**1. Introduction**

The contactless monitoring of heart rate has numerous advantages over conservative methods that use contact sensors, such as electrocardiogram (ECG) monitors, conventional photoplethysmography (PPG) sensors or piezoresistive sensors [1]. Contactless sensors offer improved mobility and obviate the need for attaching or cleaning electrodes, but also have the unique ability to be used on patients who suffer from skin irritations, painful skin damage like lacerations or burns, as well as patients who exhibit anxiety or allergic reactions to contact sensors. Furthermore, some contactless instruments, such as radar-based sensors [2], can be used for heart rate monitoring through clothes or other obstacles.

The real-time operation of heart rate monitors is required for the timely detection of potentially dangerous conditions in hospitals or in-home health care applications. The heart rate and its variability can be used for emotion, stress [3,4] or drowsiness detection [5] and real-time operation is often necessary for these applications.

In recent years, significant progress has been made in the development of radar-based heart rate monitors [6–32]. The potential for the production of compact low-power sensors, which are completely non-obstructive and harmless to human health, placed radar technology as one of the most promising options for contactless vital signs monitoring. Radar sensors are used for the detection of sub-millimeter movements of chest wall skin surface that occur due to heartbeats, whereas various signal processing methods are employed for heart rate extraction from discretized radar signals. Radar technology has shown not only great potential for heart rate estimation but also the potential for extracting ventricular ejection timing using nonlinear filtering methods [33].

The most frequent radar architectures used in heart rate estimation sensors are continuous-wave (CW) Doppler radars [6–24], frequency-modulated continuous-wave (FMCW) radars [3,25], and impulse radio ultra-wideband (IR UWB) radars [26–30]. CW Doppler and FMCW radars mostly outperform IR UWB radars in terms of power consumption and sensitivity [2]. The tracking of fine chest wall motion can be obtained by measuring the phase shift of the reflected signals of continuous-wave radars. The higher the frequency of the transmitted radar signal, the higher sensitivity can be obtained. While FMCW radars can detect both the absolute and relative displacement of the chest wall surface, CW Doppler radars are only capable of tracking relative displacement. This means that FMCW radars could be applied in multiple person heart rate estimation [25]. However, CW Doppler radars have a simpler hardware architecture and lower power consumption, and in single person applications, the relative displacement information obtained by the CW Doppler radar can be enough for a good heart rate estimation.

Many research groups have extensively investigated the monitoring of heart rate using CW Doppler radars. Most of the previous research was based on the experimental data monitored in studies with healthy participants lying or sitting in a controlled environment. The early published methods were based on (1) the simple filtering of heartbeat-related signals and applying a threshold to the filtered signals for extraction of heartbeat locations [6–8,10,12], or (2) heart rate frequency estimation using spectral analysis [9,13,15–17]. These approaches were hardly capable of fast and real-time performance and high-accuracy estimation at the same time. Simple band pass filtering would provide small latency, but the filtered output signals need further processing in order to automatically extract heartbeats. The robustness of these methods is hence very limited. The research in [12] showed that the error of the heart rate extraction can be drastically increased just when the subject changes their position from supine to still sitting. On the other hand, the frequency domain approaches would need a long window (5–30 s) of data for achieving sufficient frequency resolution for the detection of the heart rate harmonic. Additionally, they usually focus on high accuracy harmonic extraction and do not necessarily offer methods for distinguishing the heart rate harmonic from breathing and intermodulation harmonics. When the heart rate harmonic extraction is applied, the achieved accuracy is moderate (mean relative error of heat rate estimation around 10% [15]). Additionally, the testing set was limited to data recorded on a small number of human subjects (1 or 2, except in [12] where 10 subjects participated in the study).

However, more recent studies used data from more subjects (up to 10 participants in sitting position) and presented promising results in terms of detection accuracy. In [18], the ensemble empirical mode decomposition (EEMD) was used for the extraction of heartbeat information and in [22] the autocorrelation and frequency-time phase regression (FTPR) provided an algorithm robust to noisy conditions, but both of them used relatively long data windows for the heart rate assessment (10–15 s), which produced a large delay.

High-accuracy approaches capable of real-time operation with CW Doppler radar architectures, which achieve a relatively small delay, have recently been presented [19–21,23,24]. Authors in [19] and [20] used a dynamic variation of the time window for processing via the fast Fourier transform (FFT) [19] or the Wavelet transform (WT) [20]. In [21], the polyphase-basis discrete cosine transform has been used for heart rate estimation. All these methods improved heart rate detection accuracy in the frequency domain when shorter data windows were used (2–5 s). Specific heartbeat signal has been obtained using the short-time Fourier transform (STFT) analysis in [23], which was further

filtered through an adaptive band pass filter for improved quality. The control of the adaptive band pass filter was done using the information extracted from the time domain analysis of the heartbeat signal on windows of 2–3 s. In [24] it has been shown that the analysis of the frequency domain only did not give satisfactory results. Therefore, the heart rate information was extracted using frequency domain analysis (window length: 3.5 s) for the coarse estimation and time domain processing using a band pass filter bank for the refinement of the results. This approach resulted in small algorithm delay (~2.5 s) and high accuracy.

Recently, new approaches based on supervised or unsupervisedmachine-learning algorithms [31,32] were introduced in the CW Doppler radar systems, and the first results have shown promising advantages in terms of heartbeat detection delay and source separation capabilities (robustness of heartbeat detection to respiration motion or random body motion) compared to traditional approaches. Convolutional neural networks (CNN) were applied in [29] to estimate heart rate from UWB radar signals. However, due to the lack of training data, this approach was person-specific since the CNN needed to be trained for each subject separately. To the best of our knowledge, there is no published paper that used artificial neural networks for heartbeat detection using the CW Doppler radar technology. This paper focused on the development of a system for instantaneous heart rate estimation (delay of less than 1 s from heartbeat occurrence) using a shallow artificial neural network (ANN) as a main signal processing element. Additionally, the goal was to develop a detection algorithm that was person-independent. The contribution of this work is in the development of the system for detecting individual heartbeats considering the following requirements: (1) a low-complexity time domain-based algorithm (without relying on periodic occurrences of heartbeat-related chest displacements as in the case of traditional spectral approaches), (2) suitable for real-time human presence detection, (3) calibration-free (no need for I/Q imbalance, offset compensation or the usage of any demodulation techniques) and 4) testing on a separate group of subjects from those whose heart rate signals were used in the ANN model selection and training process.

## **2. Materials and Methods**
