没有合适的资源?快使用搜索试试~ 我知道了~
Mobisys2023.pdf
需积分: 0 0 下载量 110 浏览量
2023-11-06
21:22:22
上传
评论
收藏 2.47MB PDF 举报
温馨提示
试读
13页
Mobisys2023.pdf
资源推荐
资源详情
资源评论
LocRa: Enable Practical Long-Range Backsca�er Localization for
Low-Cost Tags
Jinyan Jiang, Jiliang Wang, Yijie Chen, Yihao Liu, Yunhao Liu
Tsinghua University, Beijing, P.R. China
ABSTRACT
Long-range backscatter localization is a promising technology for
the Internet of Things. Existing works cannot work well for dis-
tributed base stations and low-cost tags. We present LocRa, which
provides accurate localization for long-range backscatter with dis-
tributed base stations. We present a novel method to extract accu-
rate channel information and synchronize the phase of dierent
base stations. To compensate for the frequency and phase error
on low-cost tags, we combine multiple channel measurements and
eliminate the error by aligning dierent channels. Finally, we ex-
ploit frequency domain characteristics of the backscatter signal
to extend its bandwidth and improve the SNR, thereby enhancing
the localization accuracy. We prototype LocRa tags using custom
low-cost hardware and implement LocRa base stations on USRP.
Through extensive experiments, we show that the localization er-
ror of LocRa is 6
.
8
cm
and 88
cm
when the tag is 5m and 50 m
away from the base station, which is 3.1
⇥
and 2.3
⇥
better than the
state-of-the-arts methods.
CCS CONCEPTS
• Information systems
!
Sensor networks; Global position-
ing systems; • Networks ! Cyber-physical networks.
KEYWORDS
Backscatter, Wireless Localization, LoRa
ACM Reference Format:
Jinyan Jiang, Jiliang Wang, Yijie Chen, Yihao Liu, Yunhao Liu. 2023. LocRa:
Enable Practical Long-Range Backscatter Localization for Low-Cost Tags.
In The 21st Annual International Conference on Mobile Systems, Applications
and Services (MobiSys ’23), June 18–22, 2023, Helsinki, Finland. ACM, New
York, NY, USA, 13 pages. https://doi.org/10.1145/3581791.3596863
1 INTRODUCTION
Recent advances in wireless localization enable various new appli-
cations in the Internet of Things (IoT), such as smart homes, smart
agriculture, and smart healthcare [
1
–
13
]. Backscatter localization
MobiSys ’23, June 18–22, 2023, Helsinki, Finland
© 2023 Copyright held by the owner/author(s).
ACM ISBN 979-8-4007-0110-8/23/06.
https://doi.org/10.1145/3581791.3596863
Base-station
LocRa tag
Rx 2
Rx 1
Tx
Phase jitter
Limited bandwidth
and SNR
Ellipse 1
Ellipse 2
Figure 1: LocRa provides long-range backscatter localization
for low-cost tags with distributed base stations.
systems have brought smaller form factors and lower power con-
sumption to wireless localization. More specically, the recent de-
velopment of long-range backscatter [
8
,
13
–
16
] (e.g., LoRa backscat-
ter) is considered to be promising as it signicantly extends the
communication and localization range. However, existing long-
range backscatter localization systems still face the following key
challenges while applying to practical scenarios:
Distributed base stations. Backscatter localization systems usu-
ally require the time/phase synchornization between transmitters
(Tx) and receivers (Rx) to calculate the Time-of-Flight (ToF). To
achieve precise synchornization among transceivers, some past
works [
8
,
13
,
17
] exploit a common external reference clock shared
through RF cables. However, it is not practical for distributed base
stations because deploying such long cables is high-cost and incon-
venient in practice. For wireless synchronization techniques, such
as GPSDO [
18
] and RFclock [
19
], they need to add extra hardware to
existing base stations. Moreover, the eectiveness of these solutions
is signicantly degraded in indoor multipath environments [20].
Low-cost backscatter tags. The unstable low-cost and low-power
oscillators on backscatter tags cause severe frequency and phase
error in backscatter signal, which reduces the localization accu-
racy. External environmental factors (e.g., temperature, humidity
and pressure [
21
]) further worsen the error. Previous works, e.g.,
`
locate, propose to measure the shifting frequency of tags and
then use it to calibrate the phase shift. However, it is dicult to
accurately measure the unstable shifting frequency in a multipath
environment (details in §4).
Low SNR backscatter signal of limited bandwidth. The backscatter
signal usually has a ver y low SNR and limited bandwidth, espe-
cially in long-range localization (e.g., LoRa backscatter). Traditional
317
This work is licensed under a Creative Commons Attribution International 4.0 License.
MobiSys ’23, June 18–22, 2023, Helsinki, Finland Jinyan Jiang, Jiliang Wang, Yijie Chen, Yihao Liu, Yunhao Liu
approaches need to increase the power and duration of the excita-
tion signal and use a larger bandwidth to improve the localization
accuracy. However, this increases the cost and delay for backscatter
localization. It also cannot work for long-range backscatter net-
works of LoRa with a very low SNR and limited bandwidth (details
in §5).
To address the above challenges and make long-range backscatter
localization work in practice, we present LocRa, the rst Localization
system for Long-Range backscatter tags with distributed base sta-
tions. As shown in Fig. 1, LocRa can work for unsynchronized
single-antenna base stations, e.g., one LoRa base station as the
Tx and another as the Rx, and thus use existing network as the
infrastructure. The LocRa tag, as in many other backscatter sys-
tems [
8
,
22
–
30
], shifts the frequency of the excitation signal using
low-cost and lower-power oscillators. LocRa accurately calculates
the travel time of the backscatter signal and the total travel distance
for the backscatter signal. By combining the distance to dierent
base stations, Lo cRa can provide the localization of the backscatter
tag. In the design of LocRa, we address the following key problems:
(1) How to address the random phase shift between Tx and Rx? The
random phase shift introduced during up-/down-converting in Tx
and Rx (details explained in
§
2) results in a large lo calization error.
Since the excitation and backscatter signals should experience the
same frequency and phase oset, we leverage the excitation signal
to correct the phase shift of the backscatter signal. To obtain the
phase shift of the excitation signal, a naive approach is to use the
location of base stations to infer the channel. But this method suers
from severe distortion in real multipath environment. Existing
works can measure the channel between two transceivers based on
the excitation signal [
3
]. However, we show that they lead to too
many degrees of freedom and thus cannot work well in practice,
causing ambiguity in phase shift estimation. We propose a novel
method to eectively and accurately measure the channel between
base stations and resolve the ambiguity in phase estimation.
(2) How to address phase error in unstable backscatter tags? The
frequency and initial phase of the backscatter signal drift over
time due to the low-power, high-jitter crystal oscillator on the tag,
resulting in localization failure. In
§
4, we build three types of
tags using dierent crystal oscillators and demonstrate the phase
error for the backscatter signal. To address this, the key idea is that
the phase of the same channel at the same frequency is expected
to be the same. Therefore, we propose a method to continuously
calibrate the channel phase by combining multiple measurements
at the same frequency and then eliminate the initial drifting phase
of the backscatter signal. During the whole calibration process,
backscatter tags only need to shift the frequency. All the parameter
selection and channel calculation operations are performed at the
base stations.
(3) How to improve localization accuracy for low-SNR backscatter
signal of limited bandwidth? The backscatter signal is typically weak
and heavily distorted, making it dicult to extract localization infor-
mation from it. To address this, we use LoRa signal as the excitation
signal, which allows to concentrate the energy of the backscatter
signal in the time domain and extract the channel state information
(CSI) in the frequency domain. Moreover, the backscatter signal can
be decomposed into multiple sidebands in the frequency domain,
i.e., the upper and lower sidebands. Previous works only use one
of the sidebands. Combining multiple sidebands can improve the
signal bandwidth and SNR. However, directly combining sidebands
cannot work due to the unmatched initial phase and frequency
discontinuity in the spectrum. By analyzing the characteristics of
the upper and lower sidebands, we propose a method to combine
double-sidebands signal to extend the bandwidth and increase the
signal SNR, thereby improving the localization accuracy.
Main results and contributions.
•
We investigate the key limitations for long-range backscatter
localization systems with distribute d base stations and low-cost
tags. We show that those limitations lead to signicant localiza-
tion error and existing works cannot eectively address them.
•
We propose LocRa, a long-range backscatter localization system
with distributed base stations and low-cost tags. We design a
method to eliminate the frequency and phase oset between
distributed unsynchronized base stations and resolve the ambi-
guity in existing works. We propose a method to calibrate the
output phase of backscatter tags with low-cost and low-power
crystal oscillators. Finally, we leverage double sidebands of the
backscatter signal to improve the signal bandwidth and SNR,
and thus improve localization accuracy.
•
We implement LocRa tag on a customized PCB board using
commercial low-cost and low-power o-the-shelf components.
The evaluation results show that the localization error of LocRa
is 6
.
8
cm
and 88
cm
when the tag is 5mand 50 m away from the
base station; compared with the state-of-the-arts, LocRa achieves
3.1⇥ and 2.3⇥ higher localization accuracy.
2 LOCALIZATION MODEL
There are three main parts of LocRa, as shown in Fig. 2. The Tx (e.g.,
a LoRa base station) sends a LoRa chirp. The signal is frequency
shifted by a LocRa tag and then received by the Rx (e.g., another
LoRa base station). Based on the received excitation and backscatter
chirp, the Rx derives the location of the tag. The Tx and Rx can be
far from each other, and there is no synchronization cable between
Tx and Rx.
The Tx rst generates a signal in the baseband, as shown in
Fig. 2(a). We take a LoRa base up-chirp
B (C ) = 4
92c
(
5
0
C+
1
2
:C
2
)
as an
example, where
5
0
is the starting frequency, and
:
is the frequency
changing rate. The Tx then up-converts the baseband signal to
( (C)
using a high-frequency carrier signal:
( (C) = B ( C )·4
9
(
2c5
2
C+\
)G
)
(1)
where
5
2
and
\
)G
are the frequency and initial phase of the carrier.
Then, the signal experiences various reections through dierent
paths. For a backscatter tag, assume there are
<
dierent paths
between Tx and Tag with attenuation
{U
1
1
,U
2
1
,...,U
<
1
}
and delay
{g
1
1
,g
2
1
,...,g
<
1
}
, and there are
=
paths between Tag and Rx with
attenuation
{U
1
2
,U
2
2
,...,U
=
2
}
and delay
{g
1
2
,g
2
2
,...,g
=
2
}
. As a result,
the received backscatter signal consists of
< · =
multipaths, which
is a pairwise combination of the Tx-to-Tag path and Tag-to-Rx path.
We rst analyze one signal path of
g
8
1
and
g
:
2
, then extend it to the
multipath case.
As shown in Fig. 2(b), the tag shifts the frequency of the incoming
signal by
4
9 (2c 5C+\
)06
)
, where
5
is the shifte d frequency and
318
LocRa: Enable Practical Long-Range Backsca�er Localization for Low-Cost Tags MobiSys ’23, June 18–22, 2023, Helsinki, Finland
Base station
LocRa tag
!
"#
$
%
x
x
$
-"#
&!
!
"#
$
%
'&!
!
!
!
"
(()* +,-
(().,/0
$
&!
'&!
1
!
1
"
#$$%&
1
"
'()%&
Tx Rx
(a)
(b)
(c)
!
"
Wireless
Tag phase calibration
Bandwidth extension
IFFT (INDFT)
Frequency
shifting
"
#
"
$
+,-
23456
!
!
+ !
"
Dechirp
FFT
(d)
Figure 2: The main work�ow of LocRa.
\
)06
is the initial phase. The backscatter signal (
⌫
(C) is
(
⌫
(C) = U
8
1
· B (C g
8
1
)·4
9
(
2c5
2
(C g
8
1
)+\
)G
)
· 4
9 (2c 5C+\
)06
)
(2)
The backscatter signal then propagates with time
g
:
2
and attenua-
tion U
:
2
. Thus, Rx receives the backscatter signal '
⌫
(C):
'
⌫
(C) =U
:
2
· (
⌫
(C g
:
2
)
=U
8
1
U
:
2
· B (C g
8
1
g
:
2
)·4
9
2c5
2
(C g
8
1
g
:
2
)+\
)G
· 4
9 (2c 5 (C g
:
2
)+\
)06
)
(3)
Then, Rx down-converts the signal
'
⌫
(C)
to the baseband
A
⌫
(C)
using negative carrier signal
4
9 (2c5
0
2
+\
'G
)
, where
5
0
2
and
\
'G
are
the frequency and initial phase of the carrier. Since Tx and Rx are
not synchronized, Rx has a time oset
W
in the sampling window
to process the baseband signal:
A
⌫
(C + W ) = U
8
1
U
:
2
· B (C + W g
8
1
g
:
2
)·4
92c · 5
⇠$
·
(
C+W
)
· 4
9 2c5
2
(g
8
1
+g
:
2
)
· 4
9 (\
)G
\
'G
)
· 4
9 (2c 5 (C +W g
:
2
)+\
)06
)
⇡ U
8
1
U
:
2
· B (C )·4
92c · 5
⇠$
·
(
C+W
)
· 4
9 2c5
2
(g
8
1
+g
:
2
)
· 4
9 (\
)G
\
'G
)
· 4
9 (2c 5 (C +W g
:
2
)+\
)06
)
(4)
where
5
⇠$
= 5
2
5
0
2
. Without loss of generality, we set the starting
frequency
5
0
of the chirp signal
B (C )
as 0 and then ignore
W
and
g
in
B (C W g
8
1
g
:
2
)
as the quadratic term of this tiny delay has little
eect on the result.
To obtain the channel state information, we rst dechirp the
received chirp symbol
A
⌫
(C + W)
, i.e., multiply it with a down-chirp
B
⇤
(C)
(the conjugate of a base up-chirp). Then, we apply FFT to the
dechirp results and calculate the peak of backscatter signal in the
frequency domain (shown as the red peak in Fig. 2(c)). The value of
the corresponding FFT bin
ˆ
⌘
⌫
(
5
2
)
is
ˆ
⌘
⌫
(
5
2
)
=U
8
1
U
:
2
4
9 2c5g
:
2
· 4
9 2c5
2
(g
8
1
+g
:
2
)
· 4
9 2c ·5
⇠$
·W
· 4
9 (\
)G
\
'G
)
· 4
9⇥
)06
=⌘
⌫
(
5
2
)
· 4
92c ·5
⇠$
·W
· 4
9 (\
)G
\
'G
)
· 4
9⇥
)06
(5)
where
⌘
⌫
(
5
2
)
= U
8
1
U
:
2
4
9 2c5g
:
2
· 4
9 2c5
2
(g
8
1
+g
:
2
)
is the real channel
of the backscatter signal,
⇥
)06
= \
C06
+
2
c5W
is the phase error
introduced by the tag and
W
is the sampling window oset. Note
that the term
4
9 2c5g
:
2
remains the same across dierent carrier
frequencies, indicating that the channel has a xed phase rotation.
We can extend the channel
⌘
⌫
(
5
2
)
for a signal path to
< ·=
paths
as
⌘
⌫
(
5
2
)
=
<
’
8=1
=
’
:=1
8,:
· 4
9 2c5
2
)
8,:
(6)
where
8,:
= U
8
1
U
:
2
4
9 2c5g
:
2
and )
8,:
= g
8
1
+ g
:
2
.
Assume
g
1
1
and
g
1
2
are the delays for the LoS Tx-Tag and Tag-Rx
path, respectively. We need to calculate
)
1,1
= g
1
1
+g
1
2
for localization.
By measuring the channel at multiple carrier frequencies
5
2
and
performing IFFT operation on the channel value sequence, we can
calculate the Channel Impulse Response (CIR) 5 (C ):
5
⌫
(C) =
<
’
8=1
=
’
:=1
8,:
· X (C )
8,:
)
(7)
where
X (·)
is the delta function, as shown in Fig. 2(d). Eq. 7 describes
the attenuation (
8,:
) and delay (
)
8,:
) of each path. We calculate the
travel time
)
1,1
for the LoS path from the rst impulse
X (C )
1,1
)
and then calculate the sum of Tx-to-Tag and Tag-to-Rx distance.
We can determine an ellipse based on the location of the Tx and Rx.
With multiple Rxs, we can use the intersection of ellipses as the tag
location. We need at least two Rxs to determine two ellipses.
2.1 Challenges
Frequency and phase misalignment between Tx and Rx. In
Eq. 5, the frequency and phase oset between Tx and Rx, i.e.,
4
92c ·5
⇠$
W
and
4
9 (\
)G
\
'G
)
, introduce error to the channel esti-
mation of
⌘
⌫
(
5
2
)
. For dierent frequency
5
2
, the local oscillators
in Tx and Rx will output a random initial phase
\
)G
and
\
'G
for
the carrier. The
5
⇠$
is also dierent for dierent center frequency
5
2
. Although channel estimation between static transceivers has
been a well-studied topic in wireless communicaiton [
31
,
32
], they
cannot work in our scenario as we need to decouple the phase
shift from CFO and other factors. To address this,
`
locate [
8
] ex-
ploits an external clock connected with both Tx and Rx to provide
a reference clock to cancel the
5
⇠$
between Tx and Rx. This also
provides identical
\
)G
and
\
'G
. GPSDO [
18
] can use the GPS satel-
lite broadcast signal to synchronize the time and phase of each
transceiver outdoors. RFclock [
19
] designs an external board for
SDRs to share a common reference clo ck. Chronos [
3
] requires
Tx and Rx to exchange a packet to calibrate the frequency and
phase oset. However, Chronos cannot provide accurate multipath
information for localization (details in § 3.2).
Tag phase error. In Eq. 5, there is another error term
4
9⇥
)06
introduced by the tag. As shown in Fig. 3, suppose Tx sends the
excitation chirp of length
)
at three dierent carrier frequencies.
319
剩余12页未读,继续阅读
资源评论
2301_76640429
- 粉丝: 0
- 资源: 1
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功