Keywords

1 Introduction

The system architecture of 6G for e-healthcare includes communication and sensing as one of its primary fronts. A collaboration of wireless communication and radar technology resulting in Integrated Sensing and Communication (ISAC) is envisioned to govern the beyond 5G and 6G systems. The ISAC systems will play a crucial role in advancing vehicular technology making it fully smart in terms of in-vehicle and outside sensing as well as data processing and communication [1]. Automobiles are the primary transportation method for billions of people worldwide. Cars went through an extraordinarily long road of systematic improvements. Currently, we are observing rapid development of electric and hybrid cars, where modified engines are replacing standard petrol engines. Another branch of development is focused on automated driving, where computers or steering units navigate and drive the car from point to point or even parks [11].

Nowadays, cars are packed with different kinds of sensors, which assist the driver and serve multiple purposes. Such sensors could be divided into two groups: contact and contactless sensors. Contact sensors are placed in locations that usually make constant contact during travel such as seats, bolster, steering wheel, seatbelt in passenger cars, and driving gear (helmet, suit) in specialized cases. This attribute results in restricting their usage and accuracy. Contactless or wireless sensors can sense from a distance and therefore have the advantage of no real restrictions regarding the location of the sensors, other than not to interfere with the driver’s sight. Such sensors could be also placed on or in close proximity to previously mentioned locations such as seats or steering wheel [8].

Initial systems for vital sign monitoring in vehicles were primarily based on imaging (cameras) which have the inherent drawback of security and privacy. Other wireless sensors, which in the majority are radar-based, can provide sets of valuable information regarding the position of the driver/passengers inside the vehicle without compromising privacy. This allows inter alia to evaluate driver conditions such as tiredness or consciousness [3]. Moreover, such sensors can be used to monitor the vital signs of the driver as well as passengers such as respiration, and heart rate. In some instances, this information is encoded simultaneously allowing constant measurement with high sampling rates. Some key research articles based on vital sign minoring using mm-wave radar technology are [2, 3, 7, 8, 10,11,12].

2 Key Challenges in Radar Based In-cabin Sensing

Radar-based in-cabin sensing has many advantages but it has its challenges too. This section presents the key challenges associated with radar-based in-cabin sensing from the perspective of future joint communication and sensing system scenarios.

2.1 Interference from Other Devices

With the advancements in technology, the vehicles in the market now come fully equipped with smart sensors capable of communicating with the driver as well as with each other for seamless operation of the vehicle. Apart from this, other major causes of interference include cellular devices carried by the driver, in-built radars in a vehicle to support auto-pilot/driving assistance, and other communicating devices in the vicinity of the vehicle. Spectrum overlapping between these applications create even severe problems which tend to grow with each day. One straightforward solution to this problem is to use highly directional antennas for in-cabin sensing. But it comes at the cost of losing spatial coverage of the radar [5].

2.2 Optimum Placement and Location of Radar Chip

Even though designed for comfort, vehicles have limited space, and finding a practical yet optimum location for radar placement inside the vehicle is challenging. There are some locations that have been extensively studied in the literature such as behind the steering wheel on the speedometer, on the rooftop, pasted on the rear mirror, inside the car seat targeting from the back, on the side door, etc. The effectiveness of these locations depends largely on the build (shape) of the car, the location of other electronic components, and user preference. Therefore, choosing the optimal location, position, and radar’s angle of attack are very important parameters.

2.3 Movement of Subjects with Respect to Radar

Most of the studies in the literature present the problem of in-cabin sensing as a very simplified problem in which the subject/driver is sitting in an ideal position without any movement with respect to radar. These assumptions are rarely met in practical scenarios wherein the driver usually has restricted movements in his seat. Especially during the first few minutes of starting the journey, these movements are found to be more dominant, and they gradually decrease once the driver settles down [10].

2.4 Selection of Frequency Band of Operation for Radar

The problem of vital sign monitoring of humans with radar technology has been studied extensively in the last ten years. Researchers have proposed a variety of frequency bands starting from a few GHz to the Tera Hz range. The selection of the optimal frequency band of operation for radar has a dominant effect on the detection accuracy of the system. Especially due to the opening of the mm-wave spectrum for both communication and sensing, most of the current standards are adopting mm-wave and it is expected to rule the future sensing and communication applications.

2.5 Cabin Shape and Reflections Due to Vehicle Body

A car cabin, being a metallic hollow body, creates challenges in the effective working of radar. The reflections caused by the car body create interference for received signals thus deteriorating the performance of radar. Apart from that, the vibrations in the car body during its motion also create strong artifacts in the captured signals from radar [5].

2.6 Multi-persons In-cabin Scenarios

Usually, the driver in the car is accompanied by co-passengers whose vital sign monitoring is also equally important but challenging. Some articles [4, 8] showcase the implementation of multi-person monitoring using a single radar system. Nevertheless, the task of separating dopplers from multiple targets and then extracting vital sign information from each one of them is undoubtedly a tedious task.

Fig. 1.
figure 1

mm-wave Radar setup for vital sign monitoring of driver in a vehicular environment.

Full size image

3 Material and Methods

The hardware setup utilized in this study consists of a car seat with adjustable height along with a tripod stand to hold the radar chip which allows movement of the sensor along roll, pitch, and yaw. The measurements are taken with different radar placement options based on which the optimized position of the radar in a vehicular environment is selected. The radar setup utilized in this study is FMCW-based TI 1642 Boost which operates in 76–81 GHz W-band and offers excellent performance in terms of detection accuracy [6]. Figure 1(a),(b) and (c) showcase the measurement of heart rate and breathing rate with AWR 1642 mm-wave radar chip with different angles and radar locations. The measurement settings with possible scenarios chosen for placement of radar with respect to the subject are shown in Fig. 1(d) which has a total of 24 possible locations for radar at two distinct heights of 85 cm and 130 cm from the car floor.

The sensors are located at a horizontal distance of 20 cm and 150 cm from the sternum of the subject. These locations are chosen in accordance with the structure of a typical car environment. The subject under study is to sit on the car seat with hands resting on laps or in a position of holding a driving wheel. The radar setup is moved in three-dimensional space in between the measurements of 120 s each. There are 2–4 transmit/receiver antennae on the radar chip with peak gain >9 dBi across the operating frequency. The setup measures the rate of chest displacement to calculate the heart rate and breathing rate of a person. The results measured from the proposed setup are compared with the BioHarness 3.0 module developed by Zephyr Technology which is worn by the subject during the measurements [9]. Figure 1(e) shows the location of the Zephyr BioHarness Module attached to the chest strap of the subject. The actual setup with a human subject is shown in Fig. 1(f).

Fig. 2.
figure 2

Breathing rate and heart rate measurements of Subject 1 for three scenarios: (a), (b) Radar in front (80 cm from sternum at height= 85 cm); (c), (d) Radar in back (20 cm from sternum at height= 85 cm); (e), (f) Radar at side (90 degrees to left 80 cm from sternum at height= 85 cm).

Full size image
Fig. 3.
figure 3

SNR plots of (a) heart rate, (b) breathing rate for Subject 2 and (c) heart rate, (d) breathing rate for Subject 5 in different scenarios.

Full size image

4 Data Collection and Processing

The data collection for this work was conducted on 5 adult subjects consisting of 3 males and 2 females for variability in the dataset. The first 5 scenarios lasted for 3 min each while the remaining scenarios lasted for 2 min each. The data collection procedure lasted for approximately 2 h for a total of 25 measurements per participant. Each participant was notified about the measurement scenario. A detailed description of measurement scenarios taken in this study is depicted in Table 1. Instructions were provided concerning the procedure and informed consent was duly signed by each of the participants. The data collected from each subject was stored simultaneously in both the Zephyr device and the Radar attached to a computer.

Fig. 4.
figure 4

Bland Altman’s plots of subject 2 for Scenario 1–12.

Full size image
Fig. 5.
figure 5

Bland Altman’s plots of subject 2 for Scenario 13–24.

Full size image
Table 1. Description of measurement Scenarios.
Full size table

Each subject is instructed to tap the chest 3 times which is used as a marker for the manual alignment of the signals from the Zephyr and Radar devices. The raw ECG data with the sampling frequency of 250 Hz from the Zephyr was used which clearly shows the sharp bursts of chest tap for each scenario for the time of occurrence. The respective time of occurrence of the chest tap on the raw ECG waveform is used for segmenting the physiological parameters of the heart rate and the breath rate signals from the Zephyr device. In other to merge the two datasets from the two devices, both devices must have the same sampling frequency.

Therefore, the data from the radar was resampled to match the sampling frequency of data from Zephyr. The averaged Radar data were manually merged or aligned with the Zephyr data for the time of occurrence of the chest tap for each scenario. The parameters of measurement which include the range, height of the radar device from the floor, the tilt angle of the radar, and the azimuthal angles are captured in the measurement setup. Scenarios 1, 2, 3, 4, 5, 9, 10, 11, 12, 13, 17, 18, 19, 20, 21, and 25 (with no subject) have the Radar positioned in the front while scenarios 6, 7, 8, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, and 24 have the Radar positioned at the back. Scenario 25 with no human subject was measured once during the measurement procedure of each subject as a reference case.

5 Results and Discussion

Statistical analysis is applied for comparison of results from radar setup with reference Zephyr device. Various performance matrices are calculated including Bland Altman plots, Pearson’s correlation coefficient (r) with statistical significance of p-values, and the measurement accuracy is determined using the signal-to-noise ratio (SNR) plots. Moreover, the boxplots of individual human subjects for each scenario versus the physiological parameters of heart rate and breath rate are compared amongst all 5 adult subjects to ascertain if the results are comparable. Figure 2 shows the comparison of measured results with benchmark results from Zephyr. The heart rate and breathing rate results for the three cases are presented in Fig. 2. All results are verified by data collected from Zypher and are found to be fairly accurate. To study the accuracy of the proposed radar-based setup, the coefficient of variation is calculated for each scenario. The coefficient of variation measures the dispersion of points from the mean which is given as a ratio of the standard deviation to the mean in percentage. Further, the signal-to-noise ratio (SNR) is calculated as the inverse of the coefficient of variation. The evident occurrence of a lower coefficient of variation in heart rate and breath rate in the Radar device than in the Zephyr device is reflected in the high signal-to-noise ratio (SNR) plots in the different scenarios. The SNR plots of heart rate and breathing rate for Subject 2 and 5 for different scenarios are shown in Fig. 3.

Table 2. Correlation analysis of heart rate and breathing rate of subjects from radar and Zephyr in different scenarios.
Full size table

The Bland Altman’s plots of subject 2 for scenarios 1–12 are shown in Fig. 4. Further, Bland Altman’s plots of subject 2 for scenarios 13–24 are shown in Fig. 5. It can be visualized from Fig. 4, 5 and similar plots obtained for other subjects that the scatter plots lie within the 95\(\%\) confidence level alongside some outliers on either or both sides of the limits of agreement with small systematic bias. The acceptable bias limit is taken to be ±10 bpm and exceptions are visible in some scenarios where the heart rate and breathing rate have high systematic bias. In subject 1, the scatter plots of the heart rate event exhibit a trend from high values to low values with small bias in different scenarios while the breath rate exhibits random scatter plots reflecting the consistent differences in the measurement between the two devices with low bias occurring in several scenarios. Similarly, this general trend in the scatter plots is replicated in the remaining human subjects for most scenarios except for scenario 7 (heart rate) in subject 3 and subject 4 which show the trend from lowest value to highest values, scenario 14 (heart rate) in subject 2 and subject 3 display the same trend. Hence, the Bland Altman plots showed that the Radar device is in strong agreement with the Zephyr device with scatter plots closer to the bias and are different or divergent at extremely low or high values.

It is worthy of note that for all 5 subjects, scenario 7 with the Radar device having the parameters of measurement, displays a large bias for the heart rate and a small bias for the breath rate while scenario 11 displays a low bias for both physiological parameters. However, scenario 12 displays a small bias in both heart rate and breath rate for subject 1 and subject 4 while the remaining three subjects had a high bias in the heart rate and a low bias in the breath rate respectively.

Further, Pearson’s correlation coefficient is used to determine the linear relation as per the strength of association and direction between the results obtained from the Zephyr and Radar device. In addition, the interpretation of the correlation coefficient is based on the decision rule of statistical significance of the p-values where p \(\le \) 0.05 to satisfy the null hypothesis of a significant relationship between the results from both devices. The correlation plots with statistical significance on both heart rate and breath rate for all the subjects are observed in several scenarios ranging from the front side to the back side of each subject except for scenarios 10, 15, 23, and 24 where no simultaneous statistical significant values were obtained on both physiological parameters. Table 2 provides a summary of the occurrences of the p-values (heart rate and breath rate) with the corresponding correlation for the subject.

There are some disparities in the correlation coefficient for all 5 subjects for all the scenarios such that there are scenarios where the heart rate has a positive correlation and the breath rate has a negative correlation and vice versa. The probable reason can be attributed to the differences in height and body type of each subject which can be linked to the radar cross-section (RCS). In addition, the field of view (FOV) of the radar sensor is affected by the vertical height which determines the angular tilt orientation of the radar. Hence as the vertical height is increased the radar sensor with the antenna patch should be tilted to an angle that can accommodate the FOV for good signal quality.

6 Conclusion

Our results suggest that radars offer an effective method for vital sign monitoring of drivers and passengers in vehicular environments which should be explored further. Radars have the advantage of penetration through clothes and other materials to provide better accuracy than imaging-based methods. Further, the privacy of the subject is not compromised in this method. It is observed that the radar placed in front at a height equivalent to the sternum of the subject (Scenario 1) provides the best sensing results as compared to when it is placed at any other location. Apart from this, good results are obtained for Scenario 11 for all subjects and Scenario 12 for two subjects. A maximum deviation between standard Zephyr data and radar data is observed in Scenario 7. It can be concluded that the radar placed at a distance of 40 cm from the chest of the driver at a height of 130 cm from the car floor can be an optimum location of radar. This location corresponds to the area behind the driving wheel in the car dashboard.