Non-Contact Smart Sensing of Physical Activities during Quarantine Period Using SDR Technology

The global pandemic of the coronavirus disease (COVID-19) is dramatically changing the lives of humans and results in limitation of activities, especially physical activities, which lead to various health issues such as cardiovascular, diabetes, and gout. Physical activities are often viewed as a double-edged sword. On the one hand, it offers enormous health benefits; on the other hand, it can cause irreparable damage to health. Falls during physical activities are a significant cause of fatal and non-fatal injuries. Therefore, continuous monitoring of physical activities is crucial during the quarantine period to detect falls. Even though wearable sensors can detect and recognize human physical activities, in a pandemic crisis, it is not a realistic approach. Smart sensing with the support of smartphones and other wireless devices in a non-contact manner is a promising solution for continuously monitoring physical activities and assisting patients suffering from serious health issues. In this research, a non-contact smart sensing through the walls (TTW) platform is developed to monitor human physical activities during the quarantine period using software-defined radio (SDR) technology. The developed platform is intelligent, flexible, portable, and has multi-functional capabilities. The received orthogonal frequency division multiplexing (OFDM) signals with fine-grained 64-subcarriers wireless channel state information (WCSI) are exploited for classifying different activities by applying machine learning algorithms. The fall activity is classified separately from standing, walking, running, and bending with an accuracy of 99.7% by using a fine tree algorithm. This preliminary smart sensing opens new research directions to detect COVID-19 symptoms and monitor non-communicable and communicable diseases.


Introduction
Countries around the globe have been experiencing a pandemic situation since December 2019. The outbreak of COVID-19 set up a concerning international public health crisis. As the outbreak continues to develop, the whole world is searching for possibilities to preclude the outbreak of the virus in new places, or to stop human-to-human interaction at places where the virus that originates COVID-19 was previously mingling. The public health departments in each county have taken the necessary steps to achieve these goals, such as implementing quarantine, which entails restricting human movement, maintaining a social distance from the rest of the public, or isolating healthy individuals who may not show any symptoms, with the goal of detecting virus-infected people early. Many states have legally permitted enforcement of quarantine from time to time when a new variant of  Design of non-contact smart sensing system for monitoring falls by extracting finegrained WCSI in the presence of walls.  Enhancing the monitoring system by using portable, flexible, and multi-functional SDR technology.  Intelligent monitoring accomplished by the use of machine learning algorithms. The optimal performance is evaluated by measuring classification accuracy, prediction speed, and training time for each algorithm.
Design of non-contact smart sensing system for monitoring falls by extracting finegrained WCSI in the presence of walls.
 Design of non-contact smart sensing system for monitoring falls by extracting finegrained WCSI in the presence of walls.  Enhancing the monitoring system by using portable, flexible, and multi-functional SDR technology.  Intelligent monitoring accomplished by the use of machine learning algorithms. The optimal performance is evaluated by measuring classification accuracy, prediction speed, and training time for each algorithm.
The paper is structured as follows. In Section 2, related existing work on non-contact smart sensing for monitoring human activities by using Wi-Fi, radar, and SDR-based sensing technologies is provided for deeper insight. Section 3 provides an overview of the noncontact sensing platform used for development by exploiting the SDR technology. Section 4 is dedicated to the methodology used for extracting the WCSI data and building the classification model for physical activities. In Section 5, the accomplished results and their performance are presented. Lastly, Section 6 summarizes the performance of non-contact smart sensing using SDR technology, and future recommendations are suggested for improving the system.

Related Work
There are a lot of existing Wi-Fi technology-based platforms developed for human activities monitoring and detecting vital signs. The wireless sensor network (WSN) uses Wi-Fi technology to detect movements of the human body without a portable device in the operating area. Passive dynamic velocity moving people detection system (PADS) uses device-less detection to extract amplitude and phase information from the WCSI and exploit spatial diversity using MIMO systems; such a system uses commercial Wi-Fi devices to capture human body motions [23]. Human activity can be identified by the reflected Wi-Fi signals from the human body to create a unique pattern. A system that exploits CSI for detecting and monitoring human activities uses commercial Wi-Fi devices [24]. TTW human presence sensing systems use the Wi-Fi signals for moving and stationary people with a single Wi-Fi access point (AP). In this research, researchers have carried out the experiments in an empty room, in which a person moves or is stationary, and the channel frequency response (CFR) is analyzed for human activity [25]. Device-free-based solutions use Wi-Fi devices, which are generally available in homes and offices, to extract fine-grained CSI for analysis of human activities [26]. An untrained human vitality detection platform has been proposed that relies on basic Wi-Fi infrastructure to detect human movement in real-time. This system does not require any human effort to train offline or to calibrate manually. The platform can continuously monitor human activities for various purposes [27].
A wireless occupant activity recognition system (Wi-OAR) was developed for building management systems (BMS) to create user-friendly real-time environments for residents. The CSI extraction method based on Wi-Fi signals provides contactless user-centered services in offices to work intelligently. The fast and robust target component separation (FRTCS) algorithm is designed to evaluate both accuracy and time efficiency. This prototype was developed for different office environments with two commercial Wi-Fi devices [28]. A human activity recognition system used Wi-Fi signals to collect data from ten people doing sixteen different indoor activities. This system reduces costs and improves performance in various areas [29]. Wi-Motion used CSI data for extracting the phase and magnitude response to build the classification model for six diverse human activities [30]. A non-wearable and privacy-protective human activity detection platform used Wi-Fi signals for imminent smart buildings by extracting the images of wireless channel response [31]. The Wi-Fi-based system used deep learning algorithms with enhanced CSI features for human activity recognition [32].
Nowadays, Wi-Fi access points are very readily available everywhere, and the human presence between the access points provides a unique CSI. Machine learning is used Enhancing the monitoring system by using portable, flexible, and multi-functional SDR technology.  Design of non-contact smart sensing system for monitoring falls by extracting finegrained WCSI in the presence of walls.  Enhancing the monitoring system by using portable, flexible, and multi-functional SDR technology.  Intelligent monitoring accomplished by the use of machine learning algorithms. The optimal performance is evaluated by measuring classification accuracy, prediction speed, and training time for each algorithm.
The paper is structured as follows. In Section 2, related existing work on non-contact smart sensing for monitoring human activities by using Wi-Fi, radar, and SDR-based sensing technologies is provided for deeper insight. Section 3 provides an overview of the noncontact sensing platform used for development by exploiting the SDR technology. Section 4 is dedicated to the methodology used for extracting the WCSI data and building the classification model for physical activities. In Section 5, the accomplished results and their performance are presented. Lastly, Section 6 summarizes the performance of non-contact smart sensing using SDR technology, and future recommendations are suggested for improving the system.

Related Work
There are a lot of existing Wi-Fi technology-based platforms developed for human activities monitoring and detecting vital signs. The wireless sensor network (WSN) uses Wi-Fi technology to detect movements of the human body without a portable device in the operating area. Passive dynamic velocity moving people detection system (PADS) uses device-less detection to extract amplitude and phase information from the WCSI and exploit spatial diversity using MIMO systems; such a system uses commercial Wi-Fi devices to capture human body motions [23]. Human activity can be identified by the reflected Wi-Fi signals from the human body to create a unique pattern. A system that exploits CSI for detecting and monitoring human activities uses commercial Wi-Fi devices [24]. TTW human presence sensing systems use the Wi-Fi signals for moving and stationary people with a single Wi-Fi access point (AP). In this research, researchers have carried out the experiments in an empty room, in which a person moves or is stationary, and the channel frequency response (CFR) is analyzed for human activity [25]. Device-free-based solutions use Wi-Fi devices, which are generally available in homes and offices, to extract fine-grained CSI for analysis of human activities [26]. An untrained human vitality detection platform has been proposed that relies on basic Wi-Fi infrastructure to detect human movement in real-time. This system does not require any human effort to train offline or to calibrate manually. The platform can continuously monitor human activities for various purposes [27].
A wireless occupant activity recognition system (Wi-OAR) was developed for building management systems (BMS) to create user-friendly real-time environments for residents. The CSI extraction method based on Wi-Fi signals provides contactless user-centered services in offices to work intelligently. The fast and robust target component separation (FRTCS) algorithm is designed to evaluate both accuracy and time efficiency. This prototype was developed for different office environments with two commercial Wi-Fi devices [28]. A human activity recognition system used Wi-Fi signals to collect data from ten people doing sixteen different indoor activities. This system reduces costs and improves performance in various areas [29]. Wi-Motion used CSI data for extracting the phase and magnitude response to build the classification model for six diverse human activities [30]. A non-wearable and privacy-protective human activity detection platform used Wi-Fi signals for imminent smart buildings by extracting the images of wireless channel response [31]. The Wi-Fi-based system used deep learning algorithms with enhanced CSI features for human activity recognition [32].
Nowadays, Wi-Fi access points are very readily available everywhere, and the human presence between the access points provides a unique CSI. Machine learning is used Intelligent monitoring accomplished by the use of machine learning algorithms. The optimal performance is evaluated by measuring classification accuracy, prediction speed, and training time for each algorithm.
The paper is structured as follows. In Section 2, related existing work on non-contact smart sensing for monitoring human activities by using Wi-Fi, radar, and SDR-based sensing technologies is provided for deeper insight. Section 3 provides an overview of the non-contact sensing platform used for development by exploiting the SDR technology. Section 4 is dedicated to the methodology used for extracting the WCSI data and building the classification model for physical activities. In Section 5, the accomplished results and their performance are presented. Lastly, Section 6 summarizes the performance of noncontact smart sensing using SDR technology, and future recommendations are suggested for improving the system.

Related Work
There are a lot of existing Wi-Fi technology-based platforms developed for human activities monitoring and detecting vital signs. The wireless sensor network (WSN) uses Wi-Fi technology to detect movements of the human body without a portable device in the operating area. Passive dynamic velocity moving people detection system (PADS) uses device-less detection to extract amplitude and phase information from the WCSI and exploit spatial diversity using MIMO systems; such a system uses commercial Wi-Fi devices to capture human body motions [23]. Human activity can be identified by the reflected Wi-Fi signals from the human body to create a unique pattern. A system that exploits CSI for detecting and monitoring human activities uses commercial Wi-Fi devices [24]. TTW human presence sensing systems use the Wi-Fi signals for moving and stationary people with a single Wi-Fi access point (AP). In this research, researchers have carried out the experiments in an empty room, in which a person moves or is stationary, and the channel frequency response (CFR) is analyzed for human activity [25]. Device-freebased solutions use Wi-Fi devices, which are generally available in homes and offices, to extract fine-grained CSI for analysis of human activities [26]. An untrained human vitality detection platform has been proposed that relies on basic Wi-Fi infrastructure to detect human movement in real-time. This system does not require any human effort to train offline or to calibrate manually. The platform can continuously monitor human activities for various purposes [27].
A wireless occupant activity recognition system (Wi-OAR) was developed for building management systems (BMS) to create user-friendly real-time environments for residents. The CSI extraction method based on Wi-Fi signals provides contactless user-centered services in offices to work intelligently. The fast and robust target component separation (FRTCS) algorithm is designed to evaluate both accuracy and time efficiency. This prototype was developed for different office environments with two commercial Wi-Fi devices [28]. A human activity recognition system used Wi-Fi signals to collect data from ten people doing sixteen different indoor activities. This system reduces costs and improves performance in various areas [29]. Wi-Motion used CSI data for extracting the phase and magnitude response to build the classification model for six diverse human activities [30]. A nonwearable and privacy-protective human activity detection platform used Wi-Fi signals for imminent smart buildings by extracting the images of wireless channel response [31]. The Wi-Fi-based system used deep learning algorithms with enhanced CSI features for human activity recognition [32].
Nowadays, Wi-Fi access points are very readily available everywhere, and the human presence between the access points provides a unique CSI. Machine learning is used to extract CSI data to classify human movements [33]. Wi-Fi technology is becoming increasingly popular in mobile sensing devices for monitoring daily human activities [34]. A non-contact sensing-based Wi-Run system uses commercial Wi-Fi devices to estimate human steps [35]. The wireless detection uses 5G C-band technology to record falls and body movements of people with high precision [36]. Passive Wi-Fi sensing monitors health conditions including breathing rate and falls [37]. A passive Wi-Fi system detects twodimensional phase information for monitoring human falls [38]. Breathing and heart rate patterns are important indicators of a person's physical health. A commercial Wi-Fi-based system was designed to analyze the changes in breathing and heart rate patterns. This system is inexpensive and convenient for continuous monitoring of health conditions [39]. Wi-Fall is a real-time system used to monitor the sudden falls of persons living alone, especially in old age. This system detects the fall of the human in a non-contact manner using a commodity 802.11n network interface card (NIC). This system achieves high accuracy for the fall detection of a single person [40]. RT-Fall, a non-contact sensing system, used commodity Wi-Fi devices for fall detection. This system is inexpensive for monitoring daily activities without attaching any device to the human body [41]. The Res-Beat, a noncontact sensing system, used commodity Wi-Fi devices for monitoring real-time respiration rate. The system analyzed bimodal WCSI data for breathing abnormality information by detecting peaks to evaluate respiration rates [42]. Wi-Fi technology-based sensing has the advantage of being easily accessible and low cost, but having limitations of portability and flexibility due to the limited number of OFDM subcarriers and fixed standards.
Radar technology is also used to monitor human activities and detect vital signs in the existing literature. A system based on ambient radar has been proposed to detect human activity in an indoor environment. The 7.8 GHz operating frequency detected human activity by sending 16 pulses per second. This system can differentiate between human movements to recognize different activities [43]. The Bumble-Bee radar-based system can efficiently capture micro-Doppler signatures for human movements for recognition in indoor environments [44]. A wireless sensing approach used a passive-Doppler radar to detect human body movement's variations, recognizing abnormal respiration rate and various human physical activities to observe health condition. The wireless signals are used to detect human activity [45]. The radar technology detects large-scale body motions to improve the home life of older adults. This system classifies falls using radar spectrogram image data [46]. Radar technology is a promising solution, but has the potential risk of explosion due to released heat and is not used widely due to expensive hardware setup. Furthermore, the technology needs a line-of-sight (LOS) environment, i.e., no obstruction is recommended between radar and human, which limits the system's physical deployment.
Recently, SDR-based sensing technology has been used to detect human activities and vital signs using wireless signals. A device-free system using smart sensing recognizes different human activities by extracting WCSI in an indoor environment. Human body motions were detected in a real-time setting using SDR equipment [47]. Blueprints of WCSI present distinctive variations caused by body motions, characterizing small and large-scale motions. SDR-based sensing technology exploits radio wave signals to extract human body motion patterns [48]. The SDR sensing-based platform used a deep learning algorithm-based convolutional neural networks (CNN) model to detect ankle movements [49]. The SDR sensing-based, non-contact identification platform classifies weightlifting activities performed by humans [50]. SDR technology is portable, flexible, scalable, and has multifunction capabilities [19,22]. The existing literature can be helpful in developing a COVID-19 platform to monitor human body motion, resulting in the diagnosis of various health issues, and monitoring of human activities in a non-contact manner. A summary of classification performance of monitoring health and vital signs by using non-contact sensing technology Wi-Fi, Radar, and SDR is given in Table 1. Although Wi-Fi, Radar, and SDR technologies are viable solutions for monitoring physical activities during the quarantine period, there are still limitations. In this research, we exploit SDR technology to overcome the limitations of Wi-Fi and Radar technology. The cost of SDR technology is low because it can be redefined through modification of software without changing or adding a new hardware setup. It is flexible because it can adopt any wireless standard by redefining software. It is portable because of the self-generating abilities of radio signals, and has multiple functional capabilities that can be exploited such as Wi-Fi, Radar, GSM, FM radio, etc.

Platform
The platform contains computers, SDR devices, and omni-directional antennas. The computers used for experiments are Lenovo, Intel(R) Core (TM) i5-7500 3.40 GHz processor, 12 GB RAM and Windows 10 64-bit operating system. The SDR devices used for experiments are universal software radio peripheral (USRP) B210, and the software is MATLAB Simulink version R2019a. The main functional blocks for the platform's development are transmitter PC, transmitter USRP device, wireless channel, receiver USRP device, and receiver PC, as shown in Figure 1.

Transmitter PC
In transmitter PC operation, software-defined functionality is utilized to transmit a flexible OFDM frame. Initially, a signal of random bits is generated continuously at uninterrupted sample times and gets one channel per column. Data columns are buffered into frames by stipulating samples per frame. This input data signal used quaternary phase-shift keying (QPSK) digital modulation. Further, vector data was split into smaller subcarriers, and data of identical types concatenated to create contiguous output data. The inverse fast Fourier transform (IFFT) of all subcarriers is computed to transform frequency domain data into the time domain data with orthogonality between the subcarriers. A cyclic prefix (CP) is added to each data frame for avoiding inter-symbol interference (ISI). The adoptive gain is added to improve the strength of the transmitted signal. The softwaredefined hardware configuration block of USRP is used for flexible parameters modification, which is also the operation of the transmitter PC. The software-defined parameters are given in Table 2. These adjustable parameters can be redefined at any stage to improve the platform's performance.  Figure 1. Non-contact smart sensing system overview.

Transmitter PC
In transmitter PC operation, software-defined functionality is utilized to transmit a flexible OFDM frame. Initially, a signal of random bits is generated continuously at uninterrupted sample times and gets one channel per column. Data columns are buffered into frames by stipulating samples per frame. This input data signal used quaternary phaseshift keying (QPSK) digital modulation. Further, vector data was split into smaller subcarriers, and data of identical types concatenated to create contiguous output data. The inverse fast Fourier transform (IFFT) of all subcarriers is computed to transform frequency domain data into the time domain data with orthogonality between the subcarriers. A cyclic prefix (CP) is added to each data frame for avoiding inter-symbol interference (ISI). The adoptive gain is added to improve the strength of the transmitted signal. The software-defined hardware configuration block of USRP is used for flexible parameters modification, which is also the operation of the transmitter PC. The software-defined parameters are given in Table 2. These adjustable parameters can be redefined at any stage to improve the platform's performance.

Transmitter USRP Device
The transmitter USRP device functions are digital up-conversion (DUC), digital to analog conversion (DAC), low-pass filtering (LPF), mixer, and transmit amplification (TA). These functions of the hardware device are fixed and cannot be altered.

Wireless Channel
The wireless channel is a room environment to collect the human movements of standing, walking, running, bending, and falling activities, as shown in Figure 2. The WCSI signal is collected through multipath due to the reflection of the human body in-between the two omni-directional antennas.

Receiver USRP Device
The receiver USRP device functions are low noise amplification (LNA), mixer, LPF, analog to digital converter (ADC), and digital down converter (DDC).

Receiver PC
In the receiver PC, the USRP hardware flexible receiver configuration block is used to modify and control hardware. The frame synchronization process is used to detect when the frame begins and helps remove the CP correctly. The FFT is applied to transform the time domain data into the frequency domain data. The frame status conversion sets the sampling mode of the output data frame. The amplitude response of the data is extracted to analyze the WCSI data in the frequency domain. WCSI data is in the raw form, which is further preprocessed by cleaning, smoothing, and grouping. Additionally, features are extracted to transform the WCSI data for meaningful analysis. Finally, three machine learning algorithms are applied to classify falls separately from standing, walking, running, and bending.

Methodology
The various steps involved in developing a physical activity monitoring and fall de-

Receiver USRP Device
The receiver USRP device functions are low noise amplification (LNA), mixer, LPF, analog to digital converter (ADC), and digital down converter (DDC).

Receiver PC
In the receiver PC, the USRP hardware flexible receiver configuration block is used to modify and control hardware. The frame synchronization process is used to detect when the frame begins and helps remove the CP correctly. The FFT is applied to transform the time domain data into the frequency domain data. The frame status conversion sets the sampling mode of the output data frame. The amplitude response of the data is extracted to analyze the WCSI data in the frequency domain. WCSI data is in the raw form, which is further preprocessed by cleaning, smoothing, and grouping. Additionally, features are extracted to transform the WCSI data for meaningful analysis. Finally, three machine learning algorithms are applied to classify falls separately from standing, walking, running, and bending.

Methodology
The various steps involved in developing a physical activity monitoring and fall detection system by exploiting SDR technology and machine learning algorithms are discussed as follows:

Subject and Activities
In developing a physical activities monitoring system, we considered five healthy subjects performing multiple activities. The information about the subjects conducting the activities is given in Table 3. We considered standing, walking, running, bending, and fall activities for monitoring. Each subject performs each activity ten times.

Activities Data Collection
The activity data is collected in a small room, and the virtual experimental setup is shown in Figure 2. The distance between the antennas is 5 m and the subject performs an activity at the center position by moving his body. In standing, the position of the subject remains standing still; in walking, subject moves his legs slowly; in running, subject moves his legs quickly; in bending, subject bends his body; and in fall, the subject falls on the floor from a standing position. The reflection from the human body while doing different activities is collected as WCSI data.

Activities Data Extraction
The OFDM is used to extract fine-grained WCSI data at the receiver. The amplitudebased frequency response for each activity will be collected for 10 s. The information includes subcarriers, OFDM frames, and the time taken to perform the activity. Time and frames can be articulated as the number of frames received in a unit of time. The data sampling time is set based on the device sample rate by varying interpolation and decimation values at the transmitter and receiver, respectively. Each experiment frequency response H(jw) of WCSI data is expressed in Equation (1): 12  where k denotes the maximum number of OFDM subcarriers and s represents the total number of OFDM frames samples in a single experiment. The WCSI frequency response of a single OFDM frame can be expressed as in Equation (2): The WCSI frequency response of each subcarrier contains complex value data, so we expressed the amplitude information in Equation (3): |H(jω k )| is the amplitude of the kth subcarrier; amplitude information of WCSI helps identify the different activities performed by the subjects.

Data Preprocessing
The extracted data from WCSI is in a raw form, and it requires data preprocessing to get accurate, significant, and efficient analysis. In the first step, data is cleaned by removing and replacing missing or bad WCSI data. In the second step, the smoothing process is performed for removing noise by using low-pass filtering. In the final step, the grouping method is used to find correlations between the WCSI data values.

Features Extraction
The feature extraction method is helpful for the transformation of WCSI data, which translates collected WCSI data into significant trends in data. In addition, it is used for reducing the computation complexity and time by reducing the dimensions [51,52]. Therefore, feature extraction plays a crucial role in WCSI classification approaches. Presently, the statistical characteristics approach has been used for feature extraction. Statistical features used for developing the classification model are given in Table 4. Where the mean value gives information about the stable component of the signal, the standard deviation gives information about the degree of dispersion between the signal sampling points, the variance gives information about the fluctuations from the mean, the root mean square (RMS) value is a measure of the amplitude of a WCSI data, the peak-to-peak value is used for WCSI data amplitude range, the kurtosis is used to measure of the tailedness in the WCSI data, the skewness is used to represent symmetry of the WCSI data, the peak factor is used to detect whether there is an impact in the WCSI data, Interquartile range is used to obtain statistical dispersion and is equal to the difference between 75th and 25th percentiles, waveform factor is used to obtain the ratio of the RMS value to the average value of WCSI data, FFT functionality is used to extract frequency component with maximum and minimum values, spectral probability, signal energy, and spectrum entropy are used for the extraction of frequency domain analysis.

Classification
This research uses three popular machine learning algorithms to differentiate falls from other physical activities, and evaluates their performance. The accuracy of the machine learning algorithm depends on the type of dataset. The machine learning algorithms are used to develop models that predict physical activities based on WCSI data in the existence of uncertainty. These adaptive algorithms classify fall activity separately from standing, walking, running, and bending patterns by exploiting trends in the WCSI data. When the learning machine is trained to more experimental WCSI data, the processing machine improves its identification performance. All the experiment data are converted into a heterogeneous matrix. WCSI response data is a column vector where each row is labeled with the corresponding activity. The cross-validation (CV) model assessment technique is used to evaluate the performance of machine learning algorithms in making predictions on new WCSI datasets that have not been trained. We partition the known WCSI dataset, using a subset to train the machine learning algorithm and the left-over data for testing. A random 10-fold CV is used for original WCSI samples. These samples are randomly partitioned into 10 equal-sized WCSI data subsamples. Ninety percent of the WCSI data is used for training, while 10% is used for testing to develop a machine learning model. The accuracy of a model is used as a diagnostic measure to reflect the validated model results. The information about the conducted experiments is given in Table 5. Table 4. Statistical features expressions for classification.

Sr. No. Features Expression
Root mean square Interquartile range Frequency Max Signal Energy

Results and Discussion
The results are taken from the human physical activities experiments. The 64-subcarriers amplitude response of WCSI data is analyzed to detect physical activities. In Figure 3, the standing activity amplitude response of all the subcarriers in different colors is presented along the y-axis. The results show that WCSI amplitude response remains stable due to no human body movement in the standing activity over 8000 OFDM frames. In Figure 4, the walking activity amplitude response of all the subcarriers is presented in different colors along the y-axis. The results show that WCSI amplitude response varies slowly up and down due to slow human leg movement during the walking activity over 8000 OFDM frames. In Figure 5, the running activity amplitude response of all the subcarriers is presented in different colors along the y-axis. The results show that WCSI amplitude response varies rapidly up and down due to human leg movement during the running activity over 8000 OFDM frames. In Figure 6, the bending activity amplitude response of all the subcarriers is presented in different colors along the y-axis. The results show that WCSI amplitude response varies from top to bottom due to human upper body movement during the bending activity over 8000 OFDM frames. In Figure 7, the fall activity amplitude response of all the subcarriers is presented in different colors along the y-axis. The results show that WCSI amplitude response varies when the human body falls on the floor and then is stable during the fall activity over 8000 OFDM frames. Figure 4, the walking activity amplitude response of all the subcarriers is presented in different colors along the y-axis. The results show that WCSI amplitude response varies slowly up and down due to slow human leg movement during the walking activity over 8000 OFDM frames. In Figure 5, the running activity amplitude response of all the subcarriers is presented in different colors along the y-axis. The results show that WCSI amplitude response varies rapidly up and down due to human leg movement during the running activity over 8000 OFDM frames. In Figure 6, the bending activity amplitude response of all the subcarriers is presented in different colors along the y-axis. The results show that WCSI amplitude response varies from top to bottom due to human upper body movement during the bending activity over 8000 OFDM frames. In Figure 7, the fall activity amplitude response of all the subcarriers is presented in different colors along the y-axis. The results show that WCSI amplitude response varies when the human body falls on the floor and then is stable during the fall activity over 8000 OFDM frames.        The confusion matrix presents the performance of the algorithm for each class. The confusion matrix determines the areas where the algorithm has performed well or poorly. The rows show the actual class, and the columns show the predicted class. The diagonal values give optimal results where the actual class and predicted class match. The observations from the actual class and predicted for physical activities are shown in Table 6. The results were achieved by applying machine learning algorithms on WCSI data collected from the SDR technology-based platform for monitoring falls. The performance of the different algorithms is shown in Table 7, which includes algorithm accuracy in percentage, observations per second (obs/s) for speed prediction, and time taken for training in seconds. The fine tree algorithm is best for WCSI data to classify physical activities, with an accuracy of 99.7%. Although fine KNN is less accurate, its prediction speed is higher than other algorithms, with more observations in unit time and less time in training the model on WCSI data.  The confusion matrix presents the performance of the algorithm for each class. The confusion matrix determines the areas where the algorithm has performed well or poorly. The rows show the actual class, and the columns show the predicted class. The diagonal values give optimal results where the actual class and predicted class match. The observations from the actual class and predicted for physical activities are shown in Table 6. The results were achieved by applying machine learning algorithms on WCSI data collected from the SDR technology-based platform for monitoring falls. The performance of

Conclusions
In this research, smart sensing using SDR technology is exploited to detect falls during the quarantine period from other physical activities to reduce the chances of noncommunicable as well as communicable diseases. USRP hardware is used to collect realtime data TTW of human physical activities that take place between the two antennas. The fine-grained WCSI data is extracted using OFDM technology to develop machine learning models. The machine learning model efficiently classifies fall activity separately from other physical activities. The performance of machine learning algorithms shows promising results, with the fine tree producing a high accuracy result of 99.7%, prediction speed of nearly 72,000 obs/s, and training in almost 9 s. This proof of concept can be further investigated to detect COVID-19 symptoms like shortness of breath, coughing, and cardiac arrest issues by exploring the smart sensing SDR technology platform.
Currently, the whole world is fighting against the novel coronavirus (COVID-19) and limiting their physical activities over time. In the future, smart sensing using SDR technology can cover a larger area by increasing the system gain, sampling rate, and Multiple Input Multiple Output (MIMO) antennas. Multiple subjects' physical and sports activities can be recognized by extracting the signal reflection of each subject by examining the path of the reflected signals at multiple links. We can further reconstruct the signal profile of each subject as if only a single subject has performed an activity in the environment to facilitate multi-subjects' activity recognition. The phase response is another solution to recognize multi-user activities in the same environments by measuring phase delays. The wireless channel is robust in nature, and it is hard to predict responses under changing surrounding environments. The time and frequency domain feature extraction of WCSI data can be exploited for better recognition accuracy by deploying state-of-the-art deep and machine learning algorithms.

Institutional Review Board Statement:
The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Ethics Committee of Xidian University, China.
Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.

Conflicts of Interest:
The authors declare no conflict of interest.