Content uploaded by Vasiliki Giagka
Author content
All content in this area was uploaded by Vasiliki Giagka on Aug 07, 2023
Content may be subject to copyright.
Ultrasound for Data Transfers from Deep Implants:
an Experimental Comparison Between
Binary-Frequency-Shift-Keying and On-Off-Keying
with Backscatter Modulation
Lukas Holzapfel∗, Vasiliki Giagka∗†
∗Department of System Integration and Interconnection Technologies, Fraunhofer IZM, Berlin, Germany
†Department of Microelectronics, Delft University of Technology, Delft, The Netherlands
lukas.holzapfel@izm.fraunhofer.de, v.giagka@tudelft.nl
Abstract—Implantable devices need to communicate informa-
tion to the outside world. For deep-seated miniaturized implants
ultrasound communication can be favourable. However, implants
need to operate during movement, and the selected communica-
tion scheme should be assessed accordingly. In this work, we
implemented a simple protocol to transfer data packets based
on On-Off-Keying (OOK) and Frequency-Shift-Keying (FSK) by
backscatter modulation in ultrasonic communication links for
deep implants. We then used it to compare FSK vs OOK encoding
regarding the bit error rate during continuous ultrasound power
transfer, and while moving, in a water tank setup. Our exper-
iment shows, that sub-millimeter movements can have severe
effects on OOK communication, but not for FKS. Therefore, FSK
can allow for backscatter communication from deep implants
regardless of their position and involved movements. The protocol
can also be adapted to other backscatter modulation schemes in
the future.
Index Terms—backscatter, communication, implant, modula-
tion, movement, ultrasound, wireless
I. INTRODUCTION
In the recent years ultrasound has been researched as an
alternative power source for miniaturized implants [1]–[3].
Ultrasound power transfer is particularly attractive for deep-
seated implants, as those envisioned in bioelectronic medicine
applications [4]. Bioelectronic medicine interacts with the pe-
ripheral nervous system using miniaturized neural interfaces,
which can monitor or stimulate neural activity. These devices
often need to communicate information regarding, for instance,
stimulation efficacy, elicited neural signals [5][6] or parameters
that reveal the integrity of the implant’s encapsulation [7], [8].
In such cases a parallel ultrasonic communication channel is
a natural solution [9]–[11]. On-Off Keying (OOK) is the most
commonly used modulation protocol, because of the ease of
its hardware implementation [12]. The carrier signal has to be
simply turned on or off, depending on the current bit value,
as shown in Fig. 1 a).
To further simplify the implant’s circuitry and minimize its
size and power demands, backscatter modulation is used [13].
This work is part of the Moore4Medical project funded by the ECSEL Joint
Undertaking under grant number H2020-ECSEL-2019-IA-876190.
BFSK
OOKa)
b)
Fig. 1. Block diagrams comparing a) OOK and b) the proposed implemen-
tation for binary FSK. For OOK the sinusoidal carrier with frequency fcis
turned on or off depending on the current bit value b[k]. For FSK the carrier is
multiplied by two square wave signals with frequencies f0and f1. Depending
on the current bit, one of the two modulated signals is selected.
For ultrasonic backscatter modulation, the implant modulates
the electrical load connected to its power receiving transducer.
Changing the electrical load, in turn, modulates the amount
of reflected power, which can be detected by the external
device. In its simplest form, load modulation is implemented
by a single switch that briefly shorts the transducer. However,
interference from other -possibly larger and closer- scatterers,
like tissue-bone or tissue-gas interfaces, must be assumed to be
present. The scattering from these reflectors cannot be assumed
static and therefore the amplitude and phase of the interference
must be assumed to change over time[12]. This makes the
reception of ultrasonic backscatter signals based on amplitude
modulation inherently difficult. We therefore proposed to
combine backscatter modulation with two digital square wave
modulation signals of different frequencies fmod,0and fmod,1,
to generate sidebands, thus effectively implementing binary
Frequency-Shift Keying (FSK) [14], as shown in Fig. 1 b).
The following work describes a protocol to test the robustness
of FSK against interference from other scattering compared to
OOK and gives an overview of the results of those tests. To the
best of our knowledge, movement has not been investigated
in ultrasonic communication links so far.
II. ME TH OD S
The comparison of the modulation schemes was performed
in a water tank setup, shown in Fig. 2 a). The following
sections first give an overview of the experiment setup, and
then discuss important aspects in more detail.
A. Transducer arrangement
For the experiment three transducers where placed in a
water tank as shown in Fig. 2 b). As the external TX trans-
ducer, we used an unfocused commercial 1.27 cm submersible
transducer with a resonant frequency of 5 MHz (V309-SU,
Olympus). For the transducer at the implant side, we used one
channel of a pre-charged capacitive micromachined ultrasound
transducer (CMUT)-array, described in [15]. We placed the
CMUT at a distance of approximately 11 cm on-axis of the
externally driven transducer, because it gave us the maximum
voltage across the CMUT. The CMUT was then connected to
our prototyping platform. The backscattered ultrasound was
picked up by a 1 mm diameter needle hydrophone (NH1000,
Precision Acoustics) placed close to the TX transducer. Dur-
ing data transmission the CMUT was moved laterally out of
the focal region at a speed of ≈18 mm s−1to emulate a non-
static environment.
B. Driving the external TX transducer
To emulate the external powering device, the external
transducer was driven by a waveform generator (33622A,
Keysight), set to a nominal output voltage of 10 Vpp with a
50 Ω load. The frequency was chosen to be close to the CMUT
resonant frequency of 3.5 MHz. As the reso nant frequency of
the TX transducer is 5 MHz, the efficiency of the transducer
is decreased. However, since the focus of this work is not to
maximize the power link efficiency, but to compare modulation
schemes under the same conditions, this is not detrimental to
the experimental results.
C. Modulating the backscattered ultrasound
The CMUT was connected to a prototyping platform we
developed specifically for the evaluation of backscatter mod-
ulation schemes that can be implemented by a single load
switch. Fig. 3 shows a block diagram of this platform. It
is based on a STM32G474 microcontroller, and three Ana-
log Devices ADG1401BRMZ analog switches. Each switch
is controlled by one channel of the microcontroller’s high
resolution timer (HRTIM). For this experiment, however, only
a single timer channel and analog switch was connected to
the CMUT. The microcontroller is connected via USB to a
PC for power supply and can be controlled using a text based
protocol over a virtual COM port. Each of the three switching
channels provides the possibility to populate an impedance
a)
Oscilloscope
CH1CH2
inout
Function Gen. DC coupler
TX tranducer CMUT array
Prototyping
platform
needle
hydrophone
USB
b)
TX transducerhydrophone modulating CMUT
Fig. 2. The experimental setup used for the comparison of the two modulation
schemes. a) Block diagram of the involved components. b) Arrangement of
the transducers in the water tank.
Channel 3
Channel 2
Power & Control
via USB
Reference Clock
STM32G474
Channel 1
Matching
Network
Ultrasound
Transducer
Analog
Switch
switched
load
Fig. 3. Prototyping platform used to modulate the transducer load.
matching network and set the value of the switched load. For
this experiment, the switched load was set to nominally 0 Ω.
After measuring the impedance of the CMUT in water, we
adapted the impedance matching network accordingly. The
experiment script, written in Python and running on the host
PC, configures the prototyping platform to use OOK or FSK,
and initiates the transfer of a packet of pseudo-random data.
The packet starts with a 31-bit m-sequence preamble, followed
by a payload, corresponding to a 1 s transfer duration. That
is, for a symbol rate of 1 kBd, the payload size was set to
1 kbit. Approximately 200 ms after the transfer is initiated,
the moving stage starts to move the CMUT.
D. Acquiring the backscattered signal
As the needle hydrophone has an internal preamplifier, it is
powered by a DC coupler, which passes through the picked
up AC signal to an oscilloscope (RTA4004, Rohde&Schwarz).
The sample rate of the oscilloscope is set to 83.¯
3 MHz and
the memory depth to 100 million samples. This equates to a
duration of 1.2 s per acquisition. The oscilloscope is triggered
by the prototyping platform, when the transfer is initiated. The
signal is then transferred to a host PC, where it is processed
using the receiver implementation described in section II-E.
E. Detection of the backscattered signal
The acquired signal is then processed immediately by the re-
ceiver shown in Fig. 4, which is implemented in Python, lever-
aging the Numpy and Numba packages, to achieve acceptable
processing speeds. Due to the amount of data (400 MB per
goertzel
goertzel
goertzel
goertzel
TED
....
....
loop filter
packet
sync
slicer
parallel
to
serial
Demodulate Synchronize Detect
Fig. 4. Block diagram of the receiver implementation. The input signal y(k)
is demodulated in parallel for eight different phases, with a distance of 1
8
of the Symbol length Nbetween each phase. The parallel signals are then
converted back into a single demodulated signal Y(m). This signal is then
used for packet and symbol time synchronization, and binary slicing, resulting
in the estimated bit values ˆ
b(n).
acquisition), we chose not to save the raw signal, but only the
demodulated signal, loop filter output, and detected data. This
way we can rerun the processing for all processing blocks
working on the demodulated signal, which speeds up finding
suitable parameter values for these processing blocks.
To demodulate the signal, we use the Goertzel algorithm,
which we combine with windowing. In case of OOK, only
the carrier frequency is demodulated. For FSK the first lower
and upper modulation product for each of the two modulation
frequencies -so in total four frequencies- are demodulated.
To get a bipolar signal, the sum of the two modulation
products for a binary 0, Y0(m) = Y(f0,l, m) + Y(f0,u , m),
are subtracted from the sum of the two products of a binary
1, Y1(m) = Y(f1,l, m) + Y(f1,u , m):
Y(m) = Y1(m)−Y0(m)(1)
We use eight parallel running demodulation filters with inter-
leaved phases, to be able to synchronize further processing to
the symbol timing later. The rate of the demodulated signal
Y(m)therefore is 8
Nthe rate of the incoming signal y(k),
with Nbeing the number of samples acquired per symbol. The
demodulated signal is fed into a Gardner timing error detector
(TED), a preamble detector, and a slicer. The TED output is
fed into a proportional-integral (PI) loop filter, which tells the
slicer which sample to use for the next decision. Finding the
packet start is based on the correlation of the signal to the
preamble. Only if a packet preamble is detected, the output of
the slicer is saved.
III. RES ULTS
Fig. 5 compares a demodulated bipolar signal for OOK and
FSK. Both transfers, were performed for the same conditions
Fig. 5. Comparison of backscatter communication during movement using a)
OOK and b) binary FSK. Amplitude and phase changes of the interference
cause bit errors for OOK (insets ii and iii, respectively), while no bit errors
occur for the same scenario using FSK. The amplitude changes of the
interference are much higher than the amplitude of the useful signal for OOK.
Note, that the offset at the beginning of the signal has been subtracted, which
is why negative amplitudes seem to occur also for OOK.
involving movement. Fig. 5 a) shows the result for an OOK
transfer. Before the movement starts, the modulation scheme
performs well and no bit errors occur, as highlighted with
inset i). However, during movement, rapid amplitude changes
make it difficult for the slicer to adapt the decision boundary.
Also, the phase difference between the interference and the
useful signal cause intermittent cancellation and inversion of
the useful signal. Both of these issues cause a high bit error
rate for the OOK transfer.
Fig. 5 b) shows the result for an FSK transfer. The signal
shows amplitude changes over time due to fading. However,
these are not critical for the detection and no bit errors occur.
IV. DISCUSSION
The results clearly show the advantages of FSK over
OOK for backscatter communication. A communication link
through tissue (phantoms) would have accounted for attenua-
tion and more complex scattering. But, the underlying issues of
amplitude-based modulation can be sufficiently demonstrated
in a motorized water tank setup. Further experiments are
needed to compare these and other schemes, like Phase-
Shift Keying by load modulation. Also transfers for different
scenarios, for example with different positions, velocities and
data rates.
V. CONCLUSION
Ultrasonic backscatter communication of deep implants
suffers from non-static interference by other scatterers. We
implemented a platform to benchmark backscatter communi-
cation schemes involving non-static interference and used it
to compare FSK and OOK. The results show that amplitude-
based modulation schemes, like OOK, can have high bit error
rates due to the interference.
REFERENCES
[1] F. Mazzilli, C. Lafon, and C. Dehollain, “A 10.5 cm
ultrasound link for deep implanted medical devices,”
IEEE Transactions on Biomedical Circuits and Systems,
vol. 8, no. 5, pp. 738–750, Oct. 2014, IS SN: 1940-9990.
DOI: 10.1109/TBCAS.2013.2295403.
[2] L. Tacchetti, W. A. Serdijn, and V. Giagka, “An ultra-
sonically powered and controlled ultra-high-frequency
biphasic electrical neurostimulator,” in 2018 IEEE
Biomedical Circuits and Systems Conference (BioCAS),
Oct. 2018, pp. 1–4. DO I: 10 . 1109 / BIOCAS . 2018 .
8584718.
[3] Z. Kashani and M. Kiani, “A study on ultrasonic wire-
less power transfer with phased array for biomedical
implants,” IEEE Transactions on Biomedical Circuits
and Systems, pp. 1–11, 2023, IS SN : 1940-9990. DO I:
10.1109/TBCAS.2023.3282197.
[4] V. Giagka and W. A. Serdijn, “Realizing flexible bio-
electronic medicines for accessing the peripheral nerves
– technology considerations,” Bioelectronic Medicine,
vol. 4, no. 1, p. 8, Jun. 26, 2018, IS SN: 2332-8886.
DOI: 10.1186/s42234-018-0010- y. [Online]. Available:
https://doi.org/10.1186/s42234-018-0010-y (visited on
08/05/2022).
[5] R. R. Harrison, “The Design of Integrated Circuits
to Observe Brain Activity,” Proceedings of the IEEE,
vol. 96, no. 7, pp. 1203–1216, 2008, IS SN: 15582256.
DOI: 10.1109/JPROC.2008.922581.
[6] Y. Liu, A. Urso, R. M. Da Ponte, et al., “Bidirec-
tional bioelectronic interfaces: System design and cir-
cuit implications,” IEEE Solid-State Circuits Magazine,
vol. 12, no. 2, pp. 30–46, 2020. DOI: 10.1109/MSSC.
2020.2987506.
[7] C. O. Akgun, K. Nanbakhsh, V. Giagka, and W. A.
Serdijn, “A chip integrity monitor for evaluating mois-
ture/ion ingress in mm-sized single-chip implants,”
IEEE Transaction Biomed. Circuits Syst., vol. 14, no. 4,
pp. 658–670, Aug. 2020. DOI: 10.1109/TBCAS.2020.
3007484.
[8] G. Rodrigues, M. Neca, J. Silva, et al., “Towards a
wireless system that can monitor the encapsulation
of mm-sized active implants in vivo for bioelectronic
medicine,” in 2021 IEEE Conference on Neural En-
gineering (NER), May 2021, pp. 1–4. DOI: 10 . 1109 /
NER49283.2021.9441398.
[9] M. M. Ghanbari, D. K. Piech, K. Shen, et al., “A sub-
mm 3 ultrasonic free-floating implant for multi-mote
neural recording,” IEEE Journal of Solid-State Circuits,
vol. 54, no. 11, pp. 3017–3030, 2019. DO I: 10.1109/
JSSC.2019.2936303.
[10] M. J. Weber, Y. Yoshihara, A. Sawaby, et al., “A high-
precision 36 mm3 programmable implantable pres-
sure sensor with fully ultrasonic power-up and data
link,” in 2017 Symposium on VLSI Circuits, Jun. 2017,
pp. C104–C105. DO I: 10.23919/VLSIC.2017.8008564.
[11] S. Kawasaki, I. Subramaniam, M. Saccher, and R.
Dekker, “A microwatt telemetry protocol for targeting
deep implants,” in 2021 IEEE International Ultrasonics
Symposium (IUS), IEEE, 2021, pp. 1–4. DOI: 10.1109/
IUS52206.2021.9593603.
[12] B. Jaafar, J. Neasham, and P. Degenaar, “What ul-
trasound can and cannot do in implantable medical
device communications,” IEEE Reviews in Biomedical
Engineering, vol. 16, pp. 357–370, 2023, IS SN: 1941-
1189. DO I: 10.1109/RBME.2021.3080087.
[13] M. M. Ghanbari and R. Muller, “Optimizing volumetric
efficiency and backscatter communication in biosensing
ultrasonic implants,” IEEE Transactions on Biomedical
Circuits and Systems, vol. 14, no. 6, pp. 1381–1392,
Dec. 2020, IS SN : 1940-9990. DO I: 10 . 1109 / TBCAS.
2020.3033488.
[14] L. Holzapfel and V. Giagka, “A robust backscatter mod-
ulation scheme for uninterrupted ultrasonic powering
and back-communication of deep implants,” biorxiv,
preprint, Aug. 25, 2022. DOI: 10 . 1101 / 2022 . 08 . 23 .
503752. (visited on 07/28/2023).
[15] S. Kawasaki, Y. Westhoek, I. Subramaniam, M. Sac-
cher, and R. Dekker, “Pre-charged collapse-mode ca-
pacitive micromachined ultrasonic transducer (CMUT)
for broadband ultrasound power transfer,” in 2021
IEEE Wireless Power Transfer Conference (WPTC),
Jun. 2021, pp. 1–4. DOI: 10. 1109/WPTC51349 .2021.
9458104.
