Noise reduction method based on transfer learning, terminal device, network device and storage medium

Abstract

Embodiments of the present application provide a denoising method based on transfer learning, a terminal device, a network device and a storage medium, which are used to propose a denoising model based on transfer learning, so that the denoising model in downlink or uplink transmission can adapt to the changing reference signal measurement value in the corresponding link environment, and good denoising effect is achieved. Embodiments of the present application can include: a terminal device acquires a denoising model based on transfer learning; the terminal device performs denoising processing according to the denoising model.

Description

Technical Field

[0001] This application relates to the field of communications, and more particularly to a noise reduction method based on transfer learning, a terminal device, a network device, and a storage medium. Background Technology

[0002] For current wireless communication systems, random channel conditions and noise are the most significant factors affecting transmission performance. Channel estimation can typically be performed by transmitting pilot signals, but the impact of noise cannot be effectively eliminated. Some AI-based noise reduction networks are trained and deployed under specific signal-to-noise ratio scenarios, exhibiting poor generalization performance and thus facing many limitations in practical applications. Summary of the Invention

[0003] This application provides a noise reduction method, terminal device, network device, and storage medium based on transfer learning, which is used to propose a noise reduction model based on transfer training, so that the noise reduction model in downlink or uplink transmission can be adapted to the changing reference signal measurement values ​​in the corresponding link environment, and achieve good noise reduction effect.

[0004] A first aspect of this application provides a noise reduction method based on transfer learning, which may include: a terminal device acquiring a noise reduction model based on transfer learning; and the terminal device performing noise reduction processing according to the noise reduction model.

[0005] A second aspect of this application provides a noise reduction method based on transfer learning, which may include: a network device acquiring a current reference signal measurement value; the network device acquiring a noise reduction model corresponding to the current reference signal measurement value, or a target dataset, based on the current reference signal measurement value and a preset set of reference signal measurement value intervals, wherein the noise reduction model or the target dataset is used for noise reduction processing.

[0006] A third aspect of this invention provides a terminal device with a noise reduction model based on transfer learning, enabling the noise reduction model in data transmission to adapt to varying reference signal measurements in the corresponding link environment, thereby achieving good noise reduction performance. This function can be implemented in hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above-described function.

[0007] A fourth aspect of this invention provides a network device with a noise reduction model based on transfer learning, enabling the noise reduction model in data transmission to adapt to varying reference signal measurements in the corresponding link environment, thereby achieving good noise reduction performance. This function can be implemented in hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above-described function.

[0008] Another aspect of the present invention provides a terminal device, comprising: a memory storing executable program code; a transceiver and a processor coupled to the memory; the processor and the transceiver being used to execute the method described in the first aspect of the present invention.

[0009] Another aspect of the present invention provides a network device, comprising: a memory storing executable program code; a processor coupled to the memory; the processor being configured to execute the method described in the second aspect of the present invention.

[0010] Another aspect of the present invention provides a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the methods described in the first or second aspect of the present invention.

[0011] In another aspect, the present invention provides a chip coupled to a memory in a terminal device, such that the chip, when running, calls program instructions stored in the memory, causing the terminal device to execute the method described in the first or second aspect of the present invention.

[0012] In the technical solution provided in this application embodiment, the terminal device acquires a noise reduction model based on transfer learning; the terminal device performs noise reduction processing according to the noise reduction model. The terminal device proposes a noise reduction model based on transfer learning, so that the noise reduction model in data transmission can be adapted to the changing reference signal measurement values ​​in the corresponding link environment, achieving good noise reduction effect. Attached Figure Description

[0013] Figure 1 This is a schematic diagram illustrating the process of sending information from a source to a receiver in one implementation.

[0014] Figure 2 A schematic diagram of the neuron structure;

[0015] Figure 3 This is a schematic diagram of a neural network;

[0016] Figure 4 A schematic diagram of the basic structure of a convolutional neural network;

[0017] Figure 5 A schematic diagram of the transfer learning process;

[0018] Figure 6A This is a system architecture diagram of the communication system used in the embodiments of the present invention;

[0019] Figure 6B This is a schematic diagram of an embodiment of the noise reduction method based on transfer learning in this application.

[0020] Figure 7 This is a schematic diagram of another embodiment of the noise reduction method based on transfer learning in this application.

[0021] Figure 8A This is a schematic diagram of a wireless communication system receiver containing a noise reduction model in an embodiment of this application;

[0022] Figure 8B This is a schematic diagram of a fully connected noise reduction model in an embodiment of this application;

[0023] Figure 8C This is a schematic diagram illustrating transfer training in an embodiment of this application;

[0024] Figure 8D This is a schematic diagram of another embodiment of the noise reduction method based on transfer learning in this application.

[0025] Figure 9 This is a schematic diagram of another embodiment of the noise reduction method based on transfer learning in this application.

[0026] Figure 10 This is a schematic diagram of another embodiment of the noise reduction method based on transfer learning in this application.

[0027] Figure 11 This is a schematic diagram illustrating the transfer learning and updating of the noise reduction model of the terminal device on the network device side in an embodiment of this application.

[0028] Figure 12 This is a schematic diagram of one embodiment of the terminal device in this application;

[0029] Figure 13 This is a schematic diagram of one embodiment of the network device in this application;

[0030] Figure 14 This is a schematic diagram of another embodiment of the terminal device in this invention;

[0031] Figure 15 This is a schematic diagram of another embodiment of the network device in this application. Detailed Implementation

[0032] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0033] 1. Description of the receiver in a wireless communication system

[0034] In wireless communication systems, the basic workflow involves the transmitter encoding, modulating, and encrypting the source information to form the transmitted information. This transmitted information is then transmitted wirelessly to the receiver, where it is decoded, decrypted, and demodulated to ultimately reconstruct the source information. Figure 1 The diagram shown illustrates the processing steps of sending information from a source to a receiver in one implementation.

[0035] In the aforementioned process, the encoding, modulation, encryption, decoding, demodulation, and decryption operations at the transmitting and receiving ends are controllable. However, the channel conditions and noise in the wireless environment are uncontrollable, complex, and variable. Currently, to better recover source information, it is necessary to estimate the channel environment in the wireless space to match the appropriate algorithm for better information reception at the receiving end. For interference noise in the wireless space, there is a lack of necessary processing solutions, resulting in significant differences in source recovery performance under different signal-to-noise ratios.

[0036] 2. Neural Networks and Deep Learning

[0037] A neural network is a computational model consisting of multiple interconnected neurons. The connections between nodes represent weighted sums from the input signal to the output signal, called weights. Each node performs a weighted summation of different input signals and outputs the result through a specific activation function. For example... Figure 2 The image shown is a schematic diagram of a neuron structure. Figure 3 The diagram shown is a schematic of a neural network. A neural network consists of an input layer, hidden layers, and an output layer. Through different connection methods, weights, and activation functions of multiple neurons, different outputs can be generated, thereby fitting the mapping relationship from input to output.

[0038] Deep learning employs deep neural networks with multiple hidden layers, greatly enhancing the network's ability to learn features and fit complex nonlinear mappings from input to output. Therefore, it has found widespread application in speech and image processing. Besides deep neural networks, deep learning also includes commonly used basic structures such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) for different tasks.

[0039] The basic structure of a convolutional neural network includes: an input layer, multiple convolutional layers, multiple pooling layers, a fully connected layer, and an output layer. For example... Figure 4The diagram shown illustrates the basic structure of a convolutional neural network. Each neuron in the convolutional kernel of a convolutional layer is locally connected to its input. Pooling layers are introduced to extract local maximum or average features, effectively reducing the network's parameters and uncovering local features. This allows the convolutional neural network to converge quickly and achieve excellent performance.

[0040] 3. Transfer learning

[0041] Transfer learning, an important branch of machine learning, leverages the similarity between data, tasks, or models to apply models and knowledge learned in an old domain to a new one. For example... Figure 5 The diagram shown is a schematic representation of the transfer learning process. Figure 5 This illustrates a simple transfer learning process. Taking different datasets and tasks as examples, models A and B, built from dataset A and dataset B respectively, can be fused using transfer learning methods. The fused model is then applied to a new dataset C, thus completing the application on dataset C. Datasets A and B can be called the source domains of transfer learning, and dataset C can be called the target domain. The source domain datasets are usually labeled, while the target domain datasets are usually unlabeled. Therefore, transfer learning, by training on the source domain to obtain an initial model, can then train the source domain model to be suitable for the target domain by adding a loss function that evaluates the similarity between the target and source domains, or by employing adversarial transfer techniques, thereby completing the task in the target domain.

[0042] For current wireless communication systems, random channel conditions and noise are the most significant factors affecting transmission performance. Channel estimation can typically be performed by transmitting pilot signals, but the impact of noise cannot be effectively eliminated. Traditional communication algorithms struggle to process and eliminate noise in received signals; while some methods are beneficial for noise reduction, their algorithmic complexity is extremely high, and their effectiveness is relatively limited. Furthermore, some AI-based noise reduction networks are trained and deployed under specific signal-to-noise ratio (SNR) scenarios, exhibiting poor generalization performance and thus facing many limitations in practical applications. Therefore, designing noise reduction networks with generalization performance for multi-SNR scenarios is of great significance.

[0043] The technical solutions of this application embodiment can be applied to various communication systems, such as: Global System for Mobile Communication (GSM) system, Code Division Multiple Access (CDMA) system, Wideband Code Division Multiple Access (WCDMA) system, General Packet Radio Service (GPRS), Long Term Evolution (LTE) system, Advanced Long Term Evolution (LTE-A) system, New Radio (NR) system, evolution of NR system, LTE-based access to unlicensed spectrum (LTE-U) system, NR-based access to unlicensed spectrum (NR-U) system, Non-Terrestrial Networks (NTN) system, Universal Mobile Telecommunication System (UMTS), Wireless Local Area Networks (WLAN), and Wireless Fidelity (WF). Fidelity (WiFi), 5th-Generation (5G) communication systems, or other communication systems.

[0044] Traditional communication systems typically support a limited number of connections and are easy to implement. However, with the development of communication technology, mobile communication systems will not only support traditional communication but also, for example, device-to-device (D2D) communication, machine-to-machine (M2M) communication, machine-type communication (MTC), vehicle-to-vehicle (V2V) communication, or vehicle-to-everything (V2X) communication. The embodiments of this application can also be applied to these communication systems.

[0045] Optionally, the communication system in this application embodiment can be applied to a carrier aggregation (CA) scenario, a dual connectivity (DC) scenario, or a standalone (SA) network deployment scenario.

[0046] Optionally, the communication system in this application embodiment can be applied to unlicensed spectrum, wherein unlicensed spectrum can also be considered as shared spectrum; or, the communication system in this application embodiment can also be applied to licensed spectrum, wherein licensed spectrum can also be considered as non-shared spectrum.

[0047] This application describes various embodiments in conjunction with network devices and terminal devices. The terminal device may also be referred to as user equipment (UE), access terminal, user unit, user station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, terminal, wireless communication device, user agent, or user device, etc.

[0048] Terminal devices can be stations (STAION, ST) in WLANs, cellular phones, cordless phones, Session Initiation Protocol (SIP) phones, Wireless Local Loop (WLL) stations, Personal Digital Assistant (PDA) devices, handheld devices with wireless communication capabilities, computing devices or other processing devices connected to a wireless modem, in-vehicle devices, wearable devices, terminal devices in next-generation communication systems such as NR networks, or terminal devices in future evolved Public Land Mobile Network (PLMN) networks, etc.

[0049] In the embodiments of this application, the terminal device can be deployed on land, including indoor or outdoor, handheld, wearable or vehicle-mounted; it can also be deployed on water (such as ships); and it can also be deployed in the air (such as airplanes, balloons and satellites).

[0050] In the embodiments of this application, the terminal device may be a mobile phone, a tablet computer, a computer with wireless transceiver capabilities, a virtual reality (VR) terminal device, an augmented reality (AR) terminal device, a wireless terminal device in industrial control, a wireless terminal device in self-driving, a wireless terminal device in remote medical care, a wireless terminal device in a smart grid, a wireless terminal device in transportation safety, a wireless terminal device in a smart city, or a wireless terminal device in a smart home, etc.

[0051] By way of example and not limitation, in this embodiment, the terminal device can also be a wearable device. Wearable devices, also known as wearable smart devices, are a general term for devices that utilize wearable technology to intelligently design and develop everyday wearables, such as glasses, gloves, watches, clothing, and shoes. Wearable devices are portable devices that are worn directly on the body or integrated into the user's clothing or accessories. Wearable devices are not merely hardware devices, but also achieve powerful functions through software support, data interaction, and cloud interaction. Broadly speaking, wearable smart devices include those that are feature-rich, large in size, and can achieve complete or partial functions without relying on a smartphone, such as smartwatches or smart glasses, as well as those that focus on a specific type of application function and require the use of other devices such as smartphones, such as various smart bracelets and smart jewelry for vital sign monitoring.

[0052] In the embodiments of this application, the network device can be a device for communicating with mobile devices. The network device can be an access point (AP) in WLAN, a base station (BTS) in GSM or CDMA, a base station (NodeB, NB) in WCDMA, an evolved Node B (eNB or eNodeB) in LTE, a relay station or access point, or a vehicle-mounted device, wearable device, or a network device (gNB) in an NR network, or a network device in a future evolved PLMN network or an NTN network, etc.

[0053] By way of example and not limitation, in this embodiment, the network device may have mobility characteristics; for example, the network device may be a mobile device. Optionally, the network device may be a satellite or a balloon station. For example, the satellite may be a low Earth orbit (LEO) satellite, a medium Earth orbit (MEO) satellite, a geostationary earth orbit (GEO) satellite, a high elliptical orbit (HEO) satellite, etc. Optionally, the network device may also be a base station located on land, water, or other similar locations.

[0054] In this embodiment, the network device can provide services to a cell. The terminal device communicates with the network device through the transmission resources (e.g., frequency domain resources, or spectrum resources) used by the cell. The cell can be the cell corresponding to the network device (e.g., a base station). The cell can belong to a macro base station or to a base station corresponding to a small cell. The small cell can include: metro cell, micro cell, pico cell, femto cell, etc. These small cells have the characteristics of small coverage area and low transmission power, and are suitable for providing high-speed data transmission services.

[0055] like Figure 6A The diagram shown illustrates the system architecture of the communication system used in this embodiment of the invention. This communication system may include network devices, which can communicate with terminal devices (or communication terminals, terminals). The network devices can provide communication coverage for a specific geographical area and can communicate with terminal devices located within that coverage area. Figure 6A An exemplary embodiment shows one network device and two terminal devices. Optionally, the communication system may include multiple network devices, and each network device may include other numbers of terminal devices within its coverage area. This application embodiment does not limit this. Optionally, the communication system may also include other network entities such as a network controller and a mobility management entity. This application embodiment does not limit this.

[0056] Network equipment can be further divided into access network equipment and core network equipment. That is, the wireless communication system also includes multiple core networks used to communicate with the access network equipment. Access network equipment can be evolved Node Bs (eNBs or e-NodeBs) in Long-Term Evolution (LTE), Next-Generation Radio (NR) (mobile communication system), or Authorized Auxiliary Access Long-Term Evolution (LAA-LTE) systems, such as macro base stations, micro base stations (also called "small base stations"), pico base stations, access points (APs), transmission points (TPs), or new generation Node Bs (gNodeBs).

[0057] It should be understood that devices with communication functions in the network / system of this application embodiment can be referred to as communication devices. Figure 6A Taking the communication system shown as an example, the communication equipment may include network devices and terminal devices with communication functions. The network devices and terminal devices may be the specific devices described in the embodiments of the present invention, which will not be repeated here. The communication equipment may also include other devices in the communication system, such as network controllers, mobility management entities and other network entities, which are not limited in this application embodiment.

[0058] The technical solution of this application will be further described below by way of embodiments. Figure 6B The diagram shown is a schematic representation of an embodiment of a noise reduction method based on transfer learning in this application, which may include:

[0059] 601. The terminal device obtains the noise reduction model based on transfer learning.

[0060] Optionally, the terminal device acquiring the noise reduction model based on transfer learning may include:

[0061] (1) The terminal device trains the model based on the dataset to obtain a noise reduction model based on transfer learning; or,

[0062] (2) The terminal device receives the noise reduction model based on transfer learning issued by the network device.

[0063] Optionally, in implementation method (1), the terminal device trains the model based on the dataset to obtain a noise reduction model based on transfer learning, which may include: the terminal device obtaining the source domain dataset, the labels corresponding to the source domain dataset, and the target domain dataset; the terminal device trains the model based on the source domain dataset, the labels corresponding to the source domain dataset, and the target domain dataset to obtain a noise reduction model.

[0064] Optionally, the terminal device trains a model to obtain a denoising model based on the source domain dataset, the corresponding labels of the source domain dataset, and the target domain dataset. This may include: the terminal device determining a joint loss function based on the source domain dataset, the labels, and the target domain dataset; and the terminal device training a model to obtain a denoising model based on the joint loss function.

[0065] Optionally, the terminal device determines the joint loss function based on the source domain dataset, labels, and target domain dataset. This may include: the terminal device determining an error loss function based on the source domain dataset and labels; that is, optimizing the source domain dataset and labels to achieve convergence of the optimized error loss function; determining an adaptation loss function based on the source domain dataset and target domain dataset; and determining the joint loss function based on the adaptation loss function and the error loss function.

[0066] Optionally, in implementation method (1), the terminal device trains the model based on the dataset to obtain a noise reduction model based on transfer learning, which may include: the terminal device receiving the target dataset or a subset of the target dataset sent by the network device according to the noise reduction model update instruction and the current reference signal measurement value, and training the model based on the target dataset or a subset of the target dataset to obtain the noise reduction model.

[0067] Optionally, in implementation (2), the terminal device obtaining the noise reduction model based on transfer learning may include: the terminal device receiving the noise reduction model sent by the network device according to the noise reduction model update instruction and the current reference signal measurement value.

[0068] In other words, the denoising model obtained by the terminal device can be a denoising model obtained by the terminal device through model training based on the source domain dataset, the corresponding labels of the source domain dataset, and the target domain dataset; it can also be a denoising model received by the terminal device from the network device; it can also be a denoising model obtained by the terminal device through model training based on the target dataset or a subset of the target dataset sent by the network device; or it can be a denoising model obtained through other means, without specific limitations here.

[0069] 602. The terminal device performs noise reduction processing according to the noise reduction model.

[0070] It is understandable that the terminal device performs noise reduction processing on the downlink based on this noise reduction model.

[0071] In the technical solution provided in this application embodiment, the terminal device acquires a noise reduction model based on transfer learning; the terminal device then performs noise reduction processing based on the noise reduction model. The terminal device proposes a noise reduction model based on transfer learning, enabling the noise reduction model in data transmission to adapt to varying reference signal measurements in the corresponding link environment, thus achieving good noise reduction performance.

[0072] like Figure 7 The diagram shown is a schematic representation of another embodiment of the noise reduction method based on transfer learning in this application, which may include:

[0073] Optionally, the terminal device may acquire the noise reduction model based on transfer learning, which may include, but is not limited to, the following steps 701-703, as shown below:

[0074] 701. The terminal device obtains the source domain dataset, the corresponding labels of the source domain dataset, and the target domain dataset.

[0075] Optionally, the terminal device may acquire the source domain dataset, the labels corresponding to the source domain dataset, and the target domain dataset, which may include: when the current reference signal measurement value measured by the terminal device meets the preset conditions, the terminal device acquires the source domain dataset, the labels corresponding to the source domain dataset (also known as label data), and the target domain dataset.

[0076] It is understandable that the current reference signal measurement value measured by the terminal device is the current reference signal measurement value of the downlink measured by the terminal device.

[0077] Optionally, the current reference signal measurement includes at least one of the following: Reference Signal Received Power (RSRP), Reference Signal Received Quality (RSRQ), Received Signal Strength Indicator (RSSI), and Signal-to-Interference plus Noise Ratio (SINR).

[0078] Optionally, the current reference signal measurement value meets a preset condition, which can trigger an update of the noise reduction model. This preset condition may be that the current signal measurement value is greater than a first preset threshold, or that the absolute value of the difference between the current signal measurement value and the signal measurement value adapted to the known noise reduction model is greater than a second preset threshold, or that the absolute value of the ratio between the current signal measurement value and the signal measurement value adapted to the known noise reduction model meets a threshold range, etc.

[0079] Optionally, embodiments of this application can be applied to deep neural networks, recurrent neural networks, or convolutional neural networks, or other neural networks.

[0080] It should be noted that, Figure 7 The illustrated embodiment is one example of a downlink transmission transfer learning-based noise reduction model (also known as a noise reduction network) design method. This embodiment presents a method for terminal devices to design noise reduction models using transfer learning during downlink transmission. Figure 8A The diagram shown is a schematic of a wireless communication system receiver containing a noise reduction model in an embodiment of this application.

[0081] The denoising model takes received information after passing through the channel and noise as input and outputs the denoised received information. The denoising model can be implemented using various methods such as a fully connected deep neural network (DNN), CNN, or RNN; this embodiment does not impose specific limitations. During training, the input is the original received information (source domain dataset), and the label is the noise-free received information. By minimizing the mean square error (MSE) between the output of the denoising model and the label, the network can converge. For existing denoising model dataset designs, both the training and test sets typically use a fixed signal-to-noise ratio (SNR). For example, if the denoising model is trained on an SNR of 10dB, the model will perform well on the test set and during deployment, but when changes in the link environment cause SNR variations, the model will become less adaptable to received signals with different SNRs and will not effectively reduce noise. During offline training, a new target domain dataset (e.g., SNR = 5dB) can be acquired, and the existing denoising model can be retrained on the new dataset. However, in actual downlink transmission, since it is impossible to obtain new labels for the received data, supervised retraining is not possible. Therefore, this embodiment proposes to utilize transfer learning from the source domain to the target domain to adapt the denoising model for changing SNRs during downlink transmission.

[0082] Taking a fully connected DNN or CNN denoising model as an example, such as Figure 8BThe diagram shown is a schematic of a fully connected denoising model in an embodiment of this application. The denoising model, trained on the source domain dataset A and containing N fully connected or convolutional layers, can output denoised received data after the Nth layer during deployment by inputting the received data to be denoised. When the link quality changes, the signal-to-noise ratio changes significantly, or the terminal device moves to a new cell causing the existing denoising model to become incompatible, transfer learning of the existing denoising model on the new dataset is required. When the terminal device continuously receives a set of signals passing through a new link state, denoted as the target domain dataset B, according to... Figure 8C As shown, transfer training is performed. Figure 8C This is a schematic diagram illustrating transfer training in an embodiment of this application. In traditional supervised training, optimizing the MSE loss function on the source domain dataset A enables the denoising model to converge on the source domain dataset A. However, by adding an unlabeled target domain dataset B, and by adding adaptation layers at layers N-1 and N and designing corresponding adaptation loss functions, the target domain dataset B can be adapted to the source domain dataset A. The goal of transfer learning training is to minimize the joint loss function, so that the denoising model can learn noise and channel features on the source domain dataset A, while simultaneously achieving good results on the target domain dataset B after the adaptation layers.

[0083] 702. The terminal device determines the joint loss function based on the source domain dataset, label, and target domain dataset.

[0084] Optionally, the terminal device determines the joint loss function based on the source domain dataset, labels, and target domain dataset. This may include: the terminal device determining an error loss function based on the source domain dataset and labels; that is, optimizing the source domain dataset and labels to achieve convergence of the optimized error loss function; determining an adaptation loss function based on the source domain dataset and target domain dataset; and determining the joint loss function based on the adaptation loss function and the error loss function.

[0085] Optionally, the terminal device determines the joint loss function based on the adaptation loss function and the error loss function, which may include: the terminal device determining the joint loss function according to the first formula;

[0086] The first formula is: L 联合 =L1+λL2; L 联合 Let L1 be the joint loss function, L2 be the adaptation loss function, and λ be the weighting parameter configured by the network device or terminal device. It's understandable that λ can adjust the weighting of L1 and L2. The error loss function can include the mean squared error loss function or other error loss functions; no specific limitations are made here.

[0087] 703. The terminal device trains the model based on the joint loss function to obtain the noise reduction model.

[0088] Optionally, training may end when the number of training iterations reaches a preset number, and / or when the joint loss function of the denoising model obtained from the training reaches a preset value.

[0089] For example, such as Figure 8D The diagram shown is a schematic representation of another embodiment of the noise reduction method based on transfer learning in this application. Figure 8D In the illustration, a base station is used as an example to illustrate the network equipment. The terminal device measures at least one of RSRP, RSRQ, RSSI, and SINR on the downlink data resources distributed by the network device. When a preset condition for triggering a denoising model update is detected, such as when the absolute value of the difference between the measured RSRP and the RSRP of a known denoising model exceeds a certain preset threshold, the terminal device reports a source domain dataset download instruction to the network device. The download of source domain dataset A includes the received dataset to be denoised and its corresponding labels. Optionally, to reduce the number of datasets downloaded and the time cost of transfer training, only a subset of source domain dataset A can be downloaded for adaptation training; simultaneously, the terminal device collects the target domain dataset B to be transferred for training. The terminal device can obtain the joint loss function according to the first formula above, and then train the model based on the joint loss function to obtain the updated denoising model. For example, training is completed when the number of training iterations reaches a preset number, or when the joint loss function corresponding to the denoising model obtained from the model training reaches a preset value. The resulting denoising model is the updated denoising model.

[0090] 704. The terminal device performs noise reduction processing according to the noise reduction model.

[0091] It is understandable that the terminal device performs noise reduction processing on the downlink based on this noise reduction model.

[0092] In this embodiment, the terminal device acquires a source domain dataset, corresponding labels for the source domain dataset, and a target domain dataset. The terminal device determines a joint loss function based on the source domain dataset, labels, and target domain dataset. The terminal device trains a model using the joint loss function to obtain a denoising model. The terminal device then performs denoising processing based on the denoising model. This embodiment proposes a transfer-trained denoising model, enabling the denoising model in downlink transmission to adapt to changing reference signal measurements in the downlink environment, achieving good denoising performance.

[0093] like Figure 9 The diagram shown is a schematic representation of another embodiment of the noise reduction method based on transfer learning in this application, which may include:

[0094] 901. Network devices acquire current reference signal measurement values.

[0095] The network device acquires the current reference signal measurement value, which may include: the current reference signal measurement value obtained by the network device. It can be understood that the current reference signal measurement value measured by the network device is the current reference signal measurement value of the uplink measured by the network device.

[0096] It is understood that this application provides a method for network devices (e.g., base stations) to design noise reduction model groups using transfer learning during uplink transmission. Considering that the downlink noise reduction model is configured on the terminal device side, different terminal devices determine whether to perform transfer learning based on their own specific downlink quality measurements such as RSRP / RSRQ / RSSI / SINR. For uplink transmission, the noise reduction model is configured on the network device side. However, considering that the network device needs to receive uplink transmission data from different terminal devices, and the link conditions of different terminal devices vary greatly, configuring multiple different noise reduction models on the network device is highly complex. Therefore, this application adopts a noise reduction model group configuration method for uplink transmission noise reduction model configuration.

[0097] For example, this embodiment provides a configuration method for pre-grouping noise reduction models in a network device during uplink transmission. The network device configures computational resources for K noise reduction models, where the parameter set of the kth noise reduction model is denoted as θ. k The parameter configuration and training objective of the noise reduction model can be obtained from the second formula:

[0098] The second formula is:

[0099] Network devices measure the uplink reference signal to obtain the current reference signal measurement value, such as the current reference signal-to-noise ratio (SNR). Here, X' represents the noise-reduced received data, X represents the noise-free tag data, and {SNR} k} represents the signal-to-noise ratio (SNR). k The signal-to-noise ratio interval centered at a point with a width of δ can be expressed as: By optimizing the second formula above, the optimal parameter θ can be obtained. k This allows the noise reduction model to operate within the signal-to-noise ratio range {SNR}. k Achieving good noise reduction results. In the pre-saved group of noise reduction models, the signal-to-noise ratio (SNR) range corresponding to each noise reduction model is {SNR}. kThe noise reduction model does not overlap. When the current reference signal SNR of the uplink of a terminal device falls within a certain signal-to-noise ratio (SNR) range, the corresponding noise reduction model is invoked for noise reduction. In the design method of the noise reduction model group, the noise reduction model group can be trained and deployed offline without the need for online updates. However, the noise reduction model is required to have good generalization ability for signals within the SNR range.

[0100] 902. The network device obtains the noise reduction model corresponding to the current reference signal measurement value and / or the target dataset based on the current reference signal measurement value and the preset reference signal measurement value interval set. The noise reduction model or the target dataset is used for noise reduction processing.

[0101] Wherein, if the current reference signal measurement value is the reference signal measurement value of the uplink, the noise reduction model is a noise reduction model about the uplink.

[0102] 1. The current reference signal measurement value belongs to the target reference signal measurement value interval in the preset reference signal measurement value interval set.

[0103] Optionally, the network device obtains the denoising model corresponding to the current reference signal measurement value based on the current reference signal measurement value and a preset set of reference signal measurement value intervals. This may include: if the current reference signal measurement value belongs to a target reference signal measurement value interval in the preset set of reference signal measurement value intervals, the network device obtains the denoising model corresponding to the current reference signal measurement value based on the target reference signal measurement value interval. It is understood that the network device pre-stores denoising models corresponding to different reference signal measurement value intervals and / or a target dataset.

[0104] Optionally, the network device may obtain a noise reduction model corresponding to the current reference signal measurement value based on the target reference signal measurement value range, which may include:

[0105] (1) The network device searches for the target denoising model corresponding to the target reference signal measurement range, and uses it as the denoising model corresponding to the current reference signal measurement value. It is understandable that if the network device pre-stores denoising models corresponding to different reference signal measurement ranges, the network device can directly search for the target denoising model corresponding to the target reference signal measurement range, and use it as the updated denoising model. And / or,

[0106] (2) The network device searches for the target dataset corresponding to the target reference signal measurement range, and trains the model based on the target dataset or a subset of the target dataset to obtain the denoising model corresponding to the current reference signal measurement value. It can be understood that if the network device has pre-stored target datasets corresponding to different reference signal measurement ranges, the network device can train the model based on the target dataset or a subset of the target dataset, and the obtained target denoising model is used as the updated denoising model.

[0107] 2. The current reference signal measurement value does not belong to any reference signal measurement value interval in the preset reference signal measurement value interval set.

[0108] Optionally, the network device may obtain a noise reduction model corresponding to the current reference signal measurement value based on the current reference signal measurement value and a preset set of reference signal measurement value intervals, which may include:

[0109] (1) When the current reference signal measurement value does not belong to any reference signal measurement value interval in the preset reference signal measurement value interval set, the network device finds the target noise reduction model corresponding to the target reference signal measurement value interval closest to the current reference signal measurement value as the noise reduction model corresponding to the current reference signal measurement value. And / or,

[0110] (2) When the current reference signal measurement value does not belong to any reference signal measurement value interval in the preset reference signal measurement value interval set, the network device finds the target dataset corresponding to the target reference signal measurement value interval that is closest to the current reference signal measurement value, and trains the model based on the target dataset or a subset of the target dataset to obtain the noise reduction model corresponding to the current reference signal measurement value.

[0111] Optionally, the network device trains a model based on the target dataset or a subset of the target dataset to obtain a denoising model corresponding to the current reference signal measurement value. This may include: the network device adjusts the target denoising model corresponding to the range of target reference signal measurement values ​​closest to the current reference signal measurement value based on the target dataset or a subset of the target dataset to obtain a denoising model corresponding to the current reference signal measurement value.

[0112] For example, after the network device has configured the noise reduction model group, when the network device measures the current reference signal SNR of the uplink signal of a certain terminal device to be within the signal-to-noise ratio range {SNR}, k At this point, further transfer learning based on data weights can be performed on the uplink denoising model. Specifically, since a large number of labeled offline training datasets are stored at the base station, the base station does not need to use a label-free transfer method for the target domain dataset. The base station can select a subset of the dataset that is closest to the measured uplink SNR, for example, selecting a subset of data whose difference from the measured uplink SNR is no more than 3dB, and then transfer the already converged denoising model θ to the target domain dataset. k Transfer and fine-tuning were performed on this subset of data to make the denoising model more suitable for the SNR represented by this subset of data and to improve the denoising performance.

[0113] 903. Network devices perform noise reduction processing on the uplink based on the noise reduction model.

[0114] In this embodiment, the network device acquires the current reference signal measurement value; based on the current reference signal measurement value and a preset set of reference signal measurement value intervals, the network device acquires a denoising model corresponding to the current reference signal measurement value, and / or a target dataset. The denoising model or target dataset is used for denoising processing. This embodiment applies transfer learning to the denoising model of a wireless communication system, proposing a transfer learning design method for denoising model groups in the uplink network to improve the applicability of the uplink denoising model.

[0115] like Figure 10 The diagram shown is a schematic representation of another embodiment of the noise reduction method based on transfer learning in this application, which may include:

[0116] Optionally, the terminal device may acquire the noise reduction model based on transfer learning, which may include, but is not limited to, the following steps 1001-1003, as shown below:

[0117] 1001. When the current reference signal measurement value measured by the terminal device meets the preset conditions, the terminal device reports the noise reduction model update instruction and the current reference signal measurement value.

[0118] Understandably, the current reference signal measurement value obtained by the terminal device is the current reference signal measurement value of the downlink measured by the terminal device. Optionally, the current reference signal measurement value includes at least one of the following: reference signal received power, reference signal received quality, received signal strength indication, and signal-to-noise ratio.

[0119] Optionally, the preset condition may be that the current signal measurement value is greater than a first preset threshold, or that the absolute value of the difference between the current signal measurement value and the signal measurement value adapted by the known noise reduction model is greater than a second preset threshold, or that the absolute value of the ratio between the current signal measurement value and the signal measurement value adapted by the known noise reduction model meets the threshold range, etc.

[0120] Optionally, embodiments of this application can be applied to deep neural networks, recurrent neural networks, or convolutional neural networks, or other neural networks.

[0121] Optionally, the terminal device reporting the noise reduction model update indication and the current reference signal measurement value may include: the terminal device reporting the noise reduction model update indication and the current reference signal measurement value through an uplink control indication.

[0122] 1002. Network devices acquire current reference signal measurements and noise reduction model update instructions.

[0123] The process of obtaining the current reference signal measurement value and the noise reduction model update indication by the network device may include: the network device receiving the current reference signal measurement value and the noise reduction model update indication reported by the terminal device.

[0124] Optionally, the network device receiving the current reference signal measurement value and noise reduction model update indication reported by the terminal device may include: the network device receiving the current reference signal measurement value and noise reduction model update indication reported by the terminal device through an uplink control instruction.

[0125] 1003. The network device obtains the denoising model corresponding to the current reference signal measurement value and / or the target dataset based on the current reference signal measurement value and the preset reference signal measurement value interval set. The target dataset is used for model training to obtain the denoising model.

[0126] Understandably, the network device can train a model based on the target dataset or a subset thereof to obtain a denoising model, and then send the denoising model to the terminal device. Alternatively, the network device can send the target dataset or a subset thereof to the terminal device, which can then train a model based on the target dataset or a subset thereof to obtain a denoising model.

[0127] It should be noted that the network device obtains the denoising model corresponding to the current reference signal measurement value and / or the target dataset based on the current reference signal measurement value and a preset set of reference signal measurement value intervals. (See reference...) Figure 9 The relevant descriptions of step 902 in the illustrated embodiment will not be repeated here.

[0128] 1004. Terminal devices acquire noise reduction models based on transfer learning.

[0129] Wherein, when the current reference signal measurement value is the downlink reference signal measurement value reported by the terminal device, the noise reduction model is a noise reduction model for the downlink.

[0130] Optionally, the terminal device acquiring the noise reduction model based on transfer learning may include:

[0131] (1) The network device sends the noise reduction model to the terminal device according to the noise reduction model update instruction. The noise reduction model is used by the terminal device for noise reduction processing on the downlink. The terminal device receives the noise reduction model sent by the network device according to the noise reduction model update instruction; or,

[0132] (2) The network device sends the target dataset or a subset of the target dataset to the terminal device according to the noise reduction model update instruction. The target dataset or a subset of the target dataset is used by the terminal device to train the model and obtain the noise reduction model. The terminal device receives the target dataset or a subset of the target dataset sent by the network device according to the noise reduction model update instruction, and trains the model according to the target dataset or a subset of the target dataset to obtain the noise reduction model.

[0133] Optionally, the network device may send the noise reduction model to the terminal device according to the noise reduction model update instruction, which may include: the network device sending the noise reduction model to the terminal device via downlink control instruction according to the noise reduction model update instruction; or,

[0134] The network device may send the target dataset or a subset of the target dataset to the terminal device according to the noise reduction model update instruction. This may include: the network device sending the target dataset or a subset of the target dataset to the terminal device through downlink control instructions according to the noise reduction model update instruction.

[0135] 1005. The terminal device performs noise reduction processing based on the noise reduction model.

[0136] It is understandable that the terminal device performs noise reduction processing on the downlink based on the noise reduction model.

[0137] For example, considering that when the denoising model is complex, the limited computing power of the terminal device makes it difficult to complete the transfer learning of the denoising model in a short time. Therefore, this application embodiment provides a method for designing and downloading the transfer denoising model from the network device side to the terminal device side during downlink transmission. Figure 11 The diagram shown is a schematic of the transfer learning and updating of the noise reduction model of the terminal device on the network device side in an embodiment of this application.

[0138] exist Figure 11 As shown, the terminal device measures at least one of RSRP, RSRQ, RSSI, and SINR on the downlink data resources sent by the network device. When a preset condition that requires triggering a denoising model update is detected, such as when the absolute value of the difference between the measured RSRP and the RSRP adapted to the denoising model exceeds a certain threshold, the terminal device reports the RSRP of the denoising model input signal and a denoising model update indication to the network device. Based on the model update indication and RSRP reported by the terminal device, the network device selects a target dataset or a subset of the target dataset that matches the RSRP for transfer training. The terminal device downloads the updated denoising model. Since the network device stores the labels of the target domain dataset, unlabeled transfer training is not required; only model fine-tuning is needed on the target dataset or a subset of the target dataset that matches the signal-to-noise ratio. For example, the model update indication can be carried by an Uplink Control Indicator (UCI) or other uplink indication signals.

[0139] In this embodiment, when the current reference signal measurement value measured by the terminal device meets preset conditions, the terminal device reports a noise reduction model update instruction and the current reference signal measurement value; the network device obtains the current reference signal measurement value and the noise reduction model update instruction; the network device obtains the noise reduction model corresponding to the current reference signal measurement value based on the current reference signal measurement value and a preset set of reference signal measurement value intervals, and / or a target dataset, which is used for model training to obtain the noise reduction model; the terminal device obtains the noise reduction model based on transfer learning. Specifically, the terminal device receives the noise reduction model sent by the network device according to the noise reduction model update instruction; or, the terminal device receives the target dataset or a subset of the target dataset sent by the network device according to the noise reduction model update instruction, and performs model training based on the target dataset or a subset of the target dataset to obtain the noise reduction model. This embodiment applies transfer learning to the noise reduction model of a wireless communication system, proposes a transfer learning design method for noise reduction model groups for the downlink network, improves the applicability of the downlink noise reduction model, and reduces the computational complexity of the terminal device.

[0140] This application applies transfer learning to the denoising model of a wireless communication system. By proposing a transfer-trained denoising model, the denoising model in both downlink and uplink transmission processes can adapt to varying signal-to-noise ratios in the link environment, achieving better denoising performance. Specifically, an unlabeled transfer learning method is proposed for the downlink denoising model to improve its performance in scenarios with varying signal-to-noise ratios; a transfer learning method for denoising model groups is proposed for the uplink network to improve the applicability of the uplink denoising model; and a transfer training method for the downlink denoising model by network devices is proposed to reduce the computational complexity of terminal devices. This application does not limit the specific implementation method of the denoising model; it mainly protects the design method of using transfer learning to train the downlink and uplink denoising models.

[0141] like Figure 12 The diagram shown is a schematic representation of an embodiment of a terminal device in this application, which may include:

[0142] The acquisition module 1201 is used to acquire a noise reduction model based on transfer learning;

[0143] The processing module 1202 is used to perform noise reduction processing according to the noise reduction model.

[0144] Optionally, processing module 1202 is specifically used to train a model based on the dataset to obtain a denoising model based on transfer learning; or,

[0145] The acquisition module 1201 is specifically used to receive the noise reduction model based on transfer learning sent by the network device.

[0146] Optionally, the acquisition module 1201 is specifically used to acquire the source domain dataset, the labels corresponding to the source domain dataset, and the target domain dataset; the processing module 1202 is specifically used to train the model to obtain a noise reduction model based on the source domain dataset, the labels corresponding to the source domain dataset, and the target domain dataset.

[0147] Optionally, the processing module 1202 is specifically used to determine a joint loss function based on the source domain dataset, the label, and the target domain dataset; and to train a model based on the joint loss function to obtain a denoising model.

[0148] Optionally, the acquisition module 1201 is specifically used to acquire the source domain dataset, the label corresponding to the source domain dataset, and the target domain dataset when the current reference signal measurement value measured by the terminal device meets the preset conditions.

[0149] Optionally, the processing module 1202 is specifically used to determine an error loss function based on the source domain dataset and the label; determine an adaptation loss function based on the source domain dataset and the target domain dataset; and determine a joint loss function based on the adaptation loss function and the error loss function.

[0150] Optionally, the processing module 1202 is specifically used to determine the joint loss function according to the first formula;

[0151] The first formula is: L 联合 =L1+λL2; L 联合 L1 is the joint loss function, L2 is the error loss function, L3 is the adaptation loss function, and λ is the weight parameter configured by the network device or the terminal device.

[0152] Optionally, the processing module 1202 is further configured to end training when the number of times the model is trained reaches a preset number, and / or when the joint loss function corresponding to the denoising model obtained by the model training reaches a preset value.

[0153] Optionally, the acquisition module 1201 is used to report the noise reduction model update indication and the current reference signal measurement value to the network device.

[0154] Optionally, the acquisition module 1201 is specifically used to receive the noise reduction model sent by the network device according to the noise reduction model update instruction and the current reference signal measurement value; or,

[0155] The acquisition module 1201 is used to receive the target dataset or a subset of the target dataset sent by the network device according to the noise reduction model update instruction and the current reference signal measurement value; the processing module 1202 is used to train the model according to the target dataset or a subset of the target dataset to obtain the noise reduction model.

[0156] Optionally, the acquisition module 1201 is specifically used to report a noise reduction model update instruction and the current reference signal measurement value to the network device when the current reference signal measurement value measured by the terminal device meets the preset conditions.

[0157] Optionally, the acquisition module 1201 is specifically used to report the noise reduction model update instruction and the current reference signal measurement value to the network device via the uplink control instruction.

[0158] Optionally, the current reference signal measurement includes at least one of the following: reference signal received power, reference signal received quality, received signal strength indication, and signal-to-noise ratio.

[0159] Optionally, the method can be applied to deep neural networks, recurrent neural networks, or convolutional neural networks.

[0160] like Figure 13 The diagram shown is a schematic representation of an embodiment of a network device in this application, which may include:

[0161] The acquisition module 1301 is used to acquire the current reference signal measurement value;

[0162] Processing module 1302 is used to obtain a denoising model or a target dataset corresponding to the current reference signal measurement value based on the current reference signal measurement value and a preset set of reference signal measurement value intervals. The denoising model or the target dataset is used for denoising processing.

[0163] Optionally, the processing module 1302 is used to obtain the noise reduction model corresponding to the current reference signal measurement value according to the target reference signal measurement value interval when the current reference signal measurement value belongs to the target reference signal measurement value interval in the preset reference signal measurement value interval set.

[0164] Optionally, the processing module 1302 is used to find the target denoising model corresponding to the target reference signal measurement value range, and use it as the denoising model corresponding to the current reference signal measurement value; or,

[0165] Optionally, the processing module 1302 is used to find the target dataset corresponding to the target reference signal measurement value interval, and to perform model training based on the target dataset or a subset of the target dataset to obtain the noise reduction model corresponding to the current reference signal measurement value.

[0166] Optionally, the processing module 1302 is configured to, when the current reference signal measurement value does not belong to any reference signal measurement value interval in the preset set of reference signal measurement value intervals, find the target denoising model corresponding to the target reference signal measurement value interval closest to the current reference signal measurement value as the denoising model corresponding to the current reference signal measurement value; and / or, find the target dataset corresponding to the target reference signal measurement value interval closest to the current reference signal measurement value, and perform model training based on the target dataset or a subset of the target dataset to obtain the denoising model corresponding to the current reference signal measurement value.

[0167] Optionally, the processing module 1302 is used to adjust the target denoising model corresponding to the target reference signal measurement value interval closest to the current reference signal measurement value according to the target dataset or a subset of the target dataset, so as to obtain the denoising model corresponding to the current reference signal measurement value.

[0168] Optionally, if the current reference signal measurement value is the reference signal measurement value of the uplink, the noise reduction model is a noise reduction model with respect to the uplink.

[0169] Optionally, the processing module 1302 is further configured to perform noise reduction processing on the uplink according to the noise reduction model.

[0170] Optionally, if the current reference signal measurement value is the reference signal measurement value of the downlink reported by the terminal device, the noise reduction model is a noise reduction model for the downlink.

[0171] Optionally, the acquisition module 1301 is used to receive the current reference signal measurement value reported by the terminal device.

[0172] Optionally, the acquisition module 1301 is specifically used to receive the current reference signal measurement value reported by the terminal device through an uplink control instruction.

[0173] Optionally, the acquisition module 1301 is also used to receive the noise reduction model update indication reported by the terminal device.

[0174] Optionally, the acquisition module 1301 is specifically used to receive the noise reduction model update instruction reported by the terminal device through the uplink control instruction.

[0175] Optionally, the acquisition module 1301 is specifically used to send the noise reduction model to the terminal device according to the noise reduction model update instruction, wherein the noise reduction model is used by the terminal device to perform noise reduction processing on the downlink; or,

[0176] The acquisition module 1301 is specifically used to send the target dataset or a subset of the target dataset to the terminal device according to the noise reduction model update instruction. The target dataset or a subset of the target dataset is used by the terminal device to train the model and obtain the noise reduction model.

[0177] Optionally, the acquisition module 1301 is specifically used to send the noise reduction model to the terminal device via a downlink control instruction according to the noise reduction model update instruction; or,

[0178] The acquisition module 1301 is specifically used to send the target dataset or a subset of the target dataset to the terminal device through the downlink control instruction according to the noise reduction model update instruction.

[0179] Optionally, the current reference signal measurement includes at least one of the following: reference signal received power, reference signal received quality, received signal strength indication, and signal-to-noise ratio.

[0180] Optionally, the network device is applied to a deep neural network, a recurrent neural network, or a convolutional neural network.

[0181] Corresponding to the method of at least one embodiment applied to a terminal device described above, embodiments of this application also provide one or more terminal devices. The terminal devices of these embodiments can implement any of the above methods. For example... Figure 14 The diagram shown illustrates another embodiment of the terminal device according to the present invention. Taking a mobile phone as an example, the terminal device may include: a radio frequency (RF) circuit 1410, a memory 1420, an input unit 1430, a display unit 1440, a sensor 1450, an audio circuit 1460, a wireless fidelity (WiFi) module 1470, a processor 1480, and a power supply 1490, among other components. The RF circuit 1410 includes a receiver 1414 and a transmitter 1412. Those skilled in the art will understand that... Figure 14 The mobile phone structure shown does not constitute a limitation on the mobile phone and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0182] The following is combined Figure 14 A detailed introduction to each component of a mobile phone:

[0183] RF circuit 1410 can be used for receiving and transmitting signals during information transmission or calls. Specifically, it receives downlink information from the base station and processes it with processor 1480; additionally, it transmits uplink data to the base station. Typically, RF circuit 1410 includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low-noise amplifier (LNA), a duplexer, etc. Furthermore, RF circuit 1410 can also communicate wirelessly with networks and other devices. The aforementioned wireless communication can use any communication standard or protocol, including but not limited to Global System for Mobile Communication (GSM), General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Long Term Evolution (LTE), email, Short Message Service (SMS), etc.

[0184] The memory 1420 can be used to store software programs and modules. The processor 1480 executes various mobile phone functions and data processing by running the software programs and modules stored in the memory 1420. The memory 1420 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 1420 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0185] The input unit 1430 can be used to receive input numerical or character information, and to generate key signal inputs related to user settings and function control of the mobile phone. Specifically, the input unit 1430 may include a touch panel 1431 and other input devices 1432. The touch panel 1431, also known as a touch screen, can collect touch operations performed by the user on or near it (such as operations performed by the user using a finger, stylus, or any suitable object or accessory on or near the touch panel 1431), and drive the corresponding connection devices according to a pre-set program. Optionally, the touch panel 1431 may include two parts: a touch detection device and a touch controller. The touch detection device detects the user's touch position and the signal generated by the touch operation, and transmits the signal to the touch controller; the touch controller receives touch information from the touch detection device, converts it into touch point coordinates, and sends it to the processor 1480, and can also receive and execute commands sent by the processor 1480. In addition, the touch panel 1431 can be implemented using various types such as resistive, capacitive, infrared, and surface acoustic wave. In addition to the touch panel 1431, the input unit 1430 may also include other input devices 1432. Specifically, other input devices 1432 may include, but are not limited to, one or more of the following: physical keyboard, function keys (such as volume control buttons, power buttons, etc.), trackball, mouse, joystick, etc.

[0186] The display unit 1440 can be used to display information input by the user or information provided to the user, as well as various menus of the mobile phone. The display unit 1440 may include a display panel 1441, which may optionally be configured as a liquid crystal display (LCD), organic light-emitting diode (OLED), or similar form. Further, a touch panel 1431 may cover the display panel 1441. When the touch panel 1431 detects a touch operation on or near it, it transmits the information to the processor 1480 to determine the type of touch event. Subsequently, the processor 1480 provides corresponding visual output on the display panel 1441 according to the type of touch event. Although in Figure 14 In this embodiment, the touch panel 1431 and the display panel 1441 are two separate components to realize the input and output functions of the mobile phone. However, in some embodiments, the touch panel 1431 and the display panel 1441 can be integrated to realize the input and output functions of the mobile phone.

[0187] The mobile phone may also include at least one sensor 1450, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor. The ambient light sensor can adjust the brightness of the display panel 1441 according to the ambient light level, and the proximity sensor can turn off the display panel 1441 and / or the backlight when the phone is moved to the ear. As a type of motion sensor, an accelerometer sensor can detect the magnitude of acceleration in various directions (generally three axes). When stationary, it can detect the magnitude and direction of gravity and can be used for applications that recognize the phone's posture (such as landscape / portrait switching, related games, magnetometer posture calibration), vibration recognition-related functions (such as pedometer, taps), etc. Other sensors that may be configured in the mobile phone, such as gyroscopes, barometers, hygrometers, thermometers, and infrared sensors, will not be described in detail here.

[0188] Audio circuit 1460, speaker 1461, and microphone 1462 provide an audio interface between the user and the mobile phone. Audio circuit 1460 converts received audio data into electrical signals and transmits them to speaker 1461, where speaker 1461 converts them into sound signals for output. On the other hand, microphone 1462 converts collected sound signals into electrical signals, which are received by audio circuit 1460, converted into audio data, and then processed by processor 1480 before being transmitted via RF circuit 1410 to, for example, another mobile phone, or the audio data can be output to memory 1420 for further processing.

[0189] WiFi is a short-range wireless transmission technology. Mobile phones, through the WiFi module 1470, can help users send and receive emails, browse web pages, and access streaming media, providing users with wireless broadband internet access. Although Figure 14 WiFi module 1470 is shown, but it is understood that it is not an essential component of a mobile phone and can be omitted as needed without changing the essence of the invention.

[0190] The processor 1480 is the control center of the mobile phone, connecting various parts of the phone through various interfaces and lines. It executes software programs and / or modules stored in the memory 1420, and calls data stored in the memory 1420 to perform various functions and process data, thereby providing overall monitoring of the phone. Optionally, the processor 1480 may include one or more processing units; preferably, the processor 1480 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 1480.

[0191] The mobile phone also includes a power supply 1490 (such as a battery) that supplies power to various components. Preferably, the power supply can be logically connected to the processor 1480 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. Although not shown, the mobile phone may also include a camera, Bluetooth module, etc., which will not be described in detail here.

[0192] In this embodiment of the application, the processor 1480 is used to obtain a noise reduction model based on transfer learning and to perform noise reduction processing according to the noise reduction model.

[0193] Optionally, the model can be trained based on the dataset to obtain a noise reduction model based on transfer learning; or,

[0194] The processor 1480 is specifically used to receive the noise reduction model based on transfer learning sent by the network device.

[0195] Optionally, the processor 1480 is specifically used to acquire a source domain dataset, the labels corresponding to the source domain dataset, and a target domain dataset; and to train a model to obtain a denoising model based on the source domain dataset, the labels corresponding to the source domain dataset, and the target domain dataset.

[0196] Optionally, the processor 1480 is specifically configured to determine a joint loss function based on the source domain dataset, the labels, and the target domain dataset; and to train a model based on the joint loss function to obtain a denoising model.

[0197] Optionally, the processor 1480 is specifically configured to acquire a source domain dataset, a label corresponding to the source domain dataset, and a target domain dataset when the current reference signal measurement value measured by the terminal device meets preset conditions.

[0198] Optionally, the processor 1480 is specifically configured to determine an error loss function based on the source domain dataset and the label; determine an adaptation loss function based on the source domain dataset and the target domain dataset; and determine a joint loss function based on the adaptation loss function and the error loss function.

[0199] Optionally, processor 1480 is specifically used to determine the joint loss function according to the first formula;

[0200] The first formula is: L 联合 =L1+λL2; L 联合 L1 is the joint loss function, L2 is the error loss function, L3 is the adaptation loss function, and λ is the weight parameter configured by the network device or the terminal device.

[0201] Optionally, the processor 1480 is further configured to terminate training when the number of times the model is trained reaches a preset number, and / or when the joint loss function corresponding to the denoising model obtained by the model training reaches a preset value.

[0202] Optionally, RF circuitry 1410 is used to report noise reduction model update indications and current reference signal measurements to the network device.

[0203] Optionally, the RF circuit 1410 is specifically configured to receive a noise reduction model sent by the network device based on the noise reduction model update instruction and the current reference signal measurement value; or,

[0204] RF circuit 1410 is used to receive a target dataset or a subset of the target dataset sent by the network device according to the noise reduction model update instruction and the current reference signal measurement value; processor 1480 is used to train a model according to the target dataset or a subset of the target dataset to obtain the noise reduction model.

[0205] Optionally, the RF circuit 1410 is specifically used to report a noise reduction model update instruction and the current reference signal measurement value to the network device when the current reference signal measurement value measured by the terminal device meets the preset conditions.

[0206] Optionally, the RF circuit 1410 is specifically used to report a noise reduction model update indication and a current reference signal measurement to the network device via an uplink control indication.

[0207] Optionally, the current reference signal measurement includes at least one of the following: reference signal received power, reference signal received quality, received signal strength indication, and signal-to-noise ratio.

[0208] Optionally, the method can be applied to deep neural networks, recurrent neural networks, or convolutional neural networks.

[0209] like Figure 15 The diagram shown is a schematic representation of another embodiment of the network device in this application, which may include:

[0210] Memory 1501 storing executable program code;

[0211] A processor 1502 and a transceiver 1503 are coupled to the memory 1501;

[0212] Processor 1502 is used to acquire the current reference signal measurement value; and based on the current reference signal measurement value and a preset set of reference signal measurement value intervals, acquire the denoising model corresponding to the current reference signal measurement value, or the target dataset, wherein the denoising model or the target dataset is used for denoising processing.

[0213] Optionally, the processor 1502 is configured to, when the current reference signal measurement value belongs to a target reference signal measurement value interval in a preset set of reference signal measurement value intervals, obtain a noise reduction model corresponding to the current reference signal measurement value based on the target reference signal measurement value interval.

[0214] Optionally, the processor 1502 is configured to find the target denoising model corresponding to the target reference signal measurement range, and use it as the denoising model corresponding to the current reference signal measurement value; or,

[0215] Optionally, the processor 1502 is used to find the target dataset corresponding to the target reference signal measurement value interval, and to train the model based on the target dataset or a subset of the target dataset to obtain the noise reduction model corresponding to the current reference signal measurement value.

[0216] Optionally, the processor 1502 is configured to, when the current reference signal measurement value does not belong to any reference signal measurement value interval in the preset set of reference signal measurement value intervals, find the target denoising model corresponding to the target reference signal measurement value interval closest to the current reference signal measurement value as the denoising model corresponding to the current reference signal measurement value; and / or, find the target dataset corresponding to the target reference signal measurement value interval closest to the current reference signal measurement value, and perform model training based on the target dataset or a subset of the target dataset to obtain the denoising model corresponding to the current reference signal measurement value.

[0217] Optionally, the processor 1502 is configured to adjust the target denoising model corresponding to the target reference signal measurement value interval closest to the current reference signal measurement value based on the target dataset or a subset of the target dataset, so as to obtain the denoising model corresponding to the current reference signal measurement value.

[0218] Optionally, if the current reference signal measurement value is the reference signal measurement value of the uplink, the noise reduction model is a noise reduction model with respect to the uplink.

[0219] Optionally, the processor 1502 is also configured to perform noise reduction processing on the uplink according to the noise reduction model.

[0220] Optionally, if the current reference signal measurement value is the reference signal measurement value of the downlink reported by the terminal device, the noise reduction model is a noise reduction model for the downlink.

[0221] Optionally, transceiver 1503 is used to receive the current reference signal measurement value reported by the terminal device.

[0222] Optionally, transceiver 1503 is specifically used to receive current reference signal measurements reported by the terminal device via uplink control instructions.

[0223] Optionally, transceiver 1503 is also used to receive noise reduction model update instructions reported by the terminal device.

[0224] Optionally, transceiver 1503 is specifically used to receive the noise reduction model update instruction reported by the terminal device via the uplink control instruction.

[0225] Optionally, transceiver 1503 is specifically used to send the noise reduction model to the terminal device according to the noise reduction model update instruction, wherein the noise reduction model is used by the terminal device to perform noise reduction processing on the downlink; or,

[0226] Transceiver 1503 is specifically used to send the target dataset or a subset of the target dataset to the terminal device according to the noise reduction model update instruction. The target dataset or a subset of the target dataset is used by the terminal device to train the model and obtain the noise reduction model.

[0227] Optionally, transceiver 1503 is specifically used to send the noise reduction model to the terminal device via downlink control instructions according to the noise reduction model update instruction; or,

[0228] Transceiver 1503 is specifically used to send the target dataset or a subset of the target dataset to the terminal device through the downlink control instruction according to the noise reduction model update instruction.

[0229] Optionally, the current reference signal measurement includes at least one of the following: reference signal received power, reference signal received quality, received signal strength indication, and signal-to-noise ratio.

[0230] Optionally, the network device is applied to a deep neural network, a recurrent neural network, or a convolutional neural network.

[0231] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid state disk (SSD)).

[0232] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

Claims

1. A noise reduction method based on transfer learning, characterized in that, include: The terminal device acquires a noise reduction model based on transfer learning. When the current reference signal measurement value measured by the terminal device meets preset conditions, the terminal device acquires a source domain dataset, labels corresponding to the source domain dataset, and a target domain dataset. The terminal device determines a joint loss function based on the source domain dataset, the labels, and the target domain dataset. The terminal device trains the model using the joint loss function to obtain the noise reduction model. The terminal device performs noise reduction processing according to the noise reduction model. The input of the noise reduction model is the received information after passing through the channel and channel noise, and the output of the noise reduction model is the received information after noise reduction processing.

2. The method according to claim 1, characterized in that, The terminal device determines a joint loss function based on the source domain dataset, the labels, and the target domain dataset, including: The terminal device determines an error loss function based on the source domain dataset and the label; and determines an adaptation loss function based on the source domain dataset and the target domain dataset. The terminal device determines the joint loss function based on the adaptation loss function and the error loss function.

3. The method according to claim 2, characterized in that, The terminal device determines a joint loss function based on the adaptation loss function and the error loss function, including: The terminal device determines the joint loss function according to the first formula; The first formula is: ; Let the joint loss function be... Let be the error loss function. The adaptation loss function, Weight parameters configured for network devices or terminal devices.

4. The method according to any one of claims 1-3, characterized in that, The method further includes: Training ends when the number of training iterations of the model reaches a preset number, and / or when the joint loss function corresponding to the denoising model obtained by training the model reaches a preset value.

5. The method according to claim 1, characterized in that, The current reference signal measurement includes at least one of the following: reference signal received power, reference signal received quality, received signal strength indication, and signal-to-noise ratio.

6. A terminal device, characterized in that, include: Memory containing executable program code; A processor coupled to the memory; The processor is configured to acquire a noise reduction model based on transfer learning, wherein, when the current reference signal measurement value measured by the terminal device meets preset conditions, it acquires a source domain dataset, a label corresponding to the source domain dataset, and a target domain dataset; determines a joint loss function based on the source domain dataset, the label, and the target domain dataset; trains a model based on the joint loss function to obtain a noise reduction model; and performs noise reduction processing based on the noise reduction model, wherein the input of the noise reduction model is the received information after passing through the channel and channel noise, and the output of the noise reduction model is the received information after noise reduction processing.

7. The terminal device according to claim 6, characterized in that, The processor is specifically configured to: determine an error loss function based on the source domain dataset and the label; determine an adaptation loss function based on the source domain dataset and the target domain dataset; and determine a joint loss function based on the adaptation loss function and the error loss function.

8. The terminal device according to claim 7, characterized in that, The processor is specifically used to determine the joint loss function according to the first formula; The first formula is: ; Let the joint loss function be... Let be the error loss function. The adaptation loss function, Weight parameters configured for network devices or terminal devices.

9. The terminal device according to any one of claims 6-8, characterized in that, The processor is further configured to terminate training when the number of training iterations of the model reaches a preset number, and / or when the joint loss function corresponding to the denoising model obtained by training the model reaches a preset value.

10. The terminal device according to claim 6, characterized in that, The current reference signal measurement includes at least one of the following: reference signal received power, reference signal received quality, received signal strength indication, and signal-to-noise ratio.

11. A terminal device, characterized in that, include: The acquisition module is used to acquire a noise reduction model based on transfer learning. Specifically, when the current reference signal measurement value measured by the terminal device meets preset conditions, the terminal device acquires a source domain dataset, labels corresponding to the source domain dataset, and a target domain dataset. The terminal device determines a joint loss function based on the source domain dataset, the labels, and the target domain dataset. The terminal device trains the model based on the joint loss function to obtain the noise reduction model. The processing module is used to perform noise reduction processing according to the noise reduction model, wherein the input of the noise reduction model is the received information after passing through the channel and channel noise, and the output of the noise reduction model is the received information after noise reduction processing.

12. A computer-readable storage medium comprising instructions that, when executed on a processor, cause the processor to perform the method as described in any one of claims 1-5.

Images