WO2021249634A1

WO2021249634A1 - Signal classification

Info

Publication number: WO2021249634A1
Application number: PCT/EP2020/066060
Authority: WO
Inventors: Oana-Elena Barbu; István Zsolt KOVÁCS
Original assignee: Nokia Technologies Oy
Priority date: 2020-06-10
Filing date: 2020-06-10
Publication date: 2021-12-16

Abstract

An apparatus, method and computer program is described comprising: generating a first signal matrix of a received first signal, wherein the first signal matrix is a time-frequency matrix of the first signal, and the first signal matrix comprises zero power channel subcarriers and pilot channel subcarriers; generating, using a first model, a first noise matrix using the first model based, at least in part, on a noise behaviour of the zero power channel subcarriers; generating, using a second model, a second signal matrix of a reduced noise first signal based, at least in part, on a removal of the first noise matrix from the pilot subcarriers; and generating, using a third model, at least one classification of the first signal based, at least in part, on the second signal matrix, wherein the classification is related to line of sight properties of the first signal.

Description

Signal Classification Field

The present specification relates to classification of a signal, such as classification related to line of sight properties.

Background

Algorithms for signal classification, for example based on estimated or measured time of arrival of signals, is known. There remains a need for improvements related to signal classification, for example in multi-path channel scenarios.

Summary

In a first aspect, this specification describes an apparatus comprising means for performing: generating a first signal matrix (e.g. the matrix B described below) of a received first signal, wherein the first signal matrix is a time-frequency matrix of the first signal, and the first signal matrix comprises zero power channel subcarriers and pilot channel subcarriers; generating, using a first model (e.g. a noise reconstruction machine-learning model), a first noise matrix based, at least in part, on a noise behaviour of the zero power channel subcarriers of the first signal matrix; generating, using a second model (e.g. a denoising machine-learning model), a second signal matrix of a reduced noise first signal based, at least in part, on a removal of the first noise matrix from the pilot channel subcarriers of the first signal matrix; and generating, using a third model (e.g. a classification model, such as a line-of-sight (LOS) detector machine-learning model), at least one classification of the first signal based, at least in part, on the second signal matrix, wherein the classification is related to line of sight properties of the first signal. The first matrix maybe generated using OFDM demodulation of a signal.

In some example embodiments, the at least one classification indicates one or more of line of sight components, non-line of sight components, and/ or attenuated line of sight components of the first signal.

The apparatus may further comprise means for performing: generating, using the third model, probabilities related to the one or more of the at least one classification. The apparatus may further comprise means for performing: generating a further first signal matrix of the received first signal, wherein the first signal matrix comprises a real part of the first signal, and the further first signal matrix comprises an imaginary part of the first signal; and generating a further first noise matrix, wherein the first noise matrix is based on a noise behaviour of a real part of the zero power channel subcarriers and the further first noise matrix is based on an imaginary part of the zero power channel subcarriers, wherein the second signal matrix comprises a real part of the reduced noise first signal and an imaginary part of the reduced noise first signal.

The first and further first signal matrices may be provided to a noise estimation module of the second model, and the first and further first noise matrices may be provided to a noise filtering module of the second model. The third model (e.g. a classification model) may comprise a feature extraction module and a classification module.

The apparatus may further comprise means for performing: collecting, in a channel buffer, a plurality of signal matrices corresponding to a plurality of reduced noise signals, wherein: the plurality of signal matrices includes the second signal matrix of the reduced noise first signal, the plurality of reduced noise signals includes the reduced noise first signal, and the second signal matrix comprises a real part of the reduced noise first signal and an imaginary part of the reduced noise first signal; and generating, using a fourth model (e.g. a line-of-sight forecasting machine-learning model), a forecast for line of sight properties of a first number of future time steps and/or a first number of future signals, based, at least in part, on the plurality of signal matrices corresponding to the plurality of reduced noise signals. The said generating may be performed based on one or more triggers, wherein the one or more triggers are based on one or more of a change in a user device, change in properties of the user device, change in physical orientation of the user device, change in radio conditions experienced by the user device, and/ or change in network topology. The first number of future time steps and/ or the first number of future signals may, for example, be determined based on a type and/ or velocity of the user device.

In some example embodiments, one or more (e.g. all) of the first, second, third, and/or fourth models comprise neural network(s).

The apparatus may further comprise means for performing: estimating a time of arrival of the first signal between a user device and a base station based on the generated classification. The means may comprise: at least one processor; and at least one memory including computer program code, the at least one memory and the computer program configured, with the at least one processor, to cause the performance of the apparatus.

In a second aspect, this specification describes a method comprising: generating a first signal matrix of a received first signal, wherein the first signal matrix is a time- frequency matrix of the first signal, and the first signal matrix comprises zero power channel subcarriers and pilot channel subcarriers; generating, using a first model, a first noise matrix using the first model based, at least in part, on a noise behaviour of the zero power channel subcarriers; generating, using a second model, a second signal matrix of a reduced noise first signal based, at least in part, on a removal of the first noise matrix from the pilot subcarriers; and generating, using a third model, at least one classification of the first signal based, at least in part, on the second signal matrix, wherein the classification is related to line of sight properties of the first signal. The first matrix may be generated using OFDM demodulation of a signal.

The at least one classification may indicate one or more of line of sight components, non-line of sight components, and/or attenuated line of sight components of the first signal.

The method may further comprise: generating, using the third model, probabilities related to the one or more of the at least one classification. The method may further comprise: generating a further first signal matrix of the first signal, wherein the first signal matrix comprises a real part of the first signal, and the further first signal matrix comprises an imaginaiy part of the first signal; and determining a further first noise matrix, wherein the first noise matrix is based on a noise behaviour of a real part of the zero power channel subcarriers and the further first noise matrix is based on an imaginary part of the zero power channel subcarriers, wherein the second signal matrix comprises a real part of the reduced noise first signal and an imaginary part of the reduced noise first signal. The first and further first signal matrices may be provided to a noise estimation module of the second model, and the first and further first noise matrices are provided to a noise filtering module of the second model. The third model (e.g. a classification model) may comprise a feature extraction module and a classification module. The method may further comprise: collecting, in a channel buffer, a plurality of signal matrices corresponding to a plurality of reduced noise signals, wherein: the plurality of signal matrices includes the second signal matrix of the reduced noise first signal, the plurality of reduced noise signals includes the reduced noise first signal, and the second signal matrix comprises a real part of the reduced noise first signal and an imaginaiy part of the reduced noise first signal; and generating, using a fourth model (e.g. a line- of-sight forecasting machine-learning model), a forecast for line of sight properties of a first number of future time steps and/ or a first number of future signals, based, at least in part, on the plurality of signal matrices corresponding to the plurality of reduced noise signals. The said generating may be performed based on one or more triggers, wherein the one or more triggers are based on one or more of a change in a user device, change in properties of the user device, change in physical orientation of the user device, change in radio conditions experienced by the user device, and/ or change in network topology. The first number of future time steps and/ or the first number of future signals may be determined based on a type and/or velocity of the user device.

The method may further comprise: estimating a time of arrival of the first signal between a user device and a base station based on the generated classification.

In a third aspect, this specification describes an apparatus configured to perform (at least) any method as described with reference to the second aspect.

In a fourth aspect, this specification describes computer-readable instructions which, when executed by computing apparatus, cause the computing apparatus to perform (at least) any method as described with reference to the second aspect.

In a third aspect, this specification describes a computer-readable medium (such as a non-transitory computer-readable medium) comprising program instructions stored thereon for performing (at least) any method as described above with reference to the second aspect.

In a fourth aspect, this specification describes an apparatus comprising: at least one processor; and at least one memory including computer program code which, when executed by the at least one processor, causes the apparatus to perform (at least) any method as described above with reference to the third to fifth aspects. In a fifth aspect, this specification describes a computer program comprising instructions for causing an apparatus to perform at least the following: generating a first signal matrix of a received first signal, wherein the first signal matrix is a time- frequency matrix of the first signal, and the first signal matrix comprises zero power channel subcarriers and pilot channel subcarriers; generating, using a first model (e.g. a noise reconstruction machine-learning model), a first noise matrix based, at least in part, on a noise behaviour of the zero power channel subcarriers of the first signal matrix; generating, using a second model (e.g. a denoising machine-learning model), a second signal matrix of a reduced noise first signal based, at least in part, on a removal of the first noise matrix from the pilot channel subcarriers of the first signal matrix; and generating, using a third model (e.g. a classification model, such as a line-of-sight (LOS) detector machine-learning model), at least one classification of the first signal based, at least in part, on the second signal matrix, wherein the classification is related to line of sight properties of the first signal.

In a sixth aspect, this specification describes an apparatus comprising: means (such as a module of a user device) for generating a first signal matrix (such as the matrix B described below) of a received first signal, wherein the first signal matrix is a time- frequency matrix of the first signal, and the first signal matrix comprises zero power channel subcarriers and pilot channel subcarriers; means (e.g. a first model, such as a noise reconstruction machine-learning model) for generating a first noise matrix based, at least in part, on a noise behaviour of the zero power channel subcarriers of the first signal matrix; means (such as a second mode, such as a denoising machine-learning model) for generating a second signal matrix of a reduced noise first signal based, at least in part, on a removal of the first noise matrix from the pilot channel subcarriers of the first signal matrix; and means (such as a third model, such as a classification model, which may take the form of a LOS detector machine-learning model) for generating at least one classification of the first signal based, at least in part, on the second signal matrix, wherein the classification is related to line of sight properties of the first signal.

Brief description of the drawings

Example embodiments will now be described, by way of non-limiting examples, with reference to the following schematic drawings, in which:

FIG. 1 is a plot showing a power delay profile; FIG. 2A is a block diagram of an example scenario in accordance with an example embodiment;

FIG. 2B is a plot of an example power delay profile in accordance with an example embodiment; FIG. 3 shows a matrix in accordance with an example embodiment;

FIG. 4 is a flowchart of an algorithm in accordance with an example embodiment;

FIGs. 5 and 6 are block diagrams of systems in accordance with example embodiments; FIG. 7 is a flowchart of an algorithm in accordance with an example embodiment;

FIG. 8 shows heat maps of matrices in accordance with an example embodiment; FIGs. 9 and to are block diagrams of systems in accordance with example embodiments;

FIG. it is a flowchart of an algorithm in accordance with an example embodiment;

FIG. 12 is a block diagram of a system in accordance with an example embodiment;

FIG. 13 shows a neural network in accordance with an example embodiment; FIG. 14 is a block diagram of a system in accordance with an example embodiment; and

FIGS. 15A and 15B show tangible media, respectively a removable non-volatile memory unit and a compact disc (CD) storing computer-readable code which when run by a computer perform operations according to embodiments. Detailed description

The scope of protection sought for various embodiments of the invention is set out by the independent claims. The embodiments and features, if any, described in the specification that do not fall under the scope of the independent claims are to be interpreted as examples useful for understanding various embodiments of the invention.

In the description and drawings, like reference numerals refer to like elements throughout. FIG. 1 is an example plot, indicated generally by the reference numeral 10 showing a power delay profile of a wireless propagation channel. The plot 10 comprises a profile of power (vertical axis) of a signal with respect to delay (horizontal axis), where the delay may represent time a signal may take to travel a distance between a transmitter and a receiver. A user equipment (UE) may measure a time-of-arrival (TOA) of a signal, for example, for localization purposes. The UE may then report the measurements of the TOA to a communication node, such as a base station. The UE may also report Reference Signal Time Difference (RSTD), where the RSTD may represent differences in the TO A. The TOA is a metric for the shortest time a signal takes to travel the distance between a transmitter and a receiver. If TOA is measured correctly, then the distance (d) between the transmitter and the receiver can be accurately obtained as d=TOA x c, where c = speed of light. To compute TOA, a power-delay profile (PDP) of the wireless propagation channel may be estimated, such that a TOA may be selected as the delay at which the PDP exhibits a power peak. For example, in plot to, a peak 11 is detected. However, the peak 11 may or may not correspond to a line of sight path, as described below with reference to FIGs. 2A and 2B.

FIG. 2A is a block diagram of an example scenario, indicated generally by the reference numeral 20A, in accordance with an example embodiment. The scenario 20A shows a communication node 21 (for example, a base station) and a user equipment 25. The communication node 21 may transmit one or more signals to the user equipment 25. The scenario 20A further shows a building 23 and a tree 24. A signal may be transmitted from the communication node 21 to the user equipment 25 through a wireless communication channel composed of multiple paths, such as a path 26, and/or a path 27. The path 26 maybe a non-line of sight (NLOS) path composed of a path 26a and a path 26b. The path 27 maybe an attenuated line of sight (LOS) path, as the signal may be attenuated by the tree 24.

Wireless communication channels may be composed of multiple paths, each characterized by a complex channel gain and a delay. One or more of these paths can be attenuated or fully obstructed by the various obstacles in the network, therefore the channel seen at the receiver may not necessarily contain a LOS component. In the example scenario 20A, the tree 24 attenuates the LOS path 27 compared with the NLOS path 26 (e.g. the NLOS path may be reflected by the building 23). The signal observed by the user equipment 25 is described below with reference to FIG. 2B. FIG. 2B is a plot, indicated generally by the reference numeral 20B, of an example power delay profile in accordance with an example embodiment. The plot 20B shows an LOS component 28 and NLoS component 29 of a signal received and observed at the user equipment 25. The LoS component 28 corresponds to the attenuated LOS path 27, and the NLOS component 29 corresponds to the NLOS path 26. The user equipment 25 observing the received signal may select the TOA as the delay of the NLOS component 29 (since its power is highest of the two received signals). This selection may bias the overall location estimate, since this component does not reflect the shortest distance between the communication node and the UE.

In order to accurately determine the ToA and subsequently the distance between the communication node 21 and the user equipment 25, a LOS component may need to be identified. However, the strongest component may not always correspond to the LOS (line of sight) path, as depicted in FIGs. 2A and 2B. Furthermore, the radio environment may be dynamic, particularly in cmWave and mmWave frequency bands, such that any movement of the UE and/or radio obstacles in the environment can cause transitions from LOS to NLOS conditions. In practice, therefore, the UE may not experience pure LOS or pure NLOS propagation conditions and determining which of these conditions is dominant, and for how long, in the received signals can become a computationally intensive task. FIG. 3 shows a matrix, indicated generally by the reference numeral 30, in accordance with an example embodiment. The matrix 30 may be a time frequency matrix generated by utilizing one or more noisy time-stacked channel frequency responses (NTS-CFR) of a received signal. The matrix 30 comprises zero-power subcarriers 31 and pilot channel subcarriers 32. The signal model is described below. The zero-power subcarriers 31 may comprise zero transmit signal power, such that the receiver may only receive power corresponding to the noise, and no power corresponding to data.

The pilot channel subcarriers 32 may correspond to data transmission and/or pilot transmission. For example, an orthogonal frequency-division multiplexing (OFDM) system may be used, where the OFDM system may have a bandwidth Nf_s, where f_s is the subcarrier spacing. The transmitter generates a column-vector of K complex symbols x, which it OFDM modulates into a vector

. The vector m may be attached to a cyclic prefix (CP) that is g-samples long The signal

kTs ), is sent over the wireless propagation channel with impulse response h(j ) = The received signal y(t) = (h * s)(t) + n(t) is sampled, the cyclic prefix is

removed, and the resulting sample vector is OFDM demodulated. The signal n(t) collects noise and interference contributions. The sample vector at the K subcarriers during the p-th OFDM symbol reads y_p = h^(p) + n_p, where h^(p) is the channel frequency response (CFR) K-length column vector at the K pilot subcarriers, and n_p comprises a sampled noise and/or interference at the K pilot subcarriers. The sample vector at the remaining N-K subcarriers reads

Note that n_p may not be separated or filtered out from the y_p , as h^(p) and/or n_p maybe unknown before considering the zero power subcarriers and The K pilots need not be contiguously allocated. The extended

channel frequency response vector is built by interleaving and to obtain the N-

length channel vector during the p-th OFDM symbol , where b_p(α ) = y_p(α), if ‘a’ is a

pilot subcarrier and b_p(a ) = y^zp (a), otherwise.

A channel matrix, such as the matrix 30, may be built by stacking real and imaginary parts vertically, and C OFDM symbols horizontally:

where the real and imaginary parts of the column vector b_p are defined as

FIG. 4 is a flowchart of an algorithm, indicated generally by the reference numeral 40, in accordance with an example embodiment.

The algorithm 40 starts at operation 42, where a first signal matrix (e.g. the matrix 30) of a received first signal is generated, for example at a first module of a user device (such as the user equipment 25). The received first signal may comprise a received time-frequency radio signal. For example, the first signal matrix may be a time- frequency matrix comprising zero power channel subcarriers and pilot channel subcarriers (e.g. the zero power carriers 31 and pilot channel subcarriers 32 described above). Pilot channel subcarriers may contain information about the noisy channel frequency response, while the zero power subcarriers may be used for determining a noise vector, as described in further detail below. The first signal matrix may be based on OFDM demodulation of the received first signal.

Next, at operation 44, a first noise matrix maybe generated, using a first model, based at least in part on a noise behaviour of the zero power channel subcarriers (e.g. the subcarriers 31) of the first signal matrix. The first model may be a machine learning model, such as a neural network. In an example embodiment, the zero power channel subcarriers do not carry any data, and any power carried by the zero power channel subcarriers maybe related to noise and/or interference. As such, the zero power channel subcarriers may can be used for determining noise behaviour of the received first signal (e.g. noise behaviour of the pilot subcarriers), such that the noise behaviour may be used for removing noise from the received first signal, and identifying LOS properties of the reduced noise first signal.

At operation 46, a second signal matrix of a reduced noise first signal may be generated, using a second model, based at least in part on a removal of the first noise matrix from the pilot channel subcarriers of the first signal matrix. The second model may be a machine learning model, such as a neural network. As the first noise matrix generated at operation 44 may represent the noise in the received first signal (including the pilot channel subcarriers), removal or cancellation of the first noise matrix from the pilot channel subcarriers may generate a second signal matrix of a reduced noise first signal (e.g. noise-free first signal). At operation 48, at least one classification of the first signal may be generated, using a third model, based at least in part on the second signal matrix. The classification may be related to line of sight (LOS) properties of the first signal. The third model may be a machine learning model, such as a neural network. For example, the at least one classification indicates one or more of line of sight components, non-line of sight components, and/or attenuated line of sight components of the first signal. In an example embodiment, the third model may further be used for generating probabilities related to the classification(s).

For example, the classification may allow a user device (e.g. UE 25) to identify (or estimate) whether a received signal comprises a LOS component, a NLOS component (e.g. NLOS component 29), or an attenuated LOS component (e.g. attenuated LOS component 28). Based on the classification, the user device may then accurately determine a time of arrival, and subsequently the distance between the user device and the communication node (e.g. a distance between the UE 25 and the communication node 21).

In an example embodiment, noise cancellation may be performed on the first signal to generate a signal matrix corresponding to a reduce noise first signal by implementing a machine learning architecture as a cascade of machine learning blocks (e.g. the first model and the second model). For example, the first model may comprise a generative adversarial network and the second model may comprise a deep convolutional neural network.

In an example embodiment, the classification may be generated by the third model, where the third model may comprise a machine learning block. The machine learning block of the third model may comprise any of one or more convolutional neural network based architectures (e.g. ResNet, GAN, combination with RNN/GRU/LSTM, etc.), reinforcement learning (RL) based architectures, and/or deep unfolding architectures. FIG. 5 is a block diagram of a system, indicated generally by the reference numeral 50, in accordance with an example embodiment. For example, one or more operations (e.g. operations 42 to 46) of algorithm 40 maybe carried out using the system 50. The system 50 comprises a noise reconstruction module 51 and a denoising module 56. For example, the noise reconstruction module 51 may comprise a noise extraction module 54 and a first model 55 (e.g. the first model used in operation 44), and the denoising module 56 may comprise a second model 57 (e.g. the second model used in operation 46), and a second signal matrix 58 (e.g. generated in operation 46).

The noise extraction module 54 may receive (e.g. operation 42) as inputs the first signal matrix 53 (channel matrix B, e.g. an initial noisy matrix of a first signal) and a noise indicator (‘z’) 52. The noise extraction module 54 may determine a noise behaviour of one or more zero power subcarriers, and the noise behaviour may be used by the first model 55 to generate a first noise matrix. For example, the noise behaviour may comprise a noise-power distribution model in a frequency domain.

For example, according to the signal model described above with reference to FIG. 3, the received signal during the p-th OFDM symbol, at the a-th subcarrier may be b_p (α) = y_p(α), in the event that ‘α’ is a pilot subcarrier. In the event that ‘a’ is a zero-power subcarrier, the p-th OFDM symbol at the a-th subcarrier maybe b_p(α ) = y_p ^p (α).

The first signal matrix (e.g. the channel matrix B) maybe prepared such that, column- wise, C time-instances of the demodulated received signal may be concatenated. Rowwise, the channel matrix B contains the real and imaginary parts of the received signal at the different pilot and zero power subcarriers:

An associated column vector is built for a noise indicator z, where the column vector has the same number of rows as the channel matrix B, and the a-th entry of the column vector may be defined as z(α) = 0 x (α == pilot ) + 1 x (α == ZP ) . Matrix B and vector z maybe provided as inputs to the noise reconstruction module 51, such that a noise model may be learned, for example, using the first model 55. For example, the first model 55 may comprise a generative adversarial network, and may output the first noise matrix based on inputs 52 and 53. The first noise matrix may then be provided as inputs to the second model 57 that generates a second signal matrix 58 (B_f) corresponding to a reduced noise (e.g. noise free) first signal. The denoising module 56, produces the second signal matrix 58 (e.g. a reduced noise or noise-free version of the matrix B). The second model 57 may comprise a convolutional neural network.

FIG. 6 is a block diagram of a system, indicated generally by a reference numeral 6o, in accordance with an example embodiment. System 6o comprises classification module 6i which may receive as input a signal matrix, such as the second signal matrix 58, corresponding to a reduced noise first signal (e.g. noise free first signal). The classification module 61 may comprise a third model 63 (e.g. the third model used in operation 48) for determining at least one classification for the received first signal based on the second signal matrix 58. The third model 63 may be a convolutional neural network (CNN). The third model 63 then generates a classification 64, where the classification may indicate whether a component of the received first signal is a LOS component, a NLOS component, or an attenuated LOS component. The third model 63 may optionally further generate a probability 65 indicating the probability of the component being a LOS component, a NLOS component, and/or an attenuated LOS component. For example, the classification 64 maybe generated based on whether the probability 65 of the classification 64 being accurate is over a threshold.

FIG. 7 is a flowchart of an algorithm, indicated generally by the reference numeral 70, in accordance with an example embodiment. The algorithm 70 comprises the operations 42 to 48 described above. The algorithm 70 further describes optional operations 72, 74 and 76. At operation 72, a further first signal matrix may be generated, for example, at the first module of the user device (e.g. UE 25). For example, the first signal matrix generated at operation 42 may comprise a real part of the first signal, and the further first signal matrix may comprise an imaginary part of the first signal. At operation 74, a further first noise matrix may be generated. For example, the first noise matrix may be based on a noise behaviour of a real part of the zero power channel subcarriers and the further first noise matrix may be based on an imaginary part of the zero power channel subcarriers. For example, the OFDM demodulation may be used to separately generate a first signal matrix

and a further first signal matrix

The first model 55 may, based on the inputs of the first signal matrix and the further first signal matrix, generate a first noise matrix and a further first

noise matrix

At operation 76, the classification (e.g. generated at operation 48) may be used for estimating a time of arrival metric of the received first signal, which may subsequently be used for determining a distance between a communication node and a user device (e.g. the communication node 21 and the UE 25 of the scenario 20A described above). FIG.8 shows heat maps of matrices, indicated generally by the reference numeral 80, in accordance with an example embodiment. The heat maps 80 comprises a first heat map 81, a second heat map 82, a third heat map 83, and a fourth heat map 84 for a first signal matrix, further first signal matrix, first noise matrix, and a further first noise matrix respectively. The first heat map 81 may represent the first signal matrix (M₁ = corresponding to a real part of the estimated channel of the first signal;

the second heat map 82 may represent the further first signal matrix (M₂ =

corresponding to an imaginary part of the estimated channel of the first signal; the third heat map 83 may represent the first noise matrix (M₃ = corresponding to a real part of the estimated noise of the first signal;

and the fourth heat map 84 may represent the further first noise matrix (M₄ = corresponding to an imaginary part of the estimated noise of the first

signal. In an example embodiment, instead of stacking real and imaginary parts of a first signal in a single matrix, a channel matrix for a real part of the first signal (e.g. the first signal matrix, Ml) and a channel matrix for an imaginary part of the first signal (e.g. the further first signal matrix, M2) may separately by provided as inputs to a model (e.g. the second model 57) in a denoising machine learning module. Similarly, a matrix for the noise in a first signal (e.g. the first noise matrix, M3) and a matrix for an imaginary part of the first signal (e.g. the further first noise matrix, M4) may separately be provided as inputs to the model (e.g. the second model 57) in the denoising machine learning module. This is described in further detail below.

Fig. 9 is a block diagram of a system, indicated generally by the reference numeral 90, in accordance with an example embodiment. System 90 comprising a denoising module 95, which may comprise a noise estimation module 96 and a noise filtering module 97. The noise estimation module 96 may receive as inputs a first signal matrix 91 (Ml, e.g. corresponding to a real part of first signal), and a further first signal matrix 92 (M2, e.g. corresponding to an imaginary part of the first signal). For example, the first signal matrix 91, the further first signal matrix 92 may be generated at a noise reconstruction module (e.g. the noise reconstruction module 51).

One or more outputs of the noise estimation module 96 may be provided to the noise filtering module 97. The outputs of the noise estimation module 96 may comprise an initial noise estimation (e.g. estimation of signal interference noise ratio (SINR)). The initial noise estimation may further be refined and/ or filtered at the noise filtering module 97, for example, based on a first noise matrix 93 and a further noise matrix 94. The noise filtering module 97 receives as inputs the first noise matrix 93 (M3, e.g. corresponding to a real part of an estimated noise of the first signal), and the further first noise matrix 94 (e.g. M4, e.g. corresponding to an imaginaiy part of an estimated noise of the first signal). The noise filtering 97 may then generate a second signal matrix 98 (e.g. corresponding to a real part of a reduced noise or noise-free first signal) and a further second signal matrix 99 (e.g. corresponding to an imaginary part of the reduced noise or noise-free first signal).

FIG. 10 is a block diagram of a system, indicated generally by the reference numeral too, in accordance with an example embodiment. System too comprises a classification module 103. The classification module 103 receives as inputs the second signal matrix 98 (e.g. corresponding to a real part of a reduced noise or noise-free first signal) and a further second signal matrix 99 (e.g. corresponding to an imaginary part of the reduced noise or noise-free first signal). Based on the second signal matrix and the further second signal matrix, the classification module 103 may generate a classification 106, such that the classification 106 may be related to LOS properties of the first signal, e.g. an identification of LOS, NLOS, or attenuated LOS components. In an example embodiment, the classification module 103 may further generate a probability score, where the probability score may indicate the probability of the component being a LOS component, a NLOS component, and/or an attenuated LOS component. For example, the classification 106 may be generated based on whether the probability score of the classification 106 being accurate is over a threshold.

In an example embodiment, the classification module 103 may comprise a feature extraction model 104 and a classification model 105. The feature extraction model i04may determine features of the first signal that accurately represent the frequency- time-delay characteristics of the propagation channel of the first signal. The determined features may then be used by the classification model 105 for generating the classification 106. The feature extraction model 104 and the classification model 105 may together form the third model used at operation 48 for generating the classification.

FIG. 11 is a flowchart of an algorithm, indicated generally by the reference numeral 110, in accordance with an example embodiment. The algorithm 110 starts at operation 112, where a plurality of signal matrices corresponding to a plurality of reduced noise signals are collected (e.g. in a channel buffer). For example, the plurality of signal matrices includes the second signal matrix (e.g. generated at operation 46) of the reduced noise first signal for a plurality of reduced noise signals. The plurality of reduced noise signals includes the reduced noise first signal described above. In an example embodiment, the second signal matrix may comprise a real part of the reduced noise first signal and an imaginary part of the reduced noise first signal.

Next, at operation 114, a forecast is generated, using a fourth model, for line of sight properties of a first number of future time steps and/or a first number of future signals. The forecast may be generated based at least in part on the plurality of signal matrices corresponding to the plurality of reduced noise signals. The fourth model may be a machine learning model, such as a neural network. The forecasting is described in further detail below with reference to FIG. 12. FIG. 12 is a block diagram of a system, indicated generally by the reference numeral

120, in accordance with an example embodiment. Channel matrices 121 maybe collected in a channel buffer 122. The channel matrices may correspond to signal matrices (e.g. second signal matrix generated in operation 46) of a plurality of reduced noise signals. At the module 124, a first number of latest entries of the channel matrices may be extracted from the channel buffer 122. The channel matrices may then provided as input to the fourth model 125. The fourth model may then generate a forecast 126 for line of sight properties of a first number of future time steps (e.g. estimated line of sight properties for signals received in the next 500 milliseconds), or of a first number of future signals (e.g. estimated line of sight properties for the next ten received signals). In an example embodiment, the first number of future time steps and/ or the first number of future signals are determined based on a type (e.g. hand-held UE, machine type communications (MTC) UE, vehicular UE, relay UE, and/ or other types of user device) and/or velocity of the user device. In an example embodiment, the forecast may be generated based on one or more triggers 123. For example, the one or more triggers may be based on one or more of a change in user device, change in properties (e.g. velocity) of user equipment, change in physical orientation of the user device (e.g. user device being moved from one location to another), change in radio conditions experienced by user device, and/ or change in network topology (e.g. on a request from a location server). For example, such changes may result in a change from LOS to NLOS path or attenuated LOS path, or vice versa. For example, the user device may generate the forecast periodically or when a trigger (e.g. standardized trigger message) is received. For example, the trigger may be received when a velocity of the user device changes (e.g. from pedestrian to vehicular).

FIG. 13 shows a neural network, indicated generally by the reference numeral 130, used in some example embodiments. As described above, one or more of the first model, second model, third model, and fourth model may comprise a machine learning model, similar to the neural network 130. The neural network 130 comprises an input layer 131, one or more hidden layers 132, and an output layer 133. The hidden layers 132 may comprise a plurality of hidden nodes, where the processing may be performed based on the inputs received. At the output layer 133, one or more outputs may be generated.

For example, when the neural network 130 is used to implement as the first model, the neural network 130 may be trained, at a training stage, with inputs comprising a plurality of training noisy signals. For example, during an inference stage, at the input layer 131, the first signal matrix of the first signal may be received (e.g. operation 42). The hidden layers 132 may perform processing and outputs, such as the first noise matrix may be generated (e.g. operation 44) at the output layer 133.

In another example, when the neural network 130 is used to implement the second model, the neural network 130 may be trained, at a training stage, with inputs comprising a plurality of training noise matrix and signal matrix. For example, during an inference stage, at the input layer 131, the first noise matrix and the first signal matrix of the first signal may be received. The hidden layers 132 may perform processing and outputs, such as the second signal matrix of a reduced noise first signal maybe generated (e.g. operation 46) at the output layer 133.

In another example, when the neural network 130 is used to implement the third model, the neural network 130 may be trained, at a training stage, with inputs comprising a plurality of signal matrix of reduced noise signals. For example, during an inference stage, at the input layer 131, the second signal matrix of the reduced noise first signal may be received. The hidden layers 132 may perform processing and outputs, such as the classification (e.g. LOS, NLOS, or attenuated LOS classification) of first signal may be generated (e.g. operation 48) at the output layer 133.

For completeness, FIG. 14 is a schematic diagram of components of one or more of the example embodiments described previously, which hereafter are referred to generically as a processing system 300. The processing system 300 may, for example, be the apparatus referred to in the claims below.

The processing system 300 may have a processor 302, a memory 304 closely coupled to the processor and comprised of a RAM 314 and a ROM 312, and, optionally, a user input 310 and a display 318. The processing system 300 may comprise one or more network/apparatus interfaces 308 for connection to a network/apparatus, e.g. a modem which may be wired or wireless. The network/ apparatus interface 308 may also operate as a connection to other apparatus such as device/apparatus which is not network side apparatus. Thus, direct connection between devices/apparatus without network participation is possible. The processor 302 is connected to each of the other components in order to control operation thereof.

The memory 304 may comprise a non-volatile memory, such as a hard disk drive (HDD) or a solid state drive (SSD). The ROM 312 of the memory 304 stores, amongst other things, an operating system 315 and may store software applications 316. The RAM 314 of the memory 304 is used by the processor 302 for the temporary storage of data. The operating system 315 may contain code which, when executed by the processor implements aspects of the algorithms 40, 70, and 110 described above. Note that in the case of small device/apparatus the memory can be most suitable for small size usage i.e. not always a hard disk drive (HDD) or a solid state drive (SSD) is used.

The processor 302 may take any suitable form. For instance, it may be a microcontroller, a plurality of microcontrollers, a processor, or a plurality of processors.

The processing system 300 maybe a standalone computer, a server, a console, or a network thereof. The processing system 300 and needed structural parts may be all inside device/apparatus such as IoT device/apparatus i.e. embedded to very small size. In some example embodiments, the processing system 300 may also be associated with external software applications. These may be applications stored on a remote server device/apparatus and may run partly or exclusively on the remote server device/apparatus. These applications maybe termed cloud-hosted applications. The processing system 300 maybe in communication with the remote server device/ apparatus in order to utilize the software application stored there.

FIGS. 15A and 15B show tangible media, respectively a removable memory unit 365 and a compact disc (CD) 368, storing computer-readable code which when run by a computer may perform methods according to example embodiments described above. The removable memory unit 365 may be a memory stick, e.g. a USB memory stick, having internal memory 366 storing the computer-readable code. The internal memory 366 may be accessed by a computer system via a connector 367. The CD 368 may be a CD-ROM or a DVD or similar. Other forms of tangible storage media may be used. Tangible media can be any device/apparatus capable of storing data/information which data/information can be exchanged between devices/apparatus/network.

Embodiments of the present invention may be implemented in software, hardware, application logic or a combination of software, hardware and application logic. The software, application logic and/or hardware may reside on memory, or any computer media. In an example embodiment, the application logic, software or an instruction set is maintained on any one of various conventional computer-readable media. In the context of this document, a “memory” or “computer-readable medium” may be any non-transitory media or means that can contain, store, communicate, propagate or transport the instructions for use by or in connection with an instruction execution system, apparatus, or device, such as a computer.

Reference to, where relevant, “computer-readable medium”, “computer program product”, “tangibly embodied computer program” etc., or a “processor” or “processing circuitry” etc. should be understood to encompass not only computers having differing architectures such as single/multi-processor architectures and sequencers/parallel architectures, but also specialised circuits such as field programmable gate arrays

FPGA, application specify circuits ASIC, signal processing devices/apparatus and other devices/apparatus. References to computer program, instructions, code etc. should be understood to express software for a programmable processor firmware such as the programmable content of a hardware device/apparatus as instructions for a processor or configured or configuration settings for a fixed function device/ apparatus, gate array, programmable logic device/apparatus, etc.

If desired, the different functions discussed herein may be performed in a different order and/or concurrently with each other. Furthermore, if desired, one or more of the above-described functions may be optional or may be combined. Similarly, it will also be appreciated that the flow diagrams of Figures 4, 7, and 11 are examples only and that various operations depicted therein may be omitted, reordered and/ or combined.

It will be appreciated that the above described example embodiments are purely illustrative and are not limiting on the scope of the invention. Other variations and modifications will be apparent to persons skilled in the art upon reading the present specification.

Moreover, the disclosure of the present application should be understood to include any novel features or any novel combination of features either explicitly or implicitly disclosed herein or any generalization thereof and during the prosecution of the present application or of any application derived therefrom, new claims may be formulated to cover any such features and/ or combination of such features. Although various aspects of the invention are set out in the independent claims, other aspects of the invention comprise other combinations of features from the described example embodiments and/ or the dependent claims with the features of the independent claims, and not solely the combinations explicitly set out in the claims. It is also noted herein that while the above describes various examples, these descriptions should not be viewed in a limiting sense. Rather, there are several variations and modifications which may be made without departing from the scope of the present invention as defined in the appended claims.

Claims

1. An apparatus comprising means for performing: generating a first signal matrix of a received first signal, wherein the first signal matrix is a time-frequency matrix of the first signal, and the first signal matrix comprises zero power channel subcarriers and pilot channel subcarriers; generating, using a first model, a first noise matrix based, at least in part, on a noise behaviour of the zero power channel subcarriers of the first signal matrix; generating, using a second model, a second signal matrix of a reduced noise first signal based, at least in part, on a removal of the first noise matrix from the pilot channel subcarriers of the first signal matrix; and generating, using a third model, at least one classification of the first signal based, at least in part, on the second signal matrix, wherein the classification is related to line of sight properties of the first signal.

2. An apparatus as claimed in claim l, wherein the at least one classification indicates one or more of line of sight components, non-line of sight components, and/or attenuated line of sight components of the first signal.

3. An apparatus as claimed in any one of claim l and claim 2, further comprising means for performing: generating, using the third model, probabilities related to the one or more of the at least one classification.

4. An apparatus as claimed in any one of the preceding claims, further comprising means for performing: generating a further first signal matrix of the received first signal, wherein the first signal matrix comprises a real part of the first signal, and the further first signal matrix comprises an imaginary part of the first signal; and generating a further first noise matrix, wherein the first noise matrix is based on a noise behaviour of a real part of the zero power channel subcarriers and the further first noise matrix is based on an imaginaiy part of the zero power channel subcarriers, wherein the second signal matrix comprises a real part of the reduced noise first signal and an imaginaiy part of the reduced noise first signal.

5. An apparatus as claimed in claim 4, wherein the first and further first signal matrices are provided to a noise estimation module of the second model, and the first and further first noise matrices are provided to a noise filtering module of the second model.

6. An apparatus as claimed in any one of claims 4 and 5, wherein the third model comprises a feature extraction module and a classification module.

7. An apparatus as claimed in any one of the preceding claims, further comprising means for performing: collecting, in a channel buffer, a plurality of signal matrices corresponding to a plurality of reduced noise signals, wherein: the plurality of signal matrices includes the second signal matrix of the reduced noise first signal, the plurality of reduced noise signals includes the reduced noise first signal, and the second signal matrix comprises a real part of the reduced noise first signal and an imaginary part of the reduced noise first signal; and generating, using a fourth model, a forecast for line of sight properties of a first number of future time steps and/or a first number of future signals, based, at least in part, on the plurality of signal matrices corresponding to the plurality of reduced noise signals.

8. An apparatus as claimed in claim 7, wherein the generating is performed based on one or more triggers, wherein the one or more triggers are based on one or more of a change in a user device, change in properties of the user device, change in physical orientation of the user device, change in radio conditions experienced by the user device, and/or change in network topology.

9. An apparatus as claimed in any one of claims 7 and 8, wherein the first number of future time steps and/ or the first number of future signals are determined based on a type and / or velocity of the user device.

10. An apparatus as claimed in any one of the preceding claims, wherein one or more of the first, second, third, and/or fourth models comprise neural network(s).

11. An apparatus as claimed in any one of the preceding claims, further comprising means for performing: estimating a time of arrival of the first signal between a user device and a base station based on the generated classification.

12. An apparatus as claimed in any one of the preceding claims, wherein the means comprise: at least one processor; and at least one memory including computer program code, the at least one memory and the computer program configured, with the at least one processor, to cause the performance of the apparatus.

13. A method comprising: generating a first signal matrix of a received first signal, wherein the first signal matrix is a time-frequency matrix of the first signal, and the first signal matrix comprises zero power channel subcarriers and pilot channel subcarriers; generating, using a first model, a first noise matrix using the first model based, at least in part, on a noise behaviour of the zero power channel subcarriers; generating, using a second model, a second signal matrix of a reduced noise first signal based, at least in part, on a removal of the first noise matrix from the pilot subcarriers; and generating, using a third model, at least one classification of the first signal based, at least in part, on the second signal matrix, wherein the classification is related to line of sight properties of the first signal.

14. A method as claimed in claim 13, further comprising: generating a further first signal matrix of the first signal, wherein the first signal matrix comprises a real part of the first signal, and the further first signal matrix comprises an imaginary part of the first signal; and determining a further first noise matrix, wherein the first noise matrix is based on a noise behaviour of a real part of the zero power channel subcarriers and the further first noise matrix is based on an imaginary part of the zero power channel subcarriers, wherein the second signal matrix comprises a real part of the reduced noise first signal and an imaginaiy part of the reduced noise first signal.

15. A computer program comprising instructions for causing an apparatus to perform at least the following: generating a first signal matrix of a received first signal, wherein the first signal matrix comprises zero power channel subcarriers and pilot channel subcarriers; generating, using a first model, a first noise matrix using the first model based, at least in part, on a noise behaviour of the zero power channel subcarriers; generating, using a second model, a second signal matrix of a reduced noise first signal based, at least in part, on a subtraction of the first noise matrix from the pilot subcarriers; and generating, using a third model, at least one classification of the first signal based, at least in part, on the second signal matrix, wherein the classification is related to line of sight properties of the first signal.