Adjustment Method, Circuit, Chip, Transceiver and Storage System for Equalization Parameters

Abstract

The present application provides a method, a circuit, a chip, a transceiver and a storage system for adjusting equalization parameters. The method includes: obtaining a transmission signal transmitted by a channel; extracting a plurality of equalization features from the transmission signal; determining a required feature set for each equalizer according to the plurality of equalization features, a plurality of channel features of the channel, and current required feature requirements of an equalization network model of at least one equalizer; predicting a plurality of equalization parameters for each equalizer according to the equalization network model of each equalizer and the corresponding required feature set, and transmitting the plurality of equalization parameters to the corresponding equalizer, so that each equalizer performs signal compensation on the received signal according to the corresponding plurality of equalization parameters.

Description

Technical Field

[0001] The present application relates to the technical field of integrated circuit design, and particularly to a method, a circuit, a chip, a transceiver and a storage system for adjusting equalization parameters. Background Art

[0002] As the rate of transmitted information continues to increase, the transmission channel causes a long tail phenomenon, which increasingly attenuates the high-frequency components of the signal and deteriorates the transmission quality of the signal. At the same time, the influence of noise on the signal transmission quality is also becoming increasingly serious. At present, equalization technology has become one of the effective methods to solve this problem. In order to adapt to characteristics such as channel aging and transmission characteristic changes, an adaptive circuit has become an essential part of equalization technology, which can effectively track channel changes to compensate for high-frequency attenuated signals in real time. However, as the tail of the channel increases, more and more equalization parameters need to be adaptively adjusted by the adaptive circuit. Since the traditional adaptive circuit only supports adjusting one equalization parameter at the same time, multiple equalization parameters can only be adjusted by increasing the complexity of the adaptive circuit. However, this method has high complexity and power consumption, and cannot meet the requirements for power and performance in applications such as high-performance computing processors, high-performance AI (Artificial Intelligence) computing, Internet of Things, and wireless edge. Therefore, how to perform adaptive adjustment of multiple equalization parameters without sacrificing performance has become an urgent problem to be solved. Summary of the Invention

[0003] Embodiments of the present application creatively provide a method, a circuit, a chip, a transceiver and a storage system for adjusting equalization parameters to solve the above technical problems.

[0004] According to a first aspect of the present application, there is provided a method for adjusting equalization parameters, the method including:

[0005] Obtaining a transmission signal transmitted by a channel;

[0006] Extracting a plurality of equalization features from the transmission signal;

[0007] Determining a required feature set for each equalizer according to the plurality of equalization features, a plurality of channel features of the channel, and current required feature requirements of an equalization network model of at least one equalizer;

[0008] Predicting a plurality of equalization parameters for each equalizer according to the equalization network model of each equalizer and the corresponding required feature set, and transmitting the plurality of equalization parameters to the corresponding equalizer, so that each equalizer compensates the received signal according to the corresponding plurality of equalization parameters.

[0009] According to an embodiment of the present application, the method further includes:

[0010] Obtain multiple pulse signals transmitted successively by a channel;

[0011] Obtain multiple channel characteristics of the channel from the multiple pulse signals.

[0012] According to an embodiment of the present application, the at least one equalizer includes a first equalizer and a second equalizer, and the multiple equalization parameters of the second equalizer include multiple first set equalization parameters and multiple second set equalization parameters; correspondingly, predicting the multiple equalization parameters of each equalizer according to the equalization network model of each equalizer and the corresponding required feature set includes:

[0013] Predict the multiple equalization parameters of the first equalizer according to the equalization network model of the first equalizer and the corresponding required feature set;

[0014] Predict the multiple first set equalization parameters of the second equalizer according to the equalization network model of the second equalizer and the corresponding required feature set;

[0015] Predict the multiple second set equalization parameters of the second equalizer according to the adaptive parameter adjustment circuit of the second equalizer.

[0016] According to an embodiment of the present application, each equalizer performs signal compensation on the received signal according to the corresponding multiple equalization parameters, including:

[0017] The first equalizer receives the transmission signal transmitted by the channel and uses the corresponding multiple equalization parameters to perform signal compensation on the transmission signal to obtain a first compensation signal;

[0018] The second equalizer receives the first compensation signal and uses the corresponding multiple equalization parameters to perform signal compensation on the first compensation signal to obtain a second compensation signal.

[0019] According to an embodiment of the present application, the equalization network models of the first equalizer and the second equalizer are trained in the following manner:

[0020] Obtain multiple training pulse signals transmitted successively by the channel, extract multiple training features from the multiple training pulse signals, and store the multiple training features in a mutual information library;

[0021] Obtain the training transmission signal transmitted by the channel, and obtain the training output signal after signal compensation by the first equalizer and the second equalizer;

[0022] Extract multiple training equalization features from the training transmission signal;

[0023] Train the equalization network model of the first equalizer according to the feature optimization mechanism, multiple training features, multiple training equalization features, training output signals, multiple current equalization parameters of the first equalizer, and the required feature requirements of the first equalizer;

[0024] Train the equalization network model of the second equalizer according to the feature optimization mechanism, multiple training features, multiple training equalization features, training output signals, multiple current equalization parameters of the second equalizer, and the required feature requirements of the second equalizer;

[0025] Among them, the equalization network model includes the current required feature requirements of the equalization network model.

[0026] According to an embodiment of the present application, the feature optimization mechanism includes:

[0027] Randomly select a random feature from the multiple training features and multiple training equalization features, and gradually increase the number of random features through a scoreboard to train the equalization network model until the parameter accuracy of the equalization network model is not greater than the set threshold and the number of random features reaches the set condition or the number of random features reaches the maximum;

[0028] Among them, the set condition is that the number of random features is the minimum when the parameter accuracy of the equalization network model is not greater than the set threshold.

[0029] According to an embodiment of the present application, the randomly select a random feature from the multiple training features and multiple training equalization features, and gradually increase the number of random features through a scoreboard to train the equalization network model until the parameter accuracy of the equalization network model is not greater than the set threshold and the number of random features reaches the set condition or the number of random features reaches the maximum includes:

[0030] Randomly select a random feature from the multiple training features and the multiple training equalization features;

[0031] Based on the random feature and the current equalization network model, determine the parameter accuracy of the current equalization network model;

[0032] When the parameter accuracy is not greater than the set threshold, judge whether the random feature is the minimum feature set. If it is the minimum feature set, directly store the random feature in the mutual information library. If it is not the minimum feature set, perform feature pruning and retrain the equalization network model based on the scoreboard;

[0033] When the parameter accuracy is greater than the set threshold, judge whether the random feature is the maximum feature set;

[0034] In the case where it is not the maximum feature set, increase the number of random features, and train the equalization network model based on the scoreboard until the parameter accuracy of the equalization network model is not greater than the set threshold and the number of the random features reaches the set condition or the number of the random features reaches the maximum, and store the current multiple random features in the mutual information library as the current required feature requirements of the equalization network model;

[0035] In the case where the random features are the maximum feature set, stop training the equalization network model, and store the current random features in the mutual information library as the current required feature requirements of the equalization network model.

[0036] According to an embodiment of the present application, the first equalizer is a continuous-time linear equalizer, and the second equalizer is a decision feedback equalizer.

[0037] According to an embodiment of the present application, the multiple training pulse signals include single-bit response and long-pulse response, the training transmission signal is a pseudo-random sequence, the multiple training features include low-frequency gain, high-frequency gain, and high-frequency range, and the multiple training equalization features include eye width and eye height.

[0038] According to an embodiment of the present application, the multiple equalization parameters of the first equalizer include high-frequency gain, zero point, first pole, and second pole, and the multiple equalization parameters of the second equalizer include multiple tap coefficients.

[0039] According to a second aspect of the present application, there is provided an equalization circuit, the equalization circuit comprising:

[0040] A channel, connected to the signal sending end, for receiving a transmission signal from the signal sending end and transmitting it;

[0041] At least one equalizer, connected to the channel;

[0042] An equalizer adjustment device, connected to at least one equalizer, for obtaining the transmission signal transmitted by the channel; extracting multiple equalization features from the transmission signal; determining the required feature set of each equalizer according to the multiple equalization features, the multiple channel features of the channel, and the current required feature requirements of the equalization network model of at least one equalizer; predicting the multiple equalization parameters of each equalizer according to the equalization network model of each equalizer and the corresponding required feature set, and transmitting the multiple equalization parameters to the corresponding equalizer so that each equalizer compensates the received signal according to the corresponding multiple equalization parameters.

[0043] According to an embodiment of the present application, the equalization circuit further comprises:

[0044] An adaptive parameter adjustment circuit, connected to one of the equalizers, for adjusting the corresponding equalization parameters of the connected equalizer.

[0045] According to the third aspect of the present application, an equalization chip is provided, including the above-mentioned equalization circuit.

[0046] According to the fourth aspect of the present application, a transceiver is provided, including the above-mentioned equalization chip.

[0047] According to the fifth aspect of the present application, a storage system is provided, including the above-mentioned transceiver.

[0048] According to the sixth aspect of the present application, an electronic device is provided, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method for adjusting any one of the above-mentioned equalization parameters of the present application.

[0049] According to the seventh aspect of the present application, a computer-readable storage medium storing computer instructions is provided, and the computer instructions are used to cause the computer to execute the method for adjusting any one of the above-mentioned equalization parameters of the present application.

[0050] A method, circuit, chip, transceiver, and storage system for adjusting equalization parameters provided by an embodiment of the present application obtain a transmission signal transmitted through a channel; extract a plurality of equalization features from the transmission signal; determine a required feature set for each equalizer according to the plurality of equalization features, a plurality of channel features of the channel, and current required feature requirements of an equalization network model of at least one equalizer; predict a plurality of equalization parameters for each equalizer according to the equalization network model of each equalizer and the corresponding required feature set, and transmit the plurality of equalization parameters to the corresponding equalizer, so that each equalizer compensates the received signal according to the corresponding plurality of equalization parameters. By adapting the equalizer using a deep learning-based equalization network model, while avoiding an increase in power consumption, simultaneous adjustment of multiple equalization parameters is achieved, thereby improving signal transmission performance.

[0051] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present application, nor is it used to limit the scope of the present application. Other features of the present application will become easily understood through the following description. Description of the Drawings

[0052] By reading the following detailed description with reference to the accompanying drawings, the above and other purposes, features, and advantages of the exemplary embodiments of the present application will become easily understood. In the drawings, several embodiments of the present application are shown in an exemplary rather than restrictive manner, wherein:

[0053] In the drawings, the same or corresponding reference numerals represent the same or corresponding parts.

[0054] Figure 1 Shows a schematic diagram of the composition structure of the equalization circuit provided by the embodiments of the present application;

[0055] Figure 2 Shows a schematic diagram of the implementation process of the method for adjusting equalization parameters provided by the embodiments of the present application;

[0056] Figure 3 Shows a schematic diagram of the implementation process of the equalization parameter prediction operation of the method for adjusting equalization parameters provided by the embodiments of the present application;

[0057] Figure 4 Shows a schematic diagram of the implementation process of the signal compensation operation of the method for adjusting equalization parameters provided by the embodiments of the present application;

[0058] Figure 5 Shows a schematic diagram of the implementation process of the equalization network model training operation of the method for adjusting equalization parameters provided by the embodiments of the present application;

[0059] Figure 6 Shows a schematic diagram of the operation process of the feature optimization mechanism of the method for adjusting equalization parameters provided by the embodiments of the present application;

[0060] Figure 7 Shows a training architecture diagram of a specific application example of the equalization network model training of the method for adjusting equalization parameters provided by the embodiments of the present application;

[0061] Figure 8 Shows a schematic diagram of the composition structure of a specific application example of the equalization circuit provided by the embodiments of the present application;

[0062] Figure 9 Shows a schematic diagram of the composition structure of the electronic device provided by the embodiments of the present application. Detailed implementation manners

[0063] To make the objectives, features, and advantages of the present application more obvious and understandable, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Apparently, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts shall fall within the protection scope of the present application.

[0064] The technical solutions of the present application will be further elaborated in detail below with reference to the accompanying drawings and specific embodiments.

[0065] First, the application scenarios of the embodiments of the present application are described. When a signal is transmitted through a channel, it is often affected by channel characteristics (such as frequency-selective fading, delay spread, multipath effect, etc.), resulting in signal distortion and interference. To overcome these adverse factors and improve the quality and reliability of signal transmission, an equalizer is required to compensate for the signal transmitted through the channel. The working principle of the equalizer is to compensate for the signal based on its own equalization parameters. Since channel characteristics often change with time, frequency, and spatial position, it is usually necessary to adaptively adjust the equalizer, that is, to adaptively adjust the equalization parameters of the equalizer.

[0066] Traditional equalization parameter adjustment is generally based on an adaptive circuit, and the equalizer is made adaptive through a conventional algorithm or method built into the adaptive circuit. For example, the adaptive circuit of CTLE (Continuous Time Linear Equalizer) usually uses methods such as signal slope detection to automatically adjust the equalization coefficient of CTLE, and the adaptive circuit of DFE usually uses the LMS (Learning Management System) algorithm to automatically adjust the equalization coefficient of DFE (Decision Feedback Equalizer). However, traditional adaptive circuits cannot adjust multiple equalization parameters simultaneously. Therefore, to solve the above problems, the embodiments of the present application provide a method, circuit, chip, transceiver, and storage system for adjusting equalization parameters.

[0067] Figure 1 The schematic diagram of the composition structure of the equalization circuit provided by the embodiments of the present application is shown.

[0068] The method for adjusting the equalization parameters of the embodiments of the present application is based on Figure 1 the equalization circuit shown. Referring to Figure 1 , the equalization circuit of the embodiments of the present application includes a channel, at least one equalizer, and an equalizer adjustment device connected to the equalizer.

[0069] Among them, the channel is connected to the signal transmitter, and the channel can receive and transmit the transmission signal of the signal transmitter.

[0070] To compensate for the transmission signal transmitted through the channel, the equalization circuit is configured with at least one equalizer to perform signal compensation on the transmission signal based on the equalizer.

[0071] The equalizer adjustment device is connected to the equalizer and is used to implement the self - adaptation of the equalizer, that is, to adaptively adjust multiple equalization parameters of the equalizer. Among them, the equalizer adjustment device is configured to execute the equalization parameter adjustment method of the embodiments of the present application. That is, the equalizer adjustment device is used to obtain the transmission signal transmitted by the channel; extract multiple equalization features from the transmission signal; determine the required feature set of each equalizer according to the multiple equalization features, the multiple channel features of the channel, and the current required feature requirements of the equalization network model of at least one equalizer; predict multiple equalization parameters of each equalizer according to the equalization network model of each equalizer and the corresponding required feature set, and transmit the multiple equalization parameters to the corresponding equalizer, so that each equalizer compensates the received signal according to the corresponding multiple equalization parameters.

[0072] In an embodiment of the present application, in order to perform more accurate self - adaptation on the equalizer, an adaptive parameter adjustment circuit can also be configured for one of the equalizers to jointly adjust the equalization parameters of the equalizer in combination with the adaptive parameter adjustment circuit and the equalization network model. Among them, the adaptive parameter adjustment circuit is configured inside the equalizer adjustment device and is a part of the equalizer adjustment device.

[0073] Figure 2 The schematic diagram of the implementation process of the equalization parameter adjustment method provided by the embodiments of the present application is shown.

[0074] Reference Figure 2 , the embodiments of the present application provide an equalization parameter adjustment method, which at least includes: Operation 101, obtaining the transmission signal transmitted by the channel; Operation 102, extracting multiple equalization features from the transmission signal; Operation 103, determining the required feature set of each equalizer according to the multiple equalization features, the multiple channel features of the channel, and the current required feature requirements of the equalization network model of at least one equalizer; Operation 104, predicting multiple equalization parameters of each equalizer according to the equalization network model of each equalizer and the corresponding required feature set, and transmitting the multiple equalization parameters to the corresponding equalizer, so that each equalizer compensates the received signal according to the corresponding multiple equalization parameters.

[0075] In Operation 101, the transmission signal transmitted by the channel is obtained.

[0076] After the signal sending end connected to the channel sends the transmission signal, the channel receives the transmission signal and transmits it.

[0077] In Operation 102, multiple equalization features are extracted from the transmission signal.

[0078] After the signal is transmitted through the channel, due to the influence of the channel characteristics, there is a situation of quality degradation. Therefore, it is necessary to analyze and process the transmitted signal to extract multiple equalization features to provide a basis for subsequent signal compensation. Among them, the equalization feature refers to a feature that can reflect the interference and attenuation suffered by the transmitted signal during channel transmission, such as eye height, eye width, etc.

[0079] Among them, the eye height (Eye Height) refers to the size of the opening of the eye diagram on the vertical axis (usually the amplitude axis), which reflects the amplitude stability of the signal during transmission and is closely related to the signal-to-noise ratio. On the eye diagram, the larger the eye height, the more stable the amplitude of the signal and the less affected by noise and interference. The eye width (Eye Width) is the size of the opening of the eye diagram on the horizontal axis (usually the time axis), which reflects the total jitter of the signal, that is, the temporal instability of the signal during transmission. The larger the eye width, the more uniform the time interval of the signal and the smaller the jitter.

[0080] In operation 103, according to multiple equalization features, multiple channel features of the channel, and the current required feature requirements of the equalization network model of at least one equalizer, the required feature set of each equalizer is determined.

[0081] In an embodiment of the present application, multiple channel features of the channel are also extracted in advance, specifically including: obtaining multiple pulse signals transmitted by the channel in sequence; obtaining multiple channel features of the channel from the multiple pulse signals.

[0082] Among them, obtaining multiple pulse signals transmitted by the channel in sequence can be regarded as sending a pulse signal acquisition instruction to the signal transmitter connected to the channel, so that the signal transmitter responds to the pulse signal acquisition instruction and sends a pulse signal to the channel. Among them, when sending the acquisition instruction, the requirements of the pulse signal can be sent synchronously, so that the signal transmitter sends a pulse signal with specific waveform, frequency, and phase characteristics.

[0083] After a series of pulse signals pass through the channel, pulse signals introducing channel distortion and interference characteristics can be obtained. By analyzing and extracting features from these pulse signals, multiple channel features representing the channel characteristics can be obtained, such as low-frequency gain, high-frequency gain, high-frequency range, etc.

[0084] The equalizer adjustment device is trained with an equalization network model for each equalizer. Since the equalization network model is continuously iteratively optimized in real time, the equalization network model also needs to carry the current required feature requirements of the model. Among them, the current required feature requirements of the model refer to the type of input features of the model. For example, the current required feature requirements of the equalization network model can be that the model input is feature A.

[0085] Thus, after obtaining multiple equalization features and multiple channel features, features corresponding to the current required feature requirements can be selected from the multiple equalization features and multiple channel features to obtain the required feature set of the equalizer.

[0086] In an embodiment of the present application, the equalization network model is a deep learning network model, which is pre-trained with the ability to predict the equalization parameters of the equalizer based on the features of the signal.

[0087] In operation 104, according to the equalization network model of each equalizer and the corresponding required feature set, multiple equalization parameters of each equalizer are predicted, and the multiple equalization parameters are transmitted to the corresponding equalizer, so that each equalizer compensates the received signal according to the corresponding multiple equalization parameters.

[0088] Inputting the required feature set of the equalizer into its corresponding equalization network model, multiple equalization parameters of the equalizer are obtained.

[0089] After obtaining multiple equalization parameters, the multiple equalization parameters are transmitted to the corresponding equalizer, so that the equalizer updates its multiple equalization parameters and compensates the signal received by it based on the updated multiple equalization parameters.

[0090] Thus, the solution of the embodiment of the present application adapts the equalizer by using an equalization network model based on deep learning, realizes the simultaneous adjustment of multiple equalization parameters while avoiding power consumption increase, and further improves the signal transmission performance.

[0091] Figure 3 Shows a schematic implementation flowchart of the equalization parameter prediction operation of the equalization parameter adjustment method provided by the embodiment of the present application.

[0092] Refer to Figure 3 , in an embodiment of the present application, at least one equalizer includes a first equalizer and a second equalizer, and the multiple equalization parameters of the second equalizer include multiple first set equalization parameters and multiple second set equalization parameters; correspondingly, according to the equalization network model of each equalizer and the corresponding required feature set, predicting multiple equalization parameters of each equalizer includes: operation 201, predicting multiple equalization parameters of the first equalizer according to the equalization network model of the first equalizer and the corresponding required feature set; operation 202, predicting multiple first set equalization parameters of the second equalizer according to the equalization network model of the second equalizer and the corresponding required feature set; operation 203, predicting multiple second set equalization parameters of the second equalizer according to the adaptive parameter adjustment circuit of the second equalizer.

[0093] In one embodiment of the present application, at least one equalizer is two equalizers, including a first equalizer and a second equalizer. The second equalizer is configured with an adaptive parameter adjustment circuit, and the equalization parameters of the second equalizer are configured to be completed based on the equalization network model and the adaptive parameter adjustment circuit. Among them, the multiple equalization parameters that the second equalizer needs to adjust can be divided into multiple first set equalization parameters and multiple second set equalization parameters.

[0094] In operation 201 , a plurality of equalization parameters of the first equalizer are predicted according to an equalization network model of the first equalizer and a corresponding required feature set.

[0095] The multiple equalization parameters of the first equalizer are configured to all be performed using the equalization network model. Therefore, by inputting the required feature set of the first equalizer into the equalization network model, the corresponding multiple equalization parameters can be obtained.

[0096] In operation 202 , a plurality of first set equalization parameters of the second equalizer are predicted according to an equalization network model of the second equalizer and a corresponding required feature set.

[0097] Specifically, the plurality of first set equalization parameters of the second equalizer may continue to be acquired from the corresponding equalization network model, and the required feature set corresponding to the second equalizer is input into the equalization network model to obtain the plurality of first set equalization parameters.

[0098] In operation 203 , a plurality of second set equalization parameters of the second equalizer are predicted according to the adaptive parameter adjustment circuit of the second equalizer.

[0099] The equalizer adjustment device controls the adaptive parameter adjustment circuit to predict the multiple second set equalization parameters of the second equalizer. The process of the adaptive parameter adjustment circuit predicting the multiple second set equalization parameters can refer to the process of the conventional adaptive circuit adaptively adjusting the equalization parameters of the equalizer, for example, predicting the equalization parameters based on some adaptive algorithms (such as recursive least squares method, Kalman filter, etc.) built into the adaptive parameter adjustment circuit, which will not be described in detail here.

[0100] Therefore, the embodiment of the present application achieves the adaptation of the equalizer by cleverly combining the adaptive algorithm and deep learning, eliminating part of the adaptive circuit, reducing complexity and power consumption, improving the adaptive speed, and realizing simultaneous adaptive adjustment of multiple equalization parameters.

[0101] Figure 4 A schematic diagram of the implementation flow of the signal compensation operation of the equalization parameter adjustment method provided in an embodiment of the present application is shown.

[0102] refer to Figure 4, in an embodiment of the present application, each equalizer compensates the received signal according to a corresponding plurality of equalization parameters, including: Operation 301, the first equalizer receives the transmission signal transmitted by the channel, and compensates the transmission signal using the corresponding plurality of equalization parameters to obtain a first compensated signal; Operation 302, the second equalizer receives the first compensated signal, and compensates the first compensated signal using the corresponding plurality of equalization parameters to obtain a second compensated signal.

[0103] In the case where there are two equalizers, the two equalizers are usually connected in series to jointly achieve signal compensation. Therefore, the signal compensation by each equalizer in the embodiment of the present application can be divided into the compensation of the first equalizer and the compensation of the second equalizer. The first equalizer is defaultly connected to the channel, and the second equalizer is defaultly connected in series after the first equalizer.

[0104] The first equalizer first receives the transmission signal of the channel, and compensates the transmission signal based on the corresponding plurality of equalization parameters to obtain a first compensated signal. Then the first compensated signal flows into the second equalizer, and the second equalizer further compensates the first compensated signal based on the plurality of equalization parameters to obtain a second compensated signal.

[0105] Figure 5 The figure shows a schematic implementation flow diagram of the equalization network model training operation of the equalization parameter adjustment method provided by the embodiment of the present application.

[0106] Refer to Figure 5 , in an embodiment of the present application, the equalization network models of the first equalizer and the second equalizer can be trained through the following operations: Operation 401, obtain a plurality of training pulse signals sequentially transmitted by the channel, extract a plurality of training features from the plurality of training pulse signals, and store the plurality of training features in the mutual information library; Operation 402, obtain the training transmission signal transmitted by the channel, and obtain the training output signal after signal compensation by the first equalizer and the second equalizer; Operation 403, extract a plurality of training equalization features from the training transmission signal; Operation 404, train the equalization network model of the first equalizer according to the feature selection mechanism, the plurality of training features, the plurality of training equalization features, the training output signal, the current plurality of equalization parameters of the first equalizer, and the required feature requirements of the first equalizer; Operation 405, train the equalization network model of the second equalizer according to the feature selection mechanism, the plurality of training features, the plurality of training equalization features, the training output signal, the current plurality of equalization parameters of the second equalizer, and the required feature requirements of the second equalizer; wherein, the equalization network model includes the current required feature requirements of the equalization network model.

[0107] In Operation 401, obtain a plurality of training pulse signals sequentially transmitted by the channel, extract a plurality of training features from the plurality of training pulse signals, and store the plurality of training features in the mutual information library.

[0108] First, obtain an instruction by sending a training pulse signal to the signal sender, causing the signal sender to send multiple training pulse signals to the channel. After the multiple training pulse signals are transmitted through the channel, they have already carried the influence of the channel characteristics. At this time, based on the analysis and feature extraction of the multiple pulse signals, multiple training features representing the channel characteristics can be obtained, such as channel attenuation, low-frequency gain, or high-frequency gain, etc.

[0109] Among them, the pulse signal is the impulse response, mainly used to test the channel.

[0110] After obtaining multiple training features, store the training features in the mutual information library for subsequent model training.

[0111] In operation 402, obtain the training transmission signal transmitted by the channel, and obtain the training output signal after signal compensation by the first equalizer and the second equalizer.

[0112] After obtaining multiple training features of the channel, obtain the actual training transmission signal from the signal sender, and obtain the training output signal obtained after the training transmission signal passes through the equalizer.

[0113] In operation 403, extract multiple training equalization features from the training transmission signal.

[0114] After the training transmission signal is transmitted through the channel, affected by the channel characteristics, the training transmission signal has changed. Obtain multiple training equalization features from the training transmission signal that can represent the influence of the channel characteristics.

[0115] In operation 404, train the equalization network model of the first equalizer according to the feature optimization mechanism, multiple training features, multiple training equalization features, training output signal, multiple current equalization parameters of the first equalizer, and the required feature requirements of the first equalizer.

[0116] The equalization parameters of each equalizer are different. Correspondingly, the required feature requirements of each equalizer are also different. Among them, the required feature requirements of the equalizer can be regarded as features related to the adjustment of the equalization parameters of the current equalizer, and can be configured according to the actual equalization parameters of the equalizer, which will not be elaborated here.

[0117] Use the training features, training equalization features, current multiple equalization parameters, and training output signal together as the training data of the corresponding equalizer to train the equalization network model of the equalizer, and obtain the equalization network model.

[0118] During the training of the equalization network model, in addition to considering the required feature requirements of the equalizer, a feature optimization mechanism is also configured to improve the robustness and accuracy of the equalization network model through the feature optimization mechanism.

[0119] In this embodiment of the present application, the equalization network model of the first equalizer is trained according to the feature optimization mechanism, training features, training equalization features, training output signals, multiple current equalization parameters of the first equalizer, and the required feature requirements of the first equalizer, so that the equalization network model of the first equalizer has the ability to predict its corresponding multiple equalization parameters according to any required feature or any combination of features, and the equalization network model of the first equalizer is obtained.

[0120] In an embodiment of the present application, the model architecture of the equalization network model can be a convolutional neural network, a recurrent neural network, etc.

[0121] In operation 405, the equalization network model of the second equalizer is trained according to the feature optimization mechanism, multiple training features, multiple training equalization features, training output signals, multiple current equalization parameters of the second equalizer, and the required feature requirements of the second equalizer; wherein, the equalization network model includes the current required feature requirements of the equalization network model.

[0122] The equalization network model of the second equalizer is trained through the feature optimization mechanism, multiple training features, multiple training equalization features, training output signals, multiple current equalization parameters of the second equalizer, and the required feature requirements of the second equalizer, so that the equalization network model of the second equalizer has the ability to predict its corresponding multiple equalization parameters according to any required feature or any combination of features.

[0123] Wherein, since the model training is carried out based on the feature optimization mechanism, after the equalization network model is trained, the equalization network model needs to carry its current required feature requirements, that is, the feature type of the input of the trained equalization network model.

[0124] In an embodiment of the present application, the feature optimization mechanism includes: randomly selecting a random feature from multiple training features and multiple training equalization features, and gradually increasing the number of random features through a scoreboard to train the equalization network model until the parameter accuracy of the equalization network model is not greater than the set threshold and the number of random features reaches the set condition or the number of random features reaches the maximum; wherein, the set condition is that the number of random features is the minimum when the parameter accuracy of the equalization network model is not greater than the set threshold.

[0125] The main purpose of the feature optimization mechanism is to ensure the parameter accuracy of the equalization parameters output by the equalization network model while minimizing the number of input features of the equalization network model, so as to reduce the computational complexity and power consumption.

[0126] The feature optimization mechanism can be regarded as randomly selecting random features from multiple training features and multiple training equilibrium features, and gradually increasing the number of random features based on the control of the scoreboard until the parameter accuracies of multiple equilibrium parameters output by the equilibrium network model all reach the set threshold and the number of random features reaches the set condition or the number of random features reaches the maximum. Among them, the set condition is that the number of random features is the minimum when the parameter accuracy of the equilibrium network model is not greater than the set threshold.

[0127] In an embodiment of the present application, a parameter threshold is configured for each equilibrium parameter. The parameter accuracy of each equilibrium parameter is the ratio of the equilibrium parameter to the parameter threshold, and the parameter accuracy is used to determine whether the fluctuation range of the equilibrium parameter is within a certain requirement. The requirement for the fluctuation range of the equilibrium parameter, that is, the set threshold of the parameter accuracy, can be configured according to the actual situation, such as 5%, 6%, etc.

[0128] When the number of random features increases to the maximum, that is, multiple random features are all the features required by the equalizer, and in this case, the random features cannot be increased anymore. Therefore, it is also regarded as the completion of the training of the equilibrium network model.

[0129] Figure 6 The figure shows a schematic diagram of the operation process of the feature optimization mechanism of the equilibrium parameter adjustment method provided by the embodiment of the present application.

[0130] Reference Figure 6, in an embodiment of the present application, a random feature is randomly selected from multiple training features and multiple training balanced features, and the balanced network model is trained by gradually increasing the number of random features through a scoreboard until the parameter accuracy of the balanced network model is not greater than a set threshold and the number of random features reaches a set condition or the number of random features reaches the maximum, including: operation 501, randomly selecting a random feature from multiple training features and multiple training balanced features; operation 502, determining the parameter accuracy of the current balanced network model based on the random feature and the current balanced network model; operation 503, when the parameter accuracy is not greater than the set threshold, determining whether the random feature is the minimum feature set. If it is the minimum feature set, directly store the random feature in the mutual information library. If it is not the minimum feature set, perform feature pruning and retrain the balanced network model based on the scoreboard; operation 504, when the parameter accuracy is greater than the set threshold, determining whether the random feature is the maximum feature set; operation 505, when it is not the maximum feature set, increase the number of random features and train the balanced network model based on the scoreboard until the parameter accuracy of the balanced network model is not greater than the set threshold and the number of random features reaches the set condition or the number of random features reaches the maximum, and store the current multiple random features in the mutual information library as the current required feature requirements of the balanced network model; operation 506, when the random feature is the maximum feature set, stop training the balanced network model and store the current random feature in the mutual information library as the current required feature requirements of the balanced network model.

[0131] In operation 501, a random feature is randomly selected from multiple training features and multiple training balanced features.

[0132] The feature optimization mechanism aims to train the balanced network model by gradually increasing features. First, a random feature is randomly selected from multiple training features and multiple training balanced features as the training input data for training the balanced network model.

[0133] In operation 502, the parameter accuracy of the current balanced network model is determined based on the random feature and the current balanced network model.

[0134] Taking the current random feature as the input of the balanced network model, multiple balanced parameters output by the balanced network model based on the random feature are obtained, and the multiple balanced parameters are calculated with the corresponding parameter thresholds to obtain the parameter accuracy of each balanced parameter. Among them, the parameter accuracy of each balanced parameter constitutes the parameter accuracy of the current balanced network model.

[0135] In operation 503, when the parameter accuracy is not greater than the set threshold, it is determined whether the random features are the minimum feature set. If they are the minimum feature set, the random features are directly stored in the mutual information library. If they are not the minimum feature set, feature pruning is performed, and the balanced network model is retrained based on the scoreboard.

[0136] When the parameter accuracy corresponding to the current random features is not greater than the set threshold, it indicates that the current balanced network model has met the requirements. However, to ensure that the computational complexity of the balanced network model is optimal, it is necessary to determine whether the current random features are the minimum feature set. If so, the random features are saved in the mutual information library and used as the current required feature requirements of the current balanced network model. If they are not the minimum feature set, feature pruning is performed, and the balanced network model is controlled by the scoreboard to be retrained based on the randomly pruned features until the random features are the minimum feature set and the parameter accuracy is not greater than the set threshold. Here, the minimum feature set means that the number of random features reaches the minimum.

[0137] Among them, saving the current random features to the mutual information library can be regarded as storing the feature type of the current random features in the mutual information library, so as to use the feature type of the random features as the current required feature requirements of the balanced network model.

[0138] For example, when applying a kind of random features in the training process of the balanced network model, if the parameter accuracy is not greater than the set threshold, such as 5%, then a judgment on the minimum feature set of the random features is made. If it is the minimum feature set, it is directly stored in the mutual information library; if it is not the minimum feature set, feature pruning is performed, and the balanced network model is retrained based on the scoreboard.

[0139] In operation 504, when the parameter accuracy is greater than the set threshold, it is determined whether the random features are the maximum feature set.

[0140] If the current parameter accuracy is greater than the set threshold, it indicates that the current balanced network model cannot meet the usage requirements. At this time, it is necessary to determine whether the random features reach the maximum feature set. The maximum feature set means that the number of random features reaches the maximum. For example, when the number of features that the balanced network model can use is 5, when the number of random features reaches 5, it is determined that the random features are the maximum feature set.

[0141] For example, if the accuracy is higher than the set threshold, such as 5%, a judgment on the maximum feature set is made.

[0142] In operation 505, when it is not the maximum feature set, the number of random features is increased, and the balanced network model is trained based on the scoreboard until the parameter accuracy of the balanced network model is not greater than the set threshold and the number of random features reaches the set condition or the number of random features reaches the maximum. Then, the current multiple random features are stored in the mutual information library as the current required feature requirements of the balanced network model.

[0143] If the random features are not the maximum feature set, the random features are gradually increased, and the balanced network model is gradually trained based on the scoreboard until the parameter accuracy of the balanced network model is not greater than the set threshold and the number of random features reaches the set condition or the number of random features reaches the maximum.

[0144] After training the balanced network model is completed, the final multiple random features of the current balanced network model are also saved in the mutual information library for subsequent use.

[0145] For example, at this time, if it is not the maximum feature set, the features are extended, that is, the types of random features are increased, and the feature set is optimized until the parameter accuracy is less than or equal to the set threshold, and then the mutual information library is updated.

[0146] In an embodiment of the present application, increasing the number of random features and training the balanced network model based on the scoreboard until the parameter accuracy of the balanced network model is not greater than the set threshold and the number of random features reaches the set condition or the number of random features reaches the maximum can be regarded as:

[0147] During the process of increasing the random features, if the parameter accuracy is controlled within the set threshold, the minimum feature set of the multiple random features is judged, feature pruning is performed, and then, under the control of the scoreboard, the balanced network model is retrained. If, after feature pruning, the parameter accuracy is still within the set threshold under the minimum feature set, the feature type of the current random features is updated to the mutual information library. If the parameter accuracy exceeds the set threshold after pruning the features, the last randomly pruned feature is re-added.

[0148] During the process of increasing the number of random features, if the parameter accuracy exceeds the set threshold, the maximum feature set is judged. If the maximum feature set is reached, all the current random features are stored in the mutual information library. If it is within the maximum feature set, the features are extended, and the added random features are input into the current balanced network model based on the scoreboard, and the balanced network model is controlled to be retrained.

[0149] In operation 506, when the random features are the maximum feature set, the training of the balanced network model is stopped, and the current random features are stored in the mutual information library as the current required feature requirements of the balanced network model.

[0150] If the random features have reached the maximum feature set, stop training the equalization network model and store the current random features in the mutual information library for subsequent use.

[0151] To further illustrate the solution for training the equalization network model based on the feature optimization mechanism in the embodiments of the present application, a specific application example is described below.

[0152] Figure 7 The training architecture diagram of a specific application example of the equalization network model training for the equalization parameter adjustment method provided in the embodiments of the present application is shown.

[0153] Reference Figure 7 , in this specific application example of the embodiments of the present application, the specific process of using the feature optimization mechanism to train the equalization network model includes: preprocessing the data passing through the channel; using the preprocessed data and the output data for model training and optimization; among them, the model training and optimization are performed based on a scoreboard, a mutual information library, and parameter accuracy judgment, and the parameter accuracy judgment is based on a parameter threshold. The data passing through the channel can be regarded as the above-mentioned multiple training pulse signals and training transmission signals, the output data can be regarded as the training output signal, and the preprocessing refers to extracting multiple training features and multiple training equalization features from the multiple training pulse signals and multiple training transmission signals.

[0154] It should be noted that the description of this specific application example of the present application is similar to the above Figure 6 description of the feature optimization mechanism and has similar beneficial effects, so it will not be elaborated here.

[0155] In an embodiment of the present application, the equalization network model is also iteratively optimized in real time through Figures 5 - 7 the model training method described in any of the accompanying drawings to meet the real-time characteristic requirements of the channel. The specific process can refer to Figures 5 - 7 the description of any of the accompanying drawings. For example, in the case of receiving a new transmission signal, iterative optimization of the equalization network model is performed based on the scoreboard, the mutual information library, parameter accuracy judgment, and the new transmission signal.

[0156] In an embodiment of the present application, the first equalizer is a CTLE and the second equalizer is a DFE.

[0157] In an embodiment of the present application, the multiple training pulse signals include SBR (Single Bit Response) and LPR (Long pulse Response), the training transmission signal is PRBS (Pseudo-Random Binary Sequences), the multiple training features include low-frequency gain, high-frequency gain, and high-frequency range, and the multiple training equalization features include eye width and eye height.

[0158] In an embodiment of the present application, the multiple equalization parameters of the first equalizer include high-frequency gain, zero points, first poles, and second poles, and the multiple equalization parameters of the second equalizer include multiple tap coefficients.

[0159] In an embodiment of the present application, the multiple channel characteristics further include signal attenuation. When the proportion of the high-frequency gain predicted by the equalization network model in the channel attenuation exceeds the attenuation threshold, it indicates that the equalization effort of the CTLE is too high and the equalization effort of the DFE is too low, that is, the tap coefficients of the DFE are too low. Conversely, the tap coefficients of the DFE are too high. Therefore, the embodiments of the present application also calculate in real time the proportion of the high-frequency gain in the channel attenuation, and by comparing with the attenuation threshold, determine the strength of the CTLE equalization effort compared to the DFE. Then, the determination result and the corresponding proportion are sent to the equalizer adjustment device to adjust the equalization network models of the CTLE and the DFE through some hyperparameter optimization methods or adaptive learning rate methods, so that the equalization efforts of the CTLE and the DFE reach balance.

[0160] To facilitate the understanding of the solution of the embodiments of the present application, a specific application example is described below for illustration.

[0161] Figure 8 The composition structure diagram of a specific application example of the equalization circuit provided by the embodiments of the present application is shown.

[0162] Refer to Figure 8 , the equalization method of this specific application example of the present application is based on Figure 8 of the equalization circuit. The equalization circuit of this specific application example of the present application includes FFE (Feed Forward Equalizer), channel, CTLE, DFE, and equalizer adjustment device. The equalizer adjustment device is divided into a control unit, a deep learning module, and an adaptive parameter adjustment circuit. The deep learning module is configured to include a first equalization network model of the CTLE and a second equalization network model of the DFE. The control unit is connected to the deep learning module, the adaptive parameter adjustment circuit, and also to the signal sending end. The multiple equalization coefficients of the CTLE include high-frequency gain, zero points, first poles, and second poles, which are G, Wz, Wp1, and Wp2 respectively. The multiple equalization coefficients of the DFE are 18 tap coefficients. This specific application example of the present application includes the following operations:

[0163] S1, the control unit sends acquisition instructions to the signal sending end to sequentially acquire SBR and LPR, so that the signal sending end responds to the acquisition instructions and sequentially sends SBR and LPR to the channel.

[0164] S2, the control unit controls the deep learning module to receive the SBR and LPR transmitted by the channel, and obtains features such as low-frequency gain, high-frequency gain, and high-frequency range from the SBR and LPR, and saves the features to the mutual information library;

[0165] S3, the control unit sends an acquisition instruction for acquiring the PRBS to the signal transmitting end, so that the signal transmitting end sends the PRBS to the channel in response to the acquisition instruction.

[0166] S4, the control unit controls the deep learning module to receive the PRBS transmitted by the channel, and obtains features including eye height and eye width from the PRBS, and trains the first equalizer network model and the second equalizer network model in combination with the output of the DFE. During the training process, the equalizer network model is trained based on the features controlled by the scoreboard and the output of the DFE; wherein, the scoreboard control feature can be regarded as the scoreboard controlling the selection of features in the mutual information library according to the prediction parameter accuracy, that is, expansion or cropping, and the mutual information library is used to store the low-frequency gain, high-frequency gain, high-frequency range and other features obtained from the SBR and LPR and the features controlled by the scoreboard, and the prediction parameter accuracy is calculated based on the parameter threshold of the equalization parameter.

[0167] S5, when a new transmission signal is received, the multi-equalization parameters of the CTLE are predicted using the new transmission signal, the trained first equalization network model, and the required feature requirements of the current equalization network model stored in the current mutual information library, and the predicted equalization parameters are input to the corresponding CTLE; the first 6 tap coefficients of the DFE, i.e., c k , k≤6, and use the adaptive parameter adjustment circuit to predict the last 12 tap coefficients of DFE, that is, c j , 7≤j≤18, and the predicted 18 tap coefficients are input to DFE.

[0168] Therefore, this specific application example of the present application, by cleverly combining adaptive algorithms and deep learning, realizes a CTLE+DFE combined equalization circuit with high adaptability and high equalization efficiency, greatly reducing circuit complexity and power consumption. At the same time, it also realizes real-time update of multi-dimensional equalization parameters, significantly expands the application scenarios, and also improves the high reliability of transmission signals.

[0169] It should be noted that the description of this specific application example of the embodiment of the present application is similar to the description of the above-mentioned method embodiment and the equalization circuit embodiment, and has similar beneficial effects as the method embodiment and the equalization circuit embodiment, so it will not be repeated. Figures 1 - 7 The present invention can be understood by referring to the description of any one of the accompanying drawings.

[0170] Based on the above-mentioned equalization circuit, an embodiment of the present application further provides an equalization chip, including the above-mentioned equalization circuit.

[0171] Based on the above method, an embodiment of the present application further provides a transceiver, including the above-mentioned equalization chip. Among them, the transceiver is specifically a SerDes transceiver for a NRZ (Non-Return to Zero) high-speed link, that is, a Serializer / Deserializer transceiver.

[0172] Based on the above-mentioned equalization circuit, an embodiment of the present application further provides a storage system, including the above-mentioned transceiver.

[0173] The transceiver in the embodiment of the present application is mainly oriented to a storage system, and the storage system may include but is not limited to storage applications and storage devices.

[0174] According to an embodiment of the present application, the present application further provides an electronic device and a readable storage medium, which are built into the equalizer adjustment device and are used to execute the equalization parameter adjustment method of the embodiment of the present application.

[0175] Figure 9 A schematic block diagram of an electronic device that can be used to implement an embodiment of the present application is shown.

[0176] The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present application described and / or claimed herein.

[0177] As Figure 9 shown, the device 800 includes a computing unit 801, which can perform various appropriate actions and processes according to the computer program stored in the read-only memory ROM 802 or the computer program loaded from the storage unit 808 into the random access memory RAM 803. In the RAM 803, various programs and data required for the operation of the device 800 can also be stored. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other through a bus 804. The input / output (I / O) interface 805 is also connected to the bus 804.

[0178] Multiple components in device 800 are connected to I / O interface 805, including: input unit 806, such as a keyboard, mouse, etc.; output unit 807, such as various types of displays, speakers, etc.; storage unit 808, such as a disk, optical disc, etc.; and communication unit 809, such as a network card, modem, wireless communication transceiver, etc. Communication unit 809 allows device 800 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunications networks.

[0179] Computing unit 801 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of computing unit 801 include but are not limited to a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Computing unit 801 executes the various methods and processes described above, such as the method for adjusting equalization parameters. For example, in some embodiments, the method for adjusting equalization parameters can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed onto device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by computing unit 801, one or more steps of the method for adjusting equalization parameters described above can be executed. Alternatively, in other embodiments, computing unit 801 can be configured to execute the method for adjusting equalization parameters in any other suitable manner (e.g., by means of firmware).

[0180] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-chip systems (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special-purpose or general-purpose programmable processor, and can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0181] The program code for implementing the methods of the present application can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing devices, such that when the program codes are executed by the processor or controller, the functions / operations specified in the flowchart and / or block diagram are implemented. The program codes can be executed entirely on the machine, partially on the machine, executed partially on the machine as an independent software package and partially on a remote machine, or executed entirely on a remote machine or server.

[0182] In the context of the present application, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0183] In order to provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and a pointing device (e.g., a mouse or a trackball) through which the user can provide input to the computer. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0184] The systems and techniques described herein can be implemented in a computing system including backend components (e.g., as a data server), or a computing system including middleware components (e.g., an application server), or a computing system including frontend components (e.g., a user computer having a desktop synchronization window or a web browser through which the user can interact with embodiments of the systems and techniques described herein), or a computing system including any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected to each other by digital data communication in any form or medium (e.g., a communication network). Examples of communication networks include: local area network (LAN), wide area network (WAN), and the Internet.

[0185] A computer system can include a client and a server. The client and the server are generally far from each other and typically interact through a communication network. The client-server relationship is created by computer programs running on respective computers and having a client-server relationship with each other. The server can be a cloud server, a server of a distributed system, or a server incorporating a blockchain.

[0186] It should be understood that various forms of the processes shown above can be used, steps can be reordered, added, or deleted. For example, the steps recited in this disclosure can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this application can be achieved. There is no limitation herein.

[0187] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" can explicitly or implicitly include at least one of such features. In the description of this application, "a plurality" means two or more, unless otherwise specifically defined.

[0188] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in this application, and all of them should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. A method for adjusting equalization parameters, characterized in that: The method comprises: Acquire a transmission signal transmitted by a channel; extracting a plurality of equalization features from the transmission signal; determining a required feature set for each equalizer based on the plurality of equalization features, the plurality of channel features of the channels, and current required feature requirements of an equalization network model of at least one equalizer; Predicting multiple equalization parameters of each equalizer according to the equalization network model of each equalizer and the corresponding required feature set, and transmitting the multiple equalization parameters to the corresponding equalizer, so that each equalizer performs signal compensation on the received signal according to the corresponding multiple equalization parameters; The balanced network model is trained in the following way: Acquire multiple training pulse signals transmitted sequentially by the channel, extract multiple training features from the multiple training pulse signals, and store the multiple training features in a mutual information library; Acquire a training transmission signal transmitted by a channel, and acquire a training output signal after being compensated by a first equalizer and a second equalizer signal; extracting a plurality of training equalization features from the training transmission signal; Training the equalization network model of the first equalizer according to the feature optimization mechanism, the plurality of training features, the plurality of training equalization features, the training output signal, the plurality of current equalization parameters of the first equalizer, and the required feature requirements of the first equalizer; Training the equalization network model of the second equalizer according to the feature optimization mechanism, the plurality of training features, the plurality of training equalization features, the training output signal, the plurality of current equalization parameters of the second equalizer, and the required feature requirements of the second equalizer; The balanced network model includes current required characteristic requirements of the balanced network model; The feature selection mechanism includes: Randomly select a random feature from the multiple training features and the multiple training balanced features; Based on the random features and the current balanced network model, determining the parameter accuracy of the current balanced network model; When the parameter accuracy is not greater than the set threshold, determine whether the random feature is a minimum feature set. If it is a minimum feature set, store the random feature directly in the mutual information library. If it is not a minimum feature set, perform feature clipping and retrain the balanced network model based on the scoreboard. When the parameter accuracy is greater than the set threshold, determine whether the random feature is the maximum feature set; In the case where the feature set is not the maximum, the number of random features is increased, and the balanced network model is trained based on the scoreboard until the parameter accuracy of the balanced network model is not greater than the set threshold and the number of random features reaches the set condition or the number of random features reaches the maximum, and the current multiple random features are stored in the mutual information library as the current required feature requirements of the balanced network model; When the random features are the largest feature set, stop training the balanced network model and store the current random features in the mutual information library as the current required feature requirements of the balanced network model.

2. The method according to claim 1, characterized in that The method further comprises: Acquire multiple pulse signals transmitted sequentially by the channel; A plurality of channel characteristics of the channel are obtained from the plurality of pulse signals.

3. The method according to claim 1, characterized in that The at least one equalizer includes a first equalizer and a second equalizer, and the plurality of equalization parameters of the second equalizer include a plurality of first set equalization parameters and a plurality of second set equalization parameters; accordingly, The method predicts multiple equalization parameters of each equalizer according to the equalization network model of each equalizer and the corresponding required feature set, including: Predicting a plurality of equalization parameters of the first equalizer according to the equalization network model of the first equalizer and the corresponding required feature set; predicting a plurality of first set equalization parameters of the second equalizer according to the equalization network model of the second equalizer and the corresponding required feature set; A plurality of second set equalization parameters of the second equalizer are predicted according to the adaptive parameter adjustment circuit of the second equalizer.

4. The method according to claim 3, characterized in that Each equalizer performs signal compensation on the received signal according to the corresponding multiple equalization parameters, including: The first equalizer receives the transmission signal transmitted by the channel, and uses the corresponding multiple equalization parameters to perform signal compensation on the transmission signal to obtain a first compensated signal; The second equalizer receives the first compensation signal, and uses the corresponding multiple equalization parameters to perform signal compensation on the first compensation signal to obtain a second compensation signal.

5. The method according to claim 1, characterized in that The first equalizer is a continuous time linear equalizer, and the second equalizer is a decision feedback equalizer.

6. The method according to claim 5, characterized in that The multiple training pulse signals include single-bit responses and long pulse responses, the training transmission signal is a pseudo-random sequence, the multiple training features include low-frequency gain, high-frequency gain and high-frequency range, and the multiple training equalization features include eye width and eye height.

7. The method according to claim 6, characterized in that The multiple equalization parameters of the first equalizer include a high frequency gain, a zero point, a first pole, and a second pole, and the multiple equalization parameters of the second equalizer include a plurality of tap coefficients.

8. An equalizing circuit, characterized in that: The equalization circuit comprises: A channel, connected to the signal sending end, for receiving a transmission signal from the signal sending end and transmitting it; at least one equalizer connected to the channel; An equalizer adjustment device is connected to at least one equalizer and is used to obtain a transmission signal transmitted by a channel; extract multiple equalization features from the transmission signal; determine a required feature set for each equalizer according to the multiple equalization features, multiple channel features of the channel, and current required feature requirements of an equalization network model of at least one equalizer; predict multiple equalization parameters of each equalizer according to the equalization network model of each equalizer and the corresponding required feature set, and transmit the multiple equalization parameters to the corresponding equalizer, so that each equalizer performs signal compensation on the received signal according to the corresponding multiple equalization parameters; The balanced network model is trained in the following way: Acquire multiple training pulse signals transmitted sequentially by the channel, extract multiple training features from the multiple training pulse signals, and store the multiple training features in a mutual information library; Acquire a training transmission signal transmitted by a channel, and acquire a training output signal after being compensated by a first equalizer and a second equalizer signal; extracting a plurality of training equalization features from the training transmission signal; Training the equalization network model of the first equalizer according to the feature optimization mechanism, the plurality of training features, the plurality of training equalization features, the training output signal, the plurality of current equalization parameters of the first equalizer, and the required feature requirements of the first equalizer; Training the equalization network model of the second equalizer according to the feature optimization mechanism, the plurality of training features, the plurality of training equalization features, the training output signal, the plurality of current equalization parameters of the second equalizer, and the required feature requirements of the second equalizer; The balanced network model includes current required characteristic requirements of the balanced network model; The feature selection mechanism includes: Randomly select a random feature from the multiple training features and the multiple training balanced features; Based on the random features and the current balanced network model, determining the parameter accuracy of the current balanced network model; When the parameter accuracy is not greater than the set threshold, determine whether the random feature is a minimum feature set. If it is a minimum feature set, store the random feature directly in the mutual information library. If it is not a minimum feature set, perform feature clipping and retrain the balanced network model based on the scoreboard. When the parameter accuracy is greater than the set threshold, determine whether the random feature is the maximum feature set; In the case where the feature set is not the maximum, the number of random features is increased, and the balanced network model is trained based on the scoreboard until the parameter accuracy of the balanced network model is not greater than the set threshold and the number of random features reaches the set condition or the number of random features reaches the maximum, and the current multiple random features are stored in the mutual information library as the current required feature requirements of the balanced network model; When the random features are the largest feature set, stop training the balanced network model and store the current random features in the mutual information library as the current required feature requirements of the balanced network model.

9. The equalizing circuit according to claim 8, characterized in that: The equalization circuit further comprises: The adaptive parameter adjustment circuit is connected to one of the equalizers and is used to adjust the corresponding equalization parameters of the connected equalizer.

10. A balancing chip, characterized in that: The equalizing circuit comprises the equalizing circuit described in any one of claims 8 to 9.

11. A transceiver, characterized in that: Includes the balancing chip as claimed in claim 10.

12. A storage system, characterized in that: Comprising the transceiver of claim 11.

Images