Communication

Eyelid Dynamics Characterization with 120 GHz mmW Radar

Dominik Patscheider 1 , Ruochen Wu 2 , Antoni Broquetas 2, * , Albert Aguasca 2 and Jordi Romeu 2

1 Department of Electrical Engineering and Information Technology (ETIT), Karlsruhe Institute

of Technology (KIT), Campus Süd, Engesserstraße 13, 76131 Karlsruhe, Germany;
[email protected]
2 CommSensLab, Department of Signal Theory and Communications, Universitat Politècnica
de Catalunya (UPC), Campus Nord, Carrer de Jordi Girona, 1–3, 08034 Barcelona, Spain;
[email protected] (R.W.); [email protected] (A.A.); [email protected] (J.R.)
* Correspondence: [email protected]

Abstract: This paper presents a new approach to measuring eyelid movement using millimeter
wave (mmW) radar technology. A two-step method is proposed, involving the observation of a
small resolution cell corresponding to the monitored eye and the evaluation of the phase evolution
over the measurement period. Simulations are conducted to support radar system optimization
and data interpretation with a focus on detecting eyelid movement patterns and compensating for
interference from other parts of the body. The feasibility of using this method with eyeglasses is also
explored. The proposed technique’s advantages and limitations are discussed in comparison with
existing measurement alternatives. The characteristics of eyelid dynamics, including blink frequency,
regularity, duration, and velocity can be used to assess neurological conditions and driver drowsiness.

Keywords: millimeter wave radar; eyelid detection; eye activity; stress; radar beam

1. Introduction
Millimeter wave radar (mmW radar) technology, which uses electromagnetic waves
with wavelengths in the millimeter range, has a number of applications, including esti-
mation of the range and velocity of objects [1–3], beam forming for location detection [4],
Citation: Patscheider, D.; Wu, R.; meteorology [5], geodesy [6], air traffic control [7], and seafaring [8]. In this paper, the
Broquetas, A.; Aguasca, A.; Romeu, J. use of mmW radar for biomedical applications, specifically measuring eye blink rate, is
Eyelid Dynamics Characterization explored. The ability to remotely observe the micrometric deformation of objects is utilized
with 120 GHz mmW Radar. Sensors
to investigate the characterization of eye blinking dynamics as a potential tool for detect-
2024, 24, 7464. https://doi.org/
ing neurological disorders [9]. This research includes optimization of the radar system,
10.3390/s24237464
electromagnetic modeling, and experimental evaluation. Eye blinking is a reflex response
Academic Editor: Guangcai Sun triggered by a corneal stimulus that helps keep the cornea clean and healthy by stimulating
the secretion of mucin and meibum, which are components of the tear film [10]. Studies
Received: 15 October 2024
have categorized eye blinking as spontaneous, reflex, and voluntary [11]. Factors such as
Revised: 13 November 2024
stress can affect blink rate [12] and can be used to determine one’s fitness for tasks such
Accepted: 15 November 2024
Published: 22 November 2024
as driving. In order to accurately measure eyelid movement and gain sufficient data to
understand the dynamics of blinking, this research will employ a two-step approach using
Frequency-Modulated Continuous Wave (FMCW) radar. This method aims to overcome
the limitations of previous techniques, such as high-speed cameras and magnetic coils,
Copyright: © 2024 by the authors. which have difficulty in certain lighting conditions and require significant computational
Licensee MDPI, Basel, Switzerland. effort to extract the parameters of interest. For a successful evaluation of blinking with
This article is an open access article regard to specifying neurological disorders, it is crucial to detect the inter-blink interval
distributed under the terms and (IBI), the duration of a blink, and the dynamics of a blink (i.e., the duration and speed of
conditions of the Creative Commons opening and closing the eyelids and the duration of the eye being closed).
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).

Sensors 2024, 24, 7464. https://doi.org/10.3390/s24237464 https://www.mdpi.com/journal/sensors

Sensors 2024, 24, 7464 2 of 13

2. Theoretical Background
A RADAR (RAdio Detection And Ranging system) uses radio waves to detect and
measure the distance and velocity of objects [13,14]. In order to achieve accurate measure-
ments, the radar’s beamwidth and distance resolution must be considered. To achieve
the required micrometric displacement precision using the FMCW radar, the spectrum of
the radar receiver output at the baseband is used to obtain a high-range resolution and
the target coarse range estimation. From a sequence of periodic observations, the phase
evolution of the target resolution cell provides a very precise submillimetric range, allowing
us to estimate eyelid dynamics.
The FMCW radar system typically consists of a waveform generator that produces a
voltage to control the frequency of the transmitted signal, which is usually a sawtooth or
triangle frequency deviation. The transmitted saw tooth signal can be expressed as

s(t) = A cos(2π f 0 t + θ (t) + ϕ0 ), (1)

with the amplitude A, the center frequency of the bandpass signal f 0 , the quadratic phase
θ (t) producing a linear frequency sweep ∆ f , and ϕ0 as the initial phase. For a positive or
negative linear FM slope, the phase is

θ (t) = ±πµt2 , (2)

∆f
with µ = TM being the modulation slope within − T2M ≤ t ≤ TM
2 . For a static target, the
echo signal received is

sr (t) = αA cos[ω0 t − ω0 τ + θ (t − τ ) + ϕ0 ], (3)

with a signal-level drop α and a target delay τ.

Mixing the signal received with the signal transmitted using a single mixer, or, equiva-
lently, the In-Phase output of an I/Q mixer, results in the beating tone sb (t)with a beating fre-
quency f b (t), which is the difference between the transmitted signal and echo-instantaneous
frequencies

1 d 1 d
f b (t) = f (t) − f r (t) = [ θ (t)] − [ θ (t − τ )]. (4)
2π dt 2π dt
The resulting beating tone signal is

αA2
sb ( t ) = cos[ω0 τ + 2πµτt − πµτ 2 ], (5)
2
2R
the delay for a target located at a distance (range) R is τ = c0 . The terms consists of the
following:
• ω0 τ: The phase shift due to the target distance
• 2πµτ: The beating frequency that can be expressed as

2∆ f R
f b = µτ = (6)
TM c
• πµτ 2 : The residual video phase, which can be neglected for short distances, if πµτ 2 ≪
ω0
ω0 τ. For short-range applications of FMCW radars, τ ≪ πµ , and the residual video
phase term can be neglected.
For every FM sweep, the beating tone (5) at the radar output is digitized using
an analog-to-digital converter (ADC). Since the beating frequency for a single target is
proportional to R, for an arbitrary scene containing multiple targets, the mixer output will
provide a spectrum of different tones with frequencies proportional to the different target
Sensors 2024, 24, 7464 3 of 13

ranges. Therefore, a Fourier transform can be used to isolate and distinguish different
targets with a frequency resolution of T1M and a range resolution of ∆R = 2∆c f . This is
carried out with a Discrete Fourier Transform (DFT) of the digitized signal as follows:

NSmp −1
1 2πkn
Sb ( k + 1 ) =
NSmp ∑ sb (n + 1) · exp(−j
NSmp
), (7)
n =0

To increase the number of samples at the DFT output, the beating tone signal provided
by the radar can be extended with zeros before the DFT (zero-padding). The DFT is usually
implemented with a Fast Fourier Transform (FFT) algorithm for numerical efficiency
reasons [15]. Figure 1 shows a diagram of the signal processing of the applied algorithm,
including the phase processing described in Section 5.
The range cell of interest corresponding to the observed body spot is identified and
resolved in the frequency domain, and its complex value is extracted and stored for every
radar sweep. The micromotion of the body can be ascertained from the changing phase
ϕb (k) ∈ [0; 2π ] of the range cell of interest Sb (k) = abs(Sb (k)) · exp(j · ϕb (k )) along the
radar observations. The time between consecutive chirps Tobs is determined by the design
of the waveform generator and the bandwidth ∆ f of the radar signal.

Radar Yes Δφ ≥ π No
? Face
Eyelid dominant
sb(t,lsweep) echo
dominant
echo
ADC Estimation of
sb(n,lsweep) Face
Component
FFT
Sb(k,lsweep) MTI Face
Substraction
Eye sample k
φ(lsweep) φclean(lsweep)
search
Sb(lsweep)
Blink
Phase timing and
unwrapping dynamics

Figure 1. Signal processing algorithm.

2.1. Radar Parameters

The radar used in this study, shown in Figure 2, was developed in our research labora-
tory. The system uses a System-on-a-Chip (SoC) 120 GHz radar front-end (TRX_120_001)
from [16] feeding a plastic lens antenna. As shown in Figure 3, the instantaneous frequency
transmitted by the radar is provided by a Direct Digital Synthesizer (DDS) based on a
1 Gs/s 14-bit DDS chip, programmed as a sawtooth linear FM generator. The local oscillator
of the TRX_120_001 is phase-locked to the DDS output, generating a reference submultiple
frequency of 1/(64 × 40) the transmitted frequency in the 120 GHz band. The target range
is determined from the frequency peak in the beating tone spectrum. After applying the
FFT to the echo signal, and inverting (6), we obtain
Sensors 2024, 24, 7464 4 of 13

c T c
R= τ = M fb (8)
2 2∆ f

with f b = µτ. For a micromotion measurement we evaluate the phase evolution of the
range cell of interest containing the target. With ϕb = −ω0 τ = −2kR, we obtain

λ
R= ϕ , (9)
4π b
where ϕb is the phase of the beating tone of a backscattering point, and λ = fc is wavelength
0
corresponding to the transmitted frequency. After detecting the target in a k frequency
domain sample corresponding to the observed body range R resolution cell, the phase and
amplitude of this target are evaluated over the entire measurement of l chirps (FM sweeps).
One single chirp consists of k = 0, . . . , NSmp samples. The range determination based on
the frequency of the beating tone is not ambiguous if the sampling frequency of the ADC
fulfils the Nyquist sampling frequency corresponding to the maximum expected range.
The range resolution of a radar operating with a bandwidth ∆ f is given by
c
∆R = . (10)
2∆ f

(a) (b)

Figure 2. The radar sensor module. The radar front-end is a TRX_120_01 RFE chip mounted onto an
FR4 Printed circuit board. (a) Circuit board of the radar. (b) Prototype of the 120 GHz FMCW radar.

Since the eyelid motion is small compared to the radar range cell, eyelid monitoring
is based on phase rather than range resolution. The cyclic characteristic of the phase
introduces a maximum unambiguous phase-based change in range as follows:

λ
Run = . (11)
2
For this reason, if the range changes exceed Run , the phase must be unwrapped to
obtain the correct displacement.
Table 1 shows the radar parameters used for this study.

Table 1. System parameter for simulations and measurements.

Parameter Calculation Value

Center frequency f0 122.5 GHz
Bandwidth ∆f 1–3 GHz
Sweep time TM 1.5 ms
∆f
Slope µ = TM (1–3) GHz/1.5 ms
Sweep repetition period SRP 3 ms
Sensors 2024, 24, 7464 5 of 13

TRX_120_001
VCO
PLL
×2
𝑇𝑇𝑋𝑋

𝑅𝑅𝑋𝑋
÷ 40 1/32

DDS

ADC PC

Figure 3. Overview of the radar system. The DDS generates a frequency signal ranging from 47.65 to
48.05 MHz, which is frequency-multiplied through a phase-locked loop (PLL). A voltage-controlled
oscillator (VCO) subsequently produces a 61 to 61.5 GHz signal for further frequency multiplication
to reach a range of 122 to 123 GHz, which is used as the transmission frequency for observing the
target. The echo signal, reflected from the eyelid, is captured by the receiving antenna R X , which is
then down-converted in a mixer using the transmission signal as a reference. The resulting baseband
signal is filtered and amplified, effectively removing high-frequency noise to enhance the signal
quality. The received signal is digitized with a 12-bit ADC for further processing and calculation of
the eyelid dynamics parameters.

2.2. Assessment of the Radar’s Safety

The IEEE, in [17], defines exposure limits for electric, magnetic, and electromagnetic
fields within 0 Hz to 300 GHz. It is important to respect these safety limits to minimize
health effects and risks during measurements and interference with other radio frequency-
based equipment. In this study, the radar specifications and lens characteristics allow us to
estimate a radiated power of −4.6 dBm (0.35 mW) and an illuminated spot diameter of 3.5
cm with a −3 dB contour at a typical operation range of 1 m. The resulting illumination
power density is 36 µW/cm2 , well below the 2.3 mW/cm2 ERL of local exposure limits at
120 GHz recommended in [17] for people in unrestricted environments, which is the most
stringent limit. For this reason, the proposed technique can be considered to be of minimal
health risk, with exposure limits well below the exposure limit recommendations.

3. Simulations
Electromagnetic simulations on a simplified dielectric model of a human eye were
carried out to predict and interpret the radar backscattering experimental results. Figure 4
shows the model used in the evaluation of the radar backscattering. The eyeball was
modeled as a dielectric sphere with a radius of 13 mm, a relative permittivity ε r = 10,
and a loss tangent tan δ = 0.8. A homogeneous permittivity sphere was assumed since
the cornea, sclera, and iris have similar values of permittivity. The model values were
obtained from the Foundation for Research on Information Technologies in Society (IT’IS),
which provided the tissue properties of the eye components [18]. The lower and upper
eyelids covering the eyeball were modeled with dielectric layers of a 2 and 3 mm thickness,
respectively [19], with a relative permittivity of ε r = 8 and tan δ = 0.647. The simulations
were carried out using CST Microwave Studio.
To capture the eyelid edge from different viewing angles, the upper eyelid used in the
simulations has a thickness of 3 mm. In Figure 5a, a plane wave is illuminating the eye
 Sensors 2024, 24, 7464 6 of 13

il 9, 2023 submitted to Journal Not Specified surface with the propagation vector directed horizontally towards it. The incidence
4 of 12angle
θplanewave is also visualized in Figure 4b. The upper eyelid has a 30◦ angle with respect to
the vertical axis in this figure. In subsequent simulations, the eyelid closes until it reaches
180◦ . The strong attenuation caused by the eye with a closed eyelid on the incoming field
is shown in Figure 5b. System Parameter
Consequently, the penetration depth of the eye tissues is very
Parameter
small, and the eye components Calculation
behind the cornea haveValue
a negligible influence on the radar
Centermeasurement.
frequency Figure 6 displays f 0 the simulated phase and 120radar
GHzcross-section (RCS) during
the
Bandwidth eyelid closure progression, ∆fexpressed as the eyelid angle
1 − 3.GHz
A noticeable oscillation in the
RCS is attributed to the superposition of changing phase reflections from the eye and eyelid,
Sweep time TM 1.5 ms ◦
dependent on the angle of the eyelid’s ∆ f closure. The phase rotates more than 360 with an
Slope eye closure angle of 55◦ to 85◦µ, with (1 − 3from
= TMthe transition of rotation ) GHz/1.5 ms
+π to −π shown by the
Pulse Repetition
vertical blackFrequency
lines. AccordingPRFto (9), a phase difference3∆ϕ
msb = 4 × 2π − 2.2 corresponds to
Table 1. Systemaparameter
relative change in the distance
for Simulations of the object from the radar of ∆R = 4.5 mm. When the
and Measurements.
◦
closure angle is higher than 90 , no significant backscattering changes are observed since
the frontal area of the eye contributing to backscattering is stable.

7
6
7
6

θplanewave

3 4 5 2

3 4 5
(a) 2
(b)

Figure 4. Approximated model of the anterior (b) Front

partview
of theof the(a)eye
eye. model with
Components of the eye model:
(a) Components of the eye model.
beam direction.
(1) Upper eyelid, which will be closed in the simulation; (2) lower eyelid; (3) cornea; (4) anterior
Figure 2. Approximated model
chamber; (5) sclera;of(6)
the anterior
iris; and (7) part of (b)
pupils. theFront
eye. view
(1) upper
of theeyelid, which will be closed
eye model.
in the simulation, (2) downer eyelid, (3) cornea, (4) anterior chamber, (5) sclera, (6) iris, (7) pupils.

Thus exploring the eye and its lid by the radar needs a higher precision by using the phase 83

of the beating tone, which is the phase difference of the received and send radar signal. The 84

unambigous range of it is half the wavelength 85

λ
Run = . (11)
2
Table 1 shows the parameter used for this study. The bandwidth for an approximate 86
estimation of the target is changed for a comparison of the significance of the bandwidth. 87
For general applications the radar operation should be restricted to ISM f = 122 GHz band 88
(a) (b)
with BISM = 1 GHz bandwidth. However, in controlled radio diagnostic environments the 89
radar can be operated outside
Figure 5. Eye with lidISM restrictions
model using
in Microwave suitable
Studio. shielding
(a) A horizontal and
plane prevention
wave of 90
is directed towards
EMI/EMC problems.
the eye. (b)With a higher
Electric sweepinbandwidth
field amplitude we are able to make the resolution 91
the eye model.
cell smaller to isolate other parts of the human body, which we consider afterwards for 92
phase evaluation. 93

2.2. IEEE Standard for Safety Levels for Human Exposure 94

IEEE in [3] defines exposure limits for electric, magnetic, and electromagnetic fields 95

within 0 Hz to 300 GHz. It is important to maintain limits for avoiding adverse health effects 96

during measurements and further more, make the system work properly with allowed 97

transmitted power. In this study with a maximum transmitting power of Pt = 1 dBm of 98

our radar and an antenna footprint on the head with a radius of r = 5.1 cm it maintains the 99

safety levels. 100

3. Simulation 101
Sensors 2024, 24, 7464 7 of 13

Simulation of eyelid movement, planewave θplanewave = 0◦

3 -20

2 -22

1 -24

Phase of farfield in rad

0 -26

RCS in dBsm
-1 -28

-2 -30

-3 -32
20 40 60 80 100 120 140 160 180

Closed Eyelid in degrees

Figure 6. Phase (blue line) and RCS (green line) of the backscattering simulation results with respect
to the eyelid closure angle from open (20◦ ) to complete closure (180◦ ).

4. Experimental Validation
The lens in front of the radar TRX_120_01 SoC produces a narrow spot about 3.5 cm
in diameter at a radar distance of 1 m. The ranging radar’s performance was validated
with a small 2 cm side corner reflector controlled with a micrometric positioner. The
theoretical parameters of the range resolution and phase/range sensitivity were confirmed
experimentally. With a displacement in phase sensitivity of 3.4 µm per degree, the capability
to determine the range with submillimetric precision has been validated.
The pattern of the lens antenna was measured using a static small corner surrounded
by absorbing material as the reference target and scanning the radar antenna both in
azimuth and elevation with a dual-axis angular positioner. Figure 7 shows the antenna
pattern in Cartesian coordinates. The 3 dB angle is approximately θbeam,3dB = 2◦ in the
horizontal and vertical direction. The side lobes are suppressed to −33 dB.

0 0

-5 -5

-10 -10

-15 -15

-20 -20
Magn in dB

Magn in dB

-25 -25

-30 -30

-35 -35

-40 -40

-45 -45

-50 -50
-10 -8 -6 -4 -2 0 2 4 6 8 10 -10 -8 -6 -4 -2 0 2 4 6 8 10
Azimuth angle in degree Elevation angle in degree

(a) (b)

Figure 7. Azimuth and elevation cuts of the antenna radiation pattern. (a) Azimuth cut. (b) Elevation
cut.

5. Eyelid Results
5.1. Radar Pointing Horizontally
Taking into account the ADC board’s memory limitations, the radar was configured to
acquire echo beating tone signals along 8000 consecutive chirps. With a chirp repetition
period of 3 ms, the total acquisition time per observation was 24 s, which is enough
to evaluate the performance of the proposed eye blink detection and eyelid dynamics
characterization.
The initial eyelid dynamics measurements were conducted with the radar beam
aligned horizontally with the eye axis. Precise alignment of the radar was necessary due
Sensors 2024, 24, 7464 8 of 13

to the narrow beam of the antenna. A small corner reflector was used to confirm the
correct pointing of the radar. Figure 8 shows the magnitude and phase of the resolution
cell backscattering evolution corresponding to the observed eye. Several blink events
can be identified as short and fast transient oscillations, which are present in both the
magnitudes and the phase of the observed backscattering. The phase’s slow negative slope
trend reflects a small movement made by the subject head changing the target-radar range
along the measurement window of 24 s. Phase unwrapping is needed to correctly recover
the target motion to avoid the phase jumps observed close to the ±π phase limits. After
phase unwrapping, the value of the phase change peaks can be used to detect the blink
events and determine the average blink period and the period dispersion or irregularity.

Phase evolution of object @ 0.47721 m

4

30
2

20

Magnitude in dB
Phase in rad

0
10

-2 0

-10
-4
12 14 16 18 20 22 24
tobsin s

Figure 8. Phase (blue line) and magnitude (red line) of the eye spot backscattering. The radar is
located at a 1 m distance in front of the subject’s face, observing the eye area with a horizontal beam.
Several blinks can be observed, producing fast magnitude and phase oscillatory transients.

Figure 9a shows the magnitude and phase of one of the blinks. The ripple in mag-
nitude indicates an interference pattern between two or more scattering centers in the
same resolution cell with different phase rotation paces. To reduce the impact of the face
interference on the eyelid radar observations, we have used a two-scattering-center model.
One of the scatterers is the fast-moving eyelid, with a fast phase rotation linked to blink
closure and aperture. The second scatterer models the rest of the observed face, with
slower magnitude and phase dynamics. Under this model, every transition between the
local maxima and minima corresponds to a half-cycle relative phase turn (180 degrees)
in the eyelid phase with respect to the unwanted background of the face. The fact that
the phase in the upper part in Figure 9a rotates only with fractions of a cycle indicates
that the scattering of the face is dominant and the scattering of the eyelid has a smaller
magnitude. Exploiting this model, the face contribution to the radar echo signal can be
obtained from the evolution of maximum and minimum magnitude. For example, in both
the maxima and minima samples, the phase of the observation corresponds to the phase
of the dominant face background. The magnitude of the unwanted background can be
obtained from the average value of consecutive maxima and minima magnitudes. This
information was used to estimate the background face phasor and its evolution along each
blink event. The lower part in Figure 9b shows the same blink event after subtracting the
estimated contribution of the face background. This technique can be considered an adap-
tive extension of the classical Moving Target Indicator technique common in pulsed radars
for rejecting unwanted static clutter by cancellation [20]. After background cancellation, the
phase shows the expected multi-turn evolution, reproducing the eyelid motion in detail,
in a similar way that was predicted by the numerical simulation. Note that the eye closes
faster than it opens; in the closure step, the eyelid covers the cornea, implying a shorter
Sensors 2024, 24, 7464 9 of 13

radar range, which translates into a positive phase slope, and the opposite occurs for the
aperture motion of a blink.

(a) Phase and magnitude of eyelid with face contribution

2 20

Magnitude in dB
Phase in rad 10
0
0

-10
-2
-20
19.1 19.15 19.2 19.25 19.3 19.35 19.4
tobs in s

(b) Phase and magnitude of eyelid with face cancellation

20

2
15

Magnitude in dB
Phase in rad

0 10

5
-2

0
19.1 19.15 19.2 19.25 19.3 19.35 19.4

tobs in s

Figure 9. Detail of the phase (blue line) and magnitude (red line) of the eye spot backscattering shown
in Figure 8 at the time of one of the blinks. Before (a) and after (b) face background subtraction.

5.2. Slanted Observations

Another way to reduce the background contribution of the face is to observe the eye
at a slanted angle. In our case, the radar was relocated below the face level, observing the
eye with a slanted antenna beam pointing upwards with an angle of 55◦ with respect to the
horizontal direction. In this slanted direction of observation, the parts of the face orthogonal
to the radar observation are reduced up to the point that the face’s contribution to the
measured echo is no longer dominant. Under this condition, the phase rotation of the eyelid
can be observed without requiring background cancellation. Figure 10 shows a complete
recording of 24 s, where a blink can be observed at t = 1 s, the eye is voluntarily closed in the
interval of 3 s to 6 s, and this is followed by six normal blinks until the recording ends, for t =
24 s. The phase term was converted into a relative range offset in mm. The measurements
indicate a typical closure range offset of nearly 5 mm, which is likely influenced by the
movement of the skin around the eye. In Figure 11, it is possible to observe one blink in
more detail. Based on the details observed in this measurement, we can determine the
average blinking period using Figure 10 and examine the dynamics of a single blink using
Figure 11. For instance, the average blinking period and standard deviation were calculated
using the last six blinks measured in the subject, and additional dynamics parameters are
presented in Table 2. Furthermore, a fast-speed video was recorded in synchrony with the
radar acquisition. Slow-motion video reproduction allowed us to relate the video-recorded
eyelid closure/aperture timing with the timings of the dynamics parameters obtained from
the radar observations in order to validate the proposed technique.
 Range in m

Figure 8. A rough estimation of the position of the face with the eye is achieved by identifying the
peak in the graph.
181
term was converted to a relative offset in mm and shows a closed eye between tobs = 2 s 182
Sensors 2024, 24, 7464 and tobs = 6 s. This measurement indicates a maximum offset of nearly 5 mm, which is10 of18313
likely influenced by the movement of the skin around the eye. Following this measurement,

Phase evolution of object @ 0.84373 m

1 0

-2
0
-4
-1
-6
-2
-8

Offset in mm

Magn in dB
-3 -10

-12
-4
-14
-5
-16
-6
-18

-7 -20
0 5 10 15 20 25
tobs in s

Figure Phase
10.For
Figure 9. (blue line)ofand
the evaluation magnitude
figure 10 around(red
tobsline) ofsthe
= 10.5 it iseye backscattering
secondly recorded withusinga camera.
an ascending
It confirms the eyelid blink◦ and the moment when the eyelid closes, when it is closed and
angle of observation of 55 with respect to the horizontal direction. Blinking movements are detected when it
opens again.
at approximately 1 s, 7.6 s, 10.5 s, 13.5 s, 16.9 s, 19.7 s, and 22.5 s. The phase evolution clearly reflects
184
the interval
there were 6ofnormal
the eyelid voluntary
blinks observed,motion,
and itclosing and opening
is apparent at 2.9 s and of
that the magnitude 6.4the
s, respectively.
oscillation 185
changes due to the rotation of the phase vectors resulting from the motion of the eyelid. In 186

Figure 10 it is possible to(a) Phase evolution

observe and magnitude
one blink of eyelid Based on the details observed in
more accurate. 187
30
this measurement, we can determine the blinking frequency using Figure 9 and examine 188
2
the dynamics of a single blink using Figure 10. For instance, the blinking frequency 20 was 189

calculated using the last six simulated blinks of the patient, and all relevant parameters are

Magnitude in dB
Phase in rad

190

presented in Table 2. Furthermore a video was recorded in slow motion and synchronised 10 191
0
with the measurement as a confirmation of the radar measurement. Additionally, a0 video 192

was recorded in slow motion and synchronized with the measurement to confirm the 193
-2
accuracy of the radar measurements. The results of the video during the blink are shown -10 in 194

Figure 11. he extracted frames of the video show the same blink as in Figure 10, providing 195

a comparison10.4that illustrates
10.45 the10.5
accuracy of the measurement.
10.55 10.6 10.65 Additionally,
10.7 the dynamics
10.75 196

of a blink are easier to discern through the tobsgraph

in s in Figure 10. 197

5.3. Exploring the Impact of

(b) Various
Phase andCircumstances
eyelid aperture on Signal Processing 198
15
It0 is important to examine the feasibility of extracting data for the successful detection 199

of eyelid blinks in different environments. In the previous measurement, the sweep band-
Range change in mm

200

width was set at 3 GHz and was later reduced to 1 GHz by deleting the correct samples 10

Aperture in mm
in 201
-2
the beating tone of the radar. As a result, the resolution cell in the direction of propagation, 202

5
-4

0
10.4 10.45 10.5 10.55 10.6 10.65 10.7 10.75
tobs in s

Figure 11. The blink signal at approximately 10.5 s in Figure 10. (a) Phase evolution (blue line) and
magnitude (red line). (b) Phase of the blink process (blue line) with the sampling data of the eye
aperture (red line) during this process.

Table 2. Extracted time characteristics of the eye blink at tobs = 10.5 s in Figures 10 and 11. Dynamics
of different blinks, like closing speed, for example, can be generated from such data. The average
blink period and the standard deviation in the blink period for the whole acquisition are also shown.

Parameter Start End Duration

Duration tobs = 10.40 tobs = 10.70 300 ms
Closing tobs = 10.40 tobs = 10.47 70 ms
Closed eyes tobs = 10.47 tobs = 10.51 40 ms
Opening tobs = 10.51 tobs = 10.70 190 ms
T = 2.99 s σT = 238 ms
Sensors 2024, 24, 7464 11 of 13

5.3. Errors in the Instrument and Methodology

Ranging errors in the radar instrument are caused by the phase noise present in the
echo signal of the eyelid. In the typical eyelid observation conditions of range and Signal-
to-Noise, the standard deviation of the phase noise is σ0 = 0.0024 rad, which is equivalent
to a range estimation error of σr = σ4π
r λ0
= 0.462 µm. This is the intrinsic radar instrument
error.
However, the eyelid is not measured in isolation. The interference created by the
subject’s face introduces additional range deviations well above the instrumental error.
These deviations can be observed as a phase ripple in Figure 11, which introduces an
estimated phase-typical deviation of σmeasurement = 0.1481 rad, corresponding to a range-
typical deviation of σu = 28.486 µm.

6. Discussion
It is important to explore the impact of various circumstances on signal processing
and examine the feasibility of extracting data for the successful detection of eyelid blinks in
different environments. In the previous measurements, the sweep bandwidth was set to 3
GHz and was later reduced to 1 GHz by applying a tapered time domain window to the
beating tone of the radar. As a result, the resolution cell in the direction of propagation,
defined by the bandwidth ∆ f , increased from 5 cm to 15 cm. With accurate alignment and
cancellation of the face background, no substantial loss of information occurred due to a
reduction in bandwidth.
In recent years, the majority of the research on eyelid detection has relied on optical
and electromagnetic methods. Camera-based algorithms, however, often face challenges,
including factors such as facial features, posture, lighting conditions, the angle of view
of the lens, and distance from the camera. For instance, the advanced algorithm pro-
posed in [21] is highly dependent on a face detection system in capturing the subject’s
face. Additionally, the algorithm exhibits sensitivity to the distance of the subject, and its
performance is significantly hindered when the subject’s eyes’ natural colors and contours
are altered with makeup . The method proposed in [22] for monitoring eye movements
using electromagnetic sensing technology requires the inclusion of magnetic materials into
the eyelashes, which may pose potential risks and inconvenience. In contrast, the above
issues can be effectively mitigated using the radar technique proposed, which offer robust-
ness to illumination conditions without requiring any kind of contact or eye manipulation.
Meanwhile, it was tested whether wearing glasses would affect the measurement of eyelid
movement. This is an important consideration if a patient that uses glasses needs to be
monitored. Wearing glasses did not have a significant impact on the results.
It is worth noting that certain limitations may affect the use of radar technology to
measure eyelid dynamics signals. In our experiments, the measurements were conducted in
a controlled indoor environment that was clean, static, and free from complex interference.
Consequently, external noise and environmental factors did not significantly impact the
measurement accuracy or stability. However, in more complex scenarios, such as monitor-
ing drivers, the robustness of the algorithm may be compromised. The small size of the
eyelid as a target, combined with the potential movements of the subject, introduces chal-
lenges in terms of the radar’s precision in target alignment. This difficulty in maintaining
accurate radar pointing could reduce the reliability in complex cases.
To address these limitations, it is possible to explore the integration of multi-input,
multi-output (MIMO) radar technology in future work, which has the potential to enhance
the spatial resolution and target tracking, thereby improving robustness in challenging
environments and reducing the impact of movement of the subject on the accuracy of
measuring the eyelid dynamics.

7. Conclusions
In summary, this study confirms that using radar to detect and characterize eye blink
dynamics is an alternative to using high-speed cameras or magnetic coils, avoiding the
Sensors 2024, 24, 7464 12 of 13

need for good illumination requirements or contact with the patient. The fine beam width
and bandwidth of the FMCW radar can isolate the observation area from other parts of
the body and surrounding objects. At 120 GHz, the depth of penetration into biological
tissues is negligible. For this reason, the measured backscattering changes reflect the
alterations in the shape/geometry of the observed body surface within the antenna beam
spot. Depending on the direction of observation, the eyelid return may also be affected by
echoes from the surrounding face areas, reducing the phase sensitivity of the echo with
respect to the eyelid dynamics. A background cancellation technique has been proposed,
showing the capability to recover the eyelid echo phase in these situations. Alternatively,
observing the eye at a slanted angle with respect to the frontal face direction can also
minimize the unwanted face interference. This study also tested the impact of reducing the
radar bandwidth to 1 GHz and glasses being worn during measurement, without relevant
impacts on the results obtained. Upgrading the radar with a beam-steering capability, for
example, using a multiple-input, multiple-output (MIMO) configuration, would allow us
to track the eye area automatically, allowing for more robust operation in front-of-body
motion resulting from activities such as driving a vehicle.

Author Contributions: Conceptualization, A.B. and D.P.; methodology, A.B., J.R. and D.P.; hardware,
A.A.; software, R.W., A.B. and D.P.; validation, R.W. and J.R.; data analysis, D.P., R.W. and J.R.; results
optimization, R.W. and A.B.; writing—original draft preparation, D.P.; writing—review and editing,
R.W. and A.B.; visualization, R.W.; supervision, A.B. and A.A. All authors have read and agreed to
the published version of the manuscript.
Funding: The work of R.W. was supported by a Research Scholarship from the China Scholarship
Council (CSC) under grant 202208390068.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Written informed consent was obtained from the subject involved in
this study.
Data Availability Statement: The experimental data obtained in this work is available on request
from corresponding author due to lack of suitable data repository at this time.
Acknowledgments: The authors are indebted to the Hospital Sant Joan de Déu de Barcelona for their
help in the study of healthcare-oriented radar applications.
Conflicts of Interest: The authors declare no conflicts of interest.

References
1. Steinhauer, M.; Ruoß, H.; Irion, H.; Menzel, W. Millimeter-wave-radar sensor based on a transceiver array for automotive
applications. IEEE Trans. Microw. Theory Tech. 2008, 56, 261–269.
2. Zhao, P.; Lu, C. X.; Wang, J.; Chen, C.; Wang, W.; Trigoni, N.; Markham, A. mid: Tracking and identifying people with millimeter
wave radar. In Proceedings of the 2019 15th International Conference on Distributed Computing in Sensor Systems (DCOSS),
Santorini Island, Greece, 29–31 May 2019; pp. 33–40.
3. Venon, A.; Dupuis, Y.; Vasseur, P.; Merriaux, P. Millimeter wave fmcw radars for perception, recognition and localization in
automotive applications: A survey. IEEE Trans. Intell. Veh. 2022, 7, 533–555.
4. Peng, Z.; Ran, L.; Li, C. A K-Band Portable FMCW Radar With Beamforming Array for Short-Range Localization and Vital-Doppler
Targets Discrimination. IEEE Trans. Microw. Theory Tech. 2017, 65, 3443–3452.
5. Krofli, R.A.; Kelly, R.D. Meteorological research applications of mm-wave radar. Meteorol. Atmos. Phys. 1996, 59, 105–211.
6. Westwater, E.R.; Crewell, S.; Mätzler, C. A review of surface-based microwave and millimeter-wave radiometric remote sensing
of the troposphere. URSI Radio Sci. Bull. 2004, 2004, 59–80.
7. Melezhik, P.; Razskazovskiy, V.; Reznichenko, N.; Zuykov, V.; Varavin, A.; Sidorenko, Y.; Provalov, S.; Yanovsky, F. High-efficiency
millimeter-wave coherent radar for airport surface movement monitoring and control. Meteorol. Atmos. Phys. 2011, 15, 38–43.
8. Thombre, S.; Zhao, Z.; Ramm-Schmidt, H.; García, J.; and Malkamäki, T.; Nikolskiy, S.; Hammarberg, T.; Nuortie, H.; Bhuiyan, M.
Z. H.; Särkkä, S.; Lehtola, V. V. Sensors and AI techniques for situational awareness in autonomous ships: A review. Meteorol.
Atmos. Phys. 2020, 23, 64–83.
9. Cardillo, E.; Ferro, L.; Sapienza, G.; Li, C. Reliable eye-blinking detection with millimeter-wave radar glasses. IEEE Trans. Microw.
Theory Tech. 2024, 72, 771–779.
Sensors 2024, 24, 7464 13 of 13

10. It’s Time to Think About the Blink. Available online: https://www.reviewofophthalmology.com/article/its-time-to-think-about-
the-blink (accessed on 21 March 2023).
11. Kwon, K.A.; Shipley, R.J.; Edirisinghe, M.; Ezra, D.G.; Rose, G.; Best, S.M.; Cameron, R.E. High-speed camera characterization of
voluntary eye blinking kinematics. J. R. Soc. Interface 2013, 10, 20130227.
12. Korda, A.I.; Giannakakis, G.; Ventouras, E.; Asvestas, P.A.; Smyrnis, N.; Marias, K.; Matsopoulos, G.K. Recognition of blinks
activity patterns during stress conditions using cnn and markovian analysis. Signals 2021, 2, 55–71.
13. Richards, M.A.; Scheer, J.; Holm, W.A.; Melvin, W.L. Principles of Modern Radar; SciTech Publishing, Inc.: Raleigh, NC, USA, 2010;
Volume 3.
14. Levanon, N. Radar Principles; Wiley: New York, NY, USA, 1988.
15. Jacobsen, E.; Lyons, R. The sliding DFT. IEEE Signal Process. Mag. 2003, 20, 74–80.
16. Silicon Radar GmbH. Datasheet TRX 120 001. 120-GHz Highly Integrated IQ Transceiver with Antennas in Package in Silicon Germanium
Technology; Silicon Radar GmbH: Frankfurt, Germany, 2018.
17. IEEE C95.1; IEEE Standard for Safety Levels with Respect to Human Exposure to Electric, Magnetic and Electromagnetic Fields, 0
Hz to 300 GHz. The Institute of Electrical and Electronics Engineers: New York, NY, USA, 2019.
18. Hasgall, P.A,.; Di Gennaro, F.; Baumgartner, C.; Neufeld, E.; Lloyd, B.; Gosselin, M.C.; Payne, D.; Klingenböck, A.; Kuster, N.
IT’IS Database for Thermal and Electromagnetic Parameters of Biological Tissues. Version 4.1, 2022. Available online: Itis.swiss/
database (accessed on 10 October 2024).
19. Jeong, S.; Lemke, B.N.; Dortzbach, R.K.; Park, Y.G.; Kang, H.K. The Asian upper eyelid: An anatomical study with comparison to
the Caucasian eyelid. Arch. Ophthalmol. 1999, 117, 907–912.
20. Skolnik, M.I. Radar Handbook. Ch.15, 2nd ed.; McGrawHill: New York, NY, USA, 1990.
21. Zeng, W.; Xiao, Y.; Hu, G.; Cao, Z.; Wei, S.; Fang, Z.; Zhou, J.T.; Yuan, J. Eyelid’s Intrinsic Motion-Aware Feature Learning for
Real-Time Eyeblink Detection in the Wild. IEEE Trans. Inf. Forensics Secur. 2023, 18, 5109–5121.
22. Tang, J.; Luk, P.; Zhou, Y. Wearable and Invisible Sensor Design for Eye-Motion Monitoring Based on Ferrofluid and Electromag-
netic Sensing Technologies. Bioengineering 2023, 10, 514.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.