Detection of Quasi-Static Trapped Human Being Using Mono-Static UWB Life-Detection Radar

Abstract

In practical situations such as hostage rescue, earthquake and other similar events, the ultra-wideband (UWB) life-detection radar echo response from the respiration motion of the trapped person is always quasi-/non-periodic in respiration frequency or very weak in respiration amplitude, which can be called quasi-static vital sign. Although it is an extremely difficult task, considering the economic cost, the detection ability of the traditional UWB life-detection radars with only a pair of transceiver antennas is desired to be enhanced for locating the quasi-static trapped human being. This article proposes two different detection methods for quasi-static trapped human beings through the single/multiple observation points, which corresponds to the single-/multi-station radar operating mode, respectively. Proof-of-principle experiments were carried out by our designed radar prototypes, validating the effectiveness of the proposed methods.

1. Introduction

The ultra-wideband (UWB) life-detection radar is widely used to indicate human beings trapped in obstacles in rescue applications [1,2,3,4,5,6]. However, indicating the trapped human beings from raw radar echoes is extremely difficult due to heavy environment clutter, high signal attenuation of the obstacles, non-periodic micro-motion and low reflectivity of the human body [7,8,9,10]. The existing vital sign detection algorithms are divided into two categories: fast Fourier transform (FFT)-based methods [11,12,13,14] and constant false alarm rate (CFAR)-based methods. The former mainly aims at detecting the quasi-periodic breathing signal, while the latter mainly aims at tracking the large-scale human movements. However, under normal circumstances, the behaviors of the human body also include non-periodic, small-amplitude movements, such as body shaking back and forth, and non-periodic breathing, causing the characteristics of the vital signs between quasi-static breathing and large-scale movements. Therefore, the FFT-based method often cannot accurately determine the target position, and the CFAR-based method is difficult to distinguish the target from the background. Generally, quasi-/non-periodic respiration frequency or weak respiration amplitude is considered as a quasi-static vital sign for the trapped human beings. Nowadays, the detection of non-periodic but non-weak micro-motion and the detection of quasi-periodic but weak micro-motion are the research hotspots and difficulties in the field of life detection. Detection of the quasi-static trapped human beings is widely needed and urgent in public security applications.

Recently, a large amount of research work on the detection strategies and radar systems in the field of vital signs detection has been reported.

Table 1. The maximum detection range of typical radars around the world.

Radar Name	Developer	Max Distance (m)	Characteristics
Xaver	Camero (Israel)	8	pulse radar, portable
LifeLocator TRx	GSSI * (USA)	12	pulse radar, earlier used
CE400	NovaSky (China)	20	FMCW radar, MIMO, bulky
Jia [24]	UESTC * (China)	5	SFCW * radar, dual-station
“RF-Capture” [22]	MIT * (USA)	3	FMCW radar, MIMO, 5.46~7.24 GHz
Xia [16]	IECAS * (China)	12.5	Pseudo-random radar, portable

* GSSI: Geophysical Survey System Inc., UESTC: University of Electronic Science and Technology of China, MIT: Massachusetts Institute of Technology, IECAS: Institute of Electrics, Chinese Academy of Sciences, SFCW: Stepped-Frequency Continuous Wave.

lists the main parameters and characteristics of radars published in commercial and academic papers. The novel radar transceiver signal designs are proposed to obtain the higher signal-to-noise ratio (SNR) [15,16,17]. In through-the-wall scenes, the maximum detection distances of the doppler radar proposed in Reference [15], the frequency modulated continuous wave (FMCW) radar proposed in Reference [17] and the pseudo-random coded radar are 10, 10 and 17 m, respectively. The multiple input multiple output (MIMO)-based method uses a real aperture of multiple transceiver combinations, realizing instant imaging resolution [18,19,20,21,22], which is widely used in through-wall imaging. The multi-view and multi-station radar systems proposed in References [23,24,25,26] can improve the efficiency of rescue in low SNR scenes by combining radar data in multiple observation points. In Reference [24], the proposed dual-station radar system provides a practical integrated approach for detecting and localizing stationary human targets. However, the improvement of detection capability through the optimization of transceiver signal is limited for the mono-static radar. Meanwhile, the balance between the radar sensitivity and radar false alarm should be considered in the system design. On the other hand, due to a large number of antennas in the MIMO system, the system is complex and bulky, which is not convenient for the practical field operations. Studies show that the novel multi-view system can achieve high detection capabilities and strong scene adaptability simultaneously.

To sum up, most of the existing studies are optimized in a single aspect, such as increasing the detection range, realizing radar imaging and distinguishing between obscured targets, etc. In addition, the research on the signal detection of non-periodic but non-weak micro-motion is still an open question. So, the development of an UWB radar vital sign detection system that can flexibly perform multi-scene detection is becoming a priority, motivating the design of innovative radar systems and detection strategies in a combined way, increasing the rescue efficiency.

In this article, we aim to propose an effective detection strategy dealing with general and practical cases when there exist quasi-static trapped human beings. A novel Golay complementary coded UWB radar prototype system was designed to verify the proposed detection strategy. In the strategy, two different operating modes, including stationary operating mode and scanning operating mode, are adopted. The stationary operating mode is suitable for detecting the quasi-static trapped human beings with non-periodic but non-weak respiration, using one radar to achieve single observation points. The scanning operating mode is for the quasi-periodic but weak respiration, using several radars to achieve multiple observation points.

In the stationary operating mode, a novel detection method based on the cross-correlation in slow time domain is proposed, which is suitable for the non-periodic but non-weak micro-motion life detection scene. The cross-correlation operations on the slow-time observation data of a single radar are performed in this proposed method, which can enhance and extract components with similar echo characteristics in the measured area, achieving the accurate detection of the non-periodic but non-weak micro-motion targets. In the scanning operating mode, the novel deployed UWB radar system for rescuing trapped persons can take advantage of the distributed radar configuration, and the SNR of vital sign is improved compared to that of the traditional case, achieving the accurate detection of the quasi-periodic but weak micro-motion targets.

In summary, this article presents a flexible detection strategy including stationary and scanning operating mode, to detect two typical vital signs effectively, which are quite common in the actual rescue environments. Based on the developed UWB radar system with wireless networking, the authors make an interesting attempt in the actual rescue scene and introduce two typical vital signs in different scenarios, indicating that the UWB radar system has better scene adaptability.

2. Material and Methods

2.1. The UWB Life-Detection Radar Hardware System Design

2.1.1. The Golay Complementary Coded UWB Radar

The UWB life-detection radar has obvious advantages in the ability of anti-interference and long-range resolution. The pseudo-random coded signal modulated radar transmits a pseudo-random coded signal and performs pulse compression on the radar echo in the receiver to obtain the impulse response function of the target, achieving large transmission power and high range resolution simultaneously.

The choice of pseudo-random coded signal is a key issue in the design of this system. The existing coding methods include m-sequence, Gold sequence and Golay complementary sequence. Based on the previous research on the pseudo-random codes’ technology [16,27], the Golay complementary sequence is composed of coded signal A and coded signal B, and the autocorrelation functions of these two codes can eliminate the side lobes remarkably. The amplitude of signal peak can be double. The working process of the Golay complementary coded UWB radar is that the radar sequentially transmits the Golay coded signal A and the Golay coded signal B for the electromagnetic detection. The sum of two autocorrelations of Golay coded signal A and Golay coded signal B constitutes the radar’s fast scanning time.

2.1.2. The Radar System

A novel Golay complementary coded UWB radar prototype system was designed to verify the proposed detection strategy. The prototype system contains several radars and a control host. The generation, transmission, reception of electromagnetic signal are integrated in the Golay-coded radars, while the human-machine interface, collaborative management, and data processing are handled by the control host. Each radar in the system has a full-featured transceiver, which can work alone at stationary operating mode. For the scanning operating mode, several additional modules would be activated for collaborative work, and the most important parts are network clock module and self-positioning module. By setting the time division multiplexing (TDM) architecture, the control host coordinates several radar nodes during the multiple observation points detection process to prevent mutual interference. All communication and control between radars are carried out wirelessly, and the number of observation points can be changed flexibly for different observation environments.

2.1.3. The Design of Single Golay-Coded Radar

The key components of the Golay-coded radar are a digital transmitter, a dual-channel receiver, a clock manager, a network clock, a self-positioning module and a wireless communication module, which are shown in Figure 1. The radar uses low sidelobes sinusoidal modulation Golay complementary coded signal with a center frequency of about 1 GHz as the transmit signal, which can improve the SNR [27,28,29,30]. To reduce the system cost and improve the spurious-free dynamic range (SFDR), the equivalent-time sampling technique is adopted. A Xilinx Artix-7 field-programmable gate array (FPGA) is used as the main control unit to manage the peripherals. A 16-bit analog to digital converter (ADC) with a maximal sampling rate of 160 million bits per second (Mbps) and a full-power bandwidth of 1.4 GHz is used for the receiver. A 14-bit digital to analog converter (DAC) at 2.5 Gigabit samples per second (GSPS) update rate is used for the transmitter. A W5300 chip is used for exchanging the raw radar data and commands with the control host by Transmission Control Protocol (TCP) network protocol, and the wireless bridge realizes the wireless communication. A pair of bow-tie antennas are used for electromagnetic radiation and reception. The prototyped radar system and its antennas is presented in Figure 2.

2.1.4. The Additional Modules for Scanning Operating Mode

In the scanning operating mode strategy, radars are deployed at different observation points with TDM architecture, executing electromagnetic detection in their own time slices. The network clock module and self-positioning module are adopted as additional modules to achieve the multiple observation points.

The network clock module avoids mutual interference between Golay-coded radars during the detection process via setting a time margin and executing wireless clock synchronization. Here, the module is designed based on Texas Instruments’ CC1310 chip and a counter in FPGA. The CC1310 provides a set of timers and taggers for radio operation by Sub-1 GHz technology. By sending and receiving synchronization requests twice, the modules can measure the difference value of the local network clocks between two Golay-coded radars via the two-way synchronization algorithm [31]. The measured values are used to correct the different initial offset. By executing wireless clock synchronization and setting the time margin as ~15 ms, different initial offset and clock drift between Golay-coded radars could be corrected.

The self-positioning module is deployed in each Golay-coded radar to sense the relative positions, as the relative positions are needed for the human respiration detection strategy. The automatic measure avoided manual measurement errors and time wasting. The DWM1000 is the critical component of the self-positioning module. It provides the function of timestamping and transmission times precise controlling and can be used in the two-way ranging with an error within 10 cm [32,33,34,35,36]. Here, the two-way ranging is realized by the double-sided two-way ranging algorithm executed in the STM32 Microcontroller Unit (MCU) to achieve the self-positioning of the Golay-coded radars.

2.1.5. The Workflow for Scanning Operating Mode

The orderly collaboration between Golay-coded radars depends on the reasonable time slice management and workflow management, which are designed in accordance with the radar’s parameters in

Table 2. Key parameters of the proposed radar.

Parameters	Value
Equivalent sampling frequency ($F_S$)	16 GSPS *
Real-time sampling frequency	125 MSPS *
Sampling points ($N$)	16,384
Average times ($N_A$)	32
ADC * Resolution	16 bits

* GSPS: Gigabit samples per second, MSPS: million samples per second, ADC: analog to digital converter.

.

In the system’s workflow, each time slice for the Golay-coded radar contains an electromagnetic detection and a data transmission. According to Table 2, the corresponding time consumption is about 17 ms for UWB A-scan and about 6 ms for data transmission. It means the minimum time slice is about 23 ms. Here, we set the time slice as 32 ms, and retained the time margin for wireless transmission and the network clock module.

To understand the orderly collaboration between radars, the workflow timing is presented in Figure 3. The scanning operating detection work starts from the radars and control host netted. After receiving the “START COMMAND,” the control host orders radars’ run network clock synchronization and relative position measurement, then calculates the topology’s rationality and allocates the time slice. The radars perform electromagnetic detection and upload data within their time slice. Finally, after 16 s of detection, the host calculates the target’s position and displays the results via human–machine interface.

2.2. Vital Sign Detection Using a Mono-Static UWB Life-Detection Radar

2.2.1. Background

For one UWB radar detection, the received signal $S\left(t\right)$ with the respiration of a human target can be expressed as:

$$S\left(t\right)={\displaystyle {\displaystyle \sum }_{n=0}^{N-1}}u\left(t-nT-{\tau }_r\right)\ast h_r\left(t\right) +{\displaystyle {\displaystyle \sum }_{n=0}^{N-1}}{\displaystyle {\displaystyle \sum }_{p=1,p\ne r}^P}u\left(t-nT-{\tau }_p\right)\ast h_p\left(t\right)+\gamma \left(t\right)+\omega \left(t\right)$$

(1)

where ∗ means convolution, $u\left(t\right)$ means the radar transmitted signal, $t$ means fast-time, slow-time is discrete with $nT$, on which the reflected signal is received, and $T$ is the adequate pulse repetition time (PRT). $h_r\left(t\right)$ means the respiration response of the human target and $h_p\left(t\right)$ means the joint impulse response of transmitting antenna, receiving antenna and $P-1$ static objects. ${\tau }_r$ and ${\tau }_p$ mean the propagation time-delay of the human target and the $p\mathrm{th}$ object, respectively. $\gamma \left(t\right)$ is the non-stationary inference and $\omega \left(t\right)$ is the additive white Gaussian noise.

To avoid frequency aliasing and range ambiguity, T should satisfy the Nyquist sampling theorem and ensure that all the reflected signals of targets are received during a PRT. All discrete signals of the reflected signals can be expressed as a two-dimensional (2D) (fast-time and slow-time) $M\times N$ matrix $S\left(m, n\right)$:

$$S\left(m,n\right)=h\left(m,n\right)+c\left(m,n\right)+\gamma \left(m,n\right)+\omega \left(m,n\right)$$

(2)

where M means the number of fast-time sampling points and $m=0,1,\dots ,M-1$, $N$ means the number of slow-time sampling points and $n=0,1, \dots ,N-1$. Every term in (2) corresponds to that in (1). Matrix $S\left(m, n\right)$ is named slow-time range matrix, where the column represents the range dimension and the row represents the slow-time dimension, respectively.

2.2.2. Two Types of Vital Signs

In this section, we introduce two types of vital signs, which are explained in detail below.

The first vital sign is defined as “non-periodic non-weak respiration”. When humans stand for a long time or are emotionally stressed, humans are likely to perform body shaking, irregular micro-movements and other behaviors, the corresponding radar echo signals show aperiodic and large amplitude characteristics. A radar slow-time range matrix including this type of vital sign is show in Figure 4a (in the red frame). Micro-motion response of the target can be seen obviously, but it is aperiodic. The target position calculated by the traditional FFT method has a large error, because the spectrum of aperiodic life signals is not constant. In contrast, the location of target can be estimated accurately by the proposed stationary operating mode.

The second vital sign is defined as quasi-periodic weak respiration. When a living body is unconscious or injured, the breathing amplitude tends to be decreased, and the corresponding respiration responses show the characteristics of quasi-period and small amplitude. The slow-time range matrix including this type of vital sign is shown in Figure 4b (marked in the red frame). The micro-movement of the target cannot be observed directly from the radar echoes, and it is difficult to effectively detect the target using a single radar. Due to the low signal-to-noise ratio, the target position cannot be estimated by the traditional FFT method. Notably, the scanning operation mode proposed in this paper can enhance the weak respiration response using the echoes from multiple radar observation points, improving the signal-to-noise ratio to obtain the target’s precise position.

2.2.3. Algorithm for Stationary Operating Mode

To remove respiration response from static objects around the human test subject, the time-invariant background must be estimated and subtracted from the measured data. Different approaches for estimating the static background have been proposed in Reference [37]. However, in cases where people are detected according to their micro-movement, respiration or cardiac activity, the reflected signals contained in measured radar echoes have the same time-delay and are always misinterpreted as static clutter using the background subtraction methods. Thus, people are almost invisible under these circumstances. To overcome this drawback, the adaptive background subtraction (ABS) method based on the exponential averaging and a vector of weighting coefficient $\alpha$ can be used [38].

$$p_n\left(m\right)={\alpha }_n\left(m\right)\times p_{n-1}\left(m\right)+\left(1-{\alpha }_n\left(m\right)\right)\times q_n\left(m\right)\left[M\times 1\right]$$

(3)

Here, $p_n\left(m\right)$ and $q_n\left(m\right)$ are one-dimensional (1D) vectors with the size and contain ‘sampled background estimation’ and ‘measured impulse response’ in the raw echo matrix $S\left(m, n\right)$, respectively. M is the number of fast-time sampling points and $m=0, 1, \dots ,M-1$, $N$ is the number of slow-time sampling points and $n= 0, 1, \dots ,N-1$. The new background estimation takes a fraction of the previous estimation and a fraction of the measured radar echo. The time-variant weighted coefficient $\alpha$, which has the size $\left[M\times 1\right]$, can be adjusted between 0 and 1 adaptively. According to Reference [38], $\alpha =0.95$ has a reasonable balance between background attenuation and micro-movement reservation.

According to (2), denote S_raw(k, m) as the slow-time range matrix obtained at a single observation point, where k = 0, 1,…, K−1 is the range cell index, and m = 0, 1,…, M−1 is the slow-time index. To eliminate the responses from static objects around the trapped human being, the time-invariant clutter/interference has to be estimated and subtracted from the slow-time range matrix, S_raw(k, m), using the pre-processing method, including the adaptive background subtraction, the linear trend suppression and the advance normalization (AN) [39]. The corresponding output slow-time range matrix is S(k, m), and it can be used to estimate the almost unchanged distance from the UWB life-detection radar for the quasi-static trapped human beings with non-periodic but non-weak respiration by the proposed detection method based on the ordered statistics constant false alarm rate (OS-CFAR) method. In Figure 5, we define x_m, q_m and z_m as the mth column vector, adjacent cancellation vector and background subtraction vector of S(k, m), respectively. Define b_m as the mth background, which can be estimated by b_m−1, z_m and the scale weighted factor $\mathbf{a}m$ as follows.

$${\mathbf{q}}_m = {\mathbf{x}}_m – {\mathbf{x}}_{m-1}$$

(4)

$${\mathbf{z}}_m = {\mathbf{x}}_m – {\mathbf{b}}_{m-1}$$

(5)

$${\mathbf{b}}_m = {\mathbf{b}}_{m-1} + (1 – {\mathbf{a}}_{m-1})·{\mathbf{z}}_m$$

(6)

$$\begin{gathered} {\mathbf{a}}_m \left(k\right)= 1 , k \in {\mathbf{s}}_m \mathrm{or} k \in {\mathbf{p}}_m \\ {\mathbf{a}}_m \left(k\right)= \mathrm{a} , \mathrm{otherwise} \end{gathered}$$

(7)

where k = 0, 1,…, K–1 is the range cell index. According to Reference [38],$\alpha =0.95$ is the empirical constant scale weighted factor [39]. s_m is the correlation output between the last output s_m-1 and the OS-CFAR output s_0m. Generally, with z_m being the input of the OS-CFAR method, s_0m might contain all the possible distances of the quasi-static trapped human beings away from the UWB life-detection radar. As the OS-CFAR method is the threshold-dependent method, some distances might be produced by the clutters or other interferences. Therefore, the correlation output s_m is designed to eliminate these wrong distances. The estimated distances in s_0m are assumed to be from the quasi-static trapped human beings in the case that the correlation between s_0m and s_m-1 equals to non-zero. The calculation of the correlation output p_m follows the same way as s_m:

$$\begin{gathered} {\mathbf{s}}_m\hspace{0.33em}(\mathit{i})\hspace{0.33em}=\hspace{0.33em}{\mathbf{s}}_{0m}(\mathit{i})\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\mathrm{s}.\mathrm{t}.\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\mathrm{E}\{{\mathbf{s}}_0{}_m\hspace{0.33em}(\mathit{i})\cdot {\mathbf{s}}_{m-1}(\mathit{i})\}\ne 0 \\ {\mathbf{s}}_m \left(\mathit{i}\right) = \varnothing \hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\mathrm{s}.\mathrm{t}.\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em} \mathrm{E}\{{\mathbf{s}}_{0m} (\mathit{i})·{\mathbf{s}}_{m-1} (\mathit{i}\left)\right\}=0 \end{gathered}$$

(8)

$$\begin{gathered} {\mathbf{p}}_m\hspace{0.33em}(\mathit{j})\hspace{0.33em}=\hspace{0.33em}{\mathbf{p}}_{0m}(\mathit{j})\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\mathrm{s}.\mathrm{t}.\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\mathrm{E}\{{\mathbf{p}}_0{}_m\hspace{0.33em}(\mathit{j})\cdot {\mathbf{p}}_{m-1}(\mathit{j})\}\ne 0 \\ {\mathbf{p}}_m \left(\mathit{j}\right) = \varnothing \hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\mathrm{s}.\mathrm{t}.\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em} \mathrm{E}\{{\mathbf{p}}_{0m} (\mathit{j})·{\mathbf{p}}_{m-1} (\mathit{j}\left)\right\}=0 \end{gathered}$$

(9)

where i = 1, 2,…, S is the index of s_0m and j = 1, 2,…, P is the index of p_0m. S and P are the number of the estimated distances in s_0m and p_0m, respectively. Finally, s_m is used to determine the locations of the single/multiple quasi-static human beings dynamically by an online clustering algorithm. Clusters representing the quasi-static trapped human beings are updated as streaming samples in s_m come in, and the center of each cluster is identified as the location of each quasi-static trapped human being.

2.2.4. Algorithm for Scanning Operating Mode

For the quasi-static trapped human beings with quasi-periodic but weak respiration, we denote S_i(k, m) and S_j(k, m) as the output slow-time range matrices obtained from the ith and jth (i ≠ j) observation points after eliminating the time-invariant clutter/interference, respectively. Due to the fact that the cross-correlation function of the non-periodic noise is prone to zero, the cross-correlation between S_i(k, m) and S_j(k, m) is applied to improve the low signal-to-noise ratio (SNR). However, the quasi-periodic component and its harmonic components contained in the slow-time signal S_i(k, m) and S_j(k, m) are still preserved. Assume that the size of S_i(k, m) and S_j(k, m) are K_u × M_a and K_v × M_b, respectively. The cross-correlation function R_ij(u, v, m) is defined as:

$${\mathit{R}}_{\mathit{ij}}(\mathit{u}, \mathit{v}, m) = \mathbf{E}\left\{{\mathit{S}}_{\mathit{i}}\right({\mathit{k}}_{\mathit{i=u}} , m_{\mathit{a}})·{\mathit{S}}_{\mathit{j}}({\mathit{k}}_{\mathit{v}} , m_{\mathit{b}}\left)\right\}$$

(10)

$${\mathit{R}}_{ \mathit{ij}}^{\mathit{AN}}(\mathit{u}, \mathit{v}, m) = {\mathbf{max}}_m\left\{{\mathit{R}}_{\mathit{ij}}\right(\mathit{u}, \mathit{v}, m\left)\right\}·{\mathbf{AN}}_m\left\{{\mathit{R}}_{\mathit{ij}}\right(\mathit{u}, \mathit{v}, m\left)\right\}$$

(11)

where the range cell index k_u ∈ [0, K_u – 1] and k_v ∈ [0, K_v – 1], the slow-time cell index m_a ∈ [0, M_a – 1], m_b ∈ [0, M_b – 1] and m = m_a – m_b. Then, the AN method is applied to R_ij(u, v, m) so that the weak quasi-periodic component of the quasi-static trapped human being can be further enhanced. We denote the output result of the AN method as R^AN_ij(u, v, m). The Fourier transform of R^AN_ij(u, v, m) is taken in each slow-time dimension, and the maximum corresponds to the quasi-static trapped human being, i.e., (u_imax,v_jmax) indicates the possible range location of the quasi-static trapped human being depending on the different ith and jth observation points:

$$(u_{\mathit{i}\max },v_{\mathit{j}\max }) = {\mathbf{argmax}}_{\mathit{uv}}\{{\mathrm{FFT}}_m\left\{{\mathit{R}}_{ \mathit{ij}}^{\mathit{AN}}\right(\mathit{u}, \mathit{v}, m\left)\right\}\}, \mathit{u} \ne \mathit{v}$$

(12)

3. Experimental Results

3.1. Experimental Results for Stationary Operating Mode

In the experiment shown in Figure 6, three human test subjects with non-periodic but non-weak respiration stood at 3, 5.5 and 10 m behind a wall, of which the thickness and the measured average dielectric constant are 24 cm and 4.93, respectively. The non-periodic but non-weak echo responses (marked by three red arrows) can be seen in the pre-processing slow-time range matrix depicted in Figure 7a. Due to the non-periodic, the vital sign features obtained by FFT (vital sign features (VSFs) marked by three red arrows in Figure 7b) are blurry, which is not suitable for the subsequent VSFs decision. However, as for the non-weak echo responses in amplitude, the proposed detection method based on the OS-CFAR method for stationary operating mode works well. As shown in the left part of Figure 8, the range locations of three human test subjects are detected and updated as slow time goes on. The conclusive range locations are determined by the online clustering algorithm (see the right part of Figure 8).

3.2. Experimental Results for Scanning Operating Mode

To comprehensively compare the improvement of the measurement sensitivity and specificity with that of previous studies, two laboratory experiments (named as the type-I and type-II experiments) and two practical experiments were carried out. Two Golay-coded radars were used to verify the proposed scanning operating mode. For simplicity, the two radars are labeled as No. 1 and No. 2 radars, respectively.

3.2.1. The Type-I Experiment

As shown in Figure 9, the type-I experiment was carried out in an office building, and two Golay-coded radars were placed on the same side of the 1 cm gypsum board, 1.8 m apart. The tested object was placed at the different distances from 5 to 51 m, with a 2 m interval.

Considering the uncontrollable breathing movement of personnel, we selected the metal plate beacon as the tested object to ensure the data consistency during the collection of radar echoes. To simulate the chest movement of the human body, the metal plate beacon was controlled by a screw stepper motor to drive the metal plate to vibrate periodically, as shown in Figure 10. In the type-I experiment, the metal plate beacon’s vibrating frequency was set as 0.28 Hz, and the amplitude was set as 1 cm.

Both the proposed “scanning operating mode” and the “FFT-based respiration detection method [29]” were applied to the collected radar echoes. The corresponding measurement results are shown in

Table 3. The measurement results of the type-I experiment.

Measurement No.	The Actual Distance between No. 1 Radar and Target	The Actual Distance between No. 2 Radar and Target	FFT *-Based Respiration Detection Method		Measuring Error		Scanning Operating Mode Method		Measuring Error
			No. 1 Radar Result	No. 2 Radar Result	No. 1 Radar Result	No. 2 Radar Result	No. 1 Radar Result	No. 2 Radar Result	No. 1 Radar Result	No. 2 Radar Result
1	5 m	5.31 m	4.85 m	5.41 m	−3.00%	1.80%	4.89 m	5.19 m	−2.20%	−2.34%
2	7 m	7.23 m	6.85 m	7.04 m	−2.14%	−2.60%	6.87 m	7.07 m	−1.86%	−2.18%
3	9 m	9.18 m	8.85 m	9.24 m	−1.67%	0.67%	8.89 m	9.27 m	−1.22%	1.00%
4	11 m	11.15 m	10.76 m	10.96 m	−2.18%	−1.67%	10.79 m	10.92 m	−1.91%	−2.03%
5	13 m	13.12 m	12.95 m	12.96 m	−0.38%	−1.25%	12.98 m	12.99 m	−0.15%	−1.02%
6	15 m	15.11 m	14.96 m	14.96 m	−0.27%	−0.98%	14.99 m	14.99 m	−0.07%	−0.78%
7	17 m	17.10 m	16.84 m	16.93 m	−0.94%	−0.97%	16.88 m	16.96 m	−0.71%	−0.79%
8	19 m	19.09 m	19.24 m	19.14 m	1.26%	0.29%	19.27 m	19.17 m	1.42%	0.44%
9	21 m	21.08 m	20.86 m	21.12 m	−0.67%	0.20%	20.89 m	21.15 m	−0.52%	0.35%
10	23 m	23.07 m	22.84 m	22.93 m	−0.70%	−0.61%	22.88 m	23.03 m	−0.52%	−0.17%
11	25 m	25.06 m	25.36 m	25.08 m	1.44%	0.06%	25.39 m	25.11 m	1.56%	0.18%
12	27 m	27.06 m	26.86 m	27.08 m	−0.52%	0.07%	26.89 m	26.89 m	−0.41%	−0.63%
13	29 m	29.06 m	28.91 m	28.98 m	−0.31%	−0.26%	28.94 m	29.01 m	−0.21%	−0.16%
14	31 m	31.05 m	31.27 m	31.06 m	0.87%	0.03%	31.3 m	31.09 m	0.97%	0.12%
15	33 m	33.05 m	33.01 m	33.17 m	0.03%	0.37%	32.95 m	32.63 m	−0.15%	−1.27%
16	35 m	35.05 m	34.96 m	34.96 m	−0.11%	−0.25%	34.99 m	34.99 m	−0.03%	−0.16%
17	37 m	37.04 m	37.27 m	Failure	0.73%	-	37.24 m	36.98 m	0.65%	−0.17%
18	39 m	39.04 m	39.18 m	Failure	0.46%	-	39.22 m	39.15 m	0.56%	0.28%
19	41 m	41.04 m	Failure	Failure	-	-	41.19 m	40.98 m	0.46%	−0.14%
20	43 m	43.04 m	Failure	Failure	-	-	43 m	42.98 m	0.00%	−0.13%
21	45 m	45.04 m	Failure	Failure	-	-	44.19 m	45.15 m	−1.80%	0.25%
22	47 m	47.03 m	Failure	Failure	-	-	46.55 m	46.55 m	−0.96%	−1.03%
23	49 m	49.03 m	Failure	Failure	-	-	48.85 m	48.3 m	−0.31%	−1.50%
24	51 m	51.03 m	Failure	Failure	-	-	50.87 m	51.32 m	−0.25%	0.56%

*FFT: fast Fourier transform.

. The detection errors of the experiment for No. 1 and No. 2 radars are shown in Figure 11a,b, respectively. The blue error bars represent the detection errors of single-radar detection results via the “FFT-based respiration detection method”. The red error bars represent the detection errors via the “scanning operating mode”. In addition, the blue ‘×’ indicates the failure detection case.

3.2.2. The Type-II Experiment

To study the influence of the angle between people and radar on the observation results, we designed the type-II experiment including three measurements of A, B and C, which were carried out in our institute’s stadium. The experimental setup is shown in Figure 12, and the two radars were placed on the same side of a 37 cm brick wall, 1.8 m apart.

In the measurement A, the person under test stays still and breathes normally, standing 22 m behind the wall and at the center of the horizontal line of the two radars. Define the center of the horizontal line of the two radars as the center of the system. The center of the system and the subject are on the same horizontal line, so the angle between people and radars is 0°. The “scanning operating mode” is used to detect the vital signal, combining the data from two radars.

Figure 13a,b shows the output range-frequency images of the normal FFT-based respiration detection methods using the echoes from a single radar. Due to the low SNR in this experimental scene, the output range-frequency images are too noisy to distinguish the vital sign feature (VSF).

According to our proposed “scanning operating mode”, the cross-correlated range × range two-dimensional (2D) image is shown in Figure 13c, and the VSF is remarkable. In Figure 13d, the positioning result calculated by the radar system is compared with the actual position of the person. The red circle represents the actual position of the person, the blue × represents the estimated location through the “scanning operating mode” and the positioning error is 46 cm.

In the measurement B, the tested person with static state and normal breathing stands 22 m behind the wall and moves horizontally to the left by 3 m. The angle between people and radars is 7.84°.

Figure 14a,b shows the output range-frequency images of the normal FFT-based respiration detection methods using the echoes from a single radar. Similarly, they are too noisy to distinguish the VSF. According to our proposed “scanning operating mode”, the cross-correlated range × range 2D image is shown in Figure 14c, and the VSF is remarkable. The comparison results of the positioning result calculated by the radar system and actual position of the person are shown in Figure 14d, and the positioning error is 25 cm.

In the measurement C, the tested person stands 22 m behind the wall and moves horizontally to the left by 5 m, also in a static state and breathing normally, and the angle between people and radars is 13.14°.

Figure 15a,b shows the output range-frequency images of the normal FFT-based respiration detection methods using the echoes from a single radar. Similarly, they are too noisy to distinguish the VSF. According to our proposed “scanning operating mode”, the cross-correlated range × range 2D image is shown in Figure 15c, and the VSF is remarkable. The comparison results of the positioning result calculated by the radar system and actual position of the person are shown in Figure 15d, and the positioning error is 21 cm.

3.2.3. The Ruins Experiment

In the actual ruin scene shown in Figure 16, detecting the tested human subject buried by real ruins and debris is carried out in the China National Training Base for Urban Search and Rescue. Two Golay-coded radars are placed on the third floor with the interval of 2 m. The tested human subject lies on the first floor while breathing weakly and quasi-periodically. The distances between him and the No. 1 and No. 2 radars are 3.7 and 3.59 m, respectively. There are various bricks, metals and woods in the test environment, especially on the second floor.

For the scanning operating mode, the UWB life-detection radar was placed on two different observation points. The VSF might not be found using the FFT results for the first (Figure 17a) and second (Figure 17b) observation points. However, as the quasi-periodic characteristic, the proposed detection method for scanning operating mode gives the workable result, which makes the subsequent VSF decision easy (Figure 17d). The range locations of D1 = 3.66 m and D2 = 3.60 m corresponding to the first and the second observation points are consistent with the actual situation.

3.2.4. The Multistorey Building Experiment

Another practical experiment was carried out in a multistorey office building to test the performance of the designed radar system. The two Golay-coded radars are placed on the ground of the fifth floor of a laboratory building (Figure 18a) and the tested objects are located at the ground of the fourth floor (Figure 18b). The height between floors of this building is 5 m. The tested people lay still on the fourth floor, breathing normally. The distances between the target and the radar No. 1 and No. 2 are 5.35 and 5.41 m, respectively.

Figure 19a,b shows the output range-frequency images of the normal FFT-based respiration detection methods using the echoes from a single radar. Due to the low SNR in this experimental scene, the output range-frequency images are too noisy to distinguish the VSF. According to our proposed “scanning operating mode”, the cross-correlated range × range 2D image is shown in Figure 19c. The resultant VSF is remarkable, showing that the distances between the target and the radar No. 1 and No. 2 are D1 = 5.30 m and D2 = 5.34 m, respectively.

4. Discussion

In this article, we presented an effective detection strategy, including stationary operating mode and scanning operating mode, dealing with general and practical cases when there exist quasi-static trapped human beings.

Based on the experimental results for stationary operating mode, the proposed radar system with the “stationary operating mode” shows excellent VSF detection performance for non-periodic but non-weak micro-motion targets.

Based on the type-I experiment, the conclusions can be summarized as follows:

(1)

 When the distance between the beacon and the Golay-coded radar is within 35 m, compared to the “FFT-based respiration detection method”, the “scanning operating mode” always gives the results with higher accuracy.

(2)

 When the distance between the beacon and the Golay-coded radar ranges from 37 to 39 m, the No. 2 radar fails to locate the target using the “FFT-based respiration detection method”. Notably, the “scanning operating mode” can avoid the misjudgment to give the correct result.

(3)

 The most attractive case: When the distance between the beacon and the Golay-coded radar is longer than 41 m, the “FFT-based respiration detection method” fails to provide the correct location of the target due to the two failure observation points, while the “scanning operating mode” can still work.

Based on the type- II experiment, for the target located 22 m behind the wall, using a single radar to detect the vital signs is prone to failures and false alarms. In contrast, using the “scanning operating mode” proposed in the paper can provide accurate positioning results.

Based on the two practical experiments, the proposed radar system with the “scanning operating mode” showed excellent VSF detection performance in practical complex rescue environments.

In addition, for the scanning operating mode, based on the wireless control, the novel deployed UWB radar system for rescuing trapped persons can take advantage of the distributed radar configuration, where several Golay-coded radars are deployed on a 2D observation aperture and controlled in the network to capture information from all possible wave incident and scattering angles. Specifically, for the mono-static UWB radar at one observation point, the transmitter illuminates the target area, and the receiver collects the scattered field depending on the position of the observation point and the current posture of the trapped person. On the one hand, the SNR of vital sign is improved compared to that of the traditional case. On the other hand, different from the traditional 1D range information obtained from the observation point, the 2D azimuth locations of the trapped persons can also be obtained from the echoes received at different observation points, which can improve the rescue efficiency remarkably.

The choice of the two modes introduced in this article depends on the actual application scenario. In practical applications, the environment is completely unknown for the operators. The traditional detection strategy using the single-station quasi-static mode or the FFT method are firstly selected to detect the area of interest. If the target cannot be detected, the new detection strategy described in this paper is worth trying. The scanning mode combining multiple radars and multiple observation echoes is selected to detect the potential targets in the measured area. If the target has not been detected, it may be indicated that there is no target, or the vital signal is too weak to be extracted.

5. Conclusions

Through the stationary or scanning operating mode of the mono-static UWB life-detection radar, the two different methods proposed can deal with the cases when there exist quasi-static trapped human beings. The stationary operating mode can enhance and extract components with similar echo characteristics in the measured area by cross-correlation in the slow-time domain, achieving the accurate detection of the non-periodic but non-weak micro-motion targets. The scanning operating mode can enhance and extract components with similar echo characteristics in the measured area by distributed radar deployment, achieving the quasi-periodic but weak micro-motion targets’ accurate detection. A set of experiments show that the system has excellent performance in terms of detection range, scene adaptability and target distribution. The author made an interesting attempt in the actual rescue scene, indicating that the UWB radar system has better scene adaptability. Experimental results demonstrate the effectiveness of the proposed methods for detecting the weakly or quasi-/non-periodic echo responses of the trapped human beings.