HAL Id: hal-01466097
https://hal-ensta-bretagne.archives-ouvertes.fr/hal-01466097
Submitted on 28 May 2021
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Influence of Synchronization Impairments on an Experimental TDOA/FDOA Localization System
Hugo Seuté, Cyrille Enderli, Jean-François Grandin, Ali Khenchaf, Jean-Christophe Cexus
To cite this version:
Hugo Seuté, Cyrille Enderli, Jean-François Grandin, Ali Khenchaf, Jean-Christophe Cexus. In- fluence of Synchronization Impairments on an Experimental TDOA/FDOA Localization System.
Journal of Electrical Engineering, Politehnica Publishing House, 2017, 5, pp.1-9. �10.17265/2328-
2223/2017.01.001�. �hal-01466097�
Journal of Electrical Engineering 5 (2017) 1-9 doi: 10.17265/2328-2223/2017.01.001
Influence of Synchronization Impairments on an Experimental TDOA/FDOA Localization System
Hugo Seuté
1, Cyrille Enderli
1, Jean-François Grandin
1, Ali Khenchaf
2and Jean-Christophe Cexus
21. Thales Airborne Systems, 2 Avenue Gay Lussac, 78990 Elancourt, FRANCE
2. Lab-STICC, UMR CNRS 6285, ENSTA Bretagne, 2 Rue François Verny, 29806 Brest, FRANCE
Abstract: In Electronic Warfare, and more specifically in the domain of passive localization, accurate time synchronization between platforms is decisive, especially on systems relying on TDOA (time difference of arrival) and FDOA (frequency difference of arrival). This paper investigates this issue by presenting an analysis in terms of final localization performance of an experimental passive localization system based on off-the-shelf components. This system is detailed, as well as the methodology used to carry out the acquisition of real data. This experiment has been realized with two different kinds of clock. The results are analyzed by calculating the Allan deviation and time deviation. The choice of these metrics is explained and their properties are discussed in the scope of an airborne bi-platform passive localization context. Conclusions are drawn regarding the overall localization performance of the system.
Key words: Synchronization, Allan variance, time deviation, TDOA, FDOA, passive localization, USRP (universal software radio peripheral).
1. Introduction
Historically, most passive localization systems based on electromagnetic radiation have been developed in the context of EW (electronic warfare), and were denoted as ESM (electronic support measures). Various military applications have emerged, where ESM devices are mounted onboard different kinds of platforms: airborne [1], naval [2] or even spatial [3].
Passive localization systems use the properties of a signal received at different positions and/or dates in order to compute an estimate of the position of the radiating source. These systems differ from communication systems in that the source does not cooperate with the receiver, thus the waveform is unknown a priori and cannot be used to improve an estimator of the position of the source.
Traditionally, these localization systems have relied on interferometry to obtain AOA (angle of arrival)
Corresponding author: Ali KHENCHAF, professor, research fields: radar, sea clutter, radar cross section, sea EM scattering, complex targets EM scattering, signal processing and remote sensing.
measurements in order to triangulate the position of the source. But these measurements only have an accuracy of a few degrees and the acquisition time can be long before the location is estimated with the desired accuracy. This is why other types of measurements have been introduced. Many modern techniques used in passive localization now rely on time and frequency based measurements on different platforms. For example, many techniques are based on the calculation of T/FDOA (time/frequency difference of arrival) [1-5], and scan-based localization techniques use the dates of interception of the main lobe of a rotating emitter [6].
Since the measurements depend on the time on two or more remote platforms, several clocks are needed. All clocks have imperfections which make them drift, yet a single time base must be maintained all along the measurement time, hence the need to synchronize the devices [5].
Synchronization can be done in practice by exchanging a signal in different configurations:
one-way, two-ways, common view [7]. It is also possible to dispense with a sync signal if beacons of known positions can be seen by the receivers [8].
D
DAVID PUBLISHINGInfluence of Synchronization Impairments on an Experimental TDOA/FDOA Localization System
2
This article focuses on a system designed for combined TDOA/FDOA localization in a short-base airborne ESM context. This is a challenging scenario for a synchronized system because the measured time and frequency differences are small, thus the synchronization error must be kept as low as possible in order to have good precision [5]. In the scope of this article we will try to determine what performance in sync error we can expect in real-life situations via implementation and analysis of a full synchronization system, comprised of hardware (clocks, digital receivers) and software (delay estimation algorithm) processing. Then this will be translated in terms of localization error, in a simple scenario.
The paper is organized as follows: Section 2 presents some theoretical background on clock impairments and how to characterize sync error; Section 3 describes the methodology and the experimental process of the measurements which were carried out; in Section 4 the results for two different types of embedded clocks are shown and analyzed from an operational point of view;
Section 5 establishes the link between sync error and localization performances; finally global conclusions and perspectives are presented in Section 6.
2. Timekeeping Issues
A clock can be modeled as a device producing a sine wave output of the form [9]:
sin 2 (1)
where is nominal peak output voltage, amplitude noise, nominal frequencyand phase fluctuation (noise). In the case of time and frequency analysis, we can usually ignore the term. From there we can identify two parameters of interest [9]:
Time fluctuation:
(2) Fractional frequency, derived from the latter:
(3) Due to the non-stationary nature of , these quantities cannot be analyzed through traditional
statistics, the standard variance estimator will not converge as the number of samples increases [10]. In order to have a way to evaluate the amount of fluctuation of fractional frequency , the Allan variance was introduced [9, 11]. It measures the variance of the difference of two values of spaced by a time . An efficient estimator for the Allan variance can be expressed in terms of time data:
∑ 2 (4)
where is the time horizon on which the variance is calculated, the sample of a dataset containing
values of sampled every , and / the number of samples of contained in the time horizon ( must be an integer such as 1).
Other types of variances similar to the Allan variance were developed, like the modified Allan variance, which is capable to distinguish between more types of noise [9, 12]. The expression of its estimate in terms of time data is:
1
2 3 1
∑ ∑ 2 (5)
The time Allan variance is based on the modified (frequency) Allan variance and characterizes the time error of a clock [13]. It can be expressed as:
/3 (6)
All the considerations stated above refer to the characterization of a single clock. But in practice it is not possible to measure the absolute fluctuations of a clock ( and ) without having another clock to use as a time reference for . Therefore and do not represent the absolute fluctuation of a single clock but the fluctuations of a clock relative to another reference clock. 0 s and 0 means that at a date the two clocks are perfectly aligned with each other and their frequency is exactly the same.
can be interpreted as the standard deviation of
the time error between the clocks considering an
integration time of . For example, considering that the
Influ
time error simulation i 1/√ [13] ( lowest stand
∞, so t correction of values of
On the co a white frequ we have situation, th
when 0
synchroniza highest frequ correct the ti
In the cas the sum of d white phase function example is g synchronize
(which sliding wind means as a c Similarly, deviation of considering being directl of al white freque and for rand calculation o the best inte value of
Now we analysis and data that cou system.
3. Method
In this se
uence of Syn
is a white p illustrated on (as shown on dard deviation the synchroni ffset correspo as possibl ontrary, if the uency noise (
√ [ he lowest sta
0 i.e.
tion strategy uency and us ime base, wit se of a compl different noise e noise and may reac given in Fig. 3 d time base m is sampled e dow of width correction.
, can b f the relative f
an integratio ly linked to t lso reflects d ency noise, w om walk freq of also egration time
.
will try to d identify the
uld have bee
dology and
ection will be
nchronization
phase noise n Fig. 1a), w n Fig. 1b). I n of the error ization proce onding to the
le.
time error ca (like the simu
[13] (see Fi andard devia 0 . In this y is to sam se the vector thout any ave lex noise mo e types (for e white frequ ch a minimu 3. There, a go may be to com every with and use t
be interpreted frequency bet on time of that of ,
different typ we would hav quency noise
o makes it po in order to
o apply the e noise type o en obtained in
Measurem
e described t
n Impairments
(such as in we have
It means that r is attained w ess must app
mean of as m
an be modele ulation in Fig.
ig. 2b). In ation is achie case, the mple at of time error eraging.
odel compose xample in Fi ency noise), um . Again ood way to ha mpute the mea
h )
the series of th
d as the stand tween the clo
. Its expres , a representa pes of noise:
ve 1
√ . ossible to cho o have the low
e same type of real time e n an actual E
ments
the experime
s on an Expe
the
t the when ply a many
d by 2a), this eved best the rs to
ed of g. 3, the n, an ave a an of on a hese
dard ocks, ssion ation for 1/√
The oose west
e of error ESM
ental Fig.
(Fig time proc
Fig.
(Fig simu
Fig.
(Fig time of b
rimental TDO
. 1 Represent g. 1a, top) and
e (Fig. 1b, bo cess.
. 2 Represent g. 2a, top) an
ulated white fr
. 3 Represent g. 3a, top) and
e (Fig. 3b, botto both white phas
OA/FDOA Loc
tation of time time deviation ottom) of a si
tation of time d time deviat requency noise
tation of time time deviation om) of a simula se noise and w
calization Sys
error as a fun n as a function
imulated whit
error as a fu tion (Fig. 2b, e process.
error as a fu n as a function
ated noise proc white frequency
stem 3
nction of time of integration e phase noise
nction of time bottom) of a
nction of time of integration cess composed y noise.
3
e n e
e a
e
n
d
Influ
4
protocol use passive bi-p to evaluate h time error) o performance two remote r and softwa synchroniza
The recei two SDR (so USRP B210 card, linked This setup developmen frequencies and is availa TDOA have such as Ref.
synchroniza Here, in o time error b
Fig. 4 Diagr
Fig. 5 Diagr
uence of Syn
ed to acquire latform local here is not jus of the clocks o e of a whole receivers, inc are processin
tion protocol ivers used fo oftware defin 0 (receiver # d to a laptop
was chosen nt and expe
from 70 MH able off-the-s e already been f. [14] but the
tion error.
order to have between two
ram of the exp
ram of the dela
nchronization
e time fluctu ization system st the perform of the receive
passive syste cluding their i ng, their c l.
or this experim ed radio) plat 1) and a B2 computer to n because erimentation Hz to 6 GHz,
shelf. Experim n carried out o ey did not an
e an accurate receivers, a
erimental set-u
ay estimation p
n Impairments
uation data o m. What we w mance (in term
rs, but the ov em, compose internal hardw clock and t
mental study tforms based 200 (receiver record the d it allows q
tasks on r it is quite ch mental studie
on USRP dev nalyze thorou
e way to mea a synchroniza
up.
process.
s on an Expe
on a want ms of erall ed of ware their
y are on a r #2)
data.
quick radio heap es on vices ghly
asure ation
syst emi sign B21 loca dela dela can
T
dete reco stam hav step
reco accu T [15 Fig
rimental TDO
tem has bee ission of a pe nal is actually 10 card. In ated close to ay can be neg ay cannot be ncelled if the p The sync sign
When it i ector on the ording a fix mped file. Th ve a huge amo
ps.
The comp orded by th urately obtain This delay es
] and [16]. It . 5: First the c
OA/FDOA Loc
en developed eriodic sync s y collocated w this experim o each other glected. In the
neglected, bu platform posi nal is used for is detected e signal’s ba xed number his avoids to r ount of data to
lex envelop he two rece
n the time dif timation proc t is done in fo complex enve
calization Sys
d (Fig. 4), b signal. The em with receiver #
ment two p so that the e case of a rea
ut it can be e itions are kno r two things:
(via a simp and), the re of samples record perman
o process in t
es of the s eivers are p fference betw cess is descri our steps, as i elopes are cro
stem
based on the mitter of this
#1, inside the platforms are propagation al system this estimated and own [15].
ple threshold eceivers start into a time nently and to the following
sync signals processed to ween them.
ibed in Refs.
illustrated on ss-correlated, e s e e n s d
d t e o g
s o
. n
,
Influ
Fig. 6 A rep Table 1 Val
P. V
1 1 2 1 3 5 next the p cross-correla then interpo the parabola final estimat our experim next to each signal betw Thus the va chain direct the tim Due to th requirement
When envelope of cross-correla there must b interval of p periodic w maximum t records—the that could co
uence of Syn
presentation of ues of the para Value
10 s 143 ns 200 ms 1 kHz 3 MHz 5 GHz points aroun
ation function lated by a par a is compute te of the dela ment, the two h other so the een the two alue at the ou tly correspo me error betwe hese different s on this sign cross-correla f the signals c
ation peak be a single p possible delay with a short time offset ere will be se orrespond to
nchronization
f the envelope o ameters of inte
D S S D F F N nd the main
n are selected rabola, and fi ed, which co ay between th o receivers a difference of platforms c utput of the nds to an een each rece t purposes, th nal:
ation is com contained in t must be peak which ys. Indeed, if t period—sh
expected be everal cross-c a consistent d
n Impairments
of the sync sign erest.
Description Sync signal peri Sampling period Duration of the r Frequency of the Frequency of the Nominal frequen n peak of d, these points
inally the ape orresponds to he two signal are situated r
f path of the s an be neglec delay estima estimate iver.
here are diffe
mputed from the two files unambigu occurs inside the sync sign horter than etween the correlation p delay estimat
s on an Expe
nal with
od of emission d
record of the sy e first modulati e second modul ncy of the local
the s are ex of o the
s. In right sync cted.
ation of
erent
the , the uous:
e an nal is
the two eaks ion.
imp it is esti mea sign
whe Hen freq esti
A bee com and first
whe sign T grou
rimental TDO
and
ync signal on of the sync s lation of the syn l oscillators on t
This cross-c prove the accu s stated that th imation is pro
an square rad nals, defined
ere is nce the synch quency comp imation.
A signal com en chosen arb mposed of a c d a high frequ t and second
ere and nals and The settings
uped on Tabl
OA/FDOA Loc
, sam
signal nc signal the receivers an correlation pe uracy of dela he standard d oportional to dian frequenc
by 2 s the signal hronization s ponents to
mplying with bitrarily (repr arrier modula ency signal (s requirement)
sin 2 s
are the freq is the carrier used during le 1.
calization Sys
mpled every
nd emitter eak needs to ay estimation.
deviation of th 1/ , where cy” of the cro
∞
∞∞
∞
power densi signal needs provide acc
h these requi resented on ated in amplit satisfying res ).
sin 2
quencies of th r frequency.
g the measu
stem 5
.
be narrow to . In Ref. [16]
he time offset is the “root oss-correlated
/
(7) ity spectrum.
to have high curate delay
irements has Fig. 6). It is tude by a low spectively the
(8) he modulated
urements are 5
o ] t t d
) . h y
s s w e
) d
e
Influ
6
Now that fluctuations next step is t consistent w
4. Measur
Two exp featuring a d clock type experiment.
compensated GPSDO (GP actually bas internal cont signal from
Fig. 7 re measuremen the reception clocks type computed fr and (5). Fig.
data, using experiments limited to
for relevant (les interval is t acquisition t behavior of a it appears u the system e What we c is that the fre term time st for a long proportional random walk clocks reach minimum is period
Clocks typ
uence of Syn
we have a s measuremen to produce th with operation
rement Ana
periments ha different type is mounted Clocks type d crystal osci PS disciplined sed on the sa trol loop usin a GPS receiv epresents two nts betw n system fitte e B. Fig. 8 rom the latter 9 shows the A (X). The s was 1 hour
360 s, bec close to ac ss samples a too large). I time is often a clock, but in unlikely that t
exceeds sever can see from ee running cl tability but w
time, time lly to
/, k frequency m h a minimum s reached wh
40 s (4 mea pe B have a w
nchronization
system capab nts between tw hese data in co nal contexts.
alysis
ave been ca e of clock (A d on both r
e A are TCX illator) and c d oscillator), ame TCXO ng the 1-PPS ( ver as a refere o series of ween the two ed either with 8 shows the r measuremen Allan deviati
acquisition r. The Allan cause values cquisition tim are averaged In other app used to analy n our context the synchron ral minutes.
the time devi ocks (type A) when it is left e deviation which is ch modulation no
time deviatio hen is av asurements).
worse short te
n Impairments
le of giving t wo receivers onditions tha
arried out, e A or B). The s
receivers on XO (tempera clocks type B which are clo but joined to (pulse per sec ence.
time fluctua o platforms, w h clocks type A e time devia
nts using Eqs on from the s time for th deviation plo s of
me may not d, the confide plications, lon yze the long t
of airborne E nization perio
iation plot (F ) have good s ft unsynchron incre haracteristic o
oise [13]. Typ on of 2.5 ns.
veraged durin
erm time stabi
s on an Expe
time , the t are
each same n an
ature B are ocks o an cond)
ation with A or ation . (4) same
hese ot is and t be ence nger term ESM, od of
ig. 8) short nized eases of a pe A
This ng a
ility, Fig.
the
Fig.
plat
Fig.
plat
y()
rimental TDO
. 7 Measured two experimen
. 8 Exper tforms.
. 9 Experim tforms.
101 10-10 10-9
-1
OA/FDOA Loc
d time error nts.
rimental time
mental Allan
0
calization Sys
between p
e deviation
deviation
102
(s)
-1
1/2
stem
platforms, for
between
between
Clock A Clock B
