A channel estimation method, a channel estimation device and a storage medium

Abstract

The present disclosure relates to a channel estimation method, a channel estimation device and a storage medium. The channel estimation method comprises: determining a channel estimation neural network model corresponding to a signal-to-interference ratio (SINR), and determining a first channel response estimation value of a channel; inputting the first channel response estimation value into the channel estimation neural network model to obtain a second channel response estimation value of the channel; and determining the second channel response estimation value as an estimation value of the channel. The present disclosure can estimate the channel through the channel estimation neural network model, thereby improving the estimation accuracy of the wireless channel.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of wireless communication, and particularly relates to a channel estimation method, a channel estimation device and a storage medium. BACKGROUND

[0002] In a MIMO technology (Multiple-Input Multiple-Output)-Orthogonal Frequency Division Multiplexing (OFDM) system, a linear estimation method for solving a minimum mean square error can be used for channel estimation to achieve accurate and low-computing-overhead channel estimation.

[0003] However, in a fading channel, various nonlinear characteristics are contained, which are difficult to be estimated by a linear estimation method. SUMMARY

[0004] To overcome the problems in the prior art, the present disclosure provides a channel estimation method, a channel estimation device and a storage medium.

[0005] According to a first aspect of an embodiment of the present disclosure, a channel estimation method is provided, and the method comprises:

[0006] determining a channel estimation neural network model corresponding to a signal-to-interference ratio (SINR), and determining a first channel response estimation value of a channel; inputting the first channel response estimation value into the channel estimation neural network model to obtain a second channel response estimation value of the channel; and determining the second channel response estimation value as an estimation value of the channel.

[0007] In an implementation, the determining the channel estimation neural network model corresponding to the channel signal-to-interference ratio (SINR) comprises:

[0008] determining a corresponding relationship between a SINR interval and a channel estimation neural network model, and determining a SINR interval in which the SINR is located; and according to the corresponding relationship, determining a channel estimation neural network model corresponding to the SINR interval in which the SINR is located.

[0009] In an implementation, the determining the corresponding relationship between the SINR interval and the channel estimation neural network model comprises:

[0010] determining a SINR scale, and determining a first number of SINR intervals based on the scale; collecting channel response estimation value data of each of the first number of SINR intervals to obtain a first number of channel response estimation value training data sets; training a neural network model based on the first number of channel response estimation value training data sets to obtain a first number of channel estimation neural network models.

[0011] In an implementation, the training of the neural network model based on the first number of channel response estimation value training data sets to obtain a first number of neural network models comprises:

[0012] collecting real part training data sets and imaginary part training data sets of each of the first number of SINR intervals; determining a first channel real part training data set based on the real part training data sets, and determining labels corresponding to the first channel real part training data set; determining a first channel imaginary part training data set based on the imaginary part training data sets, and determining labels corresponding to the first channel imaginary part training data set; taking the combination data of each of the first channel real part training data sets and the corresponding labels, and the combination data of the first channel imaginary part training data sets and the corresponding labels as inputs of the neural network model, and training the neural network model to obtain a first number of channel estimation neural network models.

[0013] In an implementation, the neural network model comprises a denoising convolutional neural network (DNCNN) and a super-resolution convolutional neural network (SRCNN).

[0014] The training of the neural network model based on the combination data of each of the first channel real part training data sets and the corresponding labels, and the combination data of the first channel imaginary part training data sets and the corresponding labels comprises:

[0015] The combination of the first channel real part training data sets and the corresponding labels, and the combination of the first channel imaginary part training data sets and the corresponding labels to obtain training combination data; taking the training combination data as inputs of the DNCNN and the SRCNN in the neural network model, respectively, and training the DNCNN and the SRCNN, respectively, to obtain a first number of DNCNN models and a first number of SRCNN models after training; and combining the first number of DNCNN models and the first number of SRCNN models to obtain a first number of channel estimation neural network models.

[0016] In an implementation, the determination of the first channel response estimation value of the channel comprises:

[0017] determining a channel response estimation value of a position where a demodulation reference signal (DMRS) is located in a channel; separating a real part and an imaginary part of the channel response estimation value, and performing interpolation processing on the real part and the imaginary part respectively to obtain a first channel real part and a first channel imaginary part; and determining a complex number obtained by combining the first channel real part and the first channel imaginary part as the first channel response estimation value.

[0018] In an implementation, the inputting the first channel response estimation value into the channel estimation neural network model to obtain a second channel response estimation value of the channel comprises:

[0019] inputting the first channel real part and the first channel imaginary part of the first channel response estimation value into the channel estimation neural network model respectively to obtain a second channel real part and a second channel imaginary part; and determining a complex number obtained by combining the second channel real part and the second channel imaginary part as the second channel response estimation value.

[0020] According to a second aspect of the embodiments of the present disclosure, a channel estimation device is provided, and the device comprises:

[0021] a determination module configured to determine a channel estimation neural network model corresponding to a signal-to-interference ratio (SINR) and determine a first channel response estimation value of a channel; and an estimation module configured to input the first channel response estimation value into the channel estimation neural network model to obtain a second channel response estimation value of the channel; and the determination module is configured to determine the second channel response estimation value as an estimation value of the channel.

[0022] In an implementation, the determination module is configured to:

[0023] determine a correspondence between SINR intervals and channel estimation neural network models, and determine an SINR interval in which the SINR is located; and determine, according to the correspondence, a channel estimation neural network model corresponding to the SINR interval in which the SINR is located.

[0024] In an implementation, the determination module is configured to:

[0025] determine a SINR scale and determine a first number of SINR intervals based on the scale; collect channel response estimation value data of each of the first number of SINR intervals to obtain a first number of channel response estimation value training data sets; and train a neural network model based on the first number of channel response estimation value training data sets to obtain a first number of channel estimation neural network models.

[0026] In an implementation, the determination module is configured to:

[0027] collecting a real part training data set and an imaginary part training data set of each of the SINR intervals in the first quantity; determining a first channel real part training data set based on the real part training data set, and determining a label corresponding to the first channel real part training data set; determining a first channel imaginary part training data set based on the imaginary part training data set, and determining a label corresponding to the first channel imaginary part training data set; combining each of the first channel real part training data sets and the corresponding labels, and the first channel imaginary part training data sets and the corresponding labels in the first quantity, as training combined data, and taking the training combined data as input of a neural network model, to train the neural network model, to obtain a first quantity of channel estimation neural network models.

[0028] In an implementation form, the neural network model comprises a denoising convolutional neural network DNCNN and a super-resolution convolutional neural network SRCNN.

[0029] The determining module is configured to:

[0030] The determining module is configured to:

[0031] In an implementation form, the determining module is configured to:

[0032] The determining module is configured to:

[0033] In an implementation form, the determining module is configured to:

[0034] The determining module is configured to:

[0035] According to a third aspect of embodiments of the present disclosure, a channel estimation device is provided, comprising:

[0036] A processor; a memory for storing processor-executable instructions; wherein the processor is configured to perform the channel estimation method of the first aspect or any one of the implementation forms of the first aspect.

[0037] According to a fourth aspect of the embodiments of the present disclosure, a non-transitory computer-readable storage medium is provided, when instructions in the storage medium are executed by a processor of a mobile terminal, the mobile terminal is enabled to perform the channel estimation method of the first aspect or any one of the implementation forms of the first aspect.

[0038] The technical solutions provided by the embodiments of the present disclosure can have the following beneficial effects: by determining the SINR, a channel estimation neural network model used for channel estimation is determined, and channel estimation is performed based on the channel estimation neural network model and the first channel response estimation value of the channel to determine the estimation value of the channel. The channel can be estimated by the channel estimation neural network model, which can improve the estimation accuracy of the wireless channel and reduce the computational complexity in the estimation process, so as to improve the channel capacity of the MIMO-OFDM system and obtain greater performance gain.

[0039] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and are not limiting to the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0040] The accompanying drawings, which are incorporated into and form part of the specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the disclosure.

[0041] Figure 1 is a communication system architecture diagram of a network device and a terminal according to an exemplary embodiment.

[0042] Figure 2 is a frame structure diagram of single-symbol DMRS and CSI-RS in a 5G NR protocol in a channel estimation method according to an exemplary embodiment.

[0043] Figure 3 is a system block diagram in a channel estimation method according to an exemplary embodiment.

[0044] Figure 4 is a flowchart of a channel estimation method according to an exemplary embodiment.

[0045] Figure 5 is a flowchart of a channel estimation method according to an exemplary embodiment.

[0046] Figure 6 is a flowchart of a channel estimation method according to an exemplary embodiment.

[0047] Figure 7 is a flow diagram of a neural network model in a channel estimation method according to an example embodiment.

[0048] Figure 8 is a flow diagram of a channel estimation method according to an example embodiment.

[0049] Figure 9 is a flow diagram of a channel estimation method according to an example embodiment.

[0050] Figure 10 is a flow diagram of a channel estimation method according to an example embodiment.

[0051] Figure 11 is a specific structure diagram of a neural network model in a channel estimation method according to an example embodiment.

[0052] Figure 12 is a flow diagram of a channel estimation method according to an example embodiment.

[0053] Figure 13 is a flow diagram of a channel estimation method according to an example embodiment.

[0054] Figure 14 is a block diagram of a channel estimation device according to an example embodiment.

[0055] Figure 15 is a block diagram of a channel estimation device according to an example embodiment.

[0056] Figure 16 is a block diagram of a channel estimation device according to an example embodiment. DETAILED DESCRIPTION

[0057] The example embodiments will be described in detail below with reference to the attached drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following example embodiments are not meant to represent all implementations consistent with the present disclosure. Rather, they are merely examples that can be implemented in conjunction with some aspects of the present disclosure as detailed in the appended claims.

[0058] Figure 1 is a network device and terminal communication system architecture diagram according to an example embodiment. The communication method provided by the present disclosure can be applied to Figure 1 the communication system architecture diagram shown. As Figure 1 shown, the network side device can be based on Figure 1The illustrated architecture sends signaling.

[0059] In MIMO-OFDM systems, channel estimation can be performed using linear estimation methods that solve for the minimum mean square error. The channel estimation is mathematically based and, after obtaining an initial estimate of the channel, the channel variations are tracked using the distributed pilot tones in the time-frequency grid of the OFDM system. A rectangular pilot pattern is a commonly used pilot arrangement pattern.

[0060] According to the description of the frame structure in the NR protocol, the following parameter configuration mode is selected for the transmission frame as an example of channel estimation. Each physical resource block (PRB) contains 12 subcarriers, the subcarrier spacing is 15 kHz, there is one slot in a subframe with a time length of 1 ms, there are 14 OFDM symbols in a slot, and a cyclic prefix is added to each OFDM symbol. The upper layer parameter mapping class selects class A, the reference signal configuration class selects class 1, the demodulation reference signal (DMRS) uses a single symbol, and no additional DMRS symbol is used. Figure 1 is a class 1, single symbol DMRS configuration diagram according to an exemplary embodiment. As Figure 1 shown, it is a DMRS configuration pattern in a PRB.

[0061] Suppose that N PRBs are used for channel estimation, and there are K DMRS symbols in the N PRBs. First, the LS estimation method is used to obtain the channel response estimation value H LS at the K DMRS positions. The channel estimation values at the K DMRS positions are arranged as a column vector p with dimensions Kx1. Second, for the channel response value h n,j at any point on the orthogonal frequency division multiplexing time-frequency grid, n represents the time domain OFDM symbol index, j represents the frequency domain subcarrier index, and using the linear minimum mean square error (LMMSE) estimation method, the linear minimum mean square error estimation value at the point is:

[0062]

[0063] wherein:

[0064] r hp =E{h n,j p H}

[0065] R pp =E{pp H}

[0066] However, the channel estimation using the above embodiment, i.e., the estimation method based on the discrete pilot, relies on the statistical characteristics of the channel and the transmitted data power. For each transmission frame, the channel response estimated according to the overall characteristics always has a certain deviation from the actual channel. Moreover, the two-dimensional Wiener filtering method is used for channel interpolation of the full time-frequency domain grid using the channel estimation value at the discrete pilot, which has a high computational complexity, and the complexity increases with the increase of the number K of DMRS symbols used.

[0067] However, deep learning is a frontier field in artificial intelligence, and the network architecture proposed in this discipline is good at processing the fitting problem of a nonlinear function. Deep learning has shown excellent performance in general computer fields, especially in the field of image processing. The results obtained using the neural network architecture often exceed the original algorithm. The time-frequency domain channel structure of the MIMO-OFDM system is very similar to a picture. Exploring the use of deep learning to estimate wireless channels to obtain an estimation method with higher estimation accuracy and lower computational complexity, thereby further improving the transmission performance of the MIMO-OFDM system. Therefore, the present disclosure provides a channel estimation method based on a neural network model of deep learning to estimate the channel, improve the estimation accuracy of the wireless channel, and reduce the computational complexity in the estimation process, thereby improving the channel capacity of the MIMO-OFDM system and obtaining greater performance gain.

[0068] Among them, based on the channel estimation method provided by the present disclosure, the channel estimation device according to the present disclosure includes a sending end and a receiving end. Among them, the sending end configures the reference signal DMRS and the channel state information reference signal (CSI-RS) when doing signal processing in the physical layer, and sends the signal after the signal processing is completed. The receiving end obtains the received signal, and calculates the signal-to-interference ratio SINR of the current propagation environment using the CSI-RS in the received signal. The signal processing module of the receiving end is built-in a signal-to-interference ratio scale, which is divided into several intervals according to the dB value, and each interval has a corresponding neural network model. The neural network model corresponding to the interval is used to estimate the channel according to which interval the calculated SINR falls.

[0069] The neural network estimates the channel density in units of PRB. Take the estimation of channel response of K PRBs as an example. First, obtain the DMRS in the K PRBs, process the DMRS to obtain the channel response estimation value at the DMRS. Separate the real part and the imaginary part of the channel response estimation value at the DMRS, and do two-dimensional interpolation on them respectively, wherein the time-frequency domain position of the channel response estimation value at the DMRS is not changed in the interpolation process. After two-dimensional interpolation of the real part and the imaginary part of the channel response estimation value at the DMRS, the real part and the imaginary part of the rough channel response estimation value are obtained. The real part and the imaginary part of the rough channel response estimation value are respectively passed through the selected neural network model to obtain the real part and the imaginary part of the accurate channel response estimation value. Reassemble the accurate channel response estimation value to complete the channel estimation of the K PRBs.

[0070] According to the interval division of the signal-to-interference ratio scale built in the receiver signal processing module, a signal-to-interference ratio point is selected from each interval. When the signal-to-interference ratio of the simulation or real environment is at this point, data collection begins, and the collected data includes the channel response estimation value at the DMRS in the K PRBs and the ideal estimation value of the channel response of the K PRBs. Data is collected at the signal-to-interference ratio point selected in each interval, and is arranged into a separate data set. Each data set is preprocessed to obtain a training data set for training a neural network model. Each training data set trains a neural network model.

[0071] It can be further understood that the wireless communication system of the embodiments of the present disclosure is a network that provides wireless communication functions. The wireless communication system can use different communication technologies, such as code division multiple access (CDMA), wideband code division multiple access (WCDMA), time division multiple access (TDMA), frequency division multiple access (FDMA), orthogonal frequency division multiple access (OFDMA), single carrier frequency division multiple access (SC-FDMA), carrier sense multiple access with collision avoidance (CSMA / CA), and the like. Depending on the capacity, rate, latency, and the like of different networks, the networks can be classified into 2G (English: generation) networks, 3G networks, 4G networks, or future evolution networks, such as 5G networks. The 5G network can also be referred to as a new radio network (NR). For the convenience of description, the wireless communication network can be referred to as a network in the present disclosure.

[0072] Further, the network device involved in the present disclosure can also be referred to as a wireless access network device. The wireless access network device can be a base station, an evolved node B (eNB), a home base station, an access point (AP) in a wireless fidelity (WIFI) system, a wireless relay node, a wireless backhaul node, a transmission point (TP), a transmission and reception point (TRP), and the like. It can also be a gNB in an NR system, or it can also be a component or a part of a device constituting a base station, and the like. When it is a vehicle-to-everything (V2X) communication system, the network device can also be a vehicle-mounted device. It should be understood that the specific technology and specific device form of the network device are not limited in the embodiments of the present disclosure.

[0073] Further, the terminal involved in the present disclosure, which can also be referred to as a terminal device, a user equipment (UE), a mobile station (MS), a mobile terminal (MT), etc., is a device that provides voice and / or data connectivity for a user, for example, the terminal can be a handheld device having wireless connection function, a vehicle-mounted device, etc. At present, some examples of the terminal are: a mobile phone, a pocket personal computer (PPC), a palm computer, a personal digital assistant (PDA), a notebook computer, a tablet computer, a wearable device, or a vehicle-mounted device, etc. In addition, when it is a vehicle-to-everything (V2X) communication system, the terminal device can also be a vehicle-mounted device. It should be understood that the present disclosure embodiments do not limit the specific technology and specific device form of the terminal.

[0074] If the transmission bandwidth is 20MHz, there are 100 PRBs in total. Each PRB occupies 12 subcarriers in the frequency domain and one complete time slot in the time domain. The subcarrier spacing is 15kHz, one time slot in a subframe with a time length of 1ms, and there are 14 OFDM symbols in one time slot. The upper-layer parameter mapping class selects class A, the DMRS reference signal configuration class selects class 1, the DMRS uses a single symbol, and no additional DMRS symbol is used. The upper-layer parameter CSI-RS Row is set to 6, and the CSI-RS uses the configuration mode of class FD-CDM2. No hybrid beamforming is used, the channel between the transmitting layer and each receiving antenna port is estimated, and the number of PRBs for single channel estimation is K. Figure 2 is a frame structure diagram of a single-symbol DMRS and CSI-RS in a 5G NR protocol according to an exemplary embodiment. In the diagram, the resource particles occupied by each reference signal in a single PRB can be referred to as Figure 2 .

[0075] In the present disclosure embodiments, the system applying the channel estimation method includes a transmitting end and a receiving end, wherein the transmitting end can be a terminal, and the receiving end can be a network device; or, the transmitting end can be a network device, and the receiving end can be a terminal; in this regard, the present disclosure embodiments do not limit the specific transmitting end and receiving end devices.

[0076] Figure 3 is a system block diagram in a channel estimation method according to an exemplary embodiment. As shown in Figure 3As shown, the system is mainly divided into two parts, the sending end and the receiving end. The communication process between the sending end and the receiving end follows the NR protocol, but neither the sending end nor the receiving end uses hybrid beamforming. The sending end is responsible for generating and sending data and completing all signal processing processes before data transmission.

[0077] The receiving end is divided into two parts. The first part is a signal processing module, which needs to calculate the signal-to-interference ratio (SINR) of the current environment based on CSI-RS. According to the SINR, a neural network model suitable for the current propagation environment is selected to estimate the channel. Different models represent that they are trained using different data sets. For example, the estimation performance of neural networks trained in different signal-to-interference ratio environments is not the same. The neural network trained in a high signal-to-interference ratio environment has better estimation performance when the received signal-to-interference ratio is high, and vice versa.

[0078] The second part is a neural network training module, which is responsible for training the neural network. The training method is supervised learning. The time-frequency domain structure of the MIMO-OFDM channel is taken as an example to estimate the channel response of K PRBs. The entire time-frequency domain structure is mathematically equivalent to a complex matrix. After separating the real part and the imaginary part of the matrix elements, two real matrices with unchanged dimensions can be obtained. After obtaining the channel response estimation value at the DMRS at the receiving end, the channel response estimation value at the DMRS at other time-frequency positions is estimated based on the neural network model. Taking the estimation of the channel response of K PRBs as an example, the entire time-frequency domain structure is mathematically equivalent to a complex matrix. After separating the real part and the imaginary part of the matrix elements, two real matrices with unchanged dimensions can be obtained. Each real matrix can be approximately regarded as a "picture". After obtaining the channel response estimation value at the DMRS at the receiving end, it is equivalent to knowing part of the pixels of the two "pictures". Of course, the remaining part of the pixels of the two pictures is still unknown, and these known pixels may not be accurate. At this time, the channel estimation problem is transformed into how to restore a low-resolution and incomplete pixel "picture" to a high-resolution and complete pixel "picture". Using this method, training and learning can be carried out by referring to image processing related technologies.

[0079] Figure 4 is a flowchart of a channel estimation method according to an example embodiment. As shown in Figure 4 The channel estimation method includes the following steps S11-S13.

[0080] In step S11, a channel estimation neural network model corresponding to the SINR is determined, and a first channel response estimation value of the channel is determined.

[0081] In the embodiments of the present disclosure, the receiving end calculates the signal-to-interference ratio (SINR) of the reference signal (CSI-RS) to determine the corresponding channel estimation neural network model. The channel response estimation value at the DMRS position is determined according to the received DMRS, and the coarse channel response estimation value is determined by two-dimensional interpolation according to the channel response estimation value at the DMRS position. The coarse channel response estimation value is referred to as the first channel response estimation value of the channel for the convenience of distinction.

[0082] In step S12, the first channel response estimation value is input into the channel estimation neural network model to obtain the second channel response estimation value of the channel.

[0083] In the embodiments of the present disclosure, the first channel response estimation value is input into the determined channel estimation neural network model to obtain the accurate channel response estimation value. The accurate channel response estimation value output by the channel estimation neural network model is referred to as the second channel response estimation value of the channel for the convenience of distinction.

[0084] In step S13, the second channel response estimation value is determined as the estimation value of the channel.

[0085] In the embodiments of the present disclosure, the accurate channel response estimation value output by the channel estimation neural network model, that is, the second channel response estimation value, is determined as the estimation value of the channel.

[0086] The channel estimation method provided by the embodiments of the present disclosure can determine the coarse channel response estimation value of the channel through the channel response estimation value at the DMRS of the channel, and determine the accurate channel estimation response value through the channel estimation neural network model, so as to realize the estimation of the channel. The estimation of the channel through the channel estimation neural network model can improve the estimation accuracy of the wireless channel, reduce the calculation complexity in the estimation process, improve the channel capacity of the MIMO-OFDM system, and obtain greater performance gain.

[0087] Figure 5 FIG. 1 is a flow diagram of a method for estimating a wireless channel based on deep learning according to an exemplary embodiment. As shown in FIG. 1, the method comprises the following steps. Figure 5

[0088] In step S21, the signal-to-interference ratio (SINR) of the received signal is calculated according to the reference signal (CSI-RS) in the received signal.

[0089] In step S22, a signal-to-interference ratio scale is built in, which is divided into several intervals. If the received signal-to-interference ratio (SINR) belongs to one of the intervals, a neural network model corresponding to the interval is selected to estimate the channel. ​

[0090] In some possible implementation manners, step 22 can include sub-steps S221 and S222.

[0091] In sub-step S221, the receiver is provided with an SINR scale table S, which is divided into n intervals Si, where i = 1, 2, …, n, according to the dB value of SINR. Each SINR interval Si has a corresponding neural network model Mi, where i = 1, 2, …, n.

[0092] In sub-step S222, according to the SINR value of the received signal, it is determined which interval it belongs to. After determination, the neural network model corresponding to the interval is selected to estimate the channel.

[0093] In step S23, the single channel estimation density is K PRBs, and a vector composed of the channel response estimation values at the DMRS positions in the K PRBs is obtained. The real part and the imaginary part of the vector are separated.

[0094] In some possible implementation manners, step S23 can include sub-step S231 and sub-step S232.

[0095] In sub-step S231, a vector p is obtained by using the channel response estimation values at all DMRS positions in the K PRBs.

[0096] In sub-step S232, the real part and the imaginary part of the vector p are separated to obtain p-re and p-im, respectively.

[0097] In step S24, the real part and the imaginary part of the vector composed of the channel response estimation values at the DMRS positions are respectively subjected to two-dimensional interpolation to obtain the real part and the imaginary part of the coarse channel response estimation value.

[0098] In some possible implementation manners, step S24 can include sub-step S241 and sub-step S242.

[0099] In sub-step 241, the real part p-re of the vector p is subjected to two-dimensional interpolation to obtain the real part h-re of the coarse channel response estimation value.

[0100] In step 242, the imaginary part p-im of the vector p is subjected to two-dimensional interpolation to obtain the imaginary part h-im of the coarse channel response estimation value.

[0101] In step S25, the real part of the coarse channel response estimation value is subjected to the neural network model to estimate the real part of the accurate channel response, and the imaginary part of the coarse channel response estimation value is subjected to the neural network model to estimate the imaginary part of the accurate channel response.

[0102] In some possible implementation manners, step S25 can include sub-steps S251-S253.

[0103] In sub-step S251, assuming that the signal-to-interference ratio SINR of the received signal belongs to interval Si, a neural network model Mi is selected to estimate the channel.

[0104] In sub-step S252, h-re is obtained by the neural network model Mi to estimate the real part of the accurate channel response value.

[0105] In sub-step S253, h-im is obtained by the neural network model Mi to estimate the imaginary part of the accurate channel response value.

[0106] In step S26, the real part of the accurate channel response estimated by the neural network model and the imaginary part of the accurate channel response are combined into an accurate channel response estimation value, and the channel estimation of the K PRBs is completed.

[0107] In some possible implementation manners, step S26 can include sub-step S261.

[0108] In sub-step S261, H-re and H-im are combined into H = H-re + H-im x j. H is the accurate channel response estimation value.

[0109] In the embodiments of the present disclosure, the channel estimation can be completed through the above steps, and the above implementation steps will be further described in combination with the drawings in the following embodiments.

[0110] Figure 6 is a flowchart of a channel estimation method according to an example embodiment. As shown in Figure 6 The channel estimation method includes the following steps S31-S32.

[0111] In step S31, the correspondence between the SINR interval and the channel estimation neural network model is determined, and the SINR interval in which the SINR is located is determined.

[0112] In step S32, according to the correspondence, the channel estimation neural network model corresponding to the SINR interval in which the SINR is located is determined.

[0113] In the embodiments of the present disclosure, one SINR scale can be determined, and a first number of SINR intervals can be determined according to the determined SINR scale, and each SINR interval corresponds to one channel estimation neural network model. For example, one SINR scale table S is determined, and according to the scale of SINR, for example, one dB value is determined, and the dB value is divided into a first number of SINR intervals Si. If the first number is n, i = 1, 2,..., n. Each SINR interval Si has a corresponding channel estimation neural network model Mi, where i = 1, 2,..., n.

[0114] As in the above embodiments, after receiving the CSI-RS reference signal, the calculated SINR is determined in the SINR interval. According to the correspondence between the determined SINR interval and the channel estimation neural network model, the channel estimation neural network model corresponding to the SINR interval is determined. The determined channel estimation neural network model is used to estimate the channel.

[0115] The flow of the above-mentioned embodiments of the present disclosure can refer to Figure 7 , Figure 7 is a flowchart of a neural network model in a channel estimation method according to an exemplary embodiment. As Figure 7 shown, the following steps S41 to S44 are included.

[0116] In step S41, a given signal-to-interference ratio scale is divided into several intervals, and a signal-to-interference ratio point is selected in each interval.

[0117] In some possible implementations, step S41 can include sub-steps S411 and S412:

[0118] In sub-step S411, a SINR scale table S is given, and the SINR dB value is divided into n intervals: Si, where i = 1, 2,..., n. Note that the given SINR scale table and the way of dividing the intervals are the same as the SINR scale table built-in the receiver and the way of dividing the intervals;

[0119] In sub-step 412, a signal-to-interference ratio point qi in the interval is selected in each divided SINR interval Si (i = 1, 2,..., n), where i = 1, 2,..., n. Each signal-to-interference ratio point qi corresponds to one SINR interval Si.

[0120] In step S42, data is collected at each selected signal-to-interference ratio point, and is arranged into several independent data sets, and each data set corresponds to one signal-to-interference ratio point.

[0121] In some possible implementation, step S42 can include sub-steps S421 and S422:

[0122] In sub-step S421, the received signal-to-interference ratio in the simulation or real environment is obtained, and data is collected when the received signal-to-interference ratio is qi (where i = 1, 2, …, n). The collected data includes: a vector pi formed by the channel response estimation values at the DMRS positions in the K PRBs, and an ideal estimation value Hi of the channel response of the K PRBs;

[0123] In sub-step S422, the collected data is sorted into n data sets, and each data set Di corresponds to a signal-to-interference ratio point qi, where i = 1, 2, …, n. The data set Di includes: a vector pi formed by the channel response estimation values at the DMRS positions in the K PRBs collected at the signal-to-interference ratio point qi, and an ideal estimation value Hi of the channel response of the K PRBs.

[0124] In step S43, each data set collected is preprocessed to obtain training data sets with the same number as the data sets.

[0125] In some possible implementation, step S43 can include sub-steps S431 to S437:

[0126] In sub-step S431, the data set Di is obtained, where i = 1, 2, …, n. The vector pi in Di is obtained, and the elements in the vector pi are arranged according to the time-frequency points where the DMRS is collected;

[0127] In sub-step S432, the real part and the imaginary part of the vector pi after the time-frequency position adjustment are separated to obtain pi-re and pi-im, respectively;

[0128] In sub-step S433, the real part pi-re of the vector pi after the time-frequency position adjustment is two-dimensionally interpolated to obtain the real part hi-re of the rough channel response estimation value, and the dimension of hi-re is the same as that of the ideal estimation value Hi of the channel response of the K PRBs;

[0129] In sub-step S434, the imaginary part pi-im of the vector pi after the time-frequency position adjustment is two-dimensionally interpolated to obtain the imaginary part hi-im of the rough channel response estimation value, and the dimension of hi-im is the same as that of the ideal estimation value Hi of the channel response of the K PRBs;

[0130] In sub-step S435, the ideal estimation value Hi of the channel response of the K PRBs in Di is obtained, and the real part and the imaginary part thereof are separated to obtain Hi-re and Hi-im;

[0131] In sub-step S436, the pre-processed data is combined. hi-re is the input data for training the neural network, and Hi-re is the corresponding label thereof, hi-re and Hi-re constitute a set of training data y1_i; hi-im is the input data for training the neural network, and Hi-im is the corresponding label thereof, hi-im and Hi-im constitute a set of training data y2_i. y1_i and y2_i are combined into a training data set Zi;

[0132] In sub-step S437, sub-steps S431 to S436 are repeated to obtain n training data sets.

[0133] In step S44, a neural network model is trained independently using each training data set, and the number of trained neural network models is the same as the number of training data sets.

[0134] In some possible implementations, step S44 can include sub-steps S441 to S443.

[0135] In sub-step S441, for training data set Zi, i = 1, 2,..., n, a neural network model Mi is trained using Zi;

[0136] In sub-step S442, using a supervised learning method, the loss function is mean square error, and y1_i and y2_i in training data set Zi are used together to train neural network model Mi, wherein hi-re in y1_i is the input data, and Hi-re is the corresponding label thereof; hi-im in y2_i is the input data, and Hi-im is the corresponding label thereof.

[0137] In sub-step S443, sub-steps S431 to S436 are repeated to obtain n neural network models.

[0138] The following embodiments will be further described with reference to the accompanying drawings.

[0139] Figure 8 is a flowchart of a channel estimation method according to an exemplary embodiment. As shown in Figure 8 The channel estimation method comprises the following steps.

[0140] In step S51, the SINR scale is determined, and a first number of SINR intervals is determined based on the scale.

[0141] In the embodiments of the present disclosure, based on the determined SINR scale, a first number of SINR intervals are determined, and further, for each determined SINR point, the SINR point is used to determine the DMRS time-frequency position for collecting channel response estimation value data. For example, in each divided SINR interval Si (i = 1, 2,..., n), a SINR point qi in the interval is selected, where i = 1, 2,..., n. Each signal-to-interference ratio point qi corresponds to a same SINR interval Si.

[0142] In step S52, channel response estimation value data in each SINR interval of the first number is collected to obtain a first number of channel response estimation value training data sets.

[0143] In the embodiments of the present disclosure, a received SINR in a simulation or real environment is received, and based on the SINR point qi of each SINR interval, in a case where the received SINR is qi, channel response estimation value data of a corresponding SINR interval is collected. The collected data includes: a vector pi formed by arranging channel response estimation values at DMRS positions in K PRBs, and ideal estimation values Hi of channel responses of the K PRBs.

[0144] The collected data is arranged into n data sets Dn, each data set Di corresponding to a signal-to-interference ratio point qi, where i = 1, 2, …, n. The data set Di includes: a vector pi arranged by the channel response estimation values at the DMRS positions in the K PRBs collected at the signal-to-interference ratio point qi, and an ideal estimation value Hi of the channel response of the K PRBs. For the data set Di, where i = 1, 2, …, n. Obtain the vector pi in Di, and arrange the elements in the vector pi according to the time-frequency points where the DMRS is collected. Separate the real part and the imaginary part of the vector pi after adjusting the time-frequency position to obtain pi-re and pi-im, respectively. Perform two-dimensional interpolation on the real part pi-re of the vector pi after adjusting the time-frequency position to obtain the real part hi-re of the rough channel response estimation value, which has the same dimension as the ideal estimation value Hi of the channel response of the K PRBs. Perform two-dimensional interpolation on the imaginary part pi-im of the vector pi after adjusting the time-frequency position to obtain the imaginary part hi-im of the rough channel response estimation value, which has the same dimension as the ideal estimation value Hi of the channel response of the K PRBs. Obtain the ideal estimation value Hi of the channel response of the K PRBs in Di, separate the real part and the imaginary part to obtain Hi-re and Hi-im. Combine the preprocessed data. hi-re is the input data for training the neural network, Hi-re is the corresponding label, and hi-re and Hi-re constitute a set of training data y1_i; hi-im is the input data for training the neural network, Hi-im is the corresponding label, and hi-im and Hi-im constitute a set of training data y2_i. y1_i and y2_i are combined into a training data set Zi. If the first number is n, then i = 1, 2, …, n.

[0145] In step S53, based on the first number of channel response estimation value training data sets, the neural network model is trained to obtain the first number of channel estimation neural network models.

[0146] Based on the determined combined training data, i.e., the first number of channel response estimation value training data sets, the neural network model corresponding to each channel response estimation value training data set is determined. Based on each channel response estimation value training data set, the neural network model corresponding thereto is trained separately to obtain the first number of channel estimation neural network models.

[0147] Figure 9 is a flowchart of a channel estimation method according to an example embodiment. As shown in Figure 9 The channel estimation method includes the following steps.

[0148] In step S61, the real part training data set and the imaginary part training data set of each SINR interval in the first number are collected.

[0149] In step S62, a first channel real part training data set is determined based on the real part training data set, and a label corresponding to the first channel real part training data set is determined.

[0150] In step S63, a first channel imaginary part training data set is determined based on the imaginary part training data set, and a label corresponding to the first channel imaginary part training data set is determined.

[0151] In step S64, the combination data of each first channel real part training data set and the corresponding label, and the combination data of the first channel imaginary part training data set and the corresponding label in the first quantity are taken as the input of the neural network model, and the neural network model is trained to obtain the first quantity of channel estimation neural network models.

[0152] In the embodiment of the present disclosure, based on the first quantity of channel response estimation value training data sets, the real part training data set and the imaginary part training data set of each channel response estimation value training data set are determined. The label corresponding to the first channel real part training data set and the label corresponding to the first channel imaginary part training data set are determined in each channel response estimation value training data set. The neural network model is trained in the manner of training the neural network model independently according to each channel response estimation value training data set, that is, the training data set Zi is used to train the neural network model Mi. The method of supervised learning is used, and the loss function is the mean square error. y1_i and y2_i in the training data set Zi are used together to train the neural network model Mi, wherein hi-re in y1_i is the input data, and Hi-re is the corresponding label; hi-im in y2_i is the input data, and Hi-im is the corresponding label. The first quantity of channel estimation neural network models is obtained.

[0153] In the embodiment of the present disclosure, the neural network model can be composed of two parts in series, one part is a high-resolution convolutional neural network (Super Resolution Convolution Neural Network, SRCNN), and the other part is a denoising convolutional neural network (Denoising Convolution Neural Network, DNCNN).

[0154] Figure 10 is a flow chart of a channel estimation method according to an exemplary embodiment. As shown in Figure 10 , the channel estimation method comprises the following steps.

[0155] In step S71, the training combination data obtained by combining each first channel real part training data set and the corresponding label, and the first channel imaginary part training data set and the corresponding label in the first quantity.

[0156] In step S72, the training combination data are respectively input into the DNCNN and the SRCNN in the neural network model, and the DNCNN and the SRCNN are trained respectively to obtain the first number of DNCNN models and the first number of SRCNN models after training.

[0157] In step S73, the first number of DNCNN models and the first number of SRCNN models are combined to obtain the first number of channel estimation neural network models.

[0158] In the embodiments of the present disclosure, the DNCNN and the SRCNN can be trained separately, each training combination data in the first number is input into the DNCNN in the neural network model to train the DNCNN, and the training combination data is input into the SRCNN in the neural network model to train the SRCNN, to obtain the trained DNCNN model and the SRCNN model. The DNCNN model and the SRCNN model are combined in series to determine the channel estimation neural network model, thereby obtaining the first number of channel estimation neural network models.

[0159] In the embodiments of the present disclosure, by training the DNCNN and the SRCNN separately, time can be saved and the training speed of the model can be improved.

[0160] The present disclosure needs to be explained that, Figure 11 is a specific structure diagram of a neural network model of a channel estimation method according to an exemplary embodiment. As shown in Figure 11 SRCNN is divided into 3 layers in total. The first layer is used to capture the large block features of the blurred channel impulse response, uses a convolutional neural network, uses n1 filters, and the spatial parameters of each filter are cxf1xf1, c is the number of channels of the input layer. W [1] represents all parameters for n1 convolution operations. B [1] represents the output obtained after each convolution, and the bias is added in the form of broadcast. The activation function of this layer uses ReLu.

[0161] F1(Y)=ReLu(W [1] *Y+B [1] )

[0162] The second layer is used for spatial conversion, mapping the space where the blurred channel features are located to the space where the ideal channel estimation features are located. The convolutional neural network is operated, n2 filters are used, and the spatial parameters of each filter are n1xf2xf2, n1 is the number of channels after being extracted by the first layer. W [2] represents all parameters for n2 convolution operations. B [2]This represents the output obtained after each convolution, with bias added in a broadcast manner. The role of the neural network in this layer is to perform a non-linear mapping from an n1-dimensional vector to an n2-dimensional vector; the filter space parameter can be set to f2 = 1. The activation function for this layer is ReLU.

[0163] F2(Y)=ReLU(W) [2] *F1(Y)+B [2] )

[0164] The third layer is used to reconstruct the accurately estimated channel features, resulting in a precise channel impulse response. The convolutional neural network operation uses c filters (corresponding to the number of input layer channels), where the spatial parameters of each filter are n²×f³×f³, and n² is the number of channels after linear mapping. W [3] This represents all the parameters used in c convolution operations. B [3] This represents the output obtained after each convolution, with the bias added in a broadcast manner.

[0165] F3(Y)=W [3] *F2(Y)+B [3]

[0166] Parameter configuration list 1:

[0167] Table 1

[0168] f1 = 9 f2 = 3 f3 = 5 n1 = 64 n2 = 32 c＝1 Padding ='same' Padding ='same' Padding ='same'

[0169] It is understood that each element in the table above exists independently. These elements are listed in the same table as an example, but this does not mean that all elements in the table must exist simultaneously as shown in the table. The value of each element is independent of the values ​​of any other element in the table. Therefore, those skilled in the art will understand that the value of each element in this table is an independent embodiment.

[0170] The structure of the neural network DNCNN is described below.

[0171] The time-domain structure of the channel impulse response can be obtained using SRCNN, yielding a high-resolution, dimensionally consistent time-frequency structure Y for the same target. DNCNN has 20 layers and can be divided into three parts.

[0172] The first part consists of one layer, operating as a convolutional neural network using n1 filters. Each filter has spatial parameters of c×f1×f1, where c is the number of channels in the input layer. [1] This represents all the parameters for n1 convolution operations. B [1] This represents the output obtained after each convolution, with bias added in a broadcast manner. The activation function for this layer is ReLU.

[0173] F1(Y) = ReLu(W [1] *Y + B [1] )

[0174] The second part is 18 layers of convolutional neural network operation, using n i filters, wherein the spatial parameters of each filter are n i-1 ×f i ×f i , n i-1 is the number of channels of the previous layer. W [i] represents all parameters for n i convolution operations. B [i] represents the output after each convolution, and the bias is added in the form of broadcast. In order to prevent overfitting and speed up the training, each layer is passed through a Batch Normalization layer before being passed through an activation function, and then a ReLu activation function is used.

[0175] F i (Y) = ReLu(BN(W [i] *F i-1 (Y) + B [i] )

[0176] i = 2,..., 19

[0177] The third part is a residual network layer, which first performs convolution using c filters (corresponding to the number of channels of the input layer), wherein the spatial parameters of each filter are n 19 ×f 20 ×f 20 . W

[20] represents all parameters for c convolution operations. B

[20] represents the output after each convolution, and the bias is added in the form of broadcast. Then, the residual is calculated by subtracting the estimated value of DNCNN from the Y input to the neural network, and DNCNN is used to estimate the noise between the input time-frequency domain structure Y and the ideal time-frequency domain structure. After the picture is processed by SRCNN, a large amount of waiting time can be saved during actual training.

[0178] F 20 (Y) = Y - (W

[20] *F 19 (Y) + B

[20] )

[0179] Parameter configuration list 2:

[0180] Table 2

[0181] Layer 1 Layers 2 to 19 Layer 20 [["f1=3"]] f 2-19 = 3 f 20 = 3 n1 = 64 n 2-19 = 64 c＝1 Padding ='same' Padding ='same' Padding ='same'

[0182] It can be understood that each element in the above table exists independently, and the elements are exemplarily listed in the same table, but it does not mean that all elements in the table must exist at the same time as shown in the table. The value of each element is independent of the value of any other element in the table. Therefore, those skilled in the art can understand that the value of each element in the table is an independent embodiment.

[0183] Figure 12 is a flowchart of a channel estimation method according to an exemplary embodiment. As shown in Figure 12 , the channel estimation method comprises the following steps.

[0184] In step S81, a channel response estimation value of a position where a demodulation reference signal DMRS is located in the channel is determined.

[0185] In step S82, the real part and the imaginary part of the channel response estimation value are separated, and interpolation processing is performed on the real part and the imaginary part respectively to obtain a first channel real part and a first channel imaginary part.

[0186] In step S83, a complex number obtained by combining the first channel real part and the first channel imaginary part is determined as a first channel response estimation value.

[0187] In the embodiment of the present disclosure, channel response estimation values at all DMRS positions in the K PRBs are obtained, and a vector p is formed. The real part and the imaginary part of the vector p are separated to obtain p-re and p-im respectively. The real part and the imaginary part of the vector formed by the channel response estimation values at the DMRS positions are respectively interpolated in two dimensions to obtain the real part and the imaginary part of the coarse channel response estimation value. The real part p-re of the vector p is interpolated in two dimensions to obtain the real part h-re of the coarse channel response estimation value. The imaginary part p-im of the vector p is interpolated in two dimensions to obtain the imaginary part h-im of the coarse channel response estimation value. The real part h-re of the coarse channel response estimation value and the imaginary part h-im of the coarse channel response estimation value are combined to determine the first channel response estimation value.

[0188] Figure 13 is a flowchart of a channel estimation method according to an exemplary embodiment. As shown in Figure 13 , the channel estimation method comprises the following steps.

[0189] In step S91, the first channel real part and the first channel imaginary part of the first channel response estimation value are respectively input into a channel estimation neural network model to obtain a second channel real part and a second channel imaginary part.

[0190] In step S92, a complex number obtained by combining the second channel real part and the second channel imaginary part is determined as a second channel response estimation value.

[0191] In the embodiments of the present disclosure, taking an example of a signal-to-interference ratio SINR of a received signal belonging to an interval Si, a neural network model Mi is selected to estimate a channel. An accurate real part H-re of a channel response value is obtained by the neural network model Mi. An accurate imaginary part H-im of the channel response value is obtained by the neural network model Mi. H-re and H-im are combined into H = H-re + H-im * j. H is an accurate channel response estimation value. That is, a second channel response estimation value.

[0192] Based on the same concept, the embodiments of the present disclosure also provide a channel estimation device.

[0193] It can be understood that the channel estimation device provided by the embodiments of the present disclosure comprises a hardware structure and / or a software module corresponding to the execution of each function in order to realize the above functions. In combination with the units and algorithm steps of each example disclosed in the embodiments of the present disclosure, the embodiments of the present disclosure can be realized in the form of hardware or a combination of hardware and computer software. Whether a certain function is realized by hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the technical solutions of the embodiments of the present disclosure.

[0194] Figure 14 is a block diagram of a channel estimation device according to an example embodiment. Referring to Figure 14 The channel estimation device 100 comprises a determination module 101 and an estimation module 102.

[0195] The determination module 101 is configured to determine a channel estimation neural network model corresponding to a signal-to-interference ratio SINR, and determine a first channel response estimation value of a channel. The estimation module 102 is configured to input the first channel response estimation value into the channel estimation neural network model to obtain a second channel response estimation value of the channel. The determination module 101 is configured to determine the second channel response estimation value as an estimation value of the channel.

[0196] In the embodiments of the present disclosure, the determination module 101 is configured to determine a corresponding relationship between an SINR interval and a channel estimation neural network model, and determine an SINR interval in which the SINR is located. According to the corresponding relationship, a channel estimation neural network model corresponding to the SINR interval in which the SINR is located is determined.

[0197] In the embodiment of the present disclosure, the determination module 101 is configured to determine a SINR scale, and determine a first number of SINR intervals based on the scale. Channel response estimation value data of each of the first number of SINR intervals is collected to obtain a first number of channel response estimation value training data sets. The neural network model is trained based on the first number of channel response estimation value training data sets to obtain a first number of channel estimation neural network models.

[0198] In the embodiment of the present disclosure, the determination module 101 is configured to collect real part training data sets and imaginary part training data sets of each of the first number of SINR intervals. A first channel real part training data set is determined based on the real part training data sets, and a label corresponding to the first channel real part training data set is determined. A first channel imaginary part training data set is determined based on the imaginary part training data sets, and a label corresponding to the first channel imaginary part training data set is determined. The combination data of each of the first number of first channel real part training data sets and the corresponding labels, and the combination data of the first channel imaginary part training data sets and the corresponding labels are used as the input of the neural network model, and the neural network model is trained to obtain a first number of channel estimation neural network models.

[0199] In the embodiment of the present disclosure, the neural network model includes a denoising convolutional neural network (DNCNN) and a super-resolution convolutional neural network (SRCNN).

[0200] The determination module 101 is configured to combine the first number of first channel real part training data sets and the corresponding labels, and the first channel imaginary part training data sets and the corresponding labels to obtain training combination data. The training combination data is used as the input of the DNCNN and the SRCNN in the neural network model, respectively, and the DNCNN and the SRCNN are trained respectively to obtain a first number of DNCNN models and a first number of SRCNN models after training. The first number of DNCNN models and the first number of SRCNN models are combined correspondingly to obtain a first number of channel estimation neural network models.

[0201] In the embodiment of the present disclosure, the determination module 101 is configured to determine a channel response estimation value of a position of a demodulation reference signal (DMRS) in a channel. The real part and the imaginary part of the channel response estimation value are separated, and interpolation processing is performed on the real part and the imaginary part respectively to obtain a first channel real part and a first channel imaginary part. A complex number obtained by combining the first channel real part and the first channel imaginary part is determined as a first channel response estimation value.

[0202] In the embodiment of the present disclosure, the determination module 101 is configured to input the first channel real part and the first channel imaginary part of the first channel response estimation value into the channel estimation neural network model respectively to obtain a second channel real part and a second channel imaginary part. A complex number obtained by combining the second channel real part and the second channel imaginary part is determined as a second channel response estimation value.

[0203] With regard to the apparatus in the above-described embodiments, a specific manner in which each module performs operations has been described in detail in embodiments of the method, and thus will not be described in detail here.

[0204] Figure 15 is a block diagram of an apparatus 200 for channel estimation according to an exemplary embodiment. The apparatus 200 can be a mobile phone, a computer, a digital broadcast terminal, a message communicator, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

[0205] Referring to Figure 15 , the apparatus 200 can include one or more of the following components: a processing component 202, a memory 204, a power supply component 206, a multimedia component 208, an audio component 210, an input / output (I / O) interface 212, a sensor component 214, and a communication component 216.

[0206] The processing component 202 generally controls the overall operations of the apparatus 200, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component 202 can include one or more processors 220 to execute instructions to complete all or part of steps of the above-described methods. In addition, the processing component 202 can include one or more modules to facilitate interaction between the processing component 202 and other components. For example, the processing component 202 can include a multimedia module to facilitate the interaction between the multimedia component 208 and the processing component 202.

[0207] The memory 204 is configured to store various types of data to support operations of the apparatus 200. Examples of these data include instructions for any application or method operating on the apparatus 200, contact data, phonebook data, messages, pictures, videos, etc. The memory 204 can be implemented by any type of volatile or non-volatile storage devices or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disc or optical disc.

[0208] The power supply component 206 supplies electrical power for the various components of the apparatus 200. The power supply component 206 can include a power supply management system, one or more power sources, and other components associated with generating, managing and distributing power for the apparatus 200.

[0209] The multimedia component 208 includes a screen providing an output interface between the device 200 and a user. In some embodiments, the screen can include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive input signals from a user. The touch panel includes one or more touch sensors to sense touch, swiping, and gestures on the touch panel. The touch sensors can not only sense a boundary of a touching or swiping action, but also detect duration and pressure related to the touching or swiping action. In some embodiments, the multimedia component 208 includes a front camera and / or a rear camera. When the device 200 is in an operation mode, such as a shooting mode or a video mode, the front camera and / or the rear camera can receive external multimedia data. Each of the front and rear camera can be a fixed optical lens system or have a focal length and optical zooming capability.

[0210] The audio component 210 is configured to output and / or input audio signals. For example, the audio component 210 includes a microphone (MIC) to receive an external audio signal when the device 200 is in an operation mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signal can be further stored in the memory 204 or transmitted via the communication component 216. In some embodiments, the audio component 210 also includes a speaker to output audio signals.

[0211] The I / O interface 212 provides an interface between the processing component 202 and peripheral interface modules, such as a keypad, a click wheel, buttons, and so on. The buttons can include, but are not limited to, a home button, a volume button, a start button, and a lock button.

[0212] The sensor component 214 includes one or more sensors to provide various state assessments for the device 200. For example, the sensor component 214 can detect an open / closed position of the device 200, relative positioning of components, such as a display and a keypad of the device 200, a change in position of the device 200 or a component of the device 200, presence or absence of user contact with the device 200, a change in orientation of the device 200 or acceleration / deceleration of the device 200, and temperature changes of the device 200. The sensor component 214 can include a proximity sensor configured to detect presence of a nearby object without any physical touch. The sensor component 214 can also include a light sensor, such as a CMOS or CCD image sensor, for use in an imaging application. In some embodiments, the sensor component 214 can further include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0213] The communication component 216 is configured to facilitate wired or wireless communication between the device 200 and other devices. The device 200 can access a wireless network based on a communication standard, such as WiFi, 2G, or 3G, or a combination thereof. In an exemplary embodiment, the communication component 216 receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 216 further includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on Radio Frequency Identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0214] In an exemplary embodiment, the device 200 can be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, micro-controllers, microprocessors or other electronic elements, for performing the above-described methods.

[0215] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions, such as the memory 204 including instructions, is also provided, which can be executed by the processor 220 of the device 200 to complete the above-described methods. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disc, and an optical data storage device, etc.

[0216] Figure 16 is a block diagram of an apparatus 300 for channel estimation according to an exemplary embodiment. For example, the apparatus 300 can be provided as a server. Referring to Figure 16 , the apparatus 300 includes a processing component 322, which further includes one or more processors, and a memory resource represented by a memory 332, for storing instructions, such as an application program, executable by the processing component 322. The application program stored in the memory 332 can include one or more than one module each corresponding to a set of instructions. In addition, the processing component 322 is configured to execute the instructions to perform the above-described methods.

[0217] The apparatus 300 can also include a power supply component 326 configured to perform power management for the apparatus 300, a wired or wireless network interface 350 configured to connect the apparatus 300 to a network, and an input output (I / O) interface 358. The apparatus 300 can operate based on an operating system stored in the memory 332, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, or the like. It should be noted that in some possible embodiments of the present disclosure, the step of selecting different neural network estimation models according to the signal-to-interference ratio SINR is not necessary. The reason is that in practice, the model trained at some specific signal-to-interference ratio points has better estimation performance in all interference environments. In addition, under the premise of computing power, the data sets collected at different signal-to-interference ratio points can be used to train a neural network model, thereby enhancing the generalization ability of the network and making it sufficient to cope with all intensity interference noise within a certain range.

[0218] It should be further understood that "multiple" in the present disclosure refers to two or more, and other quantifiers are similar. "And / or" describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. The character " / " generally represents that the associated objects before and after it are in an "or" relationship. The singular form "a", "said" and "the" are also intended to include the plural form, unless the context clearly indicates otherwise.

[0219] It should be further understood that the terms "first", "second", and the like are used to describe various information, but these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other, and do not represent a specific order or importance. In fact, the expressions "first", "second", and the like can be used interchangeably. For example, the first information can also be referred to as the second information, and similarly, the second information can also be referred to as the first information without departing from the scope of the present disclosure.

[0220] It should be further understood that although the operations are described in a specific order in the accompanying drawings in the embodiments of the present disclosure, it should not be understood as requiring the specific order or serial order shown, or requiring all the shown operations to be performed to obtain the desired results. In a specific environment, multi-tasking and parallel processing can be advantageous.

[0221] Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the features disclosed herein. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

[0222] It should be understood that the present disclosure is not limited to the precise structures as herein described and illustrated in the drawings, and that various modifications and changes can be made without departing from its scope. The scope of the present disclosure is limited only by the claims that follow.

Claims

1. A method of channel estimation, characterized by, The method comprises: determining a channel estimation neural network model corresponding to a signal-to-interference ratio SINR, and determining a first channel response estimation value of a channel; inputting the first channel response estimation value into the channel estimation neural network model to obtain a second channel response estimation value of the channel; determining the second channel response estimation value as an estimation value of the channel; wherein the determination of the channel estimation neural network model corresponding to the channel signal-to-interference ratio SINR comprises: determining a correspondence between SINR intervals and channel estimation neural network models, and determining an SINR interval in which the SINR calculated based on the received channel state information reference signal is located; determining, according to the correspondence, a channel estimation neural network model corresponding to the SINR interval in which the SINR is located.

2. The channel estimation method of claim 1, wherein, The determination of the correspondence between the SINR intervals and the channel estimation neural network models comprises: determining an SINR scale, and determining a first number of SINR intervals based on the scale; collecting channel response estimation value data of each of the first number of SINR intervals to obtain a first number of channel response estimation value training data sets; training a neural network model based on the first number of channel response estimation value training data sets to obtain a first number of channel estimation neural network models.

3. The channel estimation method of claim 2, wherein The training of the neural network model based on the first number of channel response estimation value training data sets to obtain a first number of neural network models comprises: collecting real part training data sets and imaginary part training data sets of each of the first number of SINR intervals; determining a first channel real part training data set based on the real part training data set, and determining a label corresponding to the first channel real part training data set; determining a first channel imaginary part training data set based on the imaginary part training data set, and determining a label corresponding to the first channel imaginary part training data set; combining data of each of the first channel real part training data sets and the corresponding labels, and data of the first channel imaginary part training data sets and the corresponding labels, as input of the neural network model, and training the neural network model to obtain a first number of channel estimation neural network models.

4. The channel estimation method of claim 3, wherein, The neural network model comprises a denoising convolutional neural network DNCNN and a super-resolution convolutional neural network SRCNN. The combination of data of each of the first channel real part training data sets and the corresponding labels, and data of the first channel imaginary part training data sets and the corresponding labels, as input of the neural network model, and the training of the neural network model to obtain a first number of channel estimation neural network models comprise: combining the first channel real part training data sets and the corresponding labels, and the first channel imaginary part training data sets and the corresponding labels to obtain training combination data; inputting the training combination data into the DNCNN and the SRCNN in the neural network model respectively, and training the DNCNN and the SRCNN respectively to obtain a first number of DNCNN models and a first number of SRCNN models after training; After the first quantity of DNCNN models and the first quantity of SRCNN models are correspondingly combined, a first quantity of channel estimation neural network models is obtained.

5. The method for channel estimation according to claim 1, wherein, The determining the first channel response estimation value of the channel comprises: determining a channel response estimation value of a position where a demodulation reference signal (DMRS) is located in the channel; separating a real part and an imaginary part of the channel response estimation value, and performing interpolation processing on the real part and the imaginary part respectively to obtain a first channel real part and a first channel imaginary part; combining the first channel real part and the first channel imaginary part to obtain a complex number, and determining the complex number as the first channel response estimation value.

6. The channel estimation method of claim 1 or 5, wherein The inputting the first channel response estimation value into the channel estimation neural network model to obtain a second channel response estimation value of the channel comprises: inputting a first channel real part and a first channel imaginary part of the first channel response estimation value into the channel estimation neural network model respectively to obtain a second channel real part and a second channel imaginary part; combining the second channel real part and the second channel imaginary part to obtain a complex number, and determining the complex number as the second channel response estimation value.

7. A channel estimation apparatus characterized by comprising: The apparatus comprises: a determining module configured to determine a channel estimation neural network model corresponding to a signal-to-interference ratio (SINR) and determine a first channel response estimation value of a channel; an estimating module configured to input the first channel response estimation value into the channel estimation neural network model to obtain a second channel response estimation value of the channel; wherein the determining module is configured to determine the second channel response estimation value as the estimation value of the channel, determine a corresponding relationship between SINR intervals and channel estimation neural network models, determine an SINR interval in which the SINR is located according to the SINR calculated based on the received channel state information reference signal, and determine a channel estimation neural network model corresponding to the SINR interval in which the SINR is located according to the corresponding relationship.

8. The channel estimation device of claim 7, wherein, The determining module is configured to: determine a SINR scale and determine a first quantity of SINR intervals based on the scale; collect channel response estimation value data of each of the first quantity of SINR intervals to obtain a first quantity of channel response estimation value training data sets; train a neural network model based on the first quantity of channel response estimation value training data sets to obtain a first quantity of channel estimation neural network models.

9. The channel estimation device of claim 8, wherein, The determining module is configured to: collect real part training data sets and imaginary part training data sets of each of the first quantity of SINR intervals; determine a first channel real part training data set based on the real part training data set and determine a label corresponding to the first channel real part training data set; determine a first channel imaginary part training data set based on the imaginary part training data set and determine a label corresponding to the first channel imaginary part training data set; combine the first channel real part training data set and the label corresponding to the first channel real part training data set, and the first channel imaginary part training data set and the label corresponding to the first channel imaginary part training data set, as input of a neural network model, and train the neural network model to obtain a first quantity of channel estimation neural network models.

10. The channel estimation apparatus of claim 9, wherein The neural network model comprises a denoising convolutional neural network (DNCNN) and a super-resolution convolutional neural network (SRCNN); The determining module is configured to: combine the first channel real part training data set and the corresponding label, the first channel imaginary part training data set and the corresponding label to obtain training combined data; input the training combined data into the DNCNN and the SRCNN of the neural network model respectively, and train the DNCNN and the SRCNN respectively to obtain a first number of DNCNN models and a first number of SRCNN models after training; combine the first number of DNCNN models and the first number of SRCNN models to obtain a first number of channel estimation neural network models.

11. The channel estimation device of claim 7, wherein, The determining module is configured to determine a channel response estimation value of a position where a demodulation reference signal (DMRS) is located in a channel. The real part and the imaginary part of the channel response estimation value are separated, and interpolation processing is performed on the real part and the imaginary part respectively to obtain a first channel real part and a first channel imaginary part. The complex number obtained by combining the first channel real part and the first channel imaginary part is determined as the first channel response estimation value.

12. The channel estimation device according to claim 7 or 11, characterized by, The determining module is configured to: input the first channel real part and the first channel imaginary part of the first channel response estimation value into the channel estimation neural network model respectively to obtain a second channel real part and a second channel imaginary part; the complex number obtained by combining the second channel real part and the second channel imaginary part is determined as a second channel response estimation value.

13. A channel estimation apparatus, characterized by comprising: comprise: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to perform the channel estimation method of any one of claims 1-6.

14. A non-transitory computer-readable storage medium, when instructions in the storage medium are executed by a processor of a mobile terminal, enabling the mobile terminal to perform the channel estimation method of any one of claims 1-6.

Images