Detecting BrakTooth Attacks
Achyuth Nandikotkur
1 a
, Issa Traore
1 b
and Mohammad Mamun
2 c
1
ECE Department, University of Victoria, British Columbia, Canada
2
National Research Council Canada, New Brunswick, Canada
Keywords:
Bluetooth Security, Intrusion Detection, Machine Learning, BrakTooth.
Abstract:
More than 5.1 billion Bluetooth-enabled devices were shipped in the year 2022 and this trend is expected to
exceed 7.1 billion by the year 2026. A large proportion of these devices are used in smart homes designed for
older adults, to help them age in place. Monitoring vitals, climate control, illumination control, fall detection,
incontinence detection, pill dispensing, and several other functions are successfully addressed by many of these
Bluetooth-enabled devices. Therefore it becomes crucial to protect them from malicious attacks and ensure
the safety and well-being of their users. Some of these devices have only Bluetooth connectivity which makes
patching them challenging for older adults, as a result, most remain unpatched. The family of vulnerabilities
recently found in the Bluetooth Classic (BT Classic) stack called BrakTooth, poses a genuine threat to such
devices. In this study, we develop an experimental procedure to capture traffic at the Link Manager Protocol
(LMP) layer of the BT Classic stack and use machine learning algorithms to detect BrakTooth-based attacks.
1 INTRODUCTION
According to the Bluetooth Special Interest Group
(SIG, 2023), it is projected that over 7.1 billion
Bluetooth-enabled devices will be shipped in 2026.
This figure encompasses all Bluetooth technology de-
vices, including BT Classic and newer versions such
as Bluetooth Low Energy (BLE). Although the num-
ber of devices shipped with just BT Classic is ex-
pected to decline, 100% of the future Bluetooth-
enabled devices are expected to support dual-mode
(BT Classic + BLE). A substantial number of these
devices are used by aging adults, such as smart-
watches, hearing aids, fall detection devices, blood
pressure monitoring devices, weighing scales, pill-
boxes, diapers, and more, to assist them in everyday
activities (Wagner et al., 2012). Therefore, it becomes
crucial to protect these devices against attacks that in-
tend to compromise them.
In recent years, Garbelini et al. (2021) has dis-
closed a family of new security vulnerabilities in com-
mercial Bluetooth stacks that can compromise the
availability of BT Classic devices, called BrakTooth.
The BrakTooth vulnerabilities exist especially in the
link manager and baseband layers of the Bluetooth
a
https://orcid.org/0000-0001-7098-8374
b
https://orcid.org/0000-0003-2987-8047
c
https://orcid.org/0000-0002-4045-8687
stack. Several vendors producing BT Classic system-
on-chips (SoCs) were notified of the disclosures and
they have already started patching their implemen-
tations of the stack, or have already patched them.
Usually, vendors provide security patches to SoCs
through wireless over-the-air (OTA) firmware up-
dates, or wired updates (i.e, via USB) (El Jaouhari and
Bouvet, 2022), or replace the devices with patched
SoCs; the last one requires recall and is rarely done.
Typically, updating the firmware of Bluetooth-only
devices over the air requires two separate devices: a
Device Firmware Update (DFU) target and a DFU
controller. The DFU controller, which is usually a
mobile device running a vendor-specific application,
is responsible for transferring the firmware image to
the DFU target device where the update needs to be
made. Nordic Semiconductors employs this tech-
nique to update their Bluetooth-based devices over
BLE (NordicSemiconductor, 2018). Several devices
used in aging in place have only Bluetooth connec-
tivity, therefore, older adults without sufficient tech-
nical expertise find it challenging to update them. As
a result, several such devices remain vulnerable and
prone to compromise. In this study, we propose a
novel method to detect BrakTooth attacks using ma-
chine learning and inexpensive hardware.
Nandikotkur, A., Traore, I. and Mamun, M.
Detecting BrakTooth Attacks.
DOI: 10.5220/0012128000003555
In Proceedings of the 20th International Conference on Security and Cryptography (SECRYPT 2023), pages 787-792
ISBN: 978-989-758-666-8; ISSN: 2184-7711
Copyright
c
2023 by SCITEPRESS Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)
787
1.1 Background
Bluetooth is a wireless communication standard that
can be used to exchange data between stationary and
fixed devices within a range of up to 100 meters
(Ferro and Potorti, 2005). Although Bluetooth occu-
pies the ISM band of 2.4 GHz which is 83MHz wide,
it does not use the Direct Sequence Spread Spectrum
(DSSS) used by WiFi. Instead, it uses Frequency
Hopping Spread Spectrum (FHSS) to hop between 79
different 1 MHZ-wide channels in this band. Due to
its use of FHSS, interference with other devices is re-
duced. However, WiFi uses a single channel that is
22MHz wide, and when both WiFi and Bluetooth net-
works are in the same range, the 22MHz channel of
WiFi occupies 22 of the 79 Bluetooth channels lead-
ing to some interference. When a Bluetooth device
experiences interference, it addresses this problem by
hopping to the next channel and retrying. In contrast,
WiFi issues an Automatic Repeat Request and slows
down the data rate in an attempt to reduce the Bit Er-
ror rate. However, the number of channels and the
channel bandwidth differ based on the type of Blue-
tooth device used. There exist two types of Bluetooth
devices: Bluetooth Classic (BT Classic) and Blue-
tooth Low Energy (BLE). A few salient differences
between these two devices are listed in Table 1.
1.2 Sniffing Bluetooth Traffic
Several devices and tools are readily available to
monitor wireless WiFi traffic. However, monitoring
Bluetooth traffic reliably is limited to the realm of
professional equipment like the Ellisys Vanguard or
the Frontline Sodera and they can be very expensive
(Cominelli et al., 2020). In general, there exist two
types of sniffing:
Passive sniffing: It involves a device that sniffs
over-the-air Bluetooth packets between two com-
municating devices in radio range.
Active sniffing: It involves a device that connects
to a remote BT device and captures the packets
exchanged down to the lowest layer in the Blue-
tooth protocol stack.
Due to the fact that Bluetooth uses frequency-
hopping spread spectrum technology with a vendor-
specific hopping pattern, it is difficult to monitor its
connections. Despite these challenges, a few inex-
pensive passive sniffers such as Ubertooth One (Os-
smann, 2011) and nRF Sniffer (nRFSniffer, 2023)
were developed, however, they can only capture pack-
ets of BLE reliably, and not BT Classic (Nijholt et al.,
2020). To the best of our knowledge, there is still no
Bluetooth Physical Layer
Baseband and Link
Control
Link Manager Protocol
(LMP)
Host Controller
Interface (HCI)
RFCOMM
L2CAP
SDP
BrakTooth
attacks
target
Figure 1: BrakTooth attacks targeting the LMP and base-
band layers of the BT Classic stack.
non-commercial solution for monitoring BT Classic
traffic. One approach to capturing the packets per-
taining to BrakTooth attacks would be to set up an
Android device as the active sniffing target and enable
a debug functionality in its Bluetooth stack known as
the HCI Snoop Log feature (Ditton et al., 2020). This
feature captures all BT traffic above the HCI layer of
the Bluetooth stack. However, as shown in Figure 1,
since the BrakTooth attacks are aimed to exploit the
vulnerabilities in the link manager protocol (LMP)
and baseband layers of the Bluetooth stack, we need
a method to monitor the lower-level traffic. Thus,
we adopted a well-known framework called Internal-
Blue (Mantz et al., 2019), which is capable of mon-
itoring communications below the HCI layer. The
researchers who developed this framework reverse-
engineered widely used Broadcom BCM4339 Blue-
tooth Controller firmware and developed patches that
give access to the LMP layer. Note that the Nexus
5 is one of the popular mobile devices that contain
the BCM4339 controller (Schulz et al., 2015). In this
study, we utilized the InternalBlue framework to en-
able LMP monitoring on a Nexus 5 device and use it
as an active sniffer to capture LMP packets from both
benign and malicious devices.
It should be noted that this approach to capturing
LMP packets was not explored in prior works. As a
consequence, detecting BrakTooth-based attacks us-
ing machine learning and inexpensive hardware was
not feasible until now. Therefore, our study presents
only our findings without a comparative analysis.
The remainder of the paper is organized as fol-
lows. Section 2 describes the related works. Section
3 presents the experimental setup and dataset collec-
tion. Section 4 presents results from applying various
machine learning algorithms to the dataset. Finally,
Section 5 presents our conclusions and future work.
SECRYPT 2023 - 20th International Conference on Security and Cryptography
788
Table 1: Salient differences between BT Classic and BLE.
Feature Bluetooth Classic Bluetooth LE
Application Audio streaming and data transfer.
Audio streaming, data transfer,
location services and device networks.
Frequency band 2.402 - 2.480 GHz (ISM) 2.402 - 2.480 GHz (ISM)
Channels 79 40 channels
Channel bandwidth 1 MHz 2 MHz
Spread spectrum FHSS FHSS
Power Consumption 1W 0.001 W-0.5 W
Data rate 1 Mb/s, 2 Mb/s, 3Mb/s 125 Kb/s, 500Kb/s, 2 Mb/s
Device discovery Inquiry or paging Advertising
Encryption algorithm E0/SAFER+ AES-CCM
Network topology Point-to-point Point-to-point, Broadcast and Mesh
2 RELATED WORKS
The following section discusses some works that use
both commercial and non-commercial devices to de-
tect attacks utilizing various methods such as thresh-
olds in Bluetooth specifications, signatures, statistical
distributions of data, and anomaly detection.
2.1 Detecting Bluetooth Attacks
Wu et al. (2020) used the Ubertooth One device to
capture over-the-air packets in the BLE advertising
channel and extracted features such as the advertis-
ing pattern, state transitions of the BLE device, ad-
vertising interval, the RF (Radio frequency) signal
frequency offset, and the RF signal strength. They
detect spoofing attacks by creating a statistical distri-
bution of CFO (Carrier frequency offset) and RSSI
(Received signal strength indication) values and use
them to identify any value that falls out of a prede-
termined threshold. They first look for a connection
request packet to identify the detection of a new de-
vice and then use the above statistical distributions to
detect a spoofing condition. One drawback of using
the Ubertooth One device for BT classic monitoring
is that it is unable to capture all packets reliably (Ni-
jholt et al., 2020; Antonioli et al., 2019).
OConnor and Reeves (2008) presented a
Bluetooth-based network intrusion detection system
that uses the Merlin LeCroy protocol analyzer to
non-intrusively capture the Bluetooth traffic and
detect malicious behavior using pattern matching.
It decodes Bluetooth packets by synchronizing with
the master device on a piconet and following the
hopping sequence of that piconet. However, since the
detection is signature-based, novel intrusions cannot
be detected. The authors have used a commercial
protocol analyzer and hence were able to observe and
analyze up to the LMP layer of the Bluetooth stack.
Huang et al. (2018) used loopholes in the Blue-
tooth specification pertaining to low-power mode
transitions, to perform Bluetooth DOS attacks. Partic-
ularly, they exploited loopholes in the sniff and park
mode conversions to trigger DOS conditions, using
slave devices in a piconet. They also suggested that
DOS conditions can be triggered intentionally or un-
intentionally when multiple piconets pertaining to dif-
ferent devices come closer to each other. Because de-
vices operating in different piconets use their own fre-
quencies for communication and are unaware of each
other’s hopping sequences, they may end up causing
interference which can lead to a DOS condition. To
detect DOS conditions caused by interference, they
compare quality characteristic data (i.e, bit error rate
and invalid data rate) of all the channels with the
normal thresholds in the Bluetooth specification. To
detect DOS attacks by slave devices, they compared
the proportion of data transmission time to the slave
timeslots, under normal conditions and during the ac-
tivity under observation.
Satam et al. (2018) proposed an intrusion detec-
tion system that runs on a Linux server machine and
captures the Bluetooth data frames like SCO (Syn-
chronous Connection Oriented) data frame, the HCI
protocol frame, and the HCI (Host controller inter-
face) data frame. Although the paper does not men-
tion the hardware used to sniff the Bluetooth traffic,
it appears that the authors used the machine’s stock
Bluetooth adapter because captured frames represent
packets above the HCI layer. Once captured, the
frames are converted into flows of size T seconds,
which are then converted to n-grams. These n-grams
are used to train various machine learning algorithms
to establish normal behavior and thus classify a new
observation flow as normal or abnormal. They per-
formed Bluesnarfing and power-draining attacks and
achieved precision and recall of 99.6%.
Although the aforementioned works employ vari-
ous novel techniques to detect different types of Blue-
Detecting BrakTooth Attacks
789
tooth attacks, to the best of our knowledge, there
is currently no existing research that focuses on the
detection of BrakTooth-based Bluetooth attacks us-
ing affordable hardware and machine learning tech-
niques. Our work aims to fill this gap.
3 EXPERIMENTAL SETUP
The LMP protocol is used by the link managers in
different Bluetooth devices to establish links and ex-
change information about supported features, power-
saving options, and encryption techniques. Most ex-
isting works that use inexpensive, off-the-shelf Blue-
tooth adapters could only observe packets above the
Host Controller Interface (HCI) layer, as software
running on the lower layers is not accessible to the
public (Nijholt et al., 2020). To address this, we de-
veloped the experimental procedure seen in Figures 2
and 3 to capture the LMP traffic of normal data per-
taining to benign Bluetooth communication and at-
tack data pertaining to the exploitation of BrakTooth-
based vulnerabilities. For this purpose, we used two
devices:
1. Attack device: This device was used to craft and
send BrakTooth attack packets to the victim de-
vice.
2. Victim device: This device received packets from
the attacking device as well as normal Bluetooth
packets from communication with other benign
devices.
More details about these two types of devices are
provided in the following subsections.
3.1 Attack Device
To exploit BrakTooth-based vulnerabilities, a proof-
of-concept (PoC) tool (BrakToothPoC, 2021) was re-
leased. As shown in Figure 2, this tool is installed
on a Ubuntu 18.04 host machine and is used to flash
custom firmware on an ESP32-WROVER-KIT to ex-
ecute various exploits.
3.2 Victim Device
A rooted Nexus 5 mobile device containing the
Broadcom BCM4339 Bluetooth controller is used as
a victim while a Ubuntu 22.04 host machine is used
to run the InternalBlue framework. The host machine
connects to the victim device via Android Debug
Bridge (ADB). When the victim’s Bluetooth stack is
compiled with debugging enabled, two new ports are
opened by the android device as follows:
ESP
WROVER KIT
BrakTooth
Exploiter
PoC
wddissector
plugin
Victim
Ubuntu
18.04
Attacker
Flash custom
firmware and
execute exploits
Figure 2: Setup of the attack device.
1. TCP 8872: This allows for an external agent
to read the HCI packets exchanged between the
BCM4339 controller and the Bluetooth stack.
2. TCP 8873: This can be used to send HCI com-
mands to the BCM4339 controller through the
Bluetooth stack.
As mentioned in Section 1.2, the InternalBlue
framework helps in monitoring the LMP packets re-
ceived by the victim device and it does so by mod-
ifying the BCM4339 firmware such that it relays the
LMP packets it received to the host machine over TCP
8872. The framework also uses the Bluetooth H4
Broadcom Wireshark plugin to parse LMP and base-
band packets from the incoming traffic and write to
Wireshark. A setup of the victim device can be seen
in Figure 3.
Victim Device
InternalBlue
Android
Debug
Bridge
Bluetooth H4
Broadcom
Wireshark plugin
Attack
Device
Benign
Device
Ubuntu
22.04
Figure 3: Setup of the victim device.
3.3 Dataset Collection
Both normal and attack data were extracted from the
Wireshark capture files, generated by the InternalBlue
framework running over the Nexus 5 device. To gen-
erate normal data, three mobile devices and a Win-
dows laptop were paired with the Nexus 5 device, and
they were allowed to exchange several sample docu-
ments, video, and audio files over the BT Classic pro-
SECRYPT 2023 - 20th International Conference on Security and Cryptography
790
tocol. To generate attack data, we used the PoC tool
running on an Ubuntu 18.04 host machine to execute
various exploits listed by Garbelini et al. (2021), using
the ESP-32-WROVER-KIT. The Nexus 5 device was
kept as the target. Following is a sample command to
execute exploits using the PoC tool.
sudo bin/bt_exploiter --host-port=/dev/
ttyUSB1 --target=<target bdaddress>
--exploit =<exploit name>
The captured Bluetooth traffic has the following fea-
tures:
1. Protocol: This field refers to the protocol used in
the Bluetooth packet such as L2CAP, OBEX, Ser-
vice Discovery Protocol (SDP), RFCOMM, and
others.
2. Info: This field provides additional information
about each packet depending on the type of proto-
col used.
3. Length: This field indicates the length of the
packet in bytes.
4. Delta: This field indicates the time difference be-
tween the current packet and the previous packet
in the pcap file.
5. Type: This field is a target variable representing
normal vs attack conditions and is labeled manu-
ally.
The LabelEncoder was imported from the scikit-
learn library to encode the Protocol and Length fea-
tures and a standard scaler was used to standardize all
the features. The distribution of the attack and nor-
mal data is shown in Table 2. As the number of fea-
tures is limited, in this paper, we focus on classifying
only two categories, i.e, normal and attack. All en-
tries pertaining to the vulnerabilities listed in Table 2
are mapped to just one class i.e, Attack. Both the at-
tack and normal data were manually labeled, and the
dataset has been made publicly available
1
.
4 RESULTS AND DISCUSSION
For the experimental evaluation, we split the dataset
into an 80-20 ratio to train three machine learning
models: Random Forest, K-Nearest Neighbour, and
an Artificial Neural Network. Gholamy et al. (2018)
suggested that the best results are obtained if 20-30%
of the data is used for testing and the remaining 80%
for training. As for evaluation criteria, we chose the
true positive rate (TPR), false positive rate (FPR), and
1
ISOT Laboratory, BrakTooth Attack Dataset, 2023,
https://onlineacademiccommunity.uvic.ca/isot/datasets/
Table 2: Distribution of normal and attack data.
Type of vulnerability Packets
normal 6269
attack au rand flooding 655
truncated sco link request 340
duplicated iocap 299
truncated lmp accepted 274
invalid feature page execution 250
feature response flooding 216
invalid timing accuracy 211
lmp overflow dm1 159
lmp auto rate overflow 151
duplicated encapsulated payload 111
invalid setup complete 67
Table 3: Results from applying various classifiers.
Classifier TPR FPR Precision
Random
Forest
98.33% 1.47% 99.36%
ANN 97.05% 7.51% 96.74%
k-NN 97.77% 1.28% 99.43%
precision as they collectively provide a comprehen-
sive understanding of the model’s performance in at-
tack detection. TPR captures the model’s ability to
detect actual attacks, FPR assesses the rate of false
alarms, and precision ensures the reliability of pos-
itive predictions. We obtained encouraging results,
as shown in Table 3, with the Random Forest model
performing the best, achieving a precision of 99.36%
and a true positive rate (TPR) of 98.33%. As men-
tioned in Section 1.2, there are no prior works that
deal with the problem of detecting BrakTooth-based
attacks. Therefore, we will not provide a comparative
analysis.
Furthermore, due to the lack of non-commercial
Bluetooth sniffers, in this paper, we propose an ex-
perimental procedure that utilizes an active sniffer to
capture attack traffic. However, it is worth noting that
in a real-world application, this approach may not be
entirely feasible as the attacker would need to specifi-
cally target the active sniffer for it to effectively detect
any attacks. One strategy to enhance the likelihood of
an attack is to periodically change the name of the
active sniffer (i.e, victim) to match the names of com-
mon devices detected in the Bluetooth piconet being
targeted.
5 CONCLUSION AND FUTURE
WORK
In this research, we used a two-prong approach to
detect BrakTooth-based attacks. First, we set up an
attack device using ESP32-WROVER-KIT running
a proof of concept (PoC) tool. Second, we set up
Detecting BrakTooth Attacks
791
an active sniffer to capture normal and attack traffic,
using the InternalBlue framework and a Broadcom
BCM4339 Bluetooth Controller. Finally, we used the
collected data to train various machine-learning mod-
els to classify attack data and achieved good perfor-
mance with the Random Forest model. Our inexpen-
sive and simple detection setup can also be extended
to notify various stakeholders, such as caretakers and
family members as soon as an attack is detected, thus
ensuring the safety and welfare of the senior resi-
dents in a smart home. In the future, we intend to
enhance our current low-cost setup to capture addi-
tional features such as RF signal strength, RF sig-
nal frequency offset, bit error rate, invalid data rate,
and more to model and identify a broader range of
Bluetooth attacks. Further, we also plan to integrate
Ubertooth One, to extend the existing setup to detect
SweynTooth-based (Garbelini et al., 2020) attacks on
BLE devices.
ACKNOWLEDGEMENTS
This project was supported in part by collaborative
research funding from the National Research Council
of Canada’s Aging in Place Program.
REFERENCES
Antonioli, D., Tippenhauer, N. O., and Rasmussen, K. B.
(2019). The knob is broken: Exploiting low entropy
in the encryption key negotiation of bluetooth br/edr.
In USENIX Security Symposium, pages 1047–1061.
BrakToothPoC (2021). Braktooth proof of con-
cept. https://github.com/Matheus-Garbelini/
braktooth esp32 bluetooth classic attacks.
Cominelli, M., Gringoli, F., Patras, P., Lind, M., and Noubir,
G. (2020). Even black cats cannot stay hidden in the
dark: Full-band de-anonymization of bluetooth classic
devices. In 2020 IEEE Symposium on Security and
Privacy (SP), pages 534–548.
Ditton, S., Tekeoglu, A., Bekiroglu, K., and Srinivasan, S.
(2020). A proof of concept denial of service attack
against bluetooth iot devices. In 2020 IEEE Interna-
tional Conference on Pervasive Computing and Com-
munications Workshops (PerCom Workshops), pages
1–6. IEEE.
El Jaouhari, S. and Bouvet, E. (2022). Secure firmware
over-the-air updates for iot: Survey, challenges, and
discussions. Internet of Things, 18:100508.
Ferro, E. and Potorti, F. (2005). Bluetooth and wi-fi wireless
protocols: a survey and a comparison. IEEE Wireless
Communications, 12(1):12–26.
Garbelini, M. E., Chattopadhyay, S., Bedi, V., Sun, S., and
Kurniawan, E. (2021). Braktooth: Causing havoc on
bluetooth link manager.
Garbelini, M. E., Wang, C., Chattopadhyay, S., Sun, S., and
Kurniawan, E. (2020). Sweyntooth: Unleashing may-
hem over bluetooth low energy. In Proceedings of the
2020 USENIX Conference on Usenix Annual Techni-
cal Conference, pages 911–925.
Gholamy, A., Kreinovich, V., and Kosheleva, O. (2018).
Why 70/30 or 80/20 relation between training and
testing sets: A pedagogical explanation.
Huang, Y., Hong, P., and Yu, B. (2018). Design of bluetooth
dos attacks detection and defense mechanism. In 2018
IEEE 4th International Conference on Computer and
Communications (ICCC), pages 1382–1387. IEEE.
Mantz, D., Classen, J., Schulz, M., and Hollick, M. (2019).
Internalblue-bluetooth binary patching and experi-
mentation framework. In Proceedings of the 17th An-
nual International Conference on Mobile Systems, Ap-
plications, and Services, pages 79–90.
Nijholt, T., Poll, E., and Vaandrager, F. (2020). Bluespec:
Development of an lmp state machine and a stateful
black-box br/edr lmp fuzzer.
NordicSemiconductor (2018). Device firmware up-
date process. https://infocenter.nordicsemi.
com/topic/com.nordic-.infocenter.sdk5.v15.0.0/
lib bootloader dfu process.html.
nRFSniffer (2023). Nrf sniffer for bluetooth
le. https://www.nordicsemi.com/Products/
Development-tools/nrf-sniffer-for-bluetooth-le.
OConnor, T. and Reeves, D. (2008). Bluetooth network-
based misuse detection. In 2008 Annual Computer Se-
curity Applications Conference (ACSAC), pages 377–
391. IEEE.
Ossmann, M. (2011). Ubertooth one. https://github.com/
greatscottgadgets/ubertooth.
Satam, P., Satam, S., and Hariri, S. (2018). Bluetooth in-
trusion detection system (bids). In 2018 IEEE/ACS
15th International Conference on Computer Systems
and Applications (AICCSA), pages 1–7.
Schulz, M., Wegemer, D., and Hollick, M. (2015). Nex-
mon: A cookbook for firmware modifications on
smartphones to enable monitor mode. arXiv preprint
arXiv:1601.07077.
SIG, B. (2023). Bluetooth market update. https://www.
bluetooth.com/2023-market-update/.
Wagner, F., Basran, J., and Dal Bello-Haas, V. (2012).
A review of monitoring technology for use with
older adults. Journal of geriatric physical therapy,
35(1):28–34.
Wu, J., Nan, Y., Kumar, V., Payer, M., and Xu, D. (2020).
Blueshield: Detecting spoofing attacks in bluetooth
low energy networks. In RAID, pages 397–411.
SECRYPT 2023 - 20th International Conference on Security and Cryptography
792