Detection of stationary humans using time‐division UWB MIMO through‐wall radar

Existing radars for stationary human detection need to process echo data from all channels simultaneously, which requires the radar system with high performance. Another limitation of existing technology is poor positioning due to sidelobe interference. Here, a time-division ultra-wideband multiple-input multiple-output (MIMO) radar is presented for stationary human detection using the incoherence of respiratory echoes. Based on the delay-and-summation imaging method, the maximum coherence factor (MCF) is weighted to suppress sidelobe, band-pass incoherence factor is weighted to distinguish stationary objects and human targets, and the variance factor is weighted to achieve accurate positioning. Experimental results show that this method can effectively suppress the noise, filter off the echo of the stationary object, and provide the distance and azimuth information of hidden human targets accurately.


Introduction
Technology for UWB radar imaging is highly desired in urban street fighting, disaster relief, and anti-terrorism reconnaissance, and so on [1,2]. Technologies of through-the-wall radar for human detection and recognition mainly use the delay time of echo signal caused by the human breathing and heartbeat to obtain the position information of hidden human. Primitively, single-frequency continuous wave radar is used to vital signs detection [3][4][5], which only have the ability to determine whether there is a human or not, but cannot provide the position information. As a solution, UWB radar has been proved to be capable of providing distance information with a high distance resolution. Both distance and respiratory information of hidden human are obtained by using single-channel ultra-wideband (UWB) radar [6][7][8][9][10]. However, single-channel UWB radars have no ability of horizontal resolution so that it is not suitable for detection and positioning of multiple human targets. In recent years, multi-channel UWB radar has significantly advanced. In [11], a two-dimensional antenna arrays are first used for human targets detection and location. Subsequently, various algorithms have been proposed in [12][13][14] to detect respiration and heart rate, providing distance and azimuth information effectively. Furthermore, the coherent characteristics and incoherent characteristics of synthetic aperture radar signals in [15] are used to distinguish stationary objects and human targets. However, these multi-channel UWB radars detection methods need to simultaneously process the echo signals of all channels, so requirements for the performance of the radar system are higher. There are also many deficiencies such as poor positioning and high sidelobe.
In this paper, a stationary human detection method using timedivision UWB MIMO radar is proposed. First, delay-andsummation (DAS) imaging is performed on all channel echoes. Second, maximum coherence factor (MCF) is used to eliminate sidelobe interference. We calculate the corresponding band-pass incoherence factor (BICF) and variance factor (VF) of the channel data to distinguish stationary objects and human targets. Finally, we calculate the product of the DAS image, MCF, BICF, and VF with appropriate weight to achieve imaging of stationary human targets with low sidelobe. This method has a better performance in human target positioning compared to the generalised incoherence coefficient imaging algorithm (GICF) proposed in [15].

Signal model
The scene is illuminated by a time-division UWB MIMO radar, assuming the number of transmit and receive antennas are M and N, respectively. The mth transmitting antenna transmits the impulse waveform x m (t) (with a peak point t 0 ), and the echo signal received by the nth receiver can be expressed as: where the first term represents echo signal of the wall, and τ w represents the round-trip propagation delay. The second term denotes the echo of stationary objects, and τ k0 is the propagation delay of the electromagnetic wave between the transceiver and the kth stationary object. The third term indicates the echo signal of human targets, and τ p, b indicates the propagation delay between the transceiver and the pth stationary human. σ w , σ k , and σ p represents the electromagnetic reflection coefficient, respectively. For a stationary human, the distance from the transceiver to the surface of human thorax changes over time due to the intrinsic respiratory micro-motion. Assuming that the respiratory rate is f b and the maximum displacement of the chest cavity is A b (usually <2 cm). Respiratory micro-motion is assumed to follow a sinusoidal function, so the actual propagation delay of electromagnetic wave is time varying. It is as follows: where R 0 denotes the distance between the transceiver and the centre of torso, and d w represents the thickness of the wall. It can be seen that the time delay caused by human respiratory motion is periodic. Therefore, there is a range deviation between signals received from different channels.
are used to distinguish human target from stationary object and provide the location of targets. It is worth noting that the value of variance at each imaging point needs to be self-adaptive threshold decision. Finally, DAS image, MCF, BICF, and VF are multiplied by a certain weight to obtain a stationary human target image.

DAS imaging
The imaging area is divided into equally spaced grids, and the pixel value for the ith grid (x i, y i ) is expressed as: where τ m, n represents the compensated delay. In (3), for a certain grid, all the echo signals from different channels are delayed and added.

MCF weighting
During the scan, the distance from stationary object to the transceiver remains unchanged, so there is a maximum value at t 0 in all echo signals after delay compensation, and the coherence of all the echo signals is very strong. However, the distance deviation among the echo signals of human target mentioned in Chapter 2 reduces their coherence. Therefore, the pixel value corresponding to human target is very small when using coherence factor (CF) weighting directly [15]. It is likely to be misjudged in this situation. So, it needs to be corrected as follows: where ℕ i represents a set including all values of the channel echo in the range of [t 0 − Δt, t 0 + Δt], and Δt is five times as much as the fast time sampling interval and max ( ⋅ ) represents the maximum value of the set. It can be seen in (4) that the coherence characteristics of the echo signals from stationary target and human are well utilised by MCF; so, the weighted image will be clearer.

BICF weighting
Electromagnetic wave propagation delay caused by respiratory micro-motion changes periodically with time, resulting in incoherence characteristic of echo signals reflected by human targets among different channels. Thus, stationary objects and stationary humans can be distinguished effectively. Let where s j (t 0 + τ j (x i, y i )) represents the value at t 0 of the jth channel echo after delay compensation. All values form a vector, marked as B, called the aperture data of corresponding imaging point. The Fourier transform (FFT) of B is its spectrum p(k), k = 0, 1, …, MN − 1. In [15], all the high-frequency components are used to calculate the incoherence factor (GICF). However, the human respiratory frequency is generally within the range of 0.1-0.7 Hz. In this paper, the ratio of the energy in this frequency band to the total energy is taken as the incoherent factor for a certain imaging point, BICF can be given by: where k 0 and k 1 are the frequency indexes corresponding to 0.1 and 0.7 Hz, respectively. It can be seen in (6) that the BICF only uses the incoherent characteristic within the human respiratory frequency range, highlighting the difference between echoes reflected by stationary objects and human targets. Thus, only stationary human targets are left in the weighted image.

VF weighting
Since there is still some clutter in k 0 , k 1 , BICF weighting mentioned in the previous section cannot achieve accurate positioning of stationary human targets. To solve this problem, we use the VF of B for further weighted imaging, as follows: In (7), we calculate the variance coefficients of all imaging points with an adaptive threshold decision, as shown in Fig. 2. Here, Q is the total number of imaging points in the imaging area, and E is the average value of all imaging points. According to experience, the discriminant factor s and q equals to 1.5 and 1.8, respectively.

Experimental results
The time-division UWB MIMO radar system and the simulated scene are shown in Fig. 3. The six horn antennas constitute antenna array with a length of 1.5 m. All antennas are evenly spaced at 0.3 m away from the imaging area 1 m. During scanning, one of the antennas is the transmitter, and the other five antennas receive signals alternately. It is equivalent to 30 channels in one cycle. Assuming that two cycles are taken, there are 60 channels in all. In Fig. 3, the thickness of wall is 0.2 m, and the relative permittivity is 6.4. The pulse width of the transmitted signal is 1 ns and the pulse repetition frequency is 1.5 Hz. In addition, it is assumed that the stationary object is located at (2, 2) m, and the stationary human target is located at (3,3) m at a respiratory rate of 0.3 Hz and the chest cavity amplitude of 0.02 m. The results are shown in Fig. 4. DAS beamforming is obtained by DAS operation on all channel echoes, as shown in Fig. 4a. It can be seen that both the stationary object and human target are imaged successfully, but the sidelobe interference is so severe that it is impossible to get the accurate location of targets. In order to reduce sidelobe and distinguish human target from stationary object effectively, Ref. [15] uses GICF of the echo signals to filter off the echo of the stationary objects and gets stationary human target imaging, as shown in Fig. 4b. The result of stationary human respiratory detection method proposed is shown in Fig. 4c. It can be seen from the comparison between Figs. 4b and c that both methods can effectively distinguish between stationary object and human target, but in terms of positioning accuracy of the stationary human target, the method in [15] is obviously lower than the method proposed in this paper.
To further verify the validity of this method, it is assumed that there are two stationary human targets located at (3, 3) m and (1.5, 2) m, respectively, and the stationary object located at (1, 1) m in the scene. Assuming that the respiratory frequencies of human targets are 0.5 and 0.3 Hz, respectively, and the amplitudes of the chest vibration induced by breathing are 0.03 and 0.02 m, respectively. Fig. 5b is the result of method proposed in [15]. It can be seen that the pixel value of human target located at (1.5, 2) m is very small; so, it is highly possible to misjudge it as a stationary object in this case. Fig. 5c shows the result of stationary human respiratory detection method proposed in this work. It is clearly that the use of VF can not only effectively distinguish object and stationary human targets but also provide the accurate location of human targets.

Conclusion
Using time-division UWB MIMO radar to process echo signal from different channels, we can reduce the amount of instantaneous calculations of the radar system to save detection cost. In addition, MCF weight suppress sidelobe effectively, BICF and VF weight distinguish human targets from other stationary object. Experimental results using simulated data show that accurate detection of stationary human targets can be achieved by this method.