any
frequency drift will be directly observable in the relative phase
offsets at the receivers


Minimizable

by making each phase measurement as short as possible


Phase
noise (transients, shocks, etc.) is more serious


Multipath
Effects


RSSI
measurement delay jitter (in circuitry): not
noticeable
in
measurements


Signal
-
to
-
noise
ratio: dependent mainly on distances between nodes


Signal
processing error: well
-
studied and there exist good
approximations/models to deal with this


Time
synchronization error: Assuming 2kHz interference frequency
and
2us
synch accuracy, this generates a 0.4%*
2
π

phase offset
error

RIPS Implementation

RIPS Implementation


Base
Station:


Handles
scheduling of transmitting pairs, frequency calibration,
calculation of the
D
abcd

range, and actual localization


Motes:


Handles
CC1000 radio chip pure sine wave
transmission
drivers


Synchronizes
participating nodes


Handles transmission/reception
of signals


Estimates
the frequency and phase offset of the sampled RSSI
signal


Radio Characteristics


Configured
to transmit in the 433MHz frequency band


Capable
of transmitting an unmodulated sine wave in a
wide frequency band (400
-
460MHz)


Necessary
for calculating the actual range from the phase offset
differences


Ability
to tune transmitter with
high
granularity (65Hz
steps)


Necessary
to achieve separation of the two transmitters


Short
-
term
frequency stability of transmitter


Precise
capture of RSSI with small delay jitter


Capable
of
transmitting
at different power levels


Necessary
because transmitter/receiver distances can vary to the
point that a
closer transmitter
completely overwhelms the signal
from the more distant
one

Time Synchronization


No
network
-
wide synchronization


Only
the nodes participating in the current ranging round are
synchronized and only for one measurement


A
master node sends a synch message containing its local
absolute timestamp and timestamp in the near future
indicating
when measurements
should begin


All
receiving nodes convert the measurement time to their local
time, set up a start timer, and retransmit the message


In
this way, nodes outside the initial receiving range can take part in
the ranging round


It
was shown that all errors combined in this protocol still allow
for microsecond synchronization accuracy of the nodes


To
account for circuitry jitter, all received messages are immediately
timestamped before being passed to the ADC and
other measurement
circuitry


Phase
offset can then be found relative to the actual receive time
without jitter

Time Synchronization

Radio Tuning and Calibration


Calibration
is necessary to take into account temperature and
voltage effects on carrier frequency


Can
take up to 34ms and should be
performed
every time the
frequency changes by more than 1MHz


Frequency
span is separated into calibration channels with
channel 0 being
430.1 MHz
, and each separation being
0.526
MHz
apart


Within
each channel, fine
-
grained tuning can be performed in
65Hz intervals without requiring recalibration (very fast)


Nominal
tuning frequency,
f
, obtained from formula
:



One
limitation is that actual tuning can differ from nominal
f

by
up to 2kHz


Due
to measurement time constraints (29ms) and mote
sampling rates (9kHz), transmission frequencies must
differ in
the range 200
-
800 Hz

Tuning Algorithm


1 transmitter begins
transmitting at nominal
f
, while second
transmitter begins transmitting at

where
i

=
-
15,
-
14, ... 15


A
receiver node analyzes the interference signal and
determines which
i

creates an interference frequency closest to
0


It
transmits this back to one of the transmitters which adjusts its
transmission
frequency


It
has been noted that these inaccuracies are mainly due to
imprecisions in the crystal driving the node


Thus
, tuning factors are not constant between different
channels on different nodes,
but:


Tuning
factors are fairly linear across different channels of the same
node


Tuning
factors can be found for two different channels of a node and
interpolated to correct tuning factors for other channels

Frequency and Phase Estimation


Performed
on each
mote then transmitted
to the base
station along with a quality indicator value


256
samples per measurement


Very
resource
-
constrained (only about 820 CPU cycles
per sample for online processing)


Post
-
processing
a bit more relaxed (about 10000 CPU
cycles available)


No
floating point hardware, so computationally expensive
solutions,
such as Fourier analysis,
are not feasible

Frequency and Phase Estimation

Online Processing


Peak
detection performed online by the ADC ISR


Raw
samples are filtered by a moving average to smooth
results and enhance SNR


Min/max
values are acquired from the leading 24 samples
(must
contain at least one full period)


The
acquired amplitude value serves as a quality indicator of
the measurement


Samples
above a threshold of 20% of the max amplitude
measured are identified as high amplitude samples


Peaks
are defined as center points of 2 consecutive high
-
threshold crossings (not
-
high to high, then high to not
-
high)


Peaks
are discarded if the signal has not crossed the low
threshold since the last peak to minimize false positives

Post Processing


Works
exclusively on peaks identified and stored in the online
algorithm


Determines
the shortest period between subsequent peaks


Accumulates
the sum of all periods that are not longer than
130% the length of the shortest period


Frequency
is defined as the reciprocal of the average of this
sum


Phase
is defined as the average phase of accepted peaks


Small
frequency errors can result in large phase errors, so
phases are computed relative to the center of the sample
buffer
thereby
reducing accumulated phase error


The
estimated frequency, phase, and amplitude tuple is sent to
the base station

Scheduling


Two
levels:


1
) High
-
level scheduling for selecting the pair of transmitters


Should
minimize the required number of interference measurements
while producing enough to localize in 3D


2
) Low
-
level scheduling for coordinating the activities of the
transmitters and receivers


Includes
time synchronization, frequency calibration, and transmission
power
scheduling


full/full
power, full/half power, and half/full power all carried out by each
transmission pair


Currently
13 channels 5MHz apart are used between 400
-
460MHz

Range Calculation


With enough measurements at different frequencies, we
can solve for
d
ABCD


Possibility of multiple solutions differing by small integer multiples
of the wavelength


Define error function:


d
ABCD

resulting in smallest error is taken to be the final
estimate


The more frequencies used, the better the estimate


RIPS uses 10 frequencies

Localization


RIPS provides ranging estimates between sets of 4 nodes, not
pairs of nodes directly


Would require solving a large number of nonlinear equations


Uses a Genetic Algorithm (GA) as a baseline instead

Tuning Results


Good interference signals measured at double the
communication range of the radios (160:80 meters)





Frequency and

Phase Tuning

Comparisons:

Tuning Results


Mean error in phase measurements using varying
amplitude filter thresholds

Ranging Results


Algorithm for estimating interference signal frequency and
phase offsets determines the average amplitude of the signal
which correlates strongly with the error of the estimate

Better Ranging Results

Localization Results

Limitations


Maximum localizable range between transmitters and
receivers is defined by the radio range


Transmitters must be within 2 radio ranges of each other
to create a successful interference signal

Conclusions


Relies on two nodes transmitting a high
-
frequency carrier
signal at slightly different frequencies


The resulting interference signal has a low beat frequency
and can be measured at receiving nodes with cheap
hardware


Relative phase offsets measured at two receivers is a
function of the distances between the four nodes and the
carrier frequency


With at least 8 nodes, it is possible to localize each node
in 3D space


Achieves a localization accuracy of 3cm and a range of
up to 120 meters

Reference


Sha
, L. et al. “Priority Inheritance Protocols: An Approach
to Real
-
Time Synchronization.” IEEE Transactions on
Computers, Vol. 39, No. 9, Sept 1990.