Signal processing device and method for non-stationary dynamic environment

Abstract

A receiver (200) for a wireless network, a device (100, 110) for the receiver (200) and the receiver (200) comprising the device (100, 110) are disclosed. The apparatus (100, 110) is configured to obtain a received signal (101) and one or more pilot symbols (102) received by the receiver (200) over a channel. The first device (100) is configured to determine a data regularization covariance matrix (103) based on the received signal (101) and the one or more pilot symbols (102). The second device (110) is configured to train a machine learning model (111) based on the one or more pilot symbols (102). The first device (100) is then configured to determine an estimate (104) of the received signal (101) based on the received signal (101) and the data regularization covariance matrix (103), while the second device (110) is configured to determine the estimate (104) of the received signal (101) by applying the trained machine learning model (111) to the received signal (101).

Description

Technical Field

[0001] The present application relates to a receiver for a wireless network. The present application proposes devices for the receiver and the receiver comprising these devices respectively. These devices can perform signal processing on a received signal received by the receiver through a channel based on pilot symbols. These devices are specially designed to compensate for channel and hardware impairments in a non-stationary environment. Background Art

[0002] The main limiting factor of existing and future wireless networks is the harsh and often uncontrollable interference in a dynamic non-stationary environment. An example of such a dynamic non-stationary environment is the uplink from a 5th generation (5G) wireless network to a low earth orbit (LEO) satellite network with high residual Doppler shift. Possibly, due to limited satellite resources, the uplink is also affected by hardware impairments such as IQ imbalance (IQI). Another example is spectrum sharing in unlicensed spectrum, where asynchronous interference may have a different time or frequency format compared to the desired signal.

[0003] In such dynamic and non-stationary environments, interference reduction becomes a key technique. Interference reduction by antenna arrays is of particular interest since (massive) multiple input multiple output (MIMO) is considered one of the key enablers for existing and future wireless networks. Interference rejection combining (IRC) based on the minimum mean square error (MMSE) criterion is a simple antenna array interference reduction technique that uses linear beamforming at the receiver based on an estimated second order statistic (SOS) that can be applied to different wireless network configurations such as single user (SU) and multiple user (MU) MIMO (SU / MU-MIMO) systems.

[0004] On the other hand, channel and hardware impairments, such as IQ imbalance, phase noise, frequency offset, etc., may make the transmitted signal abnormal, but may introduce additional useful information (in the pseudo autocorrelation matrix), which can be advantageously used to improve system performance. For a complex random signal vector, SOS is fully characterized by its autocorrelation matrix and pseudo autocorrelation matrix. However, conventional satellite transmitters or receivers only utilize the autocorrelation of the signal, resulting in suboptimal performance.

[0005] The main feature of the above non-stationary scenario is that different symbols in the desired signal data slot can be received under different propagation and / or interference conditions. Conventional space-time-frequency equalization and interference rejection techniques in wireless communications utilize known pilot symbols (also called training symbols) to estimate the weight vector of the antenna array using the estimated propagation channel of the desired signal and the space-time-frequency parameters of the interference. The basic assumption of such techniques is that the training data is reliable because the co-channel interference (CCI) completely overlaps with the training symbols of the desired signal. This is usually the case in a slowly varying environment with synchronized CCI that has the same time-frequency structure as the desired user.

[0006] However, non-stationary dynamic environments, asynchronous cells, compressed intermittent transmissions, inter-system interference with different time / frequency formats in unlicensed spectrum, etc. may lead to more complex scenarios. The main difficulty in these scenarios is that different contents of the CCI sources may be observed during the training and data intervals. Under these conditions, an adaptive beamformer designed in the training interval occupied by a known training sequence becomes inefficient for interference reduction in the data interval, since the interfering signal may not be present in the training interval. Directly using the estimate of the covariance matrix averaged over the data interval does not result in effective useful signal extraction due to the presence of useful signals in the data used for the covariance matrix estimation, resulting in the possible rejection of both the desired signal and the interference depending on the available degrees of freedom (antenna elements).

[0007] Conventional pilot-based and data-based IRC schemes are defined in various 3GPP documents. In addition, a broad linear or (for orthogonal frequency division multiplexing (OFDM)) augmented linear IRC (AIRC) has been proposed. However, adaptive AIRC based on iterative least mean square (LMS) or recursive least square (RLS) algorithms may not be suitable for modern slot-based wireless networks.

[0008] The problem is that in a non-stationary dynamic environment, conventional pilot-based solutions may lose their efficiency because the interference content at the pilot symbols used to estimate the interference plus noise covariance matrix may be different from the interference content at the data symbols of the desired signal time slot.

[0009] Furthermore, in a non-stationary dynamic environment, conventional data-based solutions may also lose their efficiency because the interference plus noise covariance matrix estimated within a data symbol or a whole data slot contains the desired signal and the interference that causes the desired signal distortion, as well as interference rejection. Summary of the invention

[0010] In view of the above, the present application aims to improve conventional data-based and pilot-based solutions. That is, it is an object to overcome the shortcomings of conventional pilot-based and data-based solutions, and the object is generally to develop more robust signal processing at the receiver for non-stationary dynamic environments. Another object is to take into account SOS to exploit the information inherent in the inappropriate signal.

[0011] These and other objects are achieved by the solutions of the present application as described in the independent claims. Advantageous implementations are further specified in the dependent claims.

[0012] The first aspect of the present application provides a device for a receiver, wherein the device is used to: obtain a received signal and one or more pilot symbols received by the receiver through a channel; determine a data regularization covariance matrix based on both the received signal and the one or more pilot symbols; and determine an estimate of the received signal based on the received signal and the data regularization covariance matrix.

[0013] The device of the first aspect is able to overcome the disadvantages of conventional pilot-based and data-based solutions, since both the signal (data symbols) and the pilot symbols are used to determine the covariance matrix. Thus, a more robust signal processing is obtained, which can be implemented at the receiver. Therefore, the receiver is improved, especially for non-stationary dynamic environments.

[0014] In one implementation form of the first aspect, the device is used to: determine a channel estimate of the channel based on the received signal and the one or more pilot symbols; determine a pilot-based covariance matrix based on the one or more pilot symbols and the channel estimate; determine a data-based covariance matrix based on the received signal; and determine the data regularization covariance matrix by weighting the pilot-based covariance matrix and the data-based covariance matrix respectively based on one or more receiver parameters of the receiver.

[0015] In this way, the pilot-based and data-based approaches are combined to create a more robust solution that is also receiver adaptive.

[0016] In an implementation form of the first aspect, the device is used to: calculate a weight vector based on the channel estimate and the data regularization covariance matrix; and determine the estimate of the received signal by linearly combining the weight vector and the received signal.

[0017] In an implementation form of the first aspect, the one or more receiver parameters include at least one of the number of antennas of the receiver, the number of user signals in the received signal in the case of multi-user reception of signals, and a signal-to-noise ratio of the received signal.

[0018] In an implementation form of the first aspect, the device is used to: determine one or more scenario parameters based on the received signal; determine the channel estimate based on the received signal, the one or more pilot symbols and the one or more scenario parameters.

[0019] In an implementation form of the first aspect, the one or more scene parameters include at least one of a delay spread of the received signal and a Doppler shift of the received signal.

[0020] Therefore, the characteristics of actual non-stationary dynamic environments can be taken into account.

[0021] In an implementation form of the first aspect, the device is used to: calculate a mixing factor based on the one or more receiver parameters; and weight the pilot-based covariance matrix and the data-based covariance matrix respectively according to the mixing factor.

[0022] The optimal mixing factor may lead to a further improved estimate of the received signal, which is reflected, for example, in a lower bit error rate.

[0023] In an implementation form of the first aspect, the device is configured to calculate the mixing factor based on the one or more receiver parameters and the one or more scene parameters.

[0024] Therefore, the characteristics of the specific non-stationary dynamic environment and the characteristics of the receiver can be taken into account.

[0025] In an implementation form of the first aspect, the device comprises a trained model for calculating the mixing factor based on the one or more receiver parameters.

[0026] In an implementation form of the first aspect, the trained model is a trained neural network.

[0027] In an implementation form of the first aspect, the received signal is an OFDM signal.

[0028] In an implementation form of the first aspect, the received signal is a MU signal or a SU signal.

[0029] A second aspect of the present application provides a device for a receiver, wherein the device is used to: obtain a received signal and one or more pilot symbols received by the receiver through a channel; train a machine learning model based on the one or more pilot symbols; and determine an estimate of the received signal by applying the trained machine learning model to the received signal.

[0030] The device of the second aspect can overcome the shortcomings of conventional pilot-based and data-based methods because it can take into account SOS to exploit information inherent in inappropriate signals. The machine learning model can be trained using pilot symbols. Since the device of the second aspect can have information about received pilot symbols and train machine learning at the device (e.g., in a receiver), the device can obtain a more accurate estimate of the received signal.

[0031] In an implementation form of the second aspect, the trained machine learning model is an augmented complex extreme learning machine (A-CELM), wherein the A-CELM has only one hidden layer, and the hidden layer includes hidden nodes and complex conjugates of the hidden nodes.

[0032] The third aspect of the present application provides a receiver, comprising: one or more antennas; a device according to the first aspect or any one of its implementation forms, wherein the receiver is used to receive the received signal and the one or more pilot symbols through the channel using the multiple antennas, and provide the received signal and the one or more pilot symbols as input to the device, and the device is used to determine the estimate of the received signal based on the input.

[0033] Since the receiver includes the device of the first aspect or the second aspect, the receiver has the same advantages as described above.

[0034] The fourth aspect of the present application provides a method for a receiver, the method comprising: obtaining a received signal and one or more pilot symbols received by the receiver through a channel; wherein the method further comprises: determining a data regularization covariance matrix based on both the received signal and the one or more pilot symbols; determining an estimate of the received signal based on the received signal and the data regularization covariance matrix; or wherein the method further comprises: training a machine learning model based on the one or more pilot symbols; determining an estimate of the received signal by applying the trained machine learning model to the received signal.

[0035] In an implementation form of the fourth aspect, the method includes: determining a channel estimate of the channel based on the received signal and the one or more pilot symbols; determining a pilot-based covariance matrix based on the one or more pilot symbols and the channel estimate; determining a data-based covariance matrix based on the received signal; and determining the data regularization covariance matrix by weighting the pilot-based covariance matrix and the data-based covariance matrix respectively based on one or more receiver parameters of the receiver.

[0036] According to the steps respectively executed by the device of the first aspect and the device of the second aspect and their implementation forms, the method of the fourth aspect may have other implementation forms. Therefore, the method of the fourth aspect achieves the same advantages as above.

[0037] A fifth aspect of the present application provides a computer program, which includes instructions. When the program is executed by a computer, the instructions cause the computer to execute the method according to the fourth aspect or any implementation form thereof.

[0038] A sixth aspect of the present application provides a storage medium storing executable program code, which, when executed by a processor, enables execution of the method according to the fourth aspect or any implementation form thereof.

[0039] For example, in view of the above aspects and implementation forms, the present application proposes data-based regularization using a pilot-based estimate of an interference plus noise covariance matrix to introduce controllable mixing factors by means of a balance between interference reduction not present in the training interval and signal distortion caused by the presence of received signal components in the data-based estimate of the interference plus noise covariance matrix. Assuming a large number of degrees of freedom (e.g., antennas) at the receiver (e.g., a modern trend in gNB design), this data-regularized estimate of the interference plus noise covariance matrix allows all interference components present in the data slot to be effectively rejected to appropriately select the mixing factors. The present application also proposes the use of neural network (NN) function approximation with offline training to adaptively select the mixing factors in typical scenarios for a given network configuration. The proposed solution can be extended to a wide range of (enhanced) linear processing to address scenarios with hardware defects, which is particularly important for LEO satellite applications.

[0040] To remedy the shortcomings of CELM, e.g., the inability to exploit signal inappropriateness, leading to suboptimal performance, the present application further proposes to apply A-CELM (compare, e.g., 'The augmented complex-valued extreme learning machine' by H. Zhang et al., Neural Computation, 311, pp. 363-372, 2018'), which is able to fully capture SOS to obtain maximum performance improvement under hardware impairments.

[0041] It should be noted that all equipment, elements, units and devices described in this application can be implemented in software or hardware elements or any combination thereof. All steps performed by various entities described in this application and the functions to be performed by various entities described in this application mean that each entity is suitable for or used to perform the corresponding steps and functions. Although in the description of the following specific embodiments, the specific functions or steps to be performed by the external entity are not reflected in the description of the specific detailed elements of the entity performing the specific steps or functions, it should be clear to the technician that these methods and functions can be implemented by corresponding software or hardware elements or any combination thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In conjunction with the accompanying drawings, the specific embodiments described below illustrate the above aspects and implementation forms, in which:

[0043] Figure 1 Different devices according to the present application are shown.

[0044] Figure 2 A receiver according to the present application is shown, which receiver comprises a device according to the present application.

[0045] Figure 3 An exemplary device according to the present application is shown.

[0046] Figure 4 Shows Figure 3 Details of an exemplary device.

[0047] Figure 5 Another exemplary device according to the present application is shown.

[0048] Figure 6 Shown with Figure 3 The exemplary device of the present application is similar to the simulation results of the exemplary device of the present application.

[0049] Figure 7 Shows Figure 6 Additional simulation results for an exemplary device.

[0050] Figure 8 Shown with Figure 5 The exemplary device of the present application is similar to the simulation results of the exemplary device of the present application.

[0051] Fig. 9 Shows Figure 8 Additional simulation results for an exemplary device.

[0052] Fig.10 Shows Figure 8 Additional simulation results for an exemplary device.

[0053] Fig.11 A flow chart of a method according to the present application is shown. DETAILED DESCRIPTION

[0054] Figure 1 Two devices 100 and 110 according to the present application are shown. Device 100 is shown in (a) and device 110 is shown in (b). Both devices 100 and 110 can be used as a receiver, for example, Figure 2 , which will be explained below. Both devices 110, 110 may perform signal processing in the receiver 200 in order to compensate for channel and hardware impairments, especially in non-stationary environments.

[0055] The device 100 is used to obtain a received signal 101, and to obtain one or more pilot symbols 102. The received signal 101 and the pilot symbol 102 are both received by a receiver 200 through a channel, which may be a damaged channel. The receiver 200 can input the received signal 101 and the pilot symbol 102 to the device 100. The received signal 101 and the pilot symbol 102 can be received, for example, in a separate time interval or time slot or other time and / or frequency resource. The received signal 101 can include data symbols and can be received in a dedicated data interval. The pilot symbol can be received in a training interval. However, the data symbol and the pilot symbol can also be received together.

[0056] The device 100 is used to determine a data regularization covariance matrix 103 based on both the received signal 101 and one or more pilot symbols 102 (e.g., in a first optional processing block indicated by a dashed box). The device 100 can use the data symbols and the pilot symbols 102 to obtain the covariance matrix 103 accordingly. In addition, the device 100 is used to determine an estimate 104 of the received signal 101 based on the received signal 101 and the data regularization covariance matrix 103 (e.g., in a second optional processing block indicated by a dashed box). The estimate 104 of the signal can be compensated for channel and / or hardware impairments that have affected (transmission of) the received signal 101.

[0057] The device 110 is further configured to obtain a received signal 101 and one or more pilot symbols 102 received by the receiver 200 through a channel in a manner similar to that described above with respect to the device 100 .

[0058] The device 110 is further configured to train a machine learning model 111 based on one or more pilot symbols 102 (e.g., in a first optional processing block indicated by a dashed box), and determine an estimate 104 of the received signal 101 (e.g., in a second optional processing block indicated by a dashed box) by applying the trained machine learning model 111 to the received signal 101. The estimate 104 of the signal may again be compensated for channel and / or hardware impairments that have affected (transmission of) the received signal 101.

[0059] Devices 100 and / or 110 may include a processor or processing circuitry for performing, carrying out, or initiating various operations of devices 100 and / or 110 described herein. The processing circuitry may include hardware and / or the processing circuitry may be controlled by software. Specifically, the processing circuitry may include or form Figure 1Optional processing blocks indicated by dashed boxes in the figure. Hardware may include analog circuits or digital circuits, or both. Digital circuits may include components such as application-specific integrated circuits (ASICs), field-programmable arrays (FPGAs), digital signal processors (DSPs), or multi-purpose processors. Devices 100 and / or 110 may also include a memory circuit that stores one or more instructions that can be executed by a processor or processing circuit (specifically, executed under the control of software). For example, the memory circuit may include a non-transitory storage medium that stores executable software code that, when executed by a processor or processing circuit, enables the execution of various operations of devices 100 and / or 110. In one embodiment, the processing circuit includes one or more processors and a non-transitory memory connected to the one or more processors. The non-transitory memory may carry an executable program code that, when executed by one or more processors, enables devices 100 and / or 110 to perform, conduct, or initiate the operations or methods described herein.

[0060] Figure 2 A receiver 200 according to the present application is shown. The receiver 200 includes one or more antennas 201, for example, may include an antenna array. The antenna 201 may be used for MIMO and / or OFDM. The receiver 200 is used to receive a receive signal 101 and one or more pilot symbols 102 (as described above) through a channel using multiple antennas 201. There may be channel and / or hardware impairments that affect the receive signal 101. The receiver 200 may be used for OFDM and / or SU / MU-MIMO reception.

[0061] The receiver 200 also includes Figure 1 , and for providing a received signal 101 and received pilot symbols 102 as input to the device 100, 110. Then, as described above, the device 100, 110 is used to determine an estimate 104 of the signal, which the receiver 200 can use for optional further signal processing, for example, for decoding the signal, etc. It is worth noting that the estimate 104 of the signal is intended to recover the signal, which is sent by the transmitter to the receiver 200 through the channel and generates the received signal 101 received by the receiver 200, for example due to channel and / or hardware impairments.

[0062] Figure 3 A block diagram of an exemplary device 100 is shown, which is based on Figure 1The device 100 shown in (a) is constructed. The same elements are marked with the same reference symbols and have the same function.

[0063] Figure 3 The device 100 shown in FIG. 1 is used to obtain a received signal 100 and one or more pilot symbols 102 , which are received by one or more antennas 201 of a receiver 200 , for example.

[0064] The device 100 may include multiple processing blocks, for example, as shown in the figure. Specifically, the device 100 may have a "channel estimation" processing block for determining a channel estimate 301 of a channel based on the received signal 101 and one or more pilot symbols 102. In addition, the device 100 may have a "pilot-based covariance matrix estimator" processing block for determining a pilot-based covariance matrix 302 based on one or more pilot symbols 102 and the channel estimate 301. In addition, the device 100 may have a "data-based covariance matrix estimator" processing block for determining a data-based covariance matrix 303 based on the received signal 101. Then, the device 100 may also have a "data regularization covariance matrix calculator" processing block for determining a data regularization covariance matrix 103 by weighting the pilot-based covariance matrix 302 and the data-based covariance matrix 303 based on one or more receiver parameters 305 of the receiver 200, respectively. The device 100 may receive receiver parameters 305 as input from the receiver 200, or may be capable of determining or calculating these receiver parameters 305. The device 100 may comprise further optional processing blocks, as explained below.

[0065] like Figure 3 As exemplarily shown in , the received signal 101 may arrive at a "scenario analyzer" processing block, which may estimate one or more scenario parameters 307 based on the received signal 101. For example, a typical scenario parameter 307 may be a delay spread and / or a Doppler shift of the received signal 101. The received signal 101 and one or more pilot symbols 102 and optionally the output of the scenario analyzer, i.e., one or more scenario parameters 307, are also used at a "channel estimator" processing block for an estimation 301 of the propagation channel of the received signal 101. As described above, a pilot-based covariance matrix 302 (interference plus noise covariance matrix) is estimated in a "pilot-based covariance matrix estimator" processing block, and a data-based covariance matrix 303 is estimated in a "data-based covariance matrix estimator" processing block.

[0066] These pilot-based covariance matrices 302 and data-based covariance matrices 303 may also be weighted in a “data regularization covariance matrix calculator” processing block according to a mixing factor 308 in order to determine the data regularization covariance matrix 103. The mixing factor 308 may be calculated based on one or more receiver parameters 305, in particular in a “database (neural network)” processing block 309 and using, for example, the number of antennas 201 of the receiver 201 and / or the number of signals in the received signal 101 and / or the signal-to-noise ratio of the received signal 101 as receiver parameters 305.

[0067] The weight vector 306 may also be calculated in a "Weight Vector Calculator" processing block using the data regularized covariance matrix 103 and the channel estimate 301. A linear combination of the weight vector 306 and the received signal 101 may then be performed in a "Signal Estimator" processing block 304 to determine an estimate 104 of the received signal 101.

[0068] Figure 4 Shows Figure 3 Implementation of a “database (neural network)” processing block 309 shown in FIG. 1 and used in device 100 to determine a mixing factor 308. In a training (offline) mode of this processing block 309, one or more receiver parameters 305 and / or one or more scene parameters 307 are used in a “scene simulator” processing block according to Figure 3 The device 100 shown in (denoted as data regularized IRC (DRIRC)) generates several realizations of the received signal 101 for a set of mixing factors 308 generated by a "mixing factor sample" processing block. The bit error rate (BER) is estimated for each mixing factor at the output of the DRIRC. Then, the "minimizer" processing block is used to find the optimal mixing factor 308 that gives the minimum BER, which is used as the training NN output for the receiver parameters 305 and scene parameters 307 as the NN training input, thereby producing NN parameters for online generation of mixing factors 308 suitable for the given receiver parameters 305 and scene parameters 307.

[0069] Figure 3 and Figure 4 Further details of the signals and processing blocks shown in can be described as follows.

[0070] For the received signal 101, the following model of the OFDM signal received by the receiver 200 using K antennas 201 may be assumed for the nth symbol on the fth subcarrier:

[0071] x(f,n)=H(f,n)s(f,n)+z(f,n), (1)

[0072] where x(f,n)∈C K×1 is the received signal vector, H(f,n)=[h1,...,h M ]∈C K×M is the propagation channel matrix, where M>1 is the number of useful signals for MU-MIMO and M=1 is the number of useful signals for SU-MIMO, z(f,m)∈C K×1 is the interference plus noise vector. It can be assumed that the signal estimation in the data time slot contains L≥1 resource blocks (RBs) adjacent in the frequency domain. Each data time slot contains N p pilot symbols and N d data symbol.

[0073] In the "Scene Analyzer" processing block, typical estimates of delay spread and Doppler shift for interpolated channel estimation can be used, for example, as described in 'Z. Sharif, AZ Sha'ameri, "Estimation of the Doppler spread and Time Delay Spread for the Wireless Communication Channel", in Proc. ICCAIE, December 2010'.

[0074] In the "Channel Estimation" processing block, pilot-based channel estimation with time / frequency interpolation can be used, as defined in 'P. Hoeher, S. Kaiser, P. Robertson, "Two-Dimensional Pilot-Symbol-Aided Channel Estimation by Wiener Filtering", in Proc. ICASSP, April 1997'.

[0075] In the “Pilot-based Covariance Matrix Estimator” processing block, the pilot-based interference plus noise covariance matrix 302 may be estimated for the mth received signal 101 as follows:

[0076]

[0077] in(·) * is the conjugate transpose operation, f pi ,n pi is the position of the i-th pilot symbol 102 of the m-th received signal in the data slot, is the channel estimate 301 at the output of the processing block, I K is a K×K identity matrix, β≥0 are diagonal loading coefficients, which can be selected based on field experiments as in US 9,641,294 B2, or analytically selected as in 'AM Kuzminskiy, P. Xiao, R. Tafazolli, "Uniform Expected Likelihood Solution for Interference Rejection Combining Regularization", in Proc. ICASSP, March 2016'.

[0078] In the "Data-Based Covariance Matrix Estimator" block, the data-based covariance matrix 303 may be estimated as follows:

[0079]

[0080] where f di ,n di is the position of the i-th data symbol in the data slot.

[0081] In the "Data Regularization Covariance Matrix Calculator" processing block, the data regularization covariance matrix 103 may be calculated as follows:

[0082]

[0083] where 0≤δ≤1 is the mixing factor 309.

[0084] Alternatively, the data regularization covariance matrix 103 can be calculated as

[0085]

[0086] This results in another implementation of the weight vector 306 in the "Weight Vector Calculator" processing block with a simpler computational complexity.

[0087] In the "Weight Vector Calculator" processing block, the weight vector 306 of the mth received signal may be calculated as follows:

[0088]

[0089] This equation is for the data regularized covariance matrix 103 defined in equation (4), and

[0090]

[0091] This equation is for the data regularized covariance matrix 103 defined in equation (5). Compared to equation (6), equation (7) has lower computational complexity because it requires only one matrix inversion per data slot.

[0092] For δ=0, equations (6) and (7) represent pilot-based 3GPP IRC, and for δ=1, equation (6) represents data-based 3GPP IRC (see 'Y. Ohwatari, N. Miki, T. Asai, T. Abe, H. Taoka, "Advanced receiver performance for suppressing inter-cell interference in advanced LTE downlink using interference rejection combining", in Proc. VTC-Autumn, September 2011').

[0093] In the "Signal Estimator" processing block 304, the estimate 104 of the mth signal is calculated as follows:

[0094]

[0095] The "Scenario Simulator" processing block of processing block 309 may generate the received signal 101 for given scenario parameters 307 and receiver parameters 305 according to equation (1) using conventional channel generation tools such as the MALAB Communication Toolbox.

[0096] The DRIRC processing block is based on Figure 3 The device 100 shown in FIG.

[0097] The mixing factor samples in processing block 309 may contain predefined samples of the mixing factor 308 , for example, 0:0.02:1.

[0098] The "Minimizer" processing block in processing block 309 may find the value of mixing factor 308 that gives the minimum BER among the values ​​of mixing factor 308 generated by the "Mixing Factor Samples" processing block.

[0099] The NN can be defined using the MATLAB Deep Learning Toolbox, e.g., using the "net" routine for a given number of hidden nodes, and the NN using the "net" routine. NN training can be performed using the MATLAB Deep Learning Toolbox, e.g., using the "train" routine for scene parameters 307 and receiver parameters 304 as NN inputs and optimal mixing factors 308 as NN outputs at the output of a "minimizer" processing block.

[0100] To obtain a widely (enhanced) linear version of data regularized AIRC (ADRIRC) suitable for networks with hardware impairments such as IQI, the received signal vector in (1) can be replaced by the extended received signal vector using the conjugated signal from the mirrored OFDM subcarriers.

[0101]

[0102] Where N DFT is the dimension of the discrete Fourier transform used in OFDM networks, (·) C is the complex conjugate operation. Then, the dimension K of all variables in the above DRIRC description is replaced by the dimension 2K of ADRIRC.

[0103] Figure 5 A block diagram of an exemplary device 110 is shown, which is based on Figure 1 The device 110 shown in FIG. 1 is constructed in the same manner as in FIG. 1 . The same elements are marked with the same reference symbols and have the same functions.

[0104] Figure 5 An improved version of CELM is proposed as a trained machine learning model 111 of the device 110 to exploit signal inappropriateness. A complex extreme learning machine is a fast single-layer feed-forward network that performs random feature mapping using unknown parameters learned by linear parameter solution. The design of the device 110 or receiver 200 can be optimized using A-CELM, where both the covariance matrix and the pseudo-covariance matrix are considered to fully capture the second-order statistics (SOS) of the received signal 101, which is inappropriate and non-circular due to hardware impairments.

[0105] The basic structure of A-CELM is as follows Figure 5 Specifically, the A-CELM has only one hidden layer 501 , which includes a hidden node 502 and a complex conjugate 503 of the hidden node 502 .

[0106] Given It represents any different N samples. Here, x k and t k Denote the received and transmitted signal vectors respectively. In the context of A-CELM, the former is the input of the A-CELM post-distorter, while the latter represents the output target vector, which is the pilot symbol 102. Hidden layer input weights and biases is randomly generated, not tuned, and N his the number of hidden layer nodes 502. The randomly initialized hidden layer weights and biases project the vector of the received signal 101 into a random space using a nonlinear mapping function f(·). The output of the hidden layer 501 corresponding to the hidden node i is given by

[0107]

[0108] In A-CELM, in order to incorporate the SOS of the received signal 101 into the NN structure, the hidden layer 501 is modified by appending the complex conjugate of the hidden node 502. A new set of hidden node outputs is obtained as

[0109]

[0110] With the kth input x k The corresponding output is given by

[0111]

[0112] where β i and γ i represents the weights of the output layer corresponding to the hidden layer 501 and the enhanced conjugate hidden layer 501. If Then there exists β i , γ i 、w i and b i , so that for k = 1, 2, ... N, o k =t k These N equations can be compactly expressed in matrix form. Suppose the input matrix contains all the received signal vectors

[0113]

[0114] and the weight matrix

[0115]

[0116] Bias Vector

[0117]

[0118] And the bias matrix, which is obtained by superimposing the bias vector N times, that is,

[0119]

[0120] The output hidden layer matrix corresponding to (1) is

[0121]

[0122] Where f(A) represents the activation function of each element of matrix A, that is

[0123]

[0124] Let H a The enhanced hidden layer matrix incorporating the complex conjugate of H is represented as H a =[H,H * ].if and Then the effective output layer weight matrix corresponding to A-CELM is As If N target vectors t k The superposition in matrix T is Then its relationship with the hidden layer output and the output layer weight is The weight bias matrix is ​​randomly generated from any continuous probability distribution. Given the target matrix T, the parameter to be calculated by A-CELM is the output weight matrix The output weight matrix is ​​obtained by minimizing the cost function

[0125]

[0126] The cost function can be obtained by the least squares method as follows

[0127]

[0128] in Denotes the matrix H a The cost function can be modified as follows to avoid overfitting:

[0129]

[0130] The closed form solution of the above unconstrained optimization problem can be obtained as follows:

[0131]

[0132] In the following, about Figure 6-10 Some performance evaluations of devices 100 and 110 are presented.

[0133] For device 100, the application of DRIRC and ADRIRC in the uplink LEO satellite MU-MIMO scenario with IQI is as follows: Figure 6 and Figure 7 , for the following simulation assumptions:

[0134] 2GHz carrier frequency

[0135] 10MHz bandwidth (NDFT =1024)

[0136] 15kHz carrier spacing

[0137] 16QAM signaling

[0138] ·1 / 2 convolutional code rate of (133,171) code with constraint length 7

[0139] Full-band spectrum allocation

[0140] Satellite multipath channel model in accordance with 3GPP TR 38.811V15.4.0

[0141] 0.42ppm maximum residual Doppler shift, corresponding to 1200km satellite altitude, 40km beam footprint diameter, 1000km / h UE speed[1]

[0142] · For different antennas, the IQI parameters are evenly distributed with 1, ..., 3 dB and 0, ..., 20 degrees corresponding to amplitude and phase imbalance

[0143] A typical uplink RB has 12 subcarriers, 2 pilots and 12 data OFDM symbols.

[0144] L = 5 RBs adjacent in the frequency domain in a data slot

[0145] Uncorrelated channels for different receive antennas

[0146] Pilots generated using the MATLAB "ltePUSCHDRS" routine

[0147] Channel estimation using 2D interpolation on 3 frequency sliding RBs

[0148] Diagonal loading of half noise power

[0149] Average BER achieved over 500 scenarios

[0150] The simulation results of M = 4 and K = 16 are as follows Figure 6 As shown in , the high (a) and low (b) Doppler shifts are 0.42 ppm and 0.0042 ppm, respectively. Figure 6 As can be seen in Fig. 2b, zero mixing factor 308 is the optimal choice for low Doppler shift scenarios that produce conventional IRC and AIRC, which means that data-based regularization is not required for low Doppler shift. On the contrary, for a properly selected mixing factor 308, Figure 6 Significant data regularization gain for higher Doppler shifts is seen in a.

[0151] The efficiency of selecting the mixing factor based on NN is as follows Figure 7 , which shows simulation results for M=2 and K=16, and two additional co-channel interferers of 3dB and 5dB for DRIRC and ADRIRC, respectively, compared to the optimal (unachievable) mixing factor values ​​found by direct BER minimization for each scenario implementation. IRC and AIRC with δ=0 are also plotted Figure 7 It is used for comparison. It can be seen that the result based on NN is very close to the optimal result.

[0152] To evaluate the performance of the device 110, input weights and biases can be randomly selected from a complex circular region with a uniform distribution and a center at the origin. A complex inverse hyperbolic sine function is used as the activation function of the hidden layer. The hyperparameters are adjusted by gradually increasing the number of hidden nodes, i.e., the number of hidden nodes is 502, and the optimal number of hidden nodes for both CELM and A-CELM is selected to be 50 based on a cross-validation method.

[0153] Figure 8 a Comparison of the BER and SNR performance of conventional CELM and A-CELM under different LEO satellite scenarios and system parameters (numerology) under the influence of Doppler frequency shift, phase noise, I / Q imbalance and HPA nonlinearity when OBO = 0 dB. d =750Hz and f d = 1.5kHz and the maximum user equipment speed is 1000km / h, the LEO satellite altitude is h = 600km and the beam footprint diameter is d = 20km. Figure 8The nonlinearity is shown in a. AM / AM and AM / PM characteristics of a typical satellite TWT A provided by the European Space Agency are used to represent the nonlinearity. The model parameters for the amplitude and phase response of the simulated HPA are given in 'A. Jayaprakash, H. Chen, P. Xiao, B. G. Evans, Y. Zhang, J. Y. Li, and A. B. Wolfseyila, "Analysis of candidate waveforms for integrated satellite terrestrial 5G systems," in 2019 IEEE Second 5G World Forum (5GWF), September 2019, pp. 636–641'. The phase noise is calculated using '3GPP, "Key satellite parameters and simulation assumptions for NTN," 3rd Generation Partnership Project (3GPP), September 2019, pp. 636–641. rd Generation Partnership Project, 3GPP), Discussion RAN1-96, April 2019, R1-1905216' for S-band satellite phase noise mask parameters for simulation. A DFT-extended OFDM waveform corresponding to an FFT size, N = 1024 subcarriers with a spacing of 15kHz can be used, and a cyclic prefix length of 72 is modulated using 4-QAM. C-ELM and A-CELM are trained using block pilots corresponding to one DFT-s-OFDM symbol. For comparison purposes, the performance curve of an ideal predistorter is also included, which perfectly linearizes the power amplifier and has no phase noise and I / Q imbalance. It can be clearly observed that A-CELM, which exploits the inappropriateness of the received signal, exhibits significant performance gains compared to CELM. A-CELM also exhibits improved performance compared to the perfect predistorter, from which it is inferred that even if the ideal predistorter does not have nonlinearity, phase noise and I / Q imbalance, the increase in BER compared to A-CELM is attributed to the distortion caused by the residual Doppler shift. A-CELM is more effective in combating distortion caused by nonlinearity, phase noise, I / Q imbalance and Doppler shift.

[0154] In the case of multipath propagation with a LEO satellite elevation angle of 10°, d =1.5kHz maximum residual Doppler frequency shift corresponding to the BER and SNR performance comparison as shown below Figure 8 b and Figure 8c. Performance curves corresponding to an ideal pre-distorter (PD) in the absence and presence of phase noise and IQ imbalance are also shown for comparison, where zero-forcing (ZF) equalization assumes perfect channel knowledge. The multipath channel uses the TDL channel model recommended by 3GPP for non-line-of-sight satellite-to-user links in suburban environments (see '3GPP, "Study on New Radio (NR) to support non-terrestrial networks", 3rd Generation Partnership Project (3GPP) rd The simulation was performed using the 3GPP (3rd Generation Partnership Project), Technical Report (TR) RAN-80, September 2020, Version 15.4.0'). The TDL channel model is used to generate impulse responses corresponding to each tap when the LEO satellite elevation angle (EA) is 10°. The channel coding uses a half-rate convolutional code with a constraint length of 7. The BER and SNR performance comparisons for delay spreads corresponding to 10ns and 50ns are shown below. Figure 8 b and Figure 8 c. A-CELM and CELM are compared with ideal predistortion and zero-forcing equalization assuming perfect channel knowledge in the following scenarios: 1) with impairments (delay spread, Doppler, phase noise, nonlinearity, IQ imbalance), 2) without impairments (zero IQ imbalance), and 3) without impairments (zero Doppler, zero phase noise, and zero IQ imbalance). In the case of coded systems, the performance advantage of the A-CELM approach is very clear and provides significant performance gain compared to all other baseline schemes.

[0155] To demonstrate the robustness of A-CELM against inappropriate reception constellation patterns, the BER performance for different I / Q phase imbalance values ​​is compared for the S-band coded system. Fig. 9 As shown in a, when SNR = 10dB, the channel delay spread is 50ns, and the maximum residual Doppler shift f d =750. CELM degrades BER performance as the inappropriateness increases, while A-CELM shows strong resilience to signal inappropriateness. The BER curves of ideal PD and ZF do not change with the increase of I / Q phase imbalance, which corresponds to the scenario without inappropriateness.

[0156] To illustrate the robustness under different Doppler conditions, the relationship between BER and residual Doppler shift is shown in the following table for a fixed delay spread and SNR = 10 dB. Fig. 9As shown in b, even though the ideal PD and ZF methods perform better than CELM in the absence of damage, A-CELM still provides better performance.

[0157] To demonstrate the ability of A-CELM to resist various impairments under different delayed Doppler conditions, we Fig.10 The performance of different delays and Doppler spreads at SNR = 10 dB is compared in Figure 2. For all different combinations of delay spread and Doppler shift, the performance of A-CELM is better than that of CELM and ideal PD and ZF.

[0158] Fig.11 A method 1100 according to the present application is shown. The method 1100 may be performed by the device 100 or the device 110. Thus, the method 1100 may be performed by the receiver 200.

[0159] The method includes a step 1101 of obtaining a received signal 101 and one or more pilot symbols 102 received by the receiver 200 through a channel. In a first alternative that can be implemented by the device 100, the method 1100 may also include a step 1102 of determining a data regularization covariance matrix 103 based on both the received signal 101 and the one or more pilot symbols 102, and a step 1103 of determining an estimate 104 of the received signal 101 based on the received signal 101 and the data regularization covariance matrix 103. In a second alternative that can be implemented by the device 110, the method 1100 may also include a step 1104 of training a machine learning model 111 based on the one or more pilot symbols 102, and a step 105 of determining an estimate 104 of the received signal 101 by applying the trained machine learning model 111 to the received signal 101.

[0160] The present application has been described in conjunction with various embodiments as examples and implementations. However, from a study of the drawings, the present application and the independent claims, a person skilled in the art will be able to understand and implement other variations when implementing the claimed subject matter. In the claims and in the specification, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude a plurality. A single element or other unit may fulfil the functions of several entities or items described in the claims. The listing of certain measures in mutually different dependent claims does not indicate that a combination of these measures cannot be used in an advantageous implementation.

Claims

1. A device (100) for a receiver (200), characterized in that The device (100) is used for: Obtaining a received signal (101) and one or more pilot symbols (102) received by the receiver (200) through a channel; determining a data regularization covariance matrix (103) based on both the received signal (101) and the one or more pilot symbols (102); An estimate (104) of the received signal (101) is determined based on the received signal (101) and the data regularization covariance matrix (103).

2. The device (100) according to claim 1, characterized in that Used for: determining a channel estimate (301) of the channel based on the received signal (101) and the one or more pilot symbols (102); determining a pilot-based covariance matrix (302) based on the one or more pilot symbols (102) and the channel estimate (301); determining a data-based covariance matrix (303) based on the received signal (101); The data regularized covariance matrix (103) is determined by weighting the pilot-based covariance matrix (302) and the data-based covariance matrix (303) respectively based on one or more receiver parameters (305) of the receiver (200).

3. The device (100) according to claim 2, characterized in that Used for: Calculating a weight vector (306) based on the channel estimate (301) and the data regularization covariance matrix (103); The estimate (104) of the received signal (101) is determined by linearly combining the weight vector (306) and the received signal (101).

4. The device (100) according to claim 2 or 3, characterized in that The one or more receiver parameters (305) include at least one of the number of antennas (201) of the receiver (200), the number of user signals in the received signal (101) in the case of multi-user reception of signals, and a signal-to-noise ratio of the received signal (101).

5. The device (100) according to any one of claims 2 to 4, characterized in that Used for: determining one or more scene parameters (307) based on the received signal (101); The channel estimate (301) is determined based on the received signal (101), the one or more pilot symbols (102) and the one or more scene parameters (307).

6. The device (100) according to claim 5, characterized in that The one or more scene parameters (307) include at least one of a delay spread of the received signal (101) and a Doppler shift of the received signal (101).

7. The device (100) according to any one of claims 1 to 6, characterized in that Used for: calculating a mixing factor (308) based on the one or more receiver parameters (305); The pilot-based covariance matrix (302) and the data-based covariance matrix (303) are weighted respectively according to the mixing factor (308).

8. The device (100) according to claim 3 or 4 and according to claim 6 or 7, characterized in that For calculating the mixing factor (308) based on the one or more receiver parameters (305) and the one or more scene parameters (307).

9. The device (100) according to claim 7 or 8, characterized in that A trained model (309) is included for calculating the mixing factor (308) based on the one or more receiver parameters (305).

10. The device (100) according to claim 9, characterized in that The trained model (309) is a trained neural network.

11. The device (100) according to any one of claims 1 to 10, characterized in that The received signal (101) is an orthogonal frequency-division multiplexing (OFDM) signal.

12. The device (100) according to any one of claims 1 to 11, characterized in that The received signal (101) is a multi-user signal or a single-user signal.

13. A device (110) for a receiver (200), characterized in that The device (100) is used for: Obtaining a received signal (101) and one or more pilot symbols (102) received by the receiver (200) through a channel; Training a machine learning model (111) based on the one or more pilot symbols (102); An estimate (104) of the received signal (101) is determined by applying the trained machine learning model (111) to the received signal (101).

14. The device (110) according to claim 13, characterized in that The trained machine learning model (111) is an augmented complex extreme learning machine (A-CELM), wherein the A-CELM has only one hidden layer (501), and the hidden layer includes a hidden node (502) and a complex conjugate (503) of the hidden node (502).

15. A receiver (200), characterized in that: include: One or more antennas (201) ； The device (100, 110) according to any one of claims 1 to 14; The receiver (200) is configured to receive the receive signal (101) and the one or more pilot symbols (102) via the channel using the multiple antennas (201), and to provide the receive signal and the one or more pilot symbols as input to the device (100, 110), wherein the device (100, 110) is configured to determine the estimate (104) of the receive signal (101) based on the input.

16. A method (1100) for a receiver, characterized in that The method comprises: Obtaining (1101) a received signal (101) and one or more pilot symbols (102) received by the receiver (200) through a channel; The method (1100) further comprises: determining (1102) a data regularization covariance matrix (103) based on both the received signal (101) and the one or more pilot symbols (102); determining (1103) an estimate (104) of the received signal (101) based on the received signal (101) and the data regularized covariance matrix (103); Or wherein the method (1100) further comprises: Training (1104) a machine learning model (111) based on the one or more pilot symbols (102); An estimate (104) of the received signal (101) is determined (1105) by applying the trained machine learning model (111) to the received signal (101).

17. The method (1100) according to claim 16, characterized in that: The method (1100) comprises: determining a channel estimate (301) of the channel based on the received signal (101) and the one or more pilot symbols (102); determining a pilot-based covariance matrix (302) based on the one or more pilot symbols (102) and the channel estimate (301); determining a data-based covariance matrix (303) based on the received signal (101); The data regularized covariance matrix (103) is determined by weighting the pilot-based covariance matrix (302) and the data-based covariance matrix (303) respectively based on one or more receiver parameters (305) of the receiver (200).

18. A computer program, characterized in that The program comprises instructions which, when executed by a computer, cause the computer to perform the method (1100) according to claim 16 or 17.

Images