Ultra-Wideband Radar Detection of Breathing Rate: A Comparative

Evaluation

Nicole Buckingham

and Denis Gra

canin

Department of Computer Science, Virginia Tech, 2202 Kraft Drive, Blacksburg, VA 24060, U.S.A.

Keywords:

Biosensors, Breathing Rate, Human Computer Interaction, Remote Sensing.

Abstract:

The cost of a medical grade breathing rate monitors can be prohibitive. However, commodity ultra-wideband

(UWB) radar based device can be used to detect breathing rate for health monitoring applications in homes.

We identiﬁed several research challenges, including high cost and functional limitations based on the user’s

location, orientation, and movement, as well as dependency on system placement and vulnerabilities in signal

processing methods. We performed a comparative evaluation for a commodity UWB radar based device,

Walabot, to determine its feasibility for health monitoring applications. The data was processed using two

breathing rate derivation techniques: Fast Fourier Transformation (FFT) and Peak Detection. The results

support feasibility of Walabot as a commodity breathing rate monitor for health monitoring in homes.

1 INTRODUCTION

Human body produces a variety of physiological sig-

nals. Four important signals are the heart rate, breath-

ing rate, temperature and blood pressure. These sig-

nals are known as the vital signs because they provide

a well rounded indication of the overall state of the

body. Health monitoring focuses on measuring vari-

ous physiological signals and processing these signals

to determine health status.

Heart rate is controlled by the rate at which the

sinoatrial node creates electrical impulses. Resting

heart rate depends on age, gender and exercise level,

but an increase in resting heart rate can reveal declin-

ing heart health and increase in risk of heart attack.

The respiratory system is responsible for the

breathing mechanism. The rate at which breath-

ing occurs is speciﬁcally controlled by the respira-

tory pacemaker. Variation in breathing rate are com-

monly associated with conditions like asthma, anx-

iety, pneumonia, lung disease and congestive heart

failure (Cleveland Clinic, 2019).

Traditionally, heart rate is used to determine risk

of heart attack and failure. However, studies have

shown that breathing rate is a more accurate way to

detect or predict heart conditions, such as cardiac ar-

rest (Cretikos et al., 2008).

https://orcid.org/0000-0003-2558-5861

https://orcid.org/0000-0001-6831-2818

Contactless health monitoring systems use sen-

sor(s) placed in the vicinity of the user but without any

direct user contact. The goal is to provide the same ca-

pabilities of wearable technology, but with the added

beneﬁts of increased comfort. This is signiﬁcant for

health applications because the users may not always

be able to correctly use wearable devices.

The effectiveness of health monitoring systems is

affected by the environment in which they are used.

Contactless devices, in particular, can beneﬁt from en-

vironments such as Smart Built Environments (SBE)

that provide data collection and analysis services in-

tegrated within a control infrastructure (Tasooji et al.,

2018). The cost of a medical grade breathing rate

monitors can be prohibitive for use in SBEs (e.g.,

smart homes). However, there are technologies that

can be used for contactless breathing rate monitors.

Ultra-wideband (UWB) devices use a wide spec-

trum of low energy radio frequency signals to de-

termine the chest displacement caused by breathing.

The measurements can be very accurate but also can

be sensitive to noise and limited by user location.

UWB devices have a potential to enable ubiquitous

breathing rate monitoring by continuously monitor-

ing users and predicting health events in real time.

UWB based system have limitations in terms of the

coverage area and the user’s position and orienta-

tion. However, with the availability of commod-

ity UWB devices such as Walabot (Walabot, 2020),

it is important to understand those limitations. We

Buckingham, N. and Gra

canin, D.

Ultra-Wideband Radar Detection of Breathing Rate: A Comparative Evaluation.

DOI: 10.5220/0010346906770684

In Proceedings of the 14th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2021) - Volume 5: HEALTHINF, pages 677-684

ISBN: 978-989-758-490-9

677

have conducted a comparative evaluation of Walabot

and a medical grade contact breathing rate monitor

(MindWare Mobile Impedance Cardiograph (Mind-

Ware Technologies LTD, 2013)) using 4 procedures:

breathing rate variation, horizontal placement vari-

ation, vertical placement variation and user move-

ment variation. The results support feasibility of Wal-

abot as a commodity breathing rate monitor for health

monitoring in homes.

2 RELATED WORK

Breathing, or respiration, is a vital mechanism that all

humans perform subconsciously in which air circu-

lates in and out of the lungs, in order to provide oxy-

gen to the body. The rate and volume of air moved

is controlled by the respiratory pacemaker, which is

located in the medulla of the brainstem.

Tidal volume is the volume of air intake dur-

ing inspiration. Typical tidal volume is dependent

on the individual but a volume of 400–500 mL is

considered healthy (Hallett and Ashurst., 2019). In

healthy adults, tidal volume is quite consistent be-

tween breaths, but infrequent changes in air intake

due to yawns or signs are considered normal. Tidal

volume may also change in situations where breath-

ing rate changes, such as exercise (Braun, 1990).

Respiration devices typically measure breathing

rate and tidal volume. Breathing (respiration) rate is

number of breaths per minute (bpm). Typical resting

breathing rate varies per person but a rate of 12-20

bpm for an adult is considered normal, whereas a rest-

ing rate over 25 or under 12 is considered unhealthy.

Various health conditions affect resting breathing rate,

such as asthma, anxiety, pneumonia, congestive heart

failure and lung disease (Cleveland Clinic, 2019).

Changes in breathing rate in hospital patients is cor-

related with an increase in mortality rate.

One limitation of breathing rate is that the mea-

surement can be affected by the time over which is it

measured. Some studies extract breathing rate from a

shorter time, such as 15 seconds. However, this mea-

surement is subject to inaccuracy because breathing

rate can change between measurement periods due to

changes in air intake, as described in the last para-

graph. Studies have shown that the time period with

the least variability is one minute (Evans et al., 2001).

There are a variety of wearable medical devices on

the market and even more being developed and tested

currently. These devices appear as normal clothing or

accessory articles, but they consist of small sensors

that can read a variety of biometric data.

From a physiological perspective, breathing rate

offers a unique opportunity for wearable devices. In

contrast to heart rate, which is measured using inter-

nal signals, physical signals caused by breathing rate

can be recorded using audio, airﬂow or the low fre-

quency mechanical signal caused by small chest in-

ﬂections (Furtak et al., 2013). These physical effects

have made breathing rate a popular subject of wear-

able device research.

Contactless monitoring of breathing rate is a novel

topic of research. Typical contactless methods use

thermal sensors, acoustic changes, lasers or radar.

Radar methods encompass techniques that utilize mi-

crowave or radio frequency (RF). Such methods rely

on the same idea: a stationary person’s breathing

rate matches the phase shifts of signals reﬂected off

the person (Lin, 1975). These methods are typically

categorized as continuous-wave (CW), frequency-

modulated continuous-wave (FMCW), impulse radar,

and ultra-wideband (UWB) (Br

user et al., 2015).

3 METHODS

UWB radar emits a wide spectrum of low energy RF

that can be used to capture the low frequency mechan-

ical signal caused by small chest inﬂections during the

breathing cycle. Our goal is to investigate the follow-

ing question: “Can UWB radar be utilized to enable a

fully functional breathing rate monitoring system for

an active user in a conﬁned SBE?”

For this work, we deﬁne a fully functional system

as one that derives accurate breathing rate data within

10% of the true breathing rate. Further, such a sys-

tem should maintain accuracy at all times in which the

user is present in the conﬁned space and should not

be limited by user location, orientation or movement.

The speciﬁc area covered by the system is dependent

on the particular system but the minimum area should

cover a standard bedroom or living room. For the

scope of this work, the system will be restricted to

one user within range at a time.

The ideal system would collect breathing rate for

each user, detect distress events, alert the user’s con-

nected devices and even call for emergency help when

necessary. However, we focus on the breathing rate

detection system only. These potential functionalities

could be evaluated in future work on devices that are

deemed fully functional.

The reviewed UWB devices show promising re-

sults but are not considered fully functional because

they do not meet the requirements described above.

Speciﬁcally, these systems are vulnerable to noise or

limited by user. There are several important chal-

lenges to overcome. The details and severity of each

HEALTHINF 2021 - 14th International Conference on Health Informatics

678

Figure 1: Walabot coordinate systems: A)Walabot device

and its Cartesian coordinate system. B)Walabot antenna ar-

ray. C)Cartesian and spherical coordinate systems.

challenge is speciﬁc to the device and system design,

but the challenges for the current systems are:

1. Cost: They are expensive, thus not affordable for

an average household.

2. System Placement: Dependance on the place-

ment of the transmitter and receiver device(s).

3. Signal Processing: Deriving breathing rate from

the received signal is sensitive to noise.

4. User Location: The user has to be in particu-

lar locations relative to transmitter / receiver de-

vice(s).

5. User Orientation: The user has to be oriented

in particular angles relative to the transmitter and

receiver device(s).

6. User Movement: Increased sensitivity to noise or

inaccuracy when the user is moving.

3.1 Design

We conducted a comparative evaluation of Walabot

(Figure 1) and a medical grade contact breathing

rate monitor (MindWare Mobile Impedance Cardio-

graph (MindWare Technologies LTD, 2013)) to de-

termine feasibility of UWB devices for home health

monitoring systems. There are 6 design criteria.

1. Only one user can be in the designated space.

2. The user must use different breathing rates to test

the full range of possible breathing rates.

3. There must be at least one trial where the user

must remain still and there must be at least one

trial where the user must move.

4. The conﬁned space should be approximately the

size of a standard living room or bedroom.

5. Walabot must be placed in a variety of locations

and orientations with respect to the user’s chest.

6. There must be at least one trial where the user sits,

stands and walks in place.

The main goal was to analyze the effectiveness

and limitations of a commodity UWB radar based de-

vice as a breathing rate monitoring system against a

medical grade wearable breathing rate monitor. The

secondary goal was to determine how well Walabot

meets the identiﬁed requirements. The steps are:

1. Analyze and discuss of the methodology used to

gather breathing rate provided by Walabot API.

2. Develop a breathing data acquisition for Walabot.

3. Develop a signal post processing script to com-

pute the breathing rate of data simultaneously

generated by Walabot and Mobile devices

4. Design a comparative evaluation to test the accu-

racy and limitations of Walabot.

5. Provide the comparative evaluation results and

discuss the accuracy and limitations of Walabot.

6. Discuss Walabot in the context of the posed re-

search question and challenges.

The research question and daily living emphasis

informed the ﬁrst four design criteria. First: only one

user was involved in the data acquisition session due

to COVID-19 situation. Second: the user was told to

use the following breathing rates during different sets

of data acquisition: normal, deep and fast. To ensure

the deep and fast rates were within reason, the user

aimed to reach 8–12 bpm during deep breathing and

15–20 bpm during fast breathing. Third: the partic-

ipant was instructed to remain still for a set time and

to perform a predetermined movement for a set time

during the data acquisition session. The chosen posi-

tions and movement were speciﬁed in a later design

decision. Fourth: the participant and the device were

placed in a speciﬁed area, within the size of a stan-

dard living room or bedroom. This stems from the

assumption that the device would be placed in a com-

mon living area to be utilized frequently. A sketch

of the layout of the room can be seen in Figure 2 top

and a picture of the actual room is illustrated in Fig-

ure 2 bottom. The user was positioned 60 cm away

from Walabot along Z axis and the exact positioning

of Walabot was varied within this layout throughout

data acquisition.

4 EVALUATION RESULTS

The comparative evaluation was carefully designed in

such a way that allows for insight into both the re-

search question and the identiﬁed criteria. Addition-

ally, it was ensured that the testing setup should reﬂect

how the device would be theoretically used in a real

living space in order to make the data useful for real

world applications.

Ultra-Wideband Radar Detection of Breathing Rate: A Comparative Evaluation

679

Figure 2: Top:A sketch of the testing setup. Bottom:A pic-

ture of the testing setup.

4.1 Raw Data Comparison

Walabot and Mobile data were aligned relative to set

number, time and frequency. However, there are two

important differences in Walabot and Mobile: magni-

tude and stability. The magnitude of the two signals

shouldn’t be directly compared. Mobile raw signal

does not have a signiﬁcant DC component, whereas

Walabot has a large DC component (Figure 3 top).

Once the DC components are removed, the difference

in magnitude in the signals was clearly noticeable. To

eliminate the magnitude differences, the signals were

normalized signals between -1 and 1 for the best vi-

sual illustration of the signals. The shape should re-

semble the expected respiration sine wave form with

peaks and valleys according to inhaling and exhaling

respectively and the frequency should correspond to

the breathing rate.

With regards to stability, Walabot data is notice-

ably less stable than Mobile data. Kilani ﬁxed this

issue by averaging the collected energy values over a

sliding window of 5 samples (Kilani, 2017). Instead,

this signal can be stabilized by smoothing. using a

sliding window. The effects of normalization and the

smoothing are shown in Figure 3 middle.

The simplest way to compare the data is to vi-

sually compare the shape of the graphs for each de-

vice. In an effort to quantify this comparison, we de-

termined the a correlation between the two signals.

However, the correlation coefﬁcient is not a determi-

nation of signal accuracy because the concept of accu-

racy in this work is based on the breathing rate calcu-

lated from the signals. Further, the signals may have

a low correlation coefﬁcient but both be highly ac-

curate. Instead, the number of extrema within each

signal is a good indication of what the breathing rate

calculation will be.

Figure 3: Top:An example of raw Walabot and Mobile sig-

nals. Middle:An example of the effects of normalization

and smoothing of Walabot signal. Bottom:An example of

processed Walabot and Mobile signals and their extrema

and their correlation coefﬁcient.

An example of this signal comparison is shown in

Figure 3 bottom. The correlation coefﬁcient is shown

in the bottom right hand corner of the graph. In this

example, the coefﬁcient is 0.68. Visually, the signals

appear to have very similar frequencies and very sim-

ilar number of minimums and maximums between

Walabot and Mobile signals. The correlation coefﬁ-

cient is not 1 since the shapes of the two signals vary.

4.2 FFT Signal Analysis Technique

Extracting breathing rate from raw data is a complex

process that is still being researched today. In fact,

there are over 100 methods for extracting breathing

rate from ECG and PPG signals alone (Charlton et al.,

2016). Signal processing of radar data to ﬁnd breath-

ing rate is a popular topic is research currently (Li

et al., 2016; Taheri and Sant’Anna, 2014). There

are many proposed extraction methods, but there is

no gold standard technique for Walabot at this point.

HEALTHINF 2021 - 14th International Conference on Health Informatics

680

As a starting point, we attempted to replicate the cal-

culation method used by Mobile. Speciﬁcally, Mo-

bile uses a low pass ﬁlter, then calculates a respiration

trend once per second using the FFT technique over a

sliding window of 5 seconds.

Next, the sliding window was deﬁned. Initially,

the window was set as 5 times the f s value, or 5 sec-

onds, and also tested at 7 and 10 seconds.

During each iteration of the loop, a FFT was per-

formed on that window of the ﬁltered data. Com-

monly, this is done in a sliding window to see

smaller time periods and to evaluate the changes over

time (Gunasekara, 2017).

The DFT was used to ﬁnd the fundamental breath-

ing frequency. The data contains very few data points

which causes the resolution of the DFT to be very low

and the calculated breathing rate is not very accurate.

Spline interpolation was used to increase the resolu-

tion. The points before, on and after the max index

were cubically interpolated.

When the sliding window ﬁnishes, an extra ﬁlter-

ing step is performed to clean up the breathing rate

calculations. First, the ﬁrst and last bpm calculations

are deleted. Then, any calculations below 6 or above

25 are deleted to disregard any values deemed unre-

alistic. These numbers were derived from the estima-

tion that the average breathing rate for a health adult

is between 12 and 20 bpm (Cleveland Clinic, 2019).

Then, the average breathing rate was

calculated throughout the set. The ac-

curacy of the average breathing rate is

100 ∗ (1 −

|Calculated bpm−User Reported bpm|

User Reported bpm

To illustrate the results, the calculated bpm and

the averages are shown (Figure 4 top). The average

breathing rate was 12.84 bpm for Walabot and 12.53

bpm for Mobile. The trends reported by Mobile for

the breathing rate is shown (Figure 4 middle). The

average reported breathing rate was 11.21 bpm.

The calculated average breathing rate for Mobile

data was 1.33 bpm above the reported average breath-

ing rate. The user counted roughly 13.5 bpm. The

reported breathing rate was not within 10%, while

the calculated breathing rates for Walabot and Mobile

were both within 10% (Figure 4 middle).

4.3 Peak Detection

We implemented a second breathing rate extraction

method. Yang et al. proposed the use of time-domain

peak detection in order to extract breathing rate within

a window of one breathing cycle, or 5 seconds. This

technique was tested with Doppler radar, an ECG and

a respiration band and yielded highly correlated re-

sults between the two contact sensors (Yang et al.,

Figure 4: Top: An example of bpm calculated from Wal-

abot and Mobile raw data using the FFT technique. Mid-

dle: An example of bpm provided by Mobile. Bottom: An

example of bpm calculated from Walabot and Mobile data

using the Peak Detection Technique.

2017). This test is quite similar to our work as it used

both radar and a respiration band. Additionally, this

technique was deemed a viable option based on visual

observation of our collected data.

We ﬁrst implemented the same lowpass ﬁlter used

in the FFT technique. Yang et al. suggested the use

of a bandpass ﬁlter, but we used a lowpass ﬁlter to

maintain consistency between the post processing and

Mobile processing suggestions (Yang et al., 2017).

Next, the sliding window from the FFT technique

was implemented again. To fully test and compare

our processing techniques, windows with 5, 7 and 10

seconds were tested. For each window we located the

indices with the local minimum and maximum points.

The period was then calculated by doubling the time

between the minimum and maximum points. Finally,

the breathing rate was calculated as 60 ∗

f s

period

When the sliding window ﬁnishes, the extra ﬁlter-

ing step that was implemented in the FFT technique

Ultra-Wideband Radar Detection of Breathing Rate: A Comparative Evaluation

681

Table 1: Mobile reported bpm results.

Set User Mobile Mobile

Reported Reported Reported

bpm bpm bpm Accuracy

1 13 11.1502 85.77%

2 15 11.6261 77.51%

3 13.5 11.2050 83.00%

is used again here to clean up the breathing rate cal-

culations. Then the average breathing rate and cor-

responding accuracy’s are computed. This procedure

explained was used to calculate the breathing rate for

both Walabot and Mobile signals for each set. To il-

lustrate the results, the calculated bpms and the aver-

ages are shown (Figure 4 bottom). In this set, the aver-

age breathing rate for Walabot was found to be 13.38

bpm and the average breathing rate for Mobile was

13.68 bpm. With this extraction technique, the calcu-

lated breathing rate for Walabot and Mobile were both

correct within 10% of the user reported 13.5 breaths.

Further, this technique yielded breathing rate calcula-

tions with signiﬁcantly higher accuracy than the FFT

calculations. Additionally, the trends reported for the

breathing rate of Mobile data are shown (Figure 4

middle). The average reported bpm is 11.21 bpm

for this data set which is not within 10% of the user

counted breaths of 13.5 bpm.

4.4 Accuracy Comparison

Three sets of data were taken on a 23 year old male.

During each set, the user was asked to breath nor-

mally. The user reported approximately 13, 15 and

13.5 bpm for sets 1, 2 and 3 respectively.

Another accuracy metric uses Mobile to Walabot

bpm ratio as percentage to compare the similarity of

the two calculations. When this value is within 10%

of 100%, the two calculations are considered statisti-

cally equivalent and this indicates high correlation of

signals and high stability of the associated breathing

rate derivation technique.

Table 1 shows the average Mobile reported breath-

ing rate and the respective accuracy’s. Prior to test-

ing, Mobile measured breathing rate was expected to

be the most accurate. However, the accuracy’s are

all signiﬁcantly lower than 90%. Further, none of

the reported averages are within 10% of the expected

value. Due to this testing, Mobile reported breathing

rate values were not used in further testing.

Table 2 shows the FFT calculations using window

sizes of 5, 7 and 10. Calculations with a window size

of 5 seconds are within 10% of the expected value.

The calculations for Mobile data with a window size

of 7 and 10 seconds are within 10% of the expected

value. However, one calculation for Walabot with a

window size of 7 seconds and two calculations with

a window size of 10 seconds are not within 10% of

the expected value. The second set is only 55.24%

accurate when using a window size of 10 seconds.

This is an interesting observation given that increas-

ing the window size increases the resolution, which

was thought to improve the accuracy. The ratios are

within 10% for a window size of 5 seconds, but the

ratios are outside of the 10% bounds for larger win-

dows. Due to these results, a window size of 5 sec-

onds was used in further testing of the FFT technique.

Table 3 shows the Peak Detection technique results

using window sizes of 5, 7 and 10. All results are

within 10% of the expected value and the ratios are

all within 10%n. An interesting observation is that

the window size does not have a signiﬁcant impact on

the accuracy of the calculation. In order to maintain

consistency with the FFT technique, a window size of

5 seconds was used in further testing.

5 DISCUSSION

The comparative evaluation is a necessary ﬁrst step

to understand the accuracy and limitations of Walabot

breathing rate measurements. Six criteria were used:

cost, system placement, signal processing, user loca-

tion, user orientation, and user movement.

Cost: The least expensive version of Walabot is

less than $100 and is considered affordable for an av-

erage household as a health monitoring device.

System Placement: Walabot has an advantage

because the antennas are all located within the de-

vice (rather than having a separate transmitter and re-

ceiver), making it easy to deploy.

Signal Processing: Two breathing rate deriva-

tion techniques were used, FFT and Peak Detection.

The results were compared against the user reported

breathing rate (13, 15 and 13.5 bpm) to obtain an ac-

curacy measurement. A window size of 5 seconds

was chosen for both techniques because then all the

results were within 10%. During the breathing rate

variation procedure of the comparative evaluation, an

issue with the window size was uncovered. Essen-

tially, using a window size of 5 seconds is ideal for

breathing rates between 12 and 20 bpm, but this win-

dow size is too small for breathing rates lower than

12. This is caused by fact that the calculation tech-

niques require a window size of at least one breath to

accurately determine breathing rate, but less than one

breath is taken during a 5 second window when the

breathing rate is lower than 12 bpm.

When the window size was 10 seconds, the accu-

HEALTHINF 2021 - 14th International Conference on Health Informatics

682

Table 2: FFT technique results.

Set Window User Mobile Mobile Walabot Walabot Mobile to

Size Reported bpm bpm bpm bpm Walabot

bpm Calculation Accuracy Calculation Accuracy Ratio

5 13 12.37 95.16% 13.35 97.32% 92.66%

7 13 13.20 98.48% 11.70 90.03% 112.82%

10 13 12.06 92.75% 9.36 71.96% 128.85%

5 15 13.69 91.27% 13.99 93.28% 97.86%

7 15 16.49 90.04% 11.41 76.04% 144.52%

10 15 14.87 99.10% 8.29 55.24% 179.37%

5 13.5 12.53 92.84% 12.81 94.92% 97.81%

7 13.5 13.87 97.27% 13.74 98.25% 100.95%

10 13.5 12.85 95.18% 12.16 90.04% 105.67%

Table 3: Peak detection technique results.

Set Window User Mobile Mobile Walabot Walabot Mobile to

Size Reported bpm bpm bpm bpm Walabot

bpm Calculation Accuracy Calculation Accuracy Ratio

5 13 12.97 99.74% 13.26 97.97% 97.81%

7 13 12.99 99.91% 13.12 99.09% 99.01%

10 13 13.07 99.45% 13.27 97.93% 98.49%

5 15 15.05 99.68% 15.05 98.21% 100%

7 15 15.15 99.02% 14.89 99.25% 101.75%

10 15 15.25 98.33% 14.66 97.75% 104.02%

5 13.5 13.60 99.27% 13.51 99.94% 106.66%

7 13.5 13.59 99.36% 13.52 99.86% 100.52%

10 13.5 13.60 99.30% 13.51 99.96% 106.66%

racy for breathing rate lower than 12 bpm increased.

However, increasing the window size decreases the

accuracy of calculations during sets with breathing

rates above 12 bpm. The original window size of 5

seconds was used for the rest of testing during the

comparative evaluation. Adding a window size ad-

justment capability could improve the accuracy of the

calculations for any breathing rate.

User Location: Walabot is expected to be func-

tional whenever a user is within the Arena speciﬁed

within the data acquisition software by the values of

R, φ and θ. This allowed for the simpliﬁcation of the

location challenge because the R range could simply

be set based on the user’s locations during the study.

Speciﬁcally, the range of R in the Arena was set as 20

to 80 cm because the user was always positioned 60

cm away from the device in Z direction. However, the

minimum value of R is 1 cm and the maximum value

is 1000 cm, which allows Z axis to be set such that it

covers a typical living room or bedroom.

User Orientation: The horizontal and verti-

cal placement variation procedures were followed.

Within each of these procedures, Walabot was an-

gled -45, -22.5, 0, 22.5 and 45 degrees away from the

user’s chest in the X or Y axis. The results showed

relatively high accuracy of breathing rate computed

by the FFT technique. The results for all locations

were all above 85% for both horizontal and vertical

testing when the reported breathing rate was above 12

bpm. The Peak Detection technique had quite low

accuracy. This suggest that the FFT technique can

provide a more accurate breathing rate when the user

is not located directly in front of Walabot. However,

these results were not all within 10% accuracy. Con-

sequently, Walabot coupled with the signal processing

techniques developed do not overcome the user orien-

tation challenge at this point. It is important to note

that the θ and φ values determine the cone shape of the

Arena in X and Y axes. For this work, the ranges for

both θ and φ were set as -1 to 1 because this allowed

for the highest sampling frequency. Further testing

should be done with higher θ and φ values to deter-

mine if the user orientation challenge is minimized

by a larger Arena size.

User Movement: The user stood against a wall,

in an open area and walked in place. Walabot signal

was very noisy during moving trials, which caused

the Peak Detection technique accuracy to drop below

90%. The FFT technique was able to accurately com-

pute breathing rate regardless of the user’s movement

Ultra-Wideband Radar Detection of Breathing Rate: A Comparative Evaluation

683

during the trials when the reported breathing rate was

above 12 bpm. This suggests that Walabot with the

FFT techniques overcomes the user movement chal-

lenge. Results from the horizontal, vertical and user

movement variation procedures revealed the high ac-

curacy and reliability of the FFT technique when the

reported breathing rate is above 12 bpm.

FFT technique yielded results with higher accu-

racy than the Peak Detection technique. The primary

reason for this is that the FFT method is not signiﬁ-

cantly affected by the noise in the shape signal, while

the Peak Detection method is highly affected. For this

reason, the FFT method should be the primary focus

in future testing. However, the FFT technique is not

sufﬁciently robust at this point. An adjustable window

size can increase the accuracy of the FFT technique.

6 CONCLUSIONS

The development of a robust and fully functional

UWB radar based system has the potential to pro-

vide accurate monitoring of breathing rate. However,

current UWB radar based systems have issues which

hinder their accuracy or reliability. Six criteria were

identiﬁed: cost, user location, user orientation, user

movement, system placement and signal processing.

We designed and performed a comparative evaluation

in which data was collected by following four proce-

dures: breathing rate variation, horizontal placement

variation, vertical placement variation and user move-

ment variation. Results from this study were promis-

ing and suggested a high potential for Walabot cou-

pled with the FFT technique. Speciﬁcally, it was de-

termined that this system meets the cost, user loca-

tion, and system placement criteria. However, further

testing is required to determine if the system can fully

meet the user orientation, user movement and signal

processing criteria. The results support feasibility of

Walabot as a commodity breathing rate monitor for

health monitoring in homes.

REFERENCES

Braun, S. R. (1990). Respiratory rate and pattern. In Walker,

H. K., Hall, W. D., and Hurst, J. W., editors, Clinical

Methods: The History, Physical, and Laboratory Ex-

aminations, chapter 43, pages 226–230. Butterworths,

Boston, 3 edition.

user, C., Antink, C. H., Wartzek, T., Walter, M., and

Leonhardt, S. (2015). Ambient and unobtrusive car-

diorespiratory monitoring techniques. IEEE Reviews

in Biomedical Engineering, 8:30–43.

Charlton, P. H., Bonnici, T., Tarassenko, L., Clifton, D. A.,

Beale, R., and Watkinson, P. J. (2016). An assessment

of algorithms to estimate respiratory rate from the

electrocardiogram and photoplethysmogram. Physi-

ological Measurement, 37(4):610–626.

Cleveland Clinic (2019). Vital signs. https://my.cleve

landclinic.org/health/articles/10881-vital-signs. [last

accessed 17 December 2020].

Cretikos, M. A., Bellomo, R., Hillman, K., Chen, J., Finfer,

S., and Flabouris, A. (2008). Respiratory rate: the

neglected vital sign. The Medical Journal of Australia,

188:657–659.

Evans, D., Hodgkinson, B., and Berry, J. (2001). Vital signs

in hospital patients: a systematic review. International

Journal of Nursing Studies, 38(6):643–650.

Furtak, N. T., Skrzetuska, E., and Krucinska, I. (2013).

Development of screen-printed breathing rate sensors.

Fibres and Textiles in Eastern Europe, 21(6):84–88.

Gunasekara, A. (2017). Contactless estimation of breath-

ing rate using UWB radar. Master’s thesis, School of

Electrical Engineering and Computer Science Univer-

sity of Ottawa. [last accessed 17 December 2020].

Hallett, S. and Ashurst., J. V. (2019). Physiology, Tidal Vol-

ume. StatPearls Publishing, Treasure Island, FL.

Kilani, R. (2017). Testing the viability of Wal-

abot for breathing detection. https://www.

hackster.io/Ruleof2/testing-the-viability-of-walabot-

for-breathing-detection-b0a14f. [Last accessed 17

December 2020].

Li, W., Tan, B., and Piechocki, R. J. (2016). Non-contact

breathing detection using passive radar. In IEEE Inter-

national Conference on Communications, pages 1–6,

Kuala Lumpur, Malaysia.

Lin, J. C. (1975). Noninvasive microwave measurement of

respiration. Proceedings of the IEEE, 63(10):1530-

1530

MindWare Technologies LTD (2013). Mindware mobile

impedance cardiograph product manual. http://

downloads.mindwaretech.com/Hardware/MindWare

Mobile Impedance Cardiograph 1.0.0-1 Manual.pdf.

[Last accessed 17 December 2020].

Taheri, T. and Sant’Anna, A. (2014). Non-invasive breath-

ing rate detection using a very low power ultra-wide-

band radar. In IEEE International Conference on

Bioinformatics and Biomedicine, pages 78–83.

Tasooji, R., Dasgupta, A., Gra

canin, D., LaGro, M., and

Matkovi

c, K. (2018). A multi-purpose IoT framework

for smart built environments. In Rabe, M., Juan, A. A.,

Mustafee, N., Skoogh, A., Jain, S., and Johansson,

B., editors, Proceedings of the 2018 Winter Simula-

tion Conference, pages 4240–4241.

Walabot (2020). Walabot DIY. https://walabot.com/diy.

[Last accessed 17 December 2020].

Yang, X., Sun, G., and Ishibashi, K. (2017). Non-contact

acquisition of respiration and heart rates using doppler

radar with time domain peak-detection algorithm. In

2017 39th Annual International Conference of the

IEEE Engineering in Medicine and Biology Society

(EMBC), pages 2847–2850.

HEALTHINF 2021 - 14th International Conference on Health Informatics

684