Study of Reﬂection-Loss-Based Material Identiﬁcation from Common Building Surfaces

—Perceiving and recognizing material properties of surfaces and objects are fundamental aspects of new and emerging use cases such as robotic perception, virtual reality (VR) applications, digital twins, and creating a 3D digital map of an environment. In this paper, we present results from our simulation based study of reﬂection loss based material identiﬁcation from eight common building materials. The study focuses on 2.6 GHz, 28 GHz, and 60 GHz radio carrier frequencies. Analysis of simulation results indicates that combination of incident angle and reﬂection loss can be used to properly identify the common building materials. We therefore propose a novel joint communication and sensing method for material recognition using reﬂection loss of the radio signal by the scatterers around the propagation path in a wireless communication network. Com-pared to existing material identiﬁcation methods, the proposed reﬂection-loss-based method is capable of identifying materials from a signiﬁcant distance without requiring any contact with the object and without requiring dedicated sensors from the infrastructure point of view.


I. INTRODUCTION
Material recognition is the process of identifying material of an object and is a fundamental aspect of many new and emerging use cases such as robotic perception, virtual reality (VR) applications, digital twins, and creating a 3D digital map of an environment. By identifying the material of surfaces and objects, robots can behave more intelligently during manipulation or handling of the objects, for example, by recognizing the objects' fragility robot can adjust the grip of the arm appropriately. VR systems can activate additional human senses (e.g., touch and smell) by using material information in a 3D digital map for emerging 6G use cases. For example, this supplemental material information can find applications in VR games, and emerging area of virtual tourism when travel disruption due to pandemic like COVID-19 is to arise.
Some of the existing material identification approaches rely on haptic recognition and spectroscopy methods. Hapticrecognition-based method requires direct contact with the object using haptic sensor and relies on pressing and sliding  interactions of the sensor with the object to determine its material properties as shown in Fig. 1 [1]. In contrast to the haptic-recognition-based methods, spectroscopy-based methods do not require physical interaction between the object and the material detector but rely on interaction between object and electromagnetic radiation (as a function of the wavelength of the radiation) to recognize the material as shown in Fig. 2 [2]. In [2], material sensing of common objects such as wood, metal, plastic, paper, and fabric using spectroscopy was proposed. Here, authors proposed a neural network employing support vector machines (SVM) to separate the objects based on the spectrogram data. Materials and their properties can also be implied by analysing the structure of the molecules. This can be done by studying the vibrational modes and rotational frequencies. These are typically measured using timedomain spectroscopy (TDS). These methods require complex time-domain pulse generation equipment and sensors. In [3] authors proposed methods for drug property identification using terahertz (THz) TDS.
When it comes to using these methods in use cases such as robotic perception, VR applications, 3D mapping, and digital twins for material identification, haptic-sensor-based methods will require a dedicated hardware and spectroscopy-based methods will need object to be at least in centimetre level proximity to the detector for effective identification of the material. Therefore, both methods are not quite feasible when it comes to using them for realistic use cases where material detection plays a vital role.
To overcome the drawbacks of existing methods, in this paper, we propose a novel method that exploits radio signal that are typically transmitted in a wireless network for communication also for material identification. It is thus an example of joint communication and sensing (JCAS). Sensing here refers to information retrieved from received radio signals for objects to be identified in the environment surrounding the radio transceivers [4]. In 5G new radio (NR), reference signals such as demodulation reference signals (DMRS), channel state information reference signal (CSI-RS) in the downlink and sounding reference signal (SRS) in the uplink are used for channel estimation and can be optimized for the JCAS purpose [5], [6]. In [7], simulation results show that DMRS signal can be used as a primary signal for sensing and can provide decent accuracy. Apart from these reference signals, synchronization signal block (SSB) for cell search can be yet another candidate signal for radio access technology (RAT) JCAS. In general, it is evident that reference signals that are known to the receiver can be used for sensing [6].
Since its inception RAT has been extensively used to provide seamless connectivity between communication devices for high capacity and secure communication by means of wireless radio signal transmission and reception. When a radio signal is transmitted between wireless transmitter and receiver devices, the radio signal is affected by the presence of multiple scatterers in and around the propagation path. The radio signal typically undergoes phenomena such as reflection, diffraction, and refraction depending on the type and nature of the obstacles it encounters while travelling from a transmitter to a receiver. These phenomena typically results into multipath propagation of the radio signal, where the number and intensity of multi-path propagation depend heavily on the number of scatterers and their locations in the propagation path together with the radio signal waveform characteristics such as bandwidth and carrier frequency. Evaluating the strength of the signals that arrive at receiver via scatterers induced by multiple paths can therefore be used to establish an understanding of the type of scatterers that are located in and around the radio signal propagation environment. In this paper, we particularly propose a method that exploits the signals received at the receiver via a specular path to identify the material of the scatterer that are located in and around the radio signal propagation path. No additional hardware is needed for proposed method, material identification can be enabled on current wireless communication infrastructure. Furthermore, the object does not have to be in centimetre level proximity for effective material detection.
The common building surfaces like floors, exterior walls, windows, and ceilings are assumed to be electrically smooth at frequencies ranging up to mmWave because their surface height variations are smaller than the radio signal carrier wavelength [8]. At lower frequency bands, diffuse scattering is negligible and the reflection process is dominated, most energy is reflected to specular propagation path at a reflected angle that is equal to the incident angle. The strong specular reflection virtually turns the common building surfaces into "electrical mirror" by superseding the effect of scattered signal that is weaker in strength [9]. However, in the frequency bands around and above mmWave bands, the roughness of surfaces becomes comparable to the radio signal wavelength due to which the illuminated surface creates scattered signal paths, depending on the angle of incidence, there are substantial scattered signal paths. Different frequencies exhibit different intensity of specular and diffuse scattering from most building surfaces. Building surfaces appear electrically smooth at lower frequencies resulting in the specular reflections dominating the multi-paths, while same surfaces exhibit significant diffuse scattering and strong specular reflections at mmWave and THz, as shown in Fig. 3 [9].
The Rayleigh criterion is used to determine the roughness of a surface. The Rayleigh criterion defines the critical height h c of a surface as [10]: where θ i is the incident angle and λ is the wavelength. If the surface height variation h 0 is smaller than h c , the surface is considered to be smooth for the given frequency, if h 0 is larger than h c , the surface is considered to be rough for the given frequency [10].

II. NUMERICAL VALIDATION
To study the contribution of reflection loss to estimate the materials of objects and surfaces, we did an extensive simulation using simulator Axcel [11] which is implemented in Matlab and developed by Ericsson Research for radio network simulation. In this paper, we simulated narrow beam reflected by variety of common building materials namely: plasterboard, thick plasterboard, glass, heavy concrete, medium concrete, brick, metal, and wood with different incident angle θ i at frequencies 2.6 GHz, 28 GHz, and 60 GHz. The parameters of materials in the simulation including permittivity, conductivity, and roughness recommended by ITU [12] are summarized in Table I. It is worth mentioning here that higher value of roughness indicates less smooth surface of the material and lower value of transparency indicates that the object is opaque. Two-step simulation was performed and the simulation setup is shown in Fig. 4 and Fig. 5, where a transmitter (TX) device and a receiver (RX) device (TX may refer to base station and RX may refer to terminal device in wireless communication system and vice versa) are placed in a free space scenario where the direct line of sight (LoS) path (blue line) between TX and RX is not blocked by any obstacle as shown in Fig. 5. By means of beam alignment, a single-bounce reflection path between TX and RX is established, the signal reaches to RX via a specular reflection path (black line) after hitting the wall at the reflection point as shown in Fig. 4. This paper assumes single-bounce-reflection path between TX and RX and does not consider multiple-bounce paths that have larger time of arrival (TOA) and weaker received signal strength (RSS) for the evaluation purpose [13]. Wall made up of materials listed in Table I is considered to generate a single bounce path between TX and RX. The distance between TX/RX and wall is set to 15 m and TX/RX antenna heights are set to 5 m. The incident angle is maintained by changing the distance between TX and RX. For the simulation, incident angle θ i with values 1.9°, 9.5°, 20.1°, 29.5°, 39.8°, 45°, 51.7°, 59°, 67.4°, 73.6°, and 81.6°are considered. A radio wave transmitted by TX incidents at an angle θ i with respect to the normal of the wall surface as shown in Fig. 4. θ r is reflected angle such that θ i =θ r , following Snell's law. d t and d r are the distances between the wall and the TX/RX, respectively. d is the overall propagation length of specular reflection path between TX/RX and d = d t + d r . To obtain reflection loss data, two-step simulation is performed. The first step is used to measure the amount of the received power reflected from surface of the wall with an incident angle θ i determined by the TX/RX distance as shown in Fig. 4. The path loss (PL) over specular path is calculated as: where P T X is the transmitted power over specular path, P RX is the received power over specular path. The second step constitutes a LoS transmission over the distance that is equal to d. From Friis' equation, the free space where f is the carrier frequency in MHz and d is distance between TX and RX points in kilometer. Finally, the reflection loss (RL), which includes absorption loss plus scattering loss, is calculated by subtracting (3) from (2) and is given by It is worth noting that RL is independent of antenna type, antenna gain, transmitted power, and TX-RX distance. RL only corresponds to radio signal carrier frequency, scatterer properties, and the incident angle θ i .

III. RESULTS AND ANALYSIS
The simulation results of reflection losses of different materials at different incident angle θ i at carrier frequencies 2.6 GHz, 28 GHz, and 60 GHz are reported in Fig. 6 to Fig. 8. It is evident that the reflection losses of different materials are different in most of the cases when incident angle θ i is same for radio signals at carrier frequencies 2.6 GHz, 28 GHz, and 60 GHz. This clearly indicates that different materials can be effectively identified by using the radio signal and exploiting the reflection loss and incident angle information.
The simulation results show that as the frequency increases, the reflection loss increases, which can be expected according to (1), since the surfaces tend to be electrically rough (compared to the wavelength of the electromagnetic waves) as the frequency increases and rougher surfaces have higher tendency to scatter the radio signals. Take brick as an example. As shown in Fig. 9, brick (typical exterior surfaces of urban buildings) has reflection loss of less than 10 dB at 2.6 GHz, high reflection loss of approximately 30 dB and 115 dB at 28 GHz and 60 GHz, respectively. This illustrates the fact  that bricky surface is electrically smooth at lower frequency but electrically rough at higher frequency, reflection losses of bricky surface are different at different frequencies, thus providing reflection loss isolation between multiple frequencies.
The rough brick has significantly different properties than the other materials. We can see from Fig. 9 that only at 2.6 GHz, brick has a significant reflected component. A rough bricky surface has very high reflection loss, more than 110 dB at 60 GHz, indicating that the effect of reflection loss from the rough surface is significantly larger than the reflection loss due to smoother materials.
Moreover, for all simulated frequencies, the reflection loss falls off when the incident angle θ i increases. The minimum reflection loss is observed when the incident angle θ i =81.6°(maximum θ i was simulated). When the incident angle θ i is large, the surface tends to be electrically smooth according to (1), therefore the scattering loss is low or negligible, then reflection loss (absorption loss plus the scattering   loss) is low, most of the energy is concentrated to specular reflection direction. However, when the incident angle is small (e.g., incident ray impinges the surface perpendicularly), the scattering loss is not negligible [9]. The scattered power is higher at small incident angles than at large incident angles. This is the reason why the reflection loss decreases when increasing the incident angle. It is also observed from Fig. 6 to Fig. 8 that at 2.6 GHz, the reflection loss differences between most materials are quite small, especially when incident angle is large. For higher frequencies (28 GHz and 60 GHz), the reflection loss differences between most materials are pronounced, the curves are almost horizontal and most of them don't intersect each other when incident angle is small, which means the materials of the common building surfaces can be determined by reflection loss and approximate incident angle. When incident angle is large, reflection loss of materials declines with different rate, some reflection loss curves intersect with each other. Therefore, the materials of the common building surfaces can't be determined by reflection loss and approximate incident angle information at certain frequency. For example, Fig. 10 shows reflection losses of wood and concrete at 24 GHz and 28 GHz, including area where the reflection loss curves for wood and concrete intersect at incident angle 53 o in Fig. 7. At incident angle 53°, wood and concrete induce similar reflection losses, thus the material of the surface can't be effectively determined by reflection loss and incident angle. In this type of situation, a multi-frequency evaluation is needed. For multi-frequency evaluation a radio signal at 24 GHz can be transmitted following the same propagation path taken by 28 GHz. When the carrier frequency of the radio signal changes, wood and concrete will induce different reflection losses at an incident angle of 53°. With significant difference in the reflection losses at 28 GHz and 24 GHz, the material of the surface can be effectively identified. It is worth noting here that the reflection loss induced by woody surface does not change much with the change in carrier frequency, but reflection loss induced by concrete surface changes significantly over different carrier frequency. The reflection loss difference for woody surface is found to be about 0.2 dB between 24 GHz and 28 GHz when incident angle is between 45 o and 59 o whereas the reflection loss difference for concrete surface is estimated to be around 1.58 dB.
The simulation result also shows that the material properties and incident angle are influencing factors of reflection loss rather than thickness when the thickness of material exceeds a certain value. For example, from Fig. 7, reflection loss curves of medium concrete (thickness 10 cm) and heavy concrete (thickness 20 cm) completely overlap with each other. Furthermore, it is worth noting that the simulation results reported in figures from Fig. 6 to Fig. 8 verify that smoother surface (metal and glass) induce lower reflection loss.

IV. REFLECTION-LOSS-BASED MATERIAL IDENTIFICATION METHOD
Based on the simulation results, a reflection-loss-based material identification method for fast material identification in a multi-path wireless communication environment was developed. The core idea is to identify the material based on comparing measured reflection loss of an electromagnetic signal from its surface with pre-recorded reflection losses from different materials stored in a database. The proposed method comprises of the following steps: • Step 1: Measuring or simulating the reflection losses from different materials at different frequency/frequencies and incident angle/angles to create a database. • Step 2: Transmitting radio signal via beams at certain frequency/frequencies towards the material from a transmitter at certain angle/angles and measuring reflection loss based on the received signal at a receiver. • Step 3: Comparing the measured reflection loss with premeasured reflection losses from different materials that are recorded in a database to identify the material. In Step 1, a reflection loss lookup database for different materials is pre-measured or pre-simulated. The lookup database is an array including materials, frequencies, incident angles and reflection losses information, such that runtime material identification procedure is reduced by implementing a simpler array indexing operation. Pre-measurement is the act of performing an initial reflection loss measurement to generate a lookup database that can be used by proposed material identification method to avoid repeated computation every time when material identification needs to be done. The reflection losses of multiple incident angles are measured by changing the distance between TX and RX as shown in Fig. 4.
An example lookup database based on pre-simulation is shown in Table II. For simplicity, only reflection losses of wood and concrete at 28 GHz and 60 GHz are listed with 11 different incident angles.
In Step 2 and 3, by means of beam sweeping, a beam pair is established between TX and RX, and the Non-LoS (NLoS) beam is single-bounce-reflected by an object to be identified. The material identification processing unit calculates the reflection loss according to transmitted power, received power, incident angle and frequency. For example if one of the measurement sets, where reflection loss of 14 dB is measured at θ i =40°and f = 28 GHz, the material inducing this reflection loss can be identified as concrete by database look up operation where reflection loss of 13.88 dB is premeasured for 28 GHz radio signal when the incident angle is 39.8°. Moreover, the reflection point can be localized by determining the intersection point of transmitter-side beam and receiver-side beam lie on a co-plane. Therefore, a data point including its coordinate (x,y,z) and material information is obtained. By collecting enough data points in an environment, a 3D digital map with material information can be generated.

V. CONCLUSIONS
In this paper a novel material identification method based on joint communication and sensing is proposed. The proposed method does not require additional hardware but by what is anyway deployed for wireless communication purposes and relies on reflection loss evaluation of radio signal induced by the material type of the object that is encountered while the radio signal propagates from a transmission point to a reception point in a typical wireless communication network. The proposed method uses a simple array indexing approach for material identification and bears significantly less computational burden in comparison to a parametric-model-based approach. The proposed method can be effectively integrated to future RAT systems to generate a 3D digital maps with material information and support new and emerging use cases where effective material identification plays a vital role.
ACKNOWLEDGMENT This work has been partly funded by the European Commission through the H2020 project Hexa-X (Grant Agreement no. 101015956).