THZ WIDEBAND CHANNEL CHARACTERIZATION METHOD AND SYSTEM FOR INDOOR LOS / NLOS IDENTIFICATION

Abstract

The invention is related to a framework method and system for LoS / NLoS identification in THz links that decides the link state from the joint time and frequency behaviour of the wideband channel.

Description

1 DESCRIPTION THZ WIDEBAND CHANNEL CHARACTERIZATION METHOD AND SYSTEM FOR INDOOR LOS / NLOS IDENTIFICATION Technical Field 5 The invention is related to a framework method and system for LoS / NLoS identification in THz links that decides the link state from the joint time and frequency behaviour of the wideband channel. Prior Art The evolution of wireless networks is driving indoor links into the terahertz (THz) band, on the 10 order of 0.3 THz with multi-GHz bandwidth, in order to support ultra-high-rate communication, high-resolution localization, and integrated sensing and communication (ISAC). In such systems, reliable knowledge of the line-of-sight and non-line-of-sight (LoS / NLoS) state of each link is fundamental rather than auxiliary. An undetected NLoS link introduces a positive ranging bias that corrupts position estimates, whereas accurate LoS / NLoS knowledge governs beam 15 steering, link adaptation, blockage recovery, and handover decisions. Because of their short wavelength, THz links are highly susceptible to blockage by the human body, walls, and furniture, which produces frequent and abrupt transitions of the link state that must be identified in real time and on a per-link basis. The sixth generation (6G) of wireless networks aims to provide data rates in the order of terabits 20 per second, along with high-accuracy localization and integrated sensing and communication, and for this reason the THz band has attracted strong interest because of its very large available bandwidth [1]. However, THz signals suffer from severe path loss due to spreading and molecular absorption, so THz systems use large antenna arrays and highly directional narrow beams to keep the link range usable [1], [2]. 25 These narrow beams are easily blocked by the human body, walls, and furniture, which causes frequent and abrupt transitions of the link state between LoS and NLoS. Reliable knowledge of this state is therefore important, because an undetected NLoS link adds a positive bias to the range estimate and reduces the positioning accuracy, and the state is also used for beam steering, link adaptation, blockage recovery, and handover. The THz channel further differs from lower-30 2 band channels in a way that affects this problem directly, since the molecular absorption caused by water-vapor resonances makes the channel frequency-selective even for a single LoS path and introduces a group delay that broadens the LoS pulse in time, and this broadening increases with distance [3]. The existing identification methods do not consider this effect, as explained below. 5 Many existing methods identify the link state by applying supervised classifiers to channel impulse response statistics. The authors in [4] collected indoor UWB channel measurements and used a support vector machine for NLoS identification, with features such as the received energy, the maximum amplitude, the rise time, the mean excess delay, the RMS delay spread, and the kurtosis. 10 The authors in [5] used delay and angular spreads together with the Rician K-factor in a weighted random forest, and applied the LoS / NLoS decision to improve fingerprint-based positioning for 6G indoor communications. However, all these features only measure the multipath spreading of the signal, so any broadening is taken as a sign of NLoS. This does not hold at THz, where the molecular absorption broadens even a single LoS path by an amount 15 that grows with distance, and the classifier cannot separate this medium-induced broadening from the broadening caused by surface scattering. A LoS link that is broadened by the medium can therefore be identified as NLoS. The identification has also been formulated as a statistical hypothesis test. The authors in [6] modelled the probability density functions of the kurtosis, the mean excess delay, and the RMS 20 delay spread under the LoS and NLoS conditions as log-normal distributions, and used a joint likelihood-ratio test to decide between the two conditions. The authors in [7] used a soft-decision method, in which the features such as the skewness, the kurtosis, the RMS delay spread, the peak-to-average ratio, and the Rician K-factor are fitted with Gaussian mixture distributions, and the soft outputs are combined. However, these density 25 functions are built from the same delay-domain statistics, and they assume that the spreading of the signal comes only from multipath. At THz this assumption breaks, because the LoS density itself shifts with distance as the molecular absorption broadens the direct path, so the LoS and NLoS distributions overlap and the likelihood-ratio and soft-decision tests can no longer separate the two conditions reliably. 30 3 A further group of methods considers the change of the link state over time. The authors in [8] used the probability density functions of the correlation coefficient between successive channel impulse responses to decide the LoS or NLoS condition, and used this decision in a weighted particle filter for localization. The authors in [9] combined channel-based identification with inertial measurements and a 5 robust unscented Kalman filter to reduce the NLoS ranging error. These methods improve the robustness against short-term changes and lower the ranging error, but they act on top of a per- observation identification step and do not change how that step reads the channel. They still treat any broadening as multipath, and the second method also needs an additional sensor, so they do not address the dispersive broadening of the LoS path at THz. 10 The dispersive broadening and rough-surface scattering effects that are relevant to LoS / NLoS identification at THz have been documented in the THz propagation literature. Han et al. [7] developed a multi-ray channel model for the THz band and demonstrated that the channel is both distance-varying and frequency-selective, that molecular absorption gives rise to spectral windows bounded by absorption peaks, and that the RMS delay spread and the temporal 15 broadening are governed jointly by the propagation distance and the carrier frequency. These results establish that broadening in the THz band originates in part from the dispersive medium rather than from multipath alone, which is precisely the mechanism that invalidates the conventional identification rule. In [8] and [9], the specular component of the reflected signal is employed for communication 20 while the diffuse component is used for sensing, thereby establishing an NLoS link when the direct path is obstructed. These studies characterize the separation between the coherent and diffuse scattering contributions at THz surfaces, which is consistent with the physical basis of the present work. They are directed, however, at establishing and exploiting the NLoS link rather than at determining whether a given observation is LoS or NLoS, and therefore address 25 a fundamentally different problem from the one considered here. In summary, the existing methods identify the link state by setting a threshold on the magnitude of channel statistics, or by classifying these statistics, and they assume that the statistics are caused only by the multipath. They also need labelled training data or measured probability density functions. They do not consider the broadening that is introduced by the dispersive THz 30 4 medium, and they do not use the frequency-selective distortion that is caused by the molecular absorption and the rough surfaces. As a result, all of the problem mentioned above has made it necessary to provide a novelty in the related field. Brief Description and Objects of the Invention 5 The main object of the present invention is to establish a method and system that is to identify the LoS or NLoS state of each THz link from the time and frequency behaviour of the wideband channel. The present invention achieves this objective with extracts seven features from each channel observation, covering the time-domain broadening of the received signal and the frequency-10 selective signatures of the THz channel, and uses them jointly to distinguish the broadening caused by the dispersive medium from the broadening caused by surface scattering. This allows the link state to be identified reliably across the range of link distances and in the sparse, beamformed THz regime, without relying on a single statistic or a threshold fixed at one operating point. 15 Another object of the invention is to establish a method and system that is extract both time- domain features, which describe the broadening of the received signal, and frequency-domain features, which describe the molecular absorption notch depth, the coherence bandwidth, and the spectral roll-off slope. Another object of the invention is to establish a method and system that is derive the decision 20 from the characterization of the channel behaviour rather than from a fixed threshold, so that the identification remains valid across the range of link distances and geometries encountered in an environment. Another object of the invention is to establish a method and system that is use per-path spectral behaviour together with the broadening features, so it continues to separate the two link states 25 in the sparse, beamformed THz regime where richness-based statistics fail. The present invention discloses a framework method and system for LoS / NLoS identification in THz links that decides the link state from the joint time and frequency behaviour of the wideband channel. Unlike conventional identifiers, which read any broadening of the received signal as a sign of multipath and therefore of an NLoS condition, the disclosed approach 30 separates the broadening caused by the dispersive medium from the broadening caused by surface scattering, using seven features that describe both the time-domain broadening and the frequency-selective signatures of the THz channel. The problem addressed by the present invention is therefore the reliable identification, for each wideband channel observation of an indoor THz link, of whether the observation corresponds 5 to a true LoS path or to an NLoS path established through surface scattering, with this identification remaining robust across the range of transmitter to receiver separations encountered in practice. A first technical challenge arises from the breakdown, at THz frequencies, of the principle underlying conventional LoS / NLoS identifiers developed for sub-GHz and ultra-wideband 10 (UWB) systems. Such identifiers infer multipath geometry from the channel impulse response and rest on the assumption that the propagation medium is non-dispersive, so that any temporal broadening of the received signal implies the presence of multipath and therefore an NLoS condition. In a non-dispersive medium, a true LoS path corresponds to a single resolvable arrival whose delay spread is effectively zero and independent of distance, and it is precisely 15 this property that makes the broadening rule reliable at lower frequency bands. In the THz band this assumption no longer holds, because molecular absorption, dominated by water-vapor resonances, renders the channel frequency selective even for a single LoS path, while the dispersion that accompanies this absorption, in accordance with the Kramers and Kronig relations, introduces a frequency-dependent group delay. The combination of a non-flat 20 magnitude response and a non-linear phase response broadens the LoS pulse in time, and since the absorption exponent depends on the path length, this dispersive broadening increases with link distance. Consequently, the presence of broadening can no longer be interpreted as evidence of an NLoS condition, and the physical origin of the broadening, namely dispersive broadening introduced by the medium as opposed to scattering broadening introduced by a 25 rough surface, must instead be determined. A second technical challenge is that the discriminating margin between the two link states can collapse in the most operationally significant case. The temporal spread of an NLoS link is dominated by its diffuse scattering cluster, whose extent is governed by the surface roughness and the illuminated footprint rather than by the link length. When the NLoS path is near 30 specular, owing to a smooth surface, a small footprint, or a large coherent reflection fraction, the diffuse cluster is suppressed and the NLoS delay spread collapses toward the dispersive 6 floor. In this regime, a LoS link, whose molecular dispersion has accumulated over distance, and a near-specular NLoS link can exhibit comparable root-mean-square delay spread. A decision rule based on a single delay-spread threshold therefore fails, not because broadening is absent, but because two physically distinct mechanisms can produce indistinguishable magnitudes. It follows that the discriminating quantity must be the type and frequency-domain 5 signature of the distortion rather than its scalar magnitude. A third technical challenge concerns the dependence of the decision statistic on the propagation distance. Because the dispersive component of LoS broadening grows with the transmitter to receiver separation, a fixed threshold calibrated at a single separation cannot remain valid across the operating range. A robust identifier must therefore compare the broadening measured on a 10 given link against the broadening that molecular absorption alone would produce over that same link, where the latter is determined from the path length inferred from the molecular absorption notch. The decision then rests on the discrepancy between the measured and the predicted broadening, rather than on the absolute value of any single feature, which keeps it valid across the full range of separations and expresses it in terms of physical consistency. 15 A fourth technical challenge stems from the sparse and beamformed nature of THz propagation, together with the frequency-selective behaviour of THz surfaces. In massive multiple-input multiple-output systems that employ ultra-narrow beams, only a few dominant paths are present in each stream, so the multipath-richness statistics on which conventional channel-impulse- response identifiers rely are weakened, and the discriminating information shifts toward the 20 distortion imprinted on individual paths. At THz frequencies, surfaces act as frequency- selective filters rather than as flat-band mirrors, combining a Rayleigh roll-off of the coherent component with a Beckmann and Kirchhoff diffuse contribution. As a result, the channel imprints a spectral fingerprint, expressed most clearly through the depth of the molecular absorption notch, the coherence bandwidth, and the high-frequency roll-off slope, that is richer 25 than the delay-domain moments considered in isolation and that remains unexploited by prior art. Taken together, these limitations constitute the technical gap that motivated the present invention, namely the absence of an identifier that distinguishes the physical origin of channel distortion rather than thresholding its magnitude. The present invention addresses this gap by 30 reformulating the broadening-implies-NLoS rule as a test of physical consistency, in which the measured distortion is compared against the broadening that molecular absorption alone could 7 produce over the path length implied by the depth of the molecular absorption notch. The consistency is evaluated jointly across two complementary families of features. The first family captures the time-domain consequences of broadening and comprises the root-mean-square delay spread, the mean excess delay, the kurtosis of the received envelope, and the group-delay spread. The second family captures the frequency-selective signatures characteristic of the 5 dispersive THz medium and its rough surfaces, and comprises the molecular absorption notch depth, which serves as a probe of total path length, together with the coherence bandwidth and the spectral roll-off slope. Distortion that exceeds the level attributable to molecular dispersion over the path length inferred from the notch, and that is corroborated by these frequency-domain signatures, provides an unambiguous indication of surface scattering and therefore of an NLoS 10 condition. The technical significance of the present invention lies in its physics-based framework for LoS / NLoS identification in indoor THz links, which decides the link state from the physical origin of the channel behaviour rather than from the magnitude of a single statistic. The method that is disclosed in the present invention, extracts seven features from each wideband channel 15 observation, covering both the time-domain consequences of broadening, namely the RMS delay spread, the mean excess delay, the kurtosis, and the group-delay spread, and the frequency-selective signatures of the THz channel, namely the molecular absorption notch depth, the coherence bandwidth, and the spectral roll-off slope. These features together separate the dispersive broadening introduced by the medium from the scattering broadening introduced 20 by rough surfaces, since the frequency-selective signatures carry path-length and surface information that keeps the LoS and NLoS conditions separable even when the time-domain broadening features overlap. This makes the identification robust across the range of link distances and reliable in the sparse, beamformed THz regime, which makes it highly relevant to next-generation 6G communication, localization, ISAC deployments. 25 The principal areas of technical interest of the present invention are sorted follow; One of the principal areas is reliable link-state identification for THz localization. In THz localization and ranging, an undetected NLoS link introduces a positive bias in the range estimate and degrades the positioning accuracy. By identifying the link state correctly even when the delay spread of a near-specular NLoS link collapses toward that of the LoS link, the 30 invention allows biased NLoS observations to be detected and either weighted down or 8 discarded. This improves the accuracy of high-resolution indoor positioning systems that operate in the THz band. Also, another principal area is robust beam and link management. THz links use highly directional narrow beams that are frequently and abruptly blocked by the human body, walls, and furniture. Reliable knowledge of the LoS or NLoS state supports beam steering, link 5 adaptation, blockage recovery, and handover. Because the present invention provides a per- observation decision that does not depend on a distance-specific threshold, it remains valid as the link distance and geometry change, which is essential for stable beam and link management in dynamic indoor environments. Furthermore, another principal area is link-state decision from channel characterization. The 10 identification relies on the characterization of the THz channel in both the time and frequency domains, in which the seven extracted features jointly describe how the medium and the surfaces shape the received signal. The time-domain features capture the broadening of the received signal, while the frequency-domain features capture the molecular absorption notch depth, the coherence bandwidth, and the spectral roll-off slope, so that the LoS and NLoS 15 conditions are distinguished from the combined time and frequency behaviour rather than from a single statistic or a threshold fixed at one operating point. This keeps the method valid as the link distance and geometry change and makes it suitable for practical THz systems. In addition, another principal area is robustness in the sparse beamformed regime. In THz systems with ultra-narrow beams, only a few dominant paths are present, and the multipath-20 richness statistics used by conventional identifiers lose their discriminating power. Because the method draws on per-path spectral behaviour together with the broadening features, it continues to separate the two link states where richness-based statistics fail. Additionally, another principal area is reliable decision in the near-specular NLoS case. The hardest case is a near-specular NLoS link, where the diffuse cluster is weak and the delay-25 domain features move close to their LoS values. The frequency-domain features preserve the separation in this regime, so the method remains reliable exactly where a single delay-spread decision would fail. The present invention addresses the problem that the conventional rule for LoS / NLoS identification does not hold in the THz band. At lower frequencies the propagation medium is 30 treated as non-dispersive, so a LoS path appears as a single arrival with almost zero delay 9 spread, and any broadening of the received signal is taken as a sign of multipath and therefore of an NLoS condition. In the THz band the molecular absorption makes the channel frequency- selective even for a single LoS path and broadens the LoS pulse by an amount that grows with distance. As a result, a LoS link that is broadened by the medium can be confused with an NLoS link, and the broadening of the received signal can no longer be used on its own to identify the 5 link state. The scientific problem is therefore to distinguish the broadening introduced by the dispersive medium from the broadening introduced by surface scattering, so that the link state can be identified correctly at THz. The technical problems that are the present invention solves and provide solutions can be sorted as follow; 10 One of the problems is; at THz, the broadening of the received signal is caused both by the dispersive medium and by surface scattering, so a single delay-domain statistic cannot tell whether a broadened signal corresponds to a long LoS link or to an NLoS link. The margin becomes smallest in the near-specular NLoS case, where the diffuse cluster is weak and the delay-domain features move close to their LoS values. The solution that is the present invention 15 brings; The present invention extracts both time-domain features, which describe the broadening of the received signal, and frequency-domain features, which describe the molecular absorption notch depth, the coherence bandwidth, and the spectral roll-off slope. The frequency-domain features carry path-length and surface information that keeps the LoS and NLoS conditions separable even when the time-domain features overlap, so the two conditions 20 are distinguished from the combined time and frequency behaviour of the channel. Another problem is; a decision statistic that depends on the propagation distance cannot be separated from the two link states by a fixed threshold, because the LoS broadening grows with distance and a threshold calibrated at one distance is not valid at another. The solution that is the present invention brings; The invention derives the decision from the characterization of the 25 channel behaviour rather than from a fixed threshold, so that the identification remains valid across the range of link distances and geometries encountered in an environment. Also, another problem is; in THz systems with ultra-narrow beams, only a few dominant paths are present, and the multipath-richness statistics used by conventional identifiers lose their discriminating power. The solution that is the present invention brings; the invention uses per-30 path spectral behaviour together with the broadening features, so it continues to separate the two link states in the sparse, beamformed THz regime where richness-based statistics fail. Furthermore, the present invention brings some innovations to the sector. Those are can be sorted as follow; One of the innovations that is brought by the present invention is identification from the time and frequency behaviour of the THz channel. The present invention identifies the link state from seven features that jointly describe the time-domain broadening and the frequency-5 selective signatures of the THz channel, rather than from a single delay-domain statistic. Another innovation that is brought by the present invention is separation of medium and surface effects. The present invention distinguishes the broadening introduced by the dispersive medium from the broadening introduced by surface scattering, which is not done by existing identifiers and which is the effect that makes the conventional rule fail at THz. 10 Also, another innovation that is brought by the present invention is use of THz-specific frequency-selective signatures. The present invention uses the molecular absorption notch depth, the coherence bandwidth, and the spectral roll-off slope, which are available at THz, to keep the LoS and NLoS conditions separable in cases where the delay-domain features alone overlap. 15 In addition, another innovation that is distance-robust decision. The present invention provides a link-state decision that remains valid across the range of link distances and geometries, rather than a decision tied to a threshold fixed at one operating point. Furthermore, the present invention has some key advantages and unique technical elements. Those are can be sorted as follow; 20 The present invention has separation of medium and surface effects. The advantage is; the present invention identifies the link state correctly even when the broadening of the received signal is caused partly by the dispersive medium, a case in which conventional identifiers can confuse a broadened LoS link with an NLoS link. Unique element is; the broadening introduced by the molecular absorption of the medium is distinguished from the broadening introduced by 25 surface scattering, which is not addressed by existing delay-domain identifiers. Also, the present invention has joint time and frequency characterization. The advantage is; the decision is taken from the combined time and frequency behaviour of the channel, so the two link states remain separable even when any single feature overlaps between the two classes. Unique element is; seven features are used together, covering the time-domain broadening, 30 namely the RMS delay spread, the mean excess delay, the kurtosis, and the group-delay spread, 11 and the frequency-selective signatures, namely the molecular absorption notch depth, the coherence bandwidth, and the spectral roll-off slope. In addition, the present invention has use of THz-specific frequency-selective signatures. The advantage is; the present invention keeps the LoS and NLoS conditions separable in the near- specular NLoS case, where the diffuse cluster is weak and the delay-domain features move 5 close to their LoS values. Unique element is; the frequency-selective signatures that become available at THz, in particular the molecular absorption notch depth, carry path-length and surface information that is not exploited by the identifiers designed for lower frequency bands. Additionally, the present invention has distance-robust decision. The advantage is; the identification remains valid across the range of link distances and geometries encountered in an 10 indoor deployment, without recalibration at each distance. Unique element is; the decision is derived from the characterization of the channel behaviour rather than from a threshold fixed at one operating point, so it does not depend on a distance-specific calibration. Furthermore, the present invention has robustness in the sparse beamformed regime. The advantage is; the present invention continues to separate the two link states in THz systems 15 with ultra-narrow beams, where only a few dominant paths are present. Unique element is; the decision draws on per-path spectral behaviour together with the broadening features, so it does not rely on the multipath-richness statistics that lose their discriminating power in the sparse, beamformed THz regime. The strike feature of the present invention is the identification of the LoS or NLoS state from 20 the combined time and frequency behaviour of the THz channel, in a way that separates the broadening introduced by the dispersive medium from the broadening introduced by surface scattering. Unlike conventional identifiers that treat any broadening as multipath and decide from a single delay-domain statistic or a fixed threshold, the proposed approach uses the frequency-selective signatures of the THz channel to keep the LoS and NLoS conditions 25 separable even where the delay-domain features overlap, which allows the link state to be identified reliably across link distances and in the sparse, beamformed THz regime. The present invention demonstrates strong industrial applicability in indoor terahertz (THz) systems for 6G communication, localization, and integrated sensing and communication (ISAC). By identifying the LoS or NLoS state of each link from the time and frequency 30 behaviour of the wideband channel, and by separating the broadening introduced by the 12 dispersive medium from the broadening introduced by surface scattering, the present invention provides a reliable link-state decision that supports positioning, beam management, and sensing in the indoor environments identified for THz deployment. Because the decision is taken from the channel characterization and is not tied to a particular classifier or to a distance-specific threshold, it can be integrated into a THz receiver as a processing block that operates on the 5 channel estimate already available at the receiver. This positions the present invention as a practical solution for next-generation indoor THz systems. The industries and applications of the present invention can be sorted as follow; In 6G indoor positioning and localization, the present invention’s use case is high-resolution positioning and ranging in offices, meeting rooms, hallways, factories, and data centres. Benefit 10 of this use is; detects NLoS links that would otherwise inject a positive ranging bias, so biased observations can be down-weighted or discarded. Keeps the identification reliable even in the near-specular NLoS case where the delay-domain features move close to their LoS values. In THz beam and link management, the present invention’s use case is beam steering, link adaptation, blockage recovery, and handover for THz links using highly directional narrow 15 beams. Benefit of this use is; provides a per-observation link-state decision that remains valid as the link distance and geometry change. Supports fast recovery when the direct path is blocked by the human body, walls, or furniture. In 6G integrated sensing and communication (ISAC), the present invention’s use case is joint communication and sensing in indoor ISAC systems, including imaging, mapping, and 20 localization. Benefit of this use is; uses the same specular and diffuse channel behaviour that ISAC receivers exploit, so a single channel observation serves both the link-state decision and the sensing function. Supports mission-critical and immersive indoor applications that require reliable link-state knowledge. In industrial and enterprise THz networks, the present invention’s use case is real-time 25 industrial control, predictive maintenance, and ultra-high-throughput indoor connectivity in factories, data centres, and enterprise spaces. Benefit of this use is; maintains reliable link-state identification in the sparse, beamformed THz regime where multipath-richness statistics lose their discriminating power. Operates without additional sensing hardware, using only the wideband channel estimate. 30 Also, the application format of the present invention can be sorted as follow; 13 In receiver-side processing block, the present invention implemented as a processing block within the THz receiver that operates on the wideband channel estimate. Requires no change to the transmitted waveform, the bandwidth, or the frame structure. In feature extraction and interpretation, the present invention extracts the seven time-domain and frequency-domain features from each channel observation and interprets them according 5 to the THz channel behaviour to separate medium-induced from surface-induced broadening. In classifier-independent decision, the present invention produces the LoS or NLoS decision using any suitable discrimination method or AI / ML classifier applied to the feature set, so the block can be adapted to the processing available in a given system. Along with these, the present invention is applicable across the main indoor THz application 10 areas identified for 6G, including positioning and localization, beam and link management, integrated sensing and communication, and industrial and enterprise networks. In each case it provides a reliable LoS or NLoS decision that improves the accuracy and robustness of the system without additional sensing hardware and without changes to the transmitted waveform. Because it is implemented as a receiver-side processing block that operates on the available 15 channel estimate and produces its decision through any suitable discrimination method or AI / ML classifier, the invention can be integrated into a wide range of THz systems, which makes it a practical and forward-compatible solution for next-generation indoor THz deployments. Description of the Figures of the Invention The figures and related descriptions necessary for the subject matter of the invention to be understood better are given below. Figure 1. System model of the present invention. Figure 2. Disclosed LoS / NLoS identification pipeline for indoor THz links of the present 25 invention. Detailed Description of the Invention The invention is related to a method and system that is identifies the LoS or NLoS state of an indoor THz link. 30 14 Referring to Figure 1 and 2; the present invention discloses a THz wideband channel characterization method for indoor LoS / NLoS identification. In the disclosed method firstly, a wide band pulse indoor transmitted by a THz transmitter. Then, in the next step of the method that is disclosed in the present invention; the transmitted wideband pulse observed under two propagation conditions as; 5 o One of the conditions is a LoS condition which the wideband pulse reaches the THz receiver through the direct path, and o Other one is NLoS condition which the direct path is obstructed and the pulse reaches the THz receiver after reflection and scattering from a rough surface the indoor by a THz receiver. 10 In the Figure 1, as an example an indoor THz link operated around a centre frequency 𝑓𝑐 of about 0.3 THz with a multi-GHz bandwidth 𝐵, in a small office room. As shown in the system model in the Figure 1, a THz transmitter emits a wide band pulse towards the THz receiver, and the received signal is observed under two propagation conditions as explained before. In the LoS condition the pulse reaches the receiver through the direct path, while in NLoS condition 15 the direct path is obstructed and the pulse reaches the THz receiver after reflection and scattering from a rough surface in the room. The two conditions are treated as two hypotheses for the same wideband channel observation, and the objective of the model is to describe how the dispersive medium and the rough surface shape signal in each case. In the preferred embodiment of the present invention, the THz transmitter emits the wideband 20 pulse as 𝑠(𝑡) whose spectrum 𝑆(𝑓) occupies the band of 𝑓 ∈ [𝑓𝑐 − 𝐵 / 2, 𝑓𝑐 + 𝐵 / 2]. The received signal by the THz receiver in the frequency domain computed by formula of; 𝑌(𝑓) = 𝐻(𝑓) 𝑆(𝑓) + 𝑊(𝑓), 1 Were the 𝐻(𝑓) is the channel transfer function for the corresponding propagation condition and 𝑊(𝑓) is additive white Gaussian noise. The channel impulse response is obtained with 25 computed by formula of; ℎ(𝜏) = ℱ−1{𝐻(𝑓)}, 2 Where the both 𝐻(𝑓) and ℎ(𝜏) are used to characterize the wideband pulse link in the frequency and time domains, respectively. After that, in the next step of the method that is disclosed in the present invention; the THz receiver forms the wideband channel estimation from the received signal. In the preferred embodiment of the present invention, the receiver obtains 𝐻(𝑓) as the transfer 5 function and ℎ(𝜏) as the impulse response when forming wideband channel estimation from the received signal for each link. This step is the standard input available at the THz receiver, and it is the observation on which the identification operates. In the preferred embodiment of the present invention, for each channel realization the THz 10 receiver obtains the wideband transfer function 𝐻(𝑓) and the corresponding impulse response ℎ(𝜏) for the separation medium-induced broadening from surface-induced broadening under either the LoS hypothesis transfer function 𝐻𝐿(𝑓) or the NLoS hypothesis transfer function 𝐻𝑁(𝑓) that is explained later. Along the direct path of length 𝑑𝐿, the received signal is attenuated by free-space spreading and 15 by molecular absorption, and it is delayed and dispersed by the frequency-dependent behavior of the medium. In this preferred embodiment of the present invention, the LoS transfer function 𝐻𝐿(𝑓) computed by formula of; 𝐻𝐿(𝑓) = 𝑐 4𝜋𝑓𝑑𝐿  exp (− 1 2  𝑘(𝑓) 𝑑𝐿)  exp (−𝑗 2𝜋𝑓  𝑛(𝑓) 𝑑𝐿 𝑐 ) , 3 20 Wherein 𝑐 is the speed of light, 𝑘(𝑓) is the molecular absorption coefficient of the medium, and 𝑑𝐿 is the direct path of length. The coefficient 𝑘(𝑓) is obtained from the water-vapor absorption lines of the HITRAN database following the Beer-Lambert law, and it is frequency-selective, so that the absorption produces spectral windows separated by absorption peaks. The complex refractive index 𝑛(𝑓) = 𝑛′(𝑓) −25 𝑗 𝑛′′(𝑓) links the absorption to the phase through the Kramers-Kronig relations, with 𝑘(𝑓) = 4𝜋𝑓 𝑛′′(𝑓) / 𝑐, so that the dispersion associated with the absorption introduces a frequency- 16 dependent group delay. Because the exponent of the absorption term carries the path length 𝑑𝐿, both the depth of the molecular notch and the temporal broadening of the LoS pulse increase with distance. This is the effect illustrated by the broadened LoS response in the system model in Figure 1, in which the received pulse is spread and shows the oscillatory signature of the molecular absorption relative to the transmitted reference. 5 In the NLoS condition the direct path is obstructed and the signal reaches the receiver after interacting with a rough surface, over total path length 𝑑𝑁 = 𝑑1 + 𝑑2, where 𝑑1 is the distance from the transmitter to the surface and 𝑑2 is the distance from the surface to the receiver. Because the surface is reached through a two-segment path, the total NLoS path is always longer than the LoS case. The interaction with rough surface is described by a coherent specular 10 component and a diffuse scattered component. In this preferred embodiment of the present invention, the NLoS hypothesis transfer function 𝐻𝑁(𝑓) computed by formula of; 𝐻𝑁(𝑓) = 𝐻𝑁 spec (𝑓) + 𝐻𝑁 diff(𝑓). 4 Wherein, the specular component as 𝐻𝑁 spec (𝑓) follows the direction defined by the incidence 15 angle 𝜃𝑖 and the reflection angle 𝜃𝑟, and it combines the Fresnel reflection coefficient of the surface material with a Rayleigh roughness attenuation that reduces the coherent reflection as the surface becomes rough and computed by formula of; 𝐻𝑁 spec (𝑓) = 𝑐 4𝜋𝑓𝑑𝑁  Γ(𝑓, 𝜃𝑖) exp (− 1 2  𝑔(𝑓))  exp (− 1 2  𝑘(𝑓) 𝑑𝑁)  exp (−𝑗 2𝜋𝑓  𝑛(𝑓) 𝑑𝑁 𝑐 ) , 5 20 Wherein, Γ(𝑓, 𝜃𝑖) is the Fresnel reflection coefficient determined by the permittivity of the surface material, and 𝑔(𝑓) as the Rayleigh roughness factor that is computed by formula of; 𝑔(𝑓) = ( 4𝜋𝜎ℎ cos 𝜃𝑖 𝜆 ) 2 = ( 4𝜋𝜎ℎ fcos 𝜃𝑖 𝑐 ) 2 , 6 25 17 Wherein, the 𝜎ℎ is standard deviation of the surface height and 𝜆 = 𝑐 / 𝑓 is wavelength. The factor exp(− 1 2 𝑔(𝑓)) shows that the coherent reflection rolls off with frequency, so that the surface acts as a frequency-selective filter rather than a flat-band mirror. In this preferred embodiment of the present invention, the diffuse component 𝐻𝑁 diff(𝑓) accounts for the energy scattered away from the specular direction by the rough surface wherein using 5 the image method, the illuminated surface patch is represented by an ensemble of scattering elements, and each element 𝑚 contributes a delayed and attenuated copy of the signal and the 𝐻𝑁 diff(𝑓) is computed by formula of; 𝐻𝑁 diff(𝑓) = ∑ 𝑐 4𝜋𝑓𝑑𝑚 𝑚  𝜌𝑑(𝜃𝑖, 𝜃𝑠,𝑚) exp (− 1 2  𝑘(𝑓) 𝑑𝑚)  exp (−𝑗 2𝜋𝑓  𝑛(𝑓) 𝑑𝑚 𝑐 ) , 7 Wherein 𝑑𝑚 is the total path length through the 𝑚-th surface element, 𝜃𝑠,𝑚 is corresponding scattering angle, and 𝜌𝑑(𝜃𝑖, 𝜃𝑠,𝑚) is Beckmann-Krichhoff diffuse scattering coefficient that governs the angular distribution of the scattered field. Because the surface elements are distributed across the illuminated patch, the path lengths 𝑑𝑚 15 differ from one element to another, and this spread of path lengths produces the diffuse cluster that broadens the NLoS response in time. The extent of this broadening is governed by the size of the illuminated patch rather than by the surface roughness alone. This behavior corresponds to the scattered paths and the broadened NLoS response shown in the system model in Figure 1, in which the received pulse is more spread and less peaked than in the LoS case. 20 In this preferred embodiment of the present invention, the two hypotheses as the LoS hypothesis 𝐻𝐿(𝑓) and the NLoS hypothesis 𝐻𝑁(𝑓) differ in two ways, in the time domain, the NLoS response carries the additional broadening produced by the diffuse cluster, while the LoS response is broadened only by the dispersion of the medium. How this feature is used within the scope of the present invention will be explained later. 25 For each channel realization the receiver obtains the wideband transfer function 𝐻(𝑓) and the corresponding impulse response ℎ(𝜏), under either the LoS hypothesis 𝐻𝐿(𝑓) or the NLoS 18 hypothesis 𝐻𝑁(𝑓). Two hypotheses differ in two ways as explained before, while in the time domain, the NLoS response carries the additional broadening produced by the diffuse cluster, while the LoS response is broadened only by the dispersion of the medium. In addition, in the frequency domain, the longer NLoS path produces a deeper molecular notch, and the rough surface imprints a frequency-selective roll-off on the reflected signal. These time-domain and 5 frequency-domain differences from the basis of the feature set used for LoS and NLoS identification. Then, in the next step of the method that is disclosed in the present invention; THz receiver computes features from each channel observation. The features can be listed as follows; As time domain features; 10 o RMS delay spread feature, o Mean excess delay feature, o Kurtosis of the received envelope feature, o Group-delay spread feature, As frequency domain features; 15 o Molecular absorption notch depth feature, o Coherence bandwidth feature, o Spectral roll-of slope feature Are computed by the THz receiver. In this step, from each channel observation, preferably seven features are computed. Four of 20 them describe the time-domain broadening of the received signal, and three describe the frequency-selective behavior of the channel. In the preferred embodiment of the present invention, the RMS delay spread feature is the second central moment of the power delay profile. Here, it measures how far the received energy is spread in time around its centroid. 25 In the NLoS condition the diffuse cluster produced by the rough surface adds energy over a range of delays, which increases the spread, while in the LoS condition the spread is limited to the broadening produced by the dispersion of the medium. The RMS delay spread is therefore larger for NLoS than for LoS. 19 Also, in the preferred embodiment of the present invention, the mean excess delay feature is the centroid of the power delay profile measured relative to the first arrival. The additional scattered energy in the NLoS condition shits this centroid to a later delay, so the mean excess delay is larger for NLoS than for LoS. Furthermore, in the preferred embodiment of the present invention, the kurtosis of the received 5 envelope feature measures how peaked the response is. A LoS response is dominated by a single strong direct arrival and is highly peaked, which gives a high kurtosis, while an NLoS response is spread over the diffuse cluster and is less peaked, which lowers the kurtosis. The kurtosis is therefore higher for LoS than NLoS. In addition, in the preferred embodiment of the present invention, the group-delay spread 10 feature is the variation of the group delay across the band. At THx this variation is produced both by the dispersion of the medium and by the frequency- selective behavior of the rough surface, and it is larger for NLoS than for LoS, so it reflects the frequency-domain behavior of the channel rather than delay-domain multipath alone. Additionally, in the preferred embodiment of the present invention, the molecular absorption 15 notch depth feature is the depth of the water-vapor absorption feature in the received spectrum. Because the absorption grows with the path length, the notch deepens with distance, and since the NLoS path through the surface is always longer than the direct LoS path for the same transmitter and receiver positions, the notch is deeper for NLoS than for LoS. This feature has no counterpart at lower frequency bands, and it acts as a probe of the total path length, which 20 is the information that keeps the two link states separable when the time-domain broadening is similar. Furthermore, in the preferred embodiment of the present invention, the coherence bandwidth feature is the frequency separation over which the channel remains correlated, and its is inversely related to the delay spread feature. 25 The additional broadening in the NLoS condition reduces the coherence bandwidth, so it is higher for LoS than for NLoS, and at THz it also reflects the molecular absorption structure of the spectrum. Also, in the preferred embodiment of the present invention, the spectral roll-off slope feature is the rate at which the received spectrum decays across the band. The Rayleigh roughness attenuation of the rough surface rolls off with frequency, which steepens the slope in the NLoS condition, so this feature carries information about the surface interaction that is specific to the frequency-selective behavior of surfaces at THz. 5 These seven features are preferably the input to the two feature families defined in next step. The seven features are grouped into two families that respond to different physical effects. The time-domain family, comprising the RMS delay spread, the mean excess delay, the kurtosis, and the group-delay spread, responds mainly to the diffuse cluster produced by the rough surface in the NLoS condition. The frequency-domain family, comprising the molecular 10 absorption notch depth, the coherence bandwidth, and the spectral roll-off slope, responds to the molecular absorption of the medium and to the frequency-dependent behavior of the surface. This grouping organizes the features according to the physical mechanism that drives them, and it prepares the two families for the interpretation carried out in the next step. After that, in the next step of the method that is disclosed in the present invention; the THz 15 receiver separates medium-induced broadening from surface-induced broadening using the THz distance and surface signatures. In the preferred embodiment of the present invention, the two hypotheses as the LoS hypothesis 𝐻𝐿 and the NLoS hypothesis 𝐻𝑁 differ according to the propagation characteristics of the corresponding link. The LoS response is characterized by the frequency-dependent absorption 20 and dispersion of the propagation medium, whereas the NLoS response is additionally affected by the longer propagation path and the interaction with the rough surface, which introduces diffuse scattering and frequency-dependent attenuation of the reflected component. Consequently, the NLoS response exhibits additional temporal spreading and spectral distortion compared with the LoS response. 25 At sub-6 GHz and millimeter-wave frequencies the medium is effectively non-dispersive behave as flat-band reflectors, so a LoS path is a single arrival with a delay spread close to zero and a spectrum with no absorption structure. In that setting the time-domain feature respond only to multipath and the frequency-domain features structure. In that setting the time-domain 21 features respond only to multipath and the frequency-domain features carry little information. In the THz band the situation is different for two reasons. First, the molecular absorption of the medium, dominated by the water-vapor resonances, makes the channel frequency-selective even for a single LoS path and broadens the LoS pulse by an amount that grows with distance, so the medium writes a distance signature into the 5 spectrum through the molecular absorption notch. Second, because the wavelength is comparable to the surface roughness, the rough surface no longer behaves as a flat-band mirror, so the coherent reflection rolls off with frequency and the scattered energy forms a diffuse cluster whose extent is set by the illuminated patch, and the surface therefore writes a frequency- selective signature into both the spectrum and the delay 10 profile. As a result, the same seven features that are weak or ambiguous discriminators at lower bands become informative at THz. This step separates the broadening introduced by the dispersive medium from the broadening introduced by surface scattering. Then, in the next step of the method that is disclosed in the present invention; the THz receiver 15 evaluates the extracted features based on the separation between the LoS and NLoS conditions and ranks the extracted features accordingly. The features that are computed before are not equally informative, so they are preferably ranked by the THz receiver in this step. In the preferred embodiment of the present invention, for ranking the features by their 20 discriminating power, the Ficher Discriminant Ratio (FDR) used that is computed by formula of; 𝐽 = (𝜇LoS−𝜇NLoS) 2 𝜎LoS 2 + 𝜎NLoS 2 , 8 Where 𝐽 denotes the FDR of a respective feature, 𝜇LoS and 𝜇NLoS denote the mean values of the respective feature under the LoS and NLoS samples, respectively, and 𝜎LoS 2 and 𝜎NLoS 2 denote 25 the corresponding variances across the LoS and NLoS samples, respectively. In the preferred embodiment of the present invention, the FDR is computed for all features. 22 Wherein a large value of 𝐽 indicates that the class means are well separated relative to the spread within each class. The FDR is computed for all seven features, which orders them by their separability. No feature is discarded, and mutual information and the minimum-redundancy maximum-relevance criterion may be used as cross-checks on the ranking. After that, in the next step of the method that is disclosed in the present invention; the THz 5 receiver selects the features for LoS / NLoS classification based on the ranking obtained before, wherein, in a near-specular case in which the time-domain features exhibit reduced variation, the frequency-domain features are prioritized by the THz receiver. The most demanding case for identification is the near-specular NLoS link, in which the surface is smooth or the illuminated patch is small, so the diffuse cluster is weak and the time-domain 10 features move close to their LoS values. In this case a decision based on the time-domain broadening alone would fail, because the broadening of the near-specular NLoS link is similar to that of a LoS link. The frequency-domain features resolves this case, because the NLoS path is always longer than the direct LoS path for the same transmitter and receiver positions, so the molecular absorption notch remains deeper for the NLoS link regardless of the size of the 15 diffuse cluster. The two feature families are therefore complementary. The time-domain family separates the classes when the diffuse cluster is strong, and the frequency-domain family maintains the separation when the diffuse cluster is weak. This complementarity is what allows the method that is the present invention discloses to remain reliable across the range NLoS conditions, and it is the reason both families are retained. 20 Then, in the next step of the method that is disclosed in the present invention; the THz receiver determines the LoS or NLoS condition based on the selected features. In the preferred embodiment of the present invention, the combining the ranked features as the LoS or NLoS decision with using a single decision statistic, a machine-learning and / or deep- learning classifier, that is applied to the same feature set. 25 In the final stage the features are mapped to a LoS or NLoS decision output. The identification is not tied to a particular classifier, because the discriminating information is contained in the features and in the two physical signatures identified before. A simple discrimination method that combines the ranked features into a single decision statistic can be used and / or machine- 23 learning and / or deep-learning classifier can be applied to the same feature set. Any method that separates the two-feature distribution can therefore be used to produce the decision. The present invention also discloses, a data processing device containing means to perform any method steps according to present invention. Also, the present invention discloses a computer program comprising instructions which, when the program is executed by the data processing 5 device, cause the data processing device to carry out the steps of the disclosed method in the present invention. Furthermore, the present invention discloses a computer-readable medium containing the computer program. 24 REFERENCES [1] I. F. Akyildiz, C. Han and S. Nie, "Combating the Distance Problem in the Millimeter Wave and Terahertz Frequency Bands," in IEEE Communications Magazine, vol. 56, no. 6, pp. 102- 108, June 2018. [2] K. Dovelos, M. Matthaiou, H. Q. Ngo and B. Bellalta, "Channel Estimation and Hybrid 5 Combining for Wideband Terahertz Massive MIMO Systems," in IEEE Journal on Selected Areas in Communications, vol. 39, no. 6, pp. 1604-1620, June 2021. [3] C. Han, A. O. Bicen and I. F. Akyildiz, "Multi-Ray Channel Modeling and Wideband Characterization for Wireless Communications in the Terahertz Band," in IEEE Transactions on Wireless Communications, vol. 14, no. 5, pp. 2402-2412, May 2015. 10 [4] S. Maranò, W. M. Gifford, H. Wymeersch and M. Z. Win, "NLOS identification and mitigation for localization based on UWB experimental data," in IEEE Journal on Selected Areas in Communications, vol. 28, no. 7, pp. 1026-1035, September 2010. [5] Y. Wu et al., "A Novel LoS / NLoS Identification-Assisted Positioning Method for 6G Indoor MIMO Communications," in IEEE Transactions on Communications, vol. 73, no. 12, 15 pp. 13578-13594, Dec. 2025. [6] I. Guvenc, C. -C. Chong and F. Watanabe, "NLOS Identification and Mitigation for UWB Localization Systems," 2007 IEEE Wireless Communications and Networking Conference, Hong Kong, China, 2007, pp. 1571-1576. [7] J. Xie, W. Wang, X. Liu, I. Rashdan, C. Di and J. Qin, "Identification of NLOS Based on 20 Soft Decision Method," in IEEE Wireless Communications Letters, vol. 12, no. 4, pp. 703-707, April 2023. [8] Z. Zeng, R. Bai, L. Wang and S. Liu, "NLOS identification and mitigation based on CIR with particle filter," 2019 IEEE Wireless Communications and Networking Conference (WCNC), Marrakesh, Morocco, 2019, pp. 1-6. 25 [9] Q. Wang et al., "A Novel NLOS Identification and Error Mitigation Method for UWB Ranging and Positioning," in IEEE Communications Letters, vol. 28, no. 1, pp. 48-52, Jan. 2024.

Claims

CLAIMS 1. A THz wideband channel characterization method for indoor LoS / NLoS identification characterized by; a. Transmitting a wideband pulse indoor by a THz transmitter, 5 b. Observing the transmitted wideband pulse under two propagation conditions bya THz receiver, wherein; I. One of the conditions is a LoS condition which the wideband pulse reaches the THz receiver through the direct path, and II. Other one is NLoS condition which the direct path is obstructed and the 10 pulse reaches the THz receiver after reflection and scattering from a rough surface the indoor, c. Forming wideband channel estimation from the received signal for each link by the THz receiver, d. Computing features from each channel observation by the THz receiver, 15 wherein; As time domain features; A. Computing RMS delay spread, B. Computing mean excess delay C. Computing kurtosis of the received envelope, 20 D. Computing group-delay spread, As frequency domain features; E. Computing molecular absorption notch depth, F. Computing coherence bandwidth, G. Computing spectral roll-off slope, 25 e. Separating medium-induced broadening from surface-induced broadening using the THz distance and surface signatures by the THz receiver, 26 f. Ranking the features computed in step d according to their separability between the LoS and NLoS conditions using the Fisher Discriminant Ratio (FDR) by the THz receiver; g. Selecting the features for LoS / NLoS classification based on the ranking obtained in step f, wherein, in a near-specular case in which the time-domain features exhibit 5 reduced variation, the frequency-domain features are prioritized by the THz receiver. h. Determining, by the THz receiver, the LoS or NLoS condition based on the selected features.

2. A method according to Claim 1, wherein; in step c, the receiver obtains 𝐻(𝑓) as the transfer function and ℎ(𝜏) as the impulse response when forming wideband channel estimation from the received signal for each link.

3. A method according to Claim 1 or 2, wherein; in the step a, the wideband pulse as 𝑠(𝑡) 15 emitted and transmitted by the THz transmitter, wherein, the wideband pulse’s spectrum as 𝑆(𝑓) occupies the band of 𝑓 ∈ [𝑓𝑐 − 𝐵 / 2, 𝑓𝑐 + 𝐵 / 2] wherein, the received signal by the THz receiver in the step b in the frequency domain computed by formula of; 𝑌(𝑓) = 𝐻(𝑓) 𝑆(𝑓) +𝑊(𝑓), wherein The 𝐻(𝑓) is the channel transfer function for the corresponding propagation condition 20 and 𝑊(𝑓) is additive white Gaussian noise, and the channel impulse response is obtained with computed by formula of; ℎ(𝜏) = ℱ−1{𝐻(𝑓)}, wherein The both 𝐻(𝑓) and ℎ(𝜏) are used to characterize the wideband pulse link in the frequency and time domains, respectively in the step c. 25 4. A method according to Claim 1, 2 or 3, wherein; in the step c, for each channel realization the THz receiver obtains the wideband transfer function 𝐻(𝑓) and the corresponding impulse response ℎ(𝜏), for the separation medium-induced broadening from surface-induced broadening in step e, under either the LoS hypothesis transfer 30 27 function 𝐻𝐿(𝑓) or the NLoS hypothesis transfer function 𝐻𝑁(𝑓) wherein the 𝐻𝐿(𝑓) computed by formula of; 𝐻𝐿(𝑓) = 𝑐 4𝜋𝑓𝑑𝐿  exp (− 1 2  𝑘(𝑓) 𝑑𝐿)  exp (−𝑗 2𝜋𝑓  𝑛(𝑓) 𝑑𝐿 𝑐 ) , 5 wherein 𝑐 is the speed of light, 𝑘(𝑓) is the molecular absorption coefficient of the medium, and 𝑑𝐿 is the direct path of length, and 𝐻𝑁(𝑓) computed by formula of; 𝐻𝑁(𝑓) = 𝐻𝑁 spec (𝑓) + 𝐻𝑁 diff(𝑓). wherein The specular component as 𝐻𝑁 spec (𝑓) follows the direction defined by the incidence angle 𝜃𝑖 and the reflection angle 𝜃𝑟, and it combines the Fresnel reflection coefficient of the surface material with a Rayleigh roughness attenuation that reduces the coherent reflection as the surface becomes rough and computed by formula of; 10 𝐻𝑁 spec (𝑓) = 𝑐 4𝜋𝑓𝑑𝑁  Γ(𝑓, 𝜃𝑖) exp (− 1 2  𝑔(𝑓))  exp (− 1 2  𝑘(𝑓) 𝑑𝑁)  exp (−𝑗 2𝜋𝑓  𝑛(𝑓) 𝑑𝑁 𝑐 ) , wherein Γ(𝑓, 𝜃𝑖) is the Fresnel reflection coefficient determined by the permittivity of the surface material, and 𝑔(𝑓) as the Rayleigh roughness factor that is computed by formula of; 𝑔(𝑓) = ( 4𝜋𝜎ℎ cos𝜃𝑖 𝜆 ) 2 = ( 4𝜋𝜎ℎ fcos 𝜃𝑖 𝑐 ) 2 , wherein 15 The 𝜎ℎ is standard deviation of the surface height and 𝜆 = 𝑐 / 𝑓 is wavelength, and the diffuse component 𝐻𝑁 diff(𝑓) accounts for the energy scattered away from the specular direction by the rough surface wherein using the image method, the illuminated surface patch is represented by an ensemble of scattering elements, and each element 𝑚 contributes a delayed and attenuated copy of the signal and the 𝐻𝑁 diff(𝑓) is computed by 20 formula of; 𝐻𝑁 diff(𝑓) =∑ 𝑐 4𝜋𝑓𝑑𝑚 𝑚  𝜌𝑑(𝜃𝑖 , 𝜃𝑠,𝑚) exp (− 1 2  𝑘(𝑓) 𝑑𝑚)  exp (−𝑗 2𝜋𝑓  𝑛(𝑓) 𝑑𝑚 𝑐 ) , Wherein 𝑑𝑚 is the total path length through the 𝑚-th surface element, 𝜃𝑠,𝑚 is corresponding scattering angle, and 𝜌𝑑(𝜃𝑖, 𝜃𝑠,𝑚) is Beckmann-Krichhoff diffuse scattering coefficient that governs the angular distribution of the scattered field, wherein, the two hypotheses comprise a LoS hypothesis 𝐻𝐿 and an NLoS hypothesis 25 𝐻𝑁 that differ according to propagation characteristics of a corresponding link, wherein the LoS hypothesis is characterized by frequency-dependent absorption and dispersion of a propagation medium, and wherein the NLoS hypothesis is additionally 28 characterized by an increased propagation path length and interaction with one or more rough surfaces causing reflected and diffuse-scattered components, thereby resulting in additional temporal spreading and spectral distortion relative to the LoS hypothesis.

5. A method according to Claim 1, wherein; in the step d, the RMS delay spread feature is the second central moment of the power delay profile.

6. A method according to Claim 1 or 5, wherein; in the step d, the mean excess delay feature is the centroid of the power delay profile measured relative to the first arrival. 10 7. A method according to Claim 1, 5 or 6, wherein; in step d, the kurtosis of the received envelope feature measures how peaked the response is.

8. A method according to Claim 1, 5, 6 or 7, wherein; in the step d, the group-delay spread 15 feature is the variation of the group delay across the band.

9. A method according to Claim 1, 5, 6, 7 or 8, wherein; in the step d, the molecular absorption notch depth feature is the depth of the water-vapor absorption feature in the received spectrum. 20 10. A method according to Claim 1, 5, 6, 7, 8 or 9, wherein; in the step d, the coherence bandwidth feature is the frequency separation over which the channel remains correlated, and it is inversely related to the delay spread feature.

11. A method according to Claim 1, wherein; in the step d, the spectral roll-off slope feature is the rate at which the received spectrum decays across the band.

12. A method according to Claim 1, wherein; in the step f, for ranking the features by their discriminating power in the step d with using the Ficher Discriminant Ratio that is 30 computed by formula of; 𝐽 = (𝜇LoS−𝜇NLoS) 2 𝜎LoS 2 +𝜎NLoS 2 , wherein wherein 𝐽 denotes the FDR of a respective feature, 𝜇LoS and 𝜇NLoS denote the mean values of the respective feature under the LoS and NLoS samples, respectively, and 29 𝜎LoS 2 and 𝜎NLoS 2 denote the corresponding variances across the LoS and NLoS samples, respectively.

13. A method according to Claim 1 or 12, wherein; in the step f, the Fisher Discriminant Ratio is computed for all features to A to D in step d. 5 14. A method according to Claim 1, wherein; in the step h, the combining the ranked features as the LoS or NLoS decision with using a single decision statistic, a machine- learning and / or deep-learning classifier, that is applied to the same feature set.

15. A data processing device containing means to perform any method steps according to claims 1 to 14.

16. A computer program comprising instructions which, when the program is executed by the data processing device of claim 15, cause the data processing device to carry out the 15 steps of the method of claim 1 to 14.

17. A computer-readable medium containing the computer program according to Claim 16.

Images