UWB Bistatic Radar Sensor: Across Channels Evaluation

Device-free localization is a promising technique for indoor positioning in infrastructures, as it eliminates the reliance on custom and costly hardware. In this study, we present the development of a compact and cost-effective bistatic radar sensor that utilizes DW3000 ultrawide-band (UWB) transceivers. The effectiveness of the sensor was evaluated through comprehensive assessments of various channels, with particular emphasis on channel 9, employing channel impulse response analysis. To enhance the accuracy of time delay estimation for multipath components scattered by moving targets, we developed a weighting function. The experimental results demonstrate the potential of device-free localization sensors as a viable solution for indoor positioning systems, offering enhanced accessibility and cost reduction compared to traditional approaches.


I. INTRODUCTION
In recent years, there has been a growing emphasis on integrating positioning systems into Internet of Things (IoT) and Industrial IoT (IIoT) networks, mainly sparked by the technological advancements in compact and energy-efficient devices. Positioning systems can be broadly categorized into active [1], [2] and passive systems [3]. The former systems involve entities equipped with electronic devices that exchange information with the infrastructure to estimate their positions [4], [5]. In contrast, passive positioning systems involve entities that either do not carry electronic devices (for example, in healthcare applications [6], [7], [8], on-body sensors may not always be practical [9], [10]) or, if they do have electronic devices, do not actively contribute to the positioning algorithm and instead act as passive listeners. An illustrative example of the latter is [11], where the target determines its position by listening to messages broadcasted from the infrastructure, resembling a traditional GPS. Passive positioning systems, however, place a significant emphasis on device-free localization (DFL) systems, which involve entities without electronic devices [12]. While camera-based solutions [13], [14] are typical examples, their computational intensity, high data rates, and privacy concerns arise due to the requirement for high-resolution images. To address these limitations, acoustic and ultrasonic systems [15], [16] typically solve such drawbacks but utilize signals with distinct propagation characteristics and frequencies whose performance is easily corrupted by environmental noise. Alternatively, radio-frequency (RF)-based DFL systems have gained prominence as they monitor the target area by measuring changes in the radio channel spectrum leveraging received signal strength indicator, channel state information, or channel impulse responses (CIRs) that enable the detection and tracking of moving entities [17], [18]. By combining positioning and tracking systems in an IoT architecture, numerous applications have emerged in various domains, such as the RF-based DFL systems for detecting and tracking noncooperative or unauthorized targets in different environments [19], [20]. A comprehensive assessment of different technologies for DFL systems can be found in [21].
Within the standardized RF domain IEEE 802.15.4a, ultrawide-band (UWB) networks can provide detailed time-domain CIR that can be analyzed to identify the time location of multipath components (MPCs) scattered from the targets, hence facilitating the development of DFL systems, e.g., bistatic radar on low-cost commercial-off-theshelf modules. An early study in this domain for monitoring and safety applications is presented in [22], where the accumulated echogram of a multipath delayed signal is constructed. Stemming from this idea, the authors in [23] and [24] demonstrated how CIR measurements can be utilized to create a positioning infrastructure. Another study mentioned in [25], employed CIRs to train a convolutional neural network model for estimating target positions, while Cimdins et al. [26] performed a geometrical mapping of the CIRs to reveal reflection sources and derive the occupancy of passengers in the aeronautic field, thus building an IoT network bringing significant operational potentials, such as dynamic control of passenger boarding or disembarkation sequences under pandemic restrictions.
Building upon [24], the contributions of this letter are threefold.
1) A comparison between the previous generation of the QORVO UWB module and the newly released generation for radar applications address the lack of information in the existing literature and contribute to the advancement and optimization of UWB radar systems. 2) A weighting function is introduced to enhance the estimation of the time location of MPCs scattered from the target. 3) An IoT architecture efficiently handles and organizes data for the selected UWB modules. The rest of this letter is organized as follows. Section II presents the algorithm and the weighting function to estimate the time location of MPCs scattered from the target. The hardware involved in the experiments and preliminary results are described in Section III. Finally, Section IV concludes this letter.

II. METHODOLOGY
The UWB Standard IEEE 802.15.4a [27] utilizes a radio signal composed of a sequence of short pulses, where the signal's bandwidth determines the pulse duration. Upon receiving the preamble of a packet, the system accumulates the symbols, reconstructing the power envelope of the received signal. The data stored in the accumulator serves two primary purposes: calculating timestamps and analyzing the CIR, compared to the transmitted signal in the time domain. The selected UWB module provides CIR measurements with a resolution of τ s = 1/(2B s ), where B s = 499.2 MHz is the signal bandwidth. It also estimates the time delay of the first path (i.e., the time delay between transmitter and receiver), denoted as τ FP , within each CIR measurement with a resolution of τ s /64. In the time domain, the range resolution R = c/2B s determines the ability to distinguish between targets and depends on the signal bandwidth, resulting in R = 0.3 m or τ s = 1.0016 ns in the time domain [28].

A. Channel Impulse Response Reconstruction
Radio signals exhibit various propagation phenomena, such as signal reflection, diffraction, and scattering, which affect their transmission. The resulting signal y(t ) is represented by the convolution of the CIR h(t ) with the input signal s(t ) corrupted by a Gaussian white noise process ε(t ), which follows a normal distribution ε(t ) ∼ N (0, σ 2 ), i.e., y(t ) = h(t ) * s(t ) + ε(t ). The temporal behavior of the communication channel is described by the CIR h(t ) = N n=0 a n δ(t − τ n ), that is a series of N shifted, attenuated, and overlapped UWB pulses δ(t − τ n ), where a n , τ n represent the amplitude and path delay of nth MPC, respectively, and δ(·) the Dirac delta function. The perfect periodic autocorrelation property of the preamble sequences allows the receiver to accumulate repeated preambles and accurately determine the channel impulse response between the transmitter and receiver [22].
As depicted in Fig. 1(a), in a stable environment, the complex envelope of the received signal (i.e., h(t )) remains constant but sampled at different time instants [22].
Consequently, once the initial CIR measurement is obtained, the signal can be discretized up to B = 64 bins based on the available timestamping resolution. By aligning subsequent CIRs measurements relative to the time delay of the first path τ FP , each sample within the CIR can be assigned to the appropriate bin, thus acquiring B times more samples and reaching a subnanosecond scale.
To compute the time delay τ MPC of the MPC component scattered from the moving target, the algorithm track changes in the CIR variance that is represented as a piecewise constant function with coefficients h l that are time-updated as  where l ∈ {0, 1, . . . , B − 2} is such that τ l ≤ τ j ≤ τ l+1 , F β ≥ 1 is a constant scalar sensitivity factor, |Y i | is the innovation [see Fig. 1(b)] and the superscripts F · and · (k) stands for the foreground and time instant. The background variation coefficients B h l are updated using (1) as well. During the initialization phase, the parameter B β equals F β to expedite the convergence time for building the background model. Once the background model is reconstructed, the parameter B β is adjusted to be much smaller than F β. By computing the difference metric s l = F h l − B β B h l , it is possible to identify segments in the CIR that exhibit higher-than-usual variances within the intervals τ l ≤ τ ≤ τ l+1 , where l is chosen such that s ≥ 0. Denoting with s i the ith sample of the difference metric, its value is normalized by the maximum value of the CIR s max (given by the direct path, i.e., the background), by the number W of consecutive values of s j ≥ s i (a larger window size W implies the presence of multiple unaccounted multipath generated by the target) and by the ratio between the peak value s peak (occurring when an object enters the scene and creates direct multipath) and s max , i.e., This difference is related to the time difference between the time delay of the first path and the time delay of the detected MPC τ MPC , i.e., τ T = τ FP − τ MPC , which identifies the target's time of flight (ToF), as reported in Fig. 2.

III. RESULTS
To evaluate the effectiveness of our method, we designed an experimental setup consisting of a single transmitter-receiver pair aligned in the same direction, as depicted in Fig. 3(a).
Both the UWB transceivers were configured to sample the surrounding environment at a frequency of approximately 125 Hz.

A. Hardware
The DWM3120 is a second-generation, fully integrated UWB transceiver developed by QORVO as part of their DW3000 family. It follows the IEEE 802. 15.4z and has the new communication channel 9, operating at a carrier frequency of 7987.2 MHz. Moreover, it is optimized for low-power battery-operated operation, making it suitable for a wide range of mobile, consumer, and industrial applications. To ensure smooth integration into networks, we have developed a prototype whose hardware architecture enables seamless connectivity, as depicted in Fig. 3(b). The prototype is endowed with a Espressif ESP32 microcontroller, a cost-effective and energy-efficient device that comes with built-in WiFi and Bluetooth functionalities; a TP4056 handling the charge of the batteries; a 3.7 V 1900 mAh LiPo battery; two 5 V 1 W photovoltaic panels.

B. Preliminary Experimental Results
The experiments consider four baselines bl = (1.3, 2.3, 3.3, 4.3) m between the transmitter and the receiver, the communication channel 5 and 9, and two transmission power levels 28.6 dBm/MHz and 24.6 dBm/MHz, while various motion patterns are considered, as depicted in Fig. 3(a). The experiments presented in Fig. 4 report the results obtained from conducting experiments with channel 9 using both transmission power levels and two different baselines while performing a "Z"-shaped pattern. Based on these preliminary results shown in Fig. 4, the choice of the baseline impacts the target distance estimation. Note that for smaller baseline values, the distance estimation significantly deteriorates. We have determined the optimal baseline for our experimental setup through these experiments. This baseline will serve as the reference for comparing and evaluating the performance with channel. The evaluation is done in a structured indoor environment by using eight Qualisys Arqus A9 high-performance cameras. 1 Fig. 5(a) illustrates the comparison between the UWB and MoCap traces, while Fig. 5(b) displays the histogram representing the error. Note that the point tracked with motion capture does not align with the virtual point tracked by the UWB radar (i.e., the human body). Consequently, this discrepancy introduces a systematic error in the evaluation, which can be quantified to be approximately 10 cm. For the remaining experiments, raw traces are depicted to provide a comprehensive overview of the experiment's dynamics. A qualitative comparison confirms the improved estimation on channel 9, showing a more stable signal. Channel 5 produces the worst outcomes compared to channel 9, primarily because channel 9 is less susceptible to RF 1 [Online]. Available: https://www.qualisys.com/cameras/arqus/#!#tech-specs  noise in the indoor environment (e.g., WiFi), owing to its higher center frequency. Better results are achieved when the default power transmission (i.e., 24.6 dBm/MHz) is utilized. The increase in transmission power amplifies the amplitude of the noise, leading to more significant fluctuations in the estimated τ T value. It is important to note that these fluctuations are not solely caused by the estimator but also arise from the variability in timestamping due to internal circuitry sources and the outcome of the leading edge algorithm. Finally, the weighting function (2) greatly enhances the stability and reliability of target distance estimation as compared to our previous work [24] (see Fig. 6).

IV. CONCLUSION
DFL holds great potential for facilitating the widespread adoption of indoor localization infrastructures. However, existing solutions often suffer from limitations, such as reliance on custom or expensive hardware, which hinders their scalability. This study evaluates the advancements offered by a compact and cost-effective radar node system that utilizes DW3000 UWB transceivers for device-free positioning to address these challenges. The results obtained from the experiments demonstrate the effectiveness of the newly developed hardware, particularly in channel 9 with 24.6 dBm/MHz transmission power. The evaluation is conducted using CIR analysis and a bistatic radar configuration. Furthermore, integrating a weighting function has significantly improved the estimation stability and accuracy of τ T . In future research, we will focus on exploring UWB in multitarget scenarios. In this experimental configuration, we consider the presence of a single moving target. However, the same methodology can be applied to identify multiple τ T values within the measured CIR. Furthermore, we intend to investigate its innovative use in robotics, where UWB can serve as a ToF sensor, enabling the development of advanced obstacle avoidance systems.