MIMO system signal transmission method, device and system based on penalty gradient

Abstract

The invention discloses a penalty gradient-based MIMO (Multiple Input Multiple Output) system signal transmission method, device and system. The method comprises the following steps: acquiring a quantized received signal, a quantization upper limit and a quantization lower limit; performing channel estimation by using pilot frequency to obtain a channel matrix; initializing relevant parameters of a quantization penalty gradient algorithm, wherein a gradient descent strategy in the quantization penalty gradient algorithm is gradient descent, near-end projection and extrapolation interactive calculation; on the basis of the quantization penalty gradient algorithm, continuously utilizing the channel matrix, the quantized receiving signal, the quantization upper limit and the quantization lower limit to update a sending signal estimation vector to obtain a final sending signal estimation vector; and carrying out decoding judgment processing on the sending signal estimation vector to obtain final detection output. According to the method, a special gradient descent strategy is designed to optimize a sending signal estimation vector, so that final decoding judgment output is obtained.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of MIMO signal detection, and particularly relates to a MIMO system signal transmission method, device and system based on a penalized gradient. BACKGROUND

[0002] A large-scale multiple input multiple output (MIMO) system has outstanding advantages such as large capacity, high data transmission rate and superior energy efficiency, and has gradually become a key supporting technology in B5G and even 6G wireless communication. However, in a large-scale MIMO system, a base station usually needs to configure an independent analog-to-digital converter (ADC) for each antenna. With the continuous increase in the number of antennas, the demand for ADCs increases sharply; at the same time, the power consumption of a single ADC increases exponentially with the increase in quantization precision, thereby causing a substantial increase in the overall energy consumption of the system. To address the above problems, researchers have proposed low-precision ADC schemes and corresponding quantized signal detection algorithms. Among them, linear approximation algorithms based on an additive quantization noise model (AQNM), such as Bussgang-based minimum mean square error (BMMSE) and Bussgang-based successive interference cancellation (BSIC) algorithms, always have a large performance loss due to the deviation of the model under high signal-to-noise ratio conditions. Non-linear detection schemes such as near maximum likelihood (nML) and sphere decoding (SD) can improve the detection accuracy, but often come with high computational complexity, which seriously limits their application in actual systems. Therefore, in the application scenario of a large-scale MIMO with multiple quantizations, existing low-precision quantized detection algorithms are difficult to balance detection performance and computational complexity. When it is necessary to simultaneously meet the requirements of low complexity and high accuracy, these traditional low-precision quantized detection schemes are not applicable. SUMMARY

[0003] To solve the above problems, the application provides a MIMO system signal transmission method, device and system based on a penalized gradient, which optimizes a transmitted signal estimation vector by designing a special gradient descent strategy, thereby obtaining a final decoding decision output, and realizes a good compromise between performance and complexity, and is suitable for a signal transmission scheme of a quantized large-scale MIMO system.

[0004] In order to achieve the above technical purposes, achieve the above technical effects, the present application realizes through the following technical solutions:

[0005] In a first aspect, the present application provides a MIMO system signal transmission method based on a quantized gradient, applied to a receiving end, comprising:

[0006] obtaining a quantized received signal , a quantization upper limit and a quantization lower limit ;

[0007] using a pilot to perform channel estimation to obtain a channel matrix ;

[0008] initializing related parameters of a quantized gradient punishment algorithm, wherein the gradient descent strategy in the quantized gradient punishment algorithm is gradient descent, proximal projection and extrapolation interaction calculation;

[0009] based on the quantized gradient punishment algorithm, constantly updating a transmission signal estimation vector using the channel matrix , the quantized received signal , the quantization upper limit and the quantization lower limit to obtain a final transmission signal estimation vector ;

[0010] performing decoding decision processing on the transmission signal estimation vector to obtain a final detection output .

[0011] In combination with the first aspect, optionally, the initializing of the related parameters of the quantized gradient punishment algorithm comprises:

[0012] setting an initial value of the transmission signal estimation vector and an initial value of a search vector to: ;

[0013] calculating a channel calculation matrix , wherein the calculation formula of the channel calculation matrix is:

[0014] ,

[0015] wherein, is a diagonalization operation, denotes extracting the polarity of an input signal ;

[0016] calculating a step size , wherein the step size The calculation formula of the shape coefficient is:

[0017] ,

[0018] In the formula, is the square of the 2-norm of the channel matrix ;

[0019] The value of the shape coefficient and the initial value of the momentum step are set, , ;

[0020] The penalty parameter is calculated, and the calculation formula of the penalty parameter is:

[0021] ,

[0022] In the formula, represents the minimum output value.

[0023] In combination with the first aspect, optionally, based on the quantized penalty gradient algorithm, the channel matrix , the quantized received signal , the quantization upper limit and the quantization lower limit are constantly used to update the transmission signal estimation vector to obtain the final transmission signal estimation vector , including:

[0024] The initial iteration number is set as , and the maximum iteration number is ;

[0025] The following steps are repeatedly executed until :

[0026] Based on the preset target function, gradient descent calculation is performed on the search vector to obtain the forward vector ;

[0027] According to the forward vector and the selected penalty function, a proximal projection operation is performed to obtain the transmission signal estimation vector ;

[0028] According to the previous transmission signal estimation vector , the momentum update is performed on the current transmission signal estimation vector to obtain the extrapolation vector ;

[0029] Based on the transmission signal estimation vector and extrapolation vector , the cost function value of and momentum step ;

[0030] comparison and size.

[0031] In combination with the first aspect, optionally, the acquisition method of the advance vector includes:

[0032] If the number of quantization bits , the gradient of the search vector and the search vector is calculated based on the channel calculation matrix :

[0033] ,

[0034] If the number of quantization bits , the gradient of the search vector , the search vector , the upper limit of quantization and the lower limit of quantization , the gradient of the search vector is calculated, and the calculation formula of the gradient of the search vector is:

[0035] ,

[0036] wherein, indicates the gradient of the search vector , is a Sigmoid function, is a full one vector, is a preset target function;

[0037] Based on the gradient of the search vector , the search vector and the step , gradient descent calculation is performed to obtain the advance vector :

[0038] .

[0039] In combination with the first aspect, optionally, the proximal projection operation is performed according to the advance vector and the selected penalty function to obtain the transmission signal estimation vector , specifically including:

[0040] ​If a sinusoidal penalty function is chosen, then the forward vector is applied based on the chosen sinusoidal penalty function. Each component is projected one by one to obtain the transmitted signal estimation vector. The One element:

[0041] ,

[0042] If a Gaussian penalty function is chosen, then the forward vector is processed based on the chosen Gaussian penalty function. Perform a near-end projection operation to obtain the transmitted signal estimation vector. The One element:

[0043] ,

[0044] In the formula, For constellation charts, It is the transmitted signal estimation vector The One element, Forward vector The One element, For penalty parameters;

[0045] from Calculated to The complete transmitted signal estimation vector is obtained. , This indicates the total number of signals sent.

[0046] In conjunction with the first aspect, optionally, the extrapolation vector The calculation formula is:

[0047] ,

[0048] in, Indicates the first Next iteration momentum step size Indicates the first The estimated vector of the transmitted signal in the next iteration.

[0049] In conjunction with the first aspect, optionally, the step based on the transmitted signal estimation vector and extrapolation vector The cost function value determines the search vector. and momentum step size Specifically, it includes:

[0050] If the cosine penalty function is chosen, the estimated vector of the transmitted signal is calculated according to the formula for calculating the cosine penalty function. and extrapolation vector The corresponding penalty function value is calculated according to the formula:

[0051] ,

[0052] ,

[0053] If the Gaussian penalty function is selected, the sending signal estimation vector and the extrapolation vector are calculated according to the Gaussian penalty function formula.

[0054] ,

[0055] ,

[0056] Based on the channel matrix , the upper limit of quantization and the lower limit of quantization , the sending signal estimation vector and the extrapolation vector are calculated. The corresponding cost function value of the sending signal estimation vector and the extrapolation vector is calculated according to the formula:

[0057] ,

[0058] ,

[0059] Wherein, represents the corresponding penalty function value of the sending signal estimation vector , represents the corresponding penalty function value of the extrapolation vector , represents the cost function value of the sending signal estimation vector , represents the cost function value of the extrapolation vector , is a logarithmic operator, and the base is , and are the first component of the upper limit of quantization and the lower limit of quantization , respectively, is the first row of the channel information matrix ; represents the first element of the sending signal estimation vector ; Represents the extrapolation vector The One element, Indicates the total number of signals sent; The first in the constellation chart Each constellation point, Indicates the penalty parameter. Indicates the shape factor;

[0060] Compare cost function values and ;

[0061] if Then update the search vector. and momentum step size The updated formula is:

[0062] ,

[0063] ,

[0064] if Then update the search vector. and momentum step size The updated formula is:

[0065] ,

[0066] ,

[0067] in, Indicates the first Next iteration momentum step size This indicates that the minimum value should be output.

[0068] In conjunction with the first aspect, optionally, the detection output The calculation formula is:

[0069] ,

[0070] In the formula, This indicates rounding to the nearest integer. constellation points, Let represent the set of transmitted signals that conform to a constellation point mapping, where the dimension of each transmitted signal in the set is . .

[0071] Secondly, the present invention provides a signal transmission device for a MIMO system based on a penalized gradient, applied at a receiving end, comprising:

[0072] The first data acquisition module is used to acquire the quantized received signal. and quantification upper limit and a lower limit of quantization ;

[0073] a second data obtaining module, configured to obtain a channel matrix by performing channel estimation with a pilot ;

[0074] an initialization module, configured to initialize parameters of a quantized penalized gradient algorithm, wherein a gradient descent strategy in the quantized penalized gradient algorithm is gradient descent, proximal projection and extrapolation interaction calculation

[0075] an estimation module, configured to update a transmission signal estimation vector based on the quantized penalized gradient algorithm, by constantly utilizing a channel matrix , a quantized received signal , an upper limit of quantization and a lower limit of quantization to obtain a final transmission signal estimation vector ;

[0076] a decoding decision module, configured to perform decoding decision processing on the transmission signal estimation vector to obtain a final detection output .

[0077] In a third aspect, the present application provides a penalized gradient-based MIMO system signal transmission system, applied to a receiving end, comprising a storage medium and a processor

[0078] The storage medium is configured to store instructions

[0079] The processor is configured to operate according to the instructions to execute the method according to any one of the first aspect.

[0080] Compared with the prior art, the present application has the following beneficial effects:

[0081] The present application proposes a penalized gradient-based MIMO system signal transmission method, device and system, which optimizes the transmission signal estimation vector by designing a special gradient descent strategy, thereby obtaining a final decoding decision, realizes a good compromise between performance and complexity, and is suitable for a signal transmission scheme of a quantized large-scale MIMO system.

[0082] In addition, most existing low-precision quantization detection schemes have certain limitations on actual application scenarios in order to ensure convergence, for example, requiring the number of base station receiving antennas to be much larger than the number of transmitting antennas, i.e. In contrast, the quantized penalized gradient algorithm proposed in the present application has an exponential convergence speed under the condition that the step size parameter satisfies , which can adapt to more diverse network environments and significantly improve the applicability of the algorithm. BRIEF DESCRIPTION OF DRAWINGS

[0083] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed in the embodiments will be briefly introduced as follows. Obviously, the accompanying drawings in the following description are only some embodiments of the present application, and all other drawings obtained by those of ordinary skill in the art based on the accompanying drawings without creative effort should be within the protection scope of the present application.

[0084] Figure 1 is a large-scale MIMO system scene schematic diagram of the MIMO system signal transmission method in an embodiment of the present application;

[0085] Figure 2 is an uplink system structure diagram of the MIMO system signal transmission method in an embodiment of the present application;

[0086] Figure 3 is a flowchart of the MIMO system signal transmission method in an embodiment of the present application;

[0087] Figure 4 is a bit error rate comparison diagram between the MIMO system signal transmission method in an embodiment of the present application and different low-precision quantization transmission methods;

[0088] Figure 5 is a bit error rate comparison diagram of the MIMO system signal transmission method in an embodiment of the present application under different initialization iteration numbers;

[0089] Figure 6 is a bit error rate comparison diagram of the MIMO system signal transmission method in an embodiment of the present application under imperfect channel. DETAILED DESCRIPTION

[0090] The technical solutions in the embodiments of the present application will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should be within the protection scope of the present application.

[0091] In addition, if the description of "first", "second" and the like is involved in the embodiments of the present application, the description of "first", "second" and the like is only for the purpose of description, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the features limited by "first", "second" can be explicitly or implicitly included at least one of the features. In addition, the technical solutions of various embodiments can be combined with each other, but it must be based on the realization of ordinary skilled in the art, when the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, nor within the protection scope required by the present application.

[0092] The present application is further described below in conjunction with the accompanying drawings and specific embodiments of the present application.

[0093] Embodiment 1

[0094] The present application provides a kind of MIMO system signal transmission method based on penalty gradient, it is applied to receiving end, including:

[0095] obtain quantized received signal And upper limit of quantization And lower limit of quantization ;

[0096] Channel estimation is carried out using pilot, and channel matrix Is obtained ;

[0097] The related parameters of quantization penalty gradient algorithm are initialized, and the gradient descent strategy in the quantization penalty gradient algorithm is gradient descent, proximal projection and extrapolation interactive calculation

[0098] Based on the quantization penalty gradient algorithm, the channel matrix Quantized received signal Upper limit of quantization And lower limit of quantization Are used to update sending signal estimation vector constantly, and final sending signal estimation vector Is obtained ;

[0099] The sending signal estimation vector Is subjected to decoding decision processing, and final detection output Is obtained .

[0100] In combination with the first aspect, optionally, the initialization of the related parameters of quantization penalty gradient algorithm includes:

[0101] The initial value of sending signal estimation vector The initial value of search vector Is set to: ;

[0102] Calculate the channel computation matrix The channel calculation matrix The calculation formula is:

[0103] ,

[0104] In the formula, For diagonalization operation, Indicates the extraction of input signal polarity;

[0105] Calculate step size The step size The calculation formula is:

[0106] ,

[0107] In the formula, Channel matrix The square of the 2-norm;

[0108] Set shape factor The value of , and the initial value of the momentum step size. The value of ;

[0109] Calculate penalty parameters The penalty parameter The calculation formula is:

[0110] ,

[0111] In the formula, This indicates that the minimum value should be output.

[0112] In one specific embodiment of the present invention, , .

[0113] In one specific embodiment of the present invention, the method of continuously utilizing the channel matrix based on the quantization penalty gradient algorithm is described. Quantized received signal Quantification upper limit Quantization lower bound The transmitted signal estimation vector is updated to obtain the final transmitted signal estimation vector. ,include:

[0114] Set the initial number of iterations The maximum number of iterations is ;

[0115] Repeat the following steps until... :

[0116] Based on the preset objective function, the search vector Perform gradient descent calculations to obtain the forward vector. ;

[0117] Based on the forward vector Using the selected penalty function, a near-end projection operation is performed to obtain the transmitted signal estimation vector. ;

[0118] Based on the previously transmitted signal estimation vector Estimate the vector of the currently transmitted signal. Perform momentum update to obtain the extrapolation vector. ;

[0119] Based on the transmitted signal estimation vector and extrapolation vector The cost function value determines the search vector. and momentum step size ;

[0120] Compare and Size.

[0121] In one specific embodiment of the present invention, the forward vector The methods for obtaining it include:

[0122] If quantization bits Then, the matrix is ​​calculated based on the channel. and search vector For the search vector The gradient is calculated as follows:

[0123] ,

[0124] In the formula, For the Sigmoid function;

[0125] If quantization bits Based on the channel matrix Search vector Quantification upper limit Quantization lower bound Calculate the search vector The gradient of the search vector The formula for calculating the gradient is:

[0126] ,

[0127] in, Represents the search vector gradient, It is a vector of all 1s. The target function is set in advance.

[0128] Based on search vector gradient Search vector and step length Gradient descent is performed to obtain the forward vector. :

[0129] .

[0130] In one specific embodiment of the present invention, the step of using the forward vector... Using the selected penalty function, a near-end projection operation is performed to obtain the transmitted signal estimation vector. Specifically, it includes:

[0131] If a sinusoidal penalty function is chosen, then the forward vector is applied based on the chosen sinusoidal penalty function. Each component is projected one by one to obtain the transmitted signal estimation vector. The One element:

[0132] ,

[0133] If a Gaussian penalty function is chosen, then the forward vector is processed based on the chosen Gaussian penalty function. Perform a near-end projection operation to obtain the transmitted signal estimation vector. The One element:

[0134] ,

[0135] In the formula, For constellation charts, It is the transmitted signal estimation vector The One element, Forward vector The One element;

[0136] from Calculated to The complete transmitted signal estimation vector is obtained. , This indicates the total number of signals sent.

[0137] In one specific embodiment of the present invention, the extrapolation vector The calculation formula is:

[0138] ,

[0139] in, Indicates the first Next iteration momentum step size Indicates the first The estimated vector of the transmitted signal in the next iteration.

[0140] In one specific embodiment of the present invention, the step of estimating the transmitted signal vector... and extrapolation vector The cost function value determines the search vector. and momentum step size Specifically, it includes:

[0141] If the cosine penalty function is chosen, the estimated vector of the transmitted signal is calculated according to the formula for calculating the cosine penalty function. and extrapolation vector The corresponding penalty function value, the formula for calculating the penalty function value is:

[0142] ,

[0143] ,

[0144] If the Gaussian penalty function is chosen, the estimated vector of the transmitted signal is calculated according to the formula for calculating the Gaussian penalty function. and extrapolation vector The corresponding penalty function value, the formula for calculating the penalty function value is:

[0145] ,

[0146] ,

[0147] Based on channel matrix Quantification upper limit Quantization lower bound Calculate the estimated vector of the transmitted signal. and extrapolation vector The corresponding cost function value, the transmitted signal estimation vector and extrapolation vector The formula for calculating the corresponding cost function value is:

[0148] ,

[0149] ,

[0150] in, Represents the transmitted signal estimation vector The corresponding penalty function value, Represents the extrapolation vector The corresponding penalty function value, Represents the transmitted signal estimation vector The cost function value, Represents the extrapolation vector The cost function value, It is a logarithmic operator with base 1. , and Quantization upper limit Quantization lower bound The One portion, Channel information matrix The OK; Represents the transmitted signal estimation vector The One element; Represents the extrapolation vector The One element, Indicates the total number of signals sent; The first in the constellation chart Each constellation point, Indicates the penalty parameter. Indicates the shape factor;

[0151] Compare cost function values and ;

[0152] if Then update the search vector. and momentum step size The updated formula is:

[0153] ,

[0154] ,

[0155] if Then update the search vector. and momentum step size The updated formula is:

[0156] ,

[0157] ,

[0158] in, Indicates the first Momentum step size of the next iteration Indicates the first The search vector for the next iteration. Indicates the first The momentum step size of the next iteration.

[0159] In one specific embodiment of the present invention, the detection output The calculation formula is:

[0160] ,

[0161] In the formula, This indicates rounding to the nearest integer. constellation points, Let represent the set of transmitted signals that conform to a constellation point mapping, where the dimension of each transmitted signal in the set is . .

[0162] The signal transmission method of the MIMO system in this embodiment of the invention will be described in detail below with reference to a specific implementation method.

[0163] like Figure 1 As shown, a real-number MIMO system includes root transmitting antenna and The receiving antenna receives the signal. The transmitted signal sent by the transmitting antenna The relationship between them is shown in the following formula:

[0164] ,

[0165] In the formula, Represents the channel matrix, with dimension . ; This represents Gaussian noise interference with dimension . , follows the mean The variance is The complex Gaussian distribution; This represents the transmission signal sent by the user side, with the dimension being... Each element belongs to the constellation point set of M-QAM modulation. Its set of decoding candidate points is { }; This represents the received signal at the base station, with dimensions of... .

[0166] like Figure 2 As shown, the base station uses an ADC with an accuracy less than a set threshold (i.e., a low-precision ADC) to process the received signal. Quantization processing is performed. Specifically, the base station is configured with... Each of the three receiving antennas is equipped with an independent RF link and a low-precision ADC. The uplink signal acquired by the receiving antenna is first processed by the RF processing module, and then quantized by the low-precision ADC module. The quantized received signal... The mathematical expression is:

[0167] ,

[0168] In the formula, express Bit-precision quantization functions are used to quantize the received signal. The continuous amplitude values ​​are mapped to a finite number of discrete levels, thus realizing the digitization of the amplitude dimension; This represents the quantized received signal at the base station, with dimension . Based on this, the channel estimation module uses the pilot signals to estimate the local channel matrix. , dimension Subsequently, the baseband receiver uses the quantized received signal. Channel matrix Quantification upper limit Quantization lower bound The baseband processor then uses the penalty gradient-based MIMO system signal transmission method proposed in this embodiment to detect the transmitted signal. Finally, the baseband processor projects the estimated transmitted signal vector onto the nearest constellation point through a simple quantization mapping operation, thus achieving channel decoding.

[0169] Traditional quantized gradient descent detection algorithms ignore discrete constellation point constraints, thus scaling the gradient descent update calculation to the entire real number domain:

[0170] ,

[0171] ,

[0172] In the formula, It is a scalar step size. For the transmitted signal estimation vector gradient, Let represent the estimated vector of the transmitted signal obtained in the (t+1)th iteration.

[0173] However, traditional quantization gradient descent methods neglect the constraints of discrete constellation points, failing to fully utilize the discrete characteristics of the transmitted signal. This results in a significant performance gap compared to optimal maximum likelihood detection, severely limiting its detection effectiveness. Such relaxation methods only work when the number of receive antennas at the base station is much greater than the number of transmit antennas. At that time, it can achieve good performance. However, with the rapid increase in the number of user antennas, if the traditional quantization gradient descent method is continued, tens of thousands of receiving antennas and corresponding hardware processing units need to be configured on the base station side, resulting in a significant increase in cost and energy consumption, which greatly limits its application scope.

[0174] To address the aforementioned issues, the penalty gradient-based MIMO system signal transmission method proposed in this invention extends and improves upon the traditional quantization gradient descent method for low-precision quantization large-scale MIMO uplink detection scenarios. It introduces discrete constellation point constraints during the solution process, effectively approximating the optimal maximum likelihood detection performance and significantly expanding its applicability. More importantly, this invention offers faster convergence speed and significantly reduced computational complexity, thus enabling a low-overhead, high-performance low-precision quantization detection scheme.

[0175] The method in this embodiment of the invention will be described in detail below with reference to a specific example. This example uses the number of transmitting antennas... Number of receiving antennas The modulation scheme is 16-QAM, and the quantization bit depth is [value missing]. The invention is illustrated using a low-precision quantization large-scale MIMO system as an example, but it is not limited thereto.

[0176] like Figure 3 As shown, the signal transmission method for a MIMO system based on penalized gradients includes the following steps:

[0177] Step 1: For a given set of... root transmitting antenna and In a MIMO system with a root-receiver antenna, the base station uses pilot signals to perform channel estimation and obtain the channel matrix. Channel matrix The dimension is The low-precision ADC on the base station side quantizes the received signal from the radio frequency link to obtain the quantized received signal. and the corresponding quantization upper limit With quantification lower limit All of their dimensions are In this embodiment, for the transmitting antenna... Receiving antenna Quantized massive MIMO systems, channel matrix The dimension is , , and All dimensions are .

[0178] Step 2: Initialize the relevant parameters of the quantization penalty gradient algorithm: Initialize the initial value of the transmitted signal estimation vector. The initial value of the search vector Channel calculation matrix Step length Penalty coefficient Shape factor and the initial value of momentum step size The specific operating steps are as follows:

[0179] Step 2.1 Initialize the transmitted signal estimation vector Initial value of the search vector Set to: Both have dimensions. In this example, .

[0180] Step 2.2 Calculate the channel calculation matrix The channel calculation matrix The calculation formula is:

[0181] ,

[0182] in, Perform a diagonalization operation, that is, keep only the diagonal elements and set the rest of the elements to zero; This indicates the quantized received signal. Each component Extracting symbolic information, specifically defined as:

[0183] ,

[0184] The resulting channel calculation matrix The dimension is In this example, .

[0185] Step 2.3 Calculate the step size The step size The calculation formula is:

[0186] ,

[0187] in, Channel matrix The square of the 2-norm.

[0188] Step 2.4 Adjust the shape factor and the initial value of momentum step size Set as , .

[0189] Step 2.5 Initialize the penalty parameter as follows:

[0190] ,

[0191] in, Output the minimum value.

[0192] Step 3: Based on the quantization penalty gradient algorithm, continuously utilize the channel matrix. Quantized received signal Quantification upper limit Quantization lower bound The transmitted signal estimation vector is updated to obtain the final transmitted signal estimation vector. (Based on the quantization-penalized gradient algorithm, the estimated vector is iteratively updated using the received signal and channel information), the specific process includes:

[0193] Step 3.1 Set the initial number of iterations And set the maximum number of iterations to .

[0194] Step 3.2 in the In each iteration, the search vector is processed based on the preset objective function. Perform gradient descent calculations to obtain the forward vector. The specific calculation steps are as follows:

[0195] Step 3.2.1: Calculate the matrix based on the channel and search vector For the search vector The gradient is calculated as follows:

[0196] ,

[0197] in, The Sigmoid function is specifically calculated as follows:

[0198] ,

[0199] The gradient calculated at the end Dimensions In this example, .

[0200] Step 3.2.2: Gradient-based Search vector and step length Gradient descent is performed to obtain the forward vector. :

[0201]

[0202] Calculated forward vector The dimension is In this example, .

[0203] Step 3.3 Based on forward vector Using the selected penalty function, a near-end projection operation is performed to obtain the transmitted signal estimation vector. The specific calculation steps are as follows:

[0204] Step 3.3.1: Perform different proximal projections depending on the type of the selected penalty function. If a sinusoidal penalty function is selected, proceed to step 3.3.2; if a Gaussian penalty function is selected, proceed to step 3.3.3.

[0205] Step 3.3.2: When using a sinusoidal penalty function, for the forward vector Each component is projected one by one to obtain the transmitted signal estimation vector. The One element:

[0206] ,

[0207] in, It is the transmitted signal estimation vector The Each element is used to calculate the estimated vector of the transmitted signal. The dimension is In this example, Proceed to step 3.4.

[0208] Step 3.3.3: When using the Gaussian penalty function, based on the selected constellation diagram For the forward vector Each component Projecting each signal one by one yields the estimated vector of the transmitted signal. The element :

[0209] ,

[0210] The calculated transmitted signal estimation vector The dimension is In this example, The constellation chart in this example is... .

[0211] Step 3.4 After completing the projection operation, estimate the vector based on the previous transmitted signal. Estimate the vector of the currently transmitted signal. Perform momentum update to obtain the extrapolation vector. :

[0212] ,

[0213] Wherein, extrapolation vector Dimensions In this example, .

[0214] Step 3.5 After completing the momentum update, estimate the vector based on the transmitted signal. and extrapolation vector The cost function value determines the search vector. and momentum step size The specific process is as follows:

[0215] Step 3.5.1: Calculate the cost function components in different forms based on the selected penalty function. When the cosine penalty function is selected, proceed to step 3.5.2; when the Gaussian penalty function is selected, proceed to step 3.5.3.

[0216] Step 3.5.2: Calculate the estimated vector of the transmitted signal according to the formula for calculating the cosine penalty function. and extrapolation vector The corresponding penalty function value:

[0217] ,

[0218] ,

[0219] The calculated penalty function value (scalar) and All dimensions are Proceed to step 3.5.4.

[0220] Step 3.5.3: Calculate the signal estimation vector according to the Gaussian penalty function calculation formula. and extrapolation vector The corresponding penalty function value:

[0221] ,

[0222] ,

[0223] Calculated penalty function value and All dimensions are .

[0224] Step 3.5.4: Based on the channel matrix Quantification upper limit Quantization lower bound Calculate the estimated vector of the transmitted signal. and extrapolation vector The corresponding cost function value:

[0225] ,

[0226] ,

[0227] in, Represents the transmitted signal estimation vector The corresponding penalty function value, Represents the extrapolation vector The corresponding penalty function value, Represents the transmitted signal estimation vector The cost function value, Represents the extrapolation vector The cost function value, It is a logarithmic operator with base 1. , and Quantization upper limit Quantization lower bound The One portion, Channel information matrix The OK; Represents the transmitted signal estimation vector The One element; Represents the extrapolation vector The One element, Indicates the total number of signals sent; The first in the constellation chart Each constellation point, Indicates the penalty parameter. Represents the shape coefficient. Cost function value. and All are scalars, and all have the same dimension. .

[0228] Step 3.5.5: Compare the cost function values and ,if Then proceed to step 3.5.6. If If so, proceed to step 3.5.7.

[0229] Step 3.5.6: Update the search vector and momentum step size :

[0230] ,

[0231] ,

[0232] Search vector and momentum step size All are scalars, with the following dimensions: and The iteration count is t = t + 1. Proceed to step 3.6.

[0233] Step 3.5.7: Update the search vector and momentum step size :

[0234] ,

[0235] ,

[0236] The momentum step size was calculated. Search vector Dimensions In this example, The number of iterations is t = t + 1.

[0237] Step 3.6 Comparison and The size, if Then proceed to step 3.2 and continue iterating. If Continue to step 4.

[0238] Step 4: The baseband processor processes the estimated vector. Decoding and decision processing (i.e., constellation point quantization) is performed to obtain the final quantization detection result. The specific quantization operation can be represented as:

[0239] ,

[0240] in, Round to the nearest nearest integer. The constellation points. That is, comparisons. Each element With the set of decoding candidate points { The Euclidean distance between each constellation point in} and The constellation point with the smallest Euclidean distance between them is... The quantitative results.

[0241] Since the method proposed in this embodiment of the invention is an iterative algorithm, its complexity mainly depends on the complexity of the iterative formula. The computational complexity is statistically calculated based on the required number of real multiplications, and is considered to be... The complexity of the inverse of a dimension matrix is ​​O(n log n). In this embodiment of the invention, the computational complexity of a single iteration mainly comes from the following three parts:

[0242] (1) Calculate the gradient and forward vector .

[0243] (2) Perform near-end projection operation and calculate the estimated vector of the transmitted signal. .

[0244] (3) Calculate the cost function value and .

[0245] Specifically, calculating the forward vector Transmitted signal estimation vector and cost function value and The required complexity are respectively , and Therefore, the overall complexity of the method proposed in this embodiment of the invention in a single iteration is... In scenarios with a smaller number of constellation points, such as 4-QAM and 16-QAM, the complexity of a single iteration can be further simplified to... Compared with traditional low-precision quantization detection algorithms, it can significantly reduce computational overhead.

[0246] In this embodiment of the invention, a step size parameter is set. The selection of can guarantee the convergence of this algorithm.

[0247] Theoretical derivation shows that the quantization penalty gradient algorithm given in the embodiments of this invention has the following effect on the step size parameter. satisfy Under certain conditions, the algorithm can be guaranteed to converge exponentially to a stable point. Its exponential convergence rate... satisfy:

[0248]

[0249] Where d and C are both constants greater than zero. Due to the convergence rate... The value is less than 1, therefore the iterative process provided in this embodiment of the invention has the characteristic of rapid convergence at an exponential rate.

[0250] From the perspective of simply performing low-precision quantized signal detection across multiple antennas, various quantization detection algorithms such as BSIC, SD, and GD can, to some extent, replace the scheme proposed in this invention. However, these algorithms either have extremely high computational complexity or their detection performance differs significantly from optimal maximum likelihood detection, making them difficult to deploy flexibly in practical applications. Considering both detection performance and computational complexity, the MIMO system signal transmission method based on penalized gradients proposed in this invention achieves an optimal trade-off between performance and efficiency, demonstrating strong practicality and applicability.

[0251] For ease of understanding, a comparison table 1 is provided here, where QPG represents the proposed large-scale MIMO detection method based on quantization-penalized gradients. This refers to the number of antennas on the large-scale MIMO base station side. Total number of user antennas This represents the total number of iterations. This refers to the size of the constellation map.

[0252]

[0253] Among them, Traditional MMSE is the conventional minimum mean square error method, while BMMSE is the minimum mean square error method based on Bussgang decomposition. One-stage nML and Two-stage nML are approximate maximum likelihood methods for one-stage and two-stage systems, respectively. GD is the conventional gradient descent method. Proposed QPG-C and QPG-G are the signal transmission methods for MIMO systems based on penalized gradients proposed in this invention. Figure 4 It can be seen that the method proposed in this invention achieves approximately the highest detection accuracy with significantly lower computational complexity. Based on Figure 5 It can be seen that the method proposed in this invention converges rapidly with the increase of the number of iterations, and its performance far exceeds that of the BMMSE method, bringing significant gains in nonlinear performance. Based on Figure 6 It can be seen that the method proposed in this invention still works efficiently under various channel error conditions, demonstrating its excellent robustness.

[0254] Example 2

[0255] Based on the same inventive concept as Embodiment 1, this embodiment of the invention provides a signal transmission device for a MIMO system based on a penalized gradient, comprising:

[0256] The first data acquisition module is used to acquire the quantized received signal. and quantification upper limit Quantization lower bound ;

[0257] The second data acquisition module is used to perform channel estimation using pilot signals and obtain the channel matrix. ;

[0258] An initialization module is used to initialize the relevant parameters of the quantization penalty gradient algorithm, wherein the gradient descent strategy in the quantization penalty gradient algorithm is gradient descent, near-end projection and extrapolation interactive calculation.

[0259] The estimation module is used to continuously utilize the channel matrix based on the quantization penalty gradient algorithm. Quantized received signal Quantification upper limit Quantization lower bound The transmitted signal estimation vector is updated to obtain the final transmitted signal estimation vector. ;

[0260] The decoding and decision module is used to estimate the vector of the transmitted signal. Perform decoding and decision processing to obtain the final detection output. .

[0261] The rest are the same as in Example 1.

[0262] Example 3

[0263] Based on the same inventive concept as in Embodiment 1, this embodiment of the invention provides a MIMO system signal transmission system based on penalized gradient, including a storage medium and a processor;

[0264] The storage medium is used to store instructions;

[0265] The processor is configured to operate according to the instructions to execute the method according to any one of Embodiment 1.

[0266] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0267] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0268] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0269] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0270] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

[0271] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for signal transmission of a MIMO system based on penalized gradient, applied to a receiving end, and characterized in that, The method comprises: acquiring the quantized received signal and an upper quantization limit and a lower quantization limit ; Channel matrix is obtained by using pilot for channel estimation ; initializing parameters of a quantization-penalized gradient algorithm, wherein a gradient descent strategy in the quantization-penalized gradient algorithm is gradient descent, proximal projection and extrapolation interaction calculation; based on the quantized penalty gradient algorithm, the channel matrix , the quantized received signal , the quantization upper limit and the quantization lower limit is updated to obtain a final transmit signal estimation vector ; estimating a vector of the transmission signal performing a decoding decision process to obtain a final detection output .

2. The method of claim 1, wherein the method is based on penalized gradient. the initializing parameters of the quantization-penalized gradient algorithm comprises: Setting the initial value of the transmission signal estimation vector , search vector initial value Setting to: ; A channel calculation matrix is calculated The calculation formula of the channel calculation matrix is as follows: ， wherein is a diagonalization operation, denotes extracting the polarity of the input signal ; The calculation step The calculation formula of the step is ， wherein is the square of the 2-norm of the channel matrix H. Setting the shape factor and the initial value of the momentum step size , , ; Computing a penalty parameter , the penalty parameter is computed as ， In the formula, represents the minimum output value.

3. The method of claim 1, wherein the method further comprises: The quantized penalty gradient algorithm is based on the channel matrix , the quantized received signal , the quantization upper limit , and the quantization lower limit The sending signal estimation vector is updated to obtain a final sending signal estimation vector , comprising: Set initial iteration number , maximum iteration number is ; The following steps are repeatedly performed until : Based on the preset objective function, the search vector Perform gradient descent calculations to obtain the forward vector. ; According to the advance vector and the selected penalty function, a proximal projection operation is performed to obtain a transmit signal estimate vector ; According to the last transmitted signal estimation vector , a current transmitted signal estimation vector is updated with momentum to obtain an extrapolated vector ; Estimation of a transmit signal based on a vector of cost function values and extrapolation of a vector ; determination of a search vector and a momentum step size ; Comparing and the size of.

4. The method of claim 3, wherein the method further comprises: The forward vector The acquisition method comprises: If the number of quantization bits is 2, the channel calculation matrix is and the search vector is [0070] The gradient of the search vector [0071] is calculated as follows: [0072] ， If the number of quantization bits , the search vector , the upper limit of quantization , the lower limit of quantization and the channel matrix , the gradient of the search vector is calculated, and the calculation formula of the gradient of the search vector is ， wherein, denotes a gradient of a search vector , is a Sigmoid function, is an all-one vector, is a preset objective function; Based on the search vector Gradient , search vector and step size , gradient descent calculation is performed to obtain the advance vector : 。 5. The method of claim 4, wherein the method further comprises: The forward vector and the selected penalty function, a proximal projection operation is performed to obtain a transmit signal estimation vector , and specifically comprises: If the sinusoidal penalty function is selected, each component of the advance vector is projected based on the selected sinusoidal penalty function to obtain the element of the transmit signal estimate vector ​ ， If a Gaussian penalty function is chosen, then the forward vector is processed based on the chosen Gaussian penalty function. Perform a near-end projection operation to obtain the transmitted signal estimation vector. The One element: ， wherein is a constellation, is a transmit signal estimate vector is a first element of an advance vector is a first element of an advance vector is a first is a penalty parameter; From the calculation to the complete transmitted signal estimate vector , denotes the total number of transmitted signals.

6. The method of claim 5, wherein: The extrapolation vector The calculation formula is: ， wherein, denotes the th iteration momentum step size, denotes the th iteration transmitted signal estimate vector.

7. The method of claim 6, wherein the method further comprises: The cost function value of the sending signal estimation vector and the extrapolation vector determines the search vector and the momentum step , specifically comprising: If the cosine penalty function is selected, the sending signal estimation vector is calculated according to the cosine penalty function calculation formula and the extrapolation vector The corresponding penalty function value is calculated according to the following formula: ， ， If the Gaussian penalty function is selected, the sending signal estimation vector is calculated according to the Gaussian penalty function calculation formula and the extrapolation vector corresponding to the penalty function value, the calculation formula of which is: ， ， based on a channel matrix , a quantization upper limit and a quantization lower limit , a transmit signal estimation vector and an extrapolation vector corresponding cost function values, the transmit signal estimation vector and the extrapolation vector corresponding cost function values are calculated according to the following formula: ， ， wherein denotes a transmit signal estimate vector corresponding penalty function value, denotes an extrapolation vector corresponding penalty function value, denotes a transmit signal estimate vector cost function value, denotes an extrapolation vector cost function value, is a logarithm operator with base , and are the first components of a quantization upper bound and a quantization lower bound , respectively, is the first row of a channel information matrix ; denotes the first element of a transmit signal estimate vector ; denotes the first element of an extrapolation vector , denotes a total number of transmit signals; denotes the first constellation point in a constellation diagram, denotes a penalty parameter, denotes a shape coefficient; Comparing cost function values and ; If , then update the search vector and the momentum step , the update formula is: ， ， If , then update the search vector and the momentum step , the update formula is: ， ， wherein, denotes the denotes the denotes the output minimum.

8. The method of claim 1, wherein the method is a method of penalized gradient based MIMO system signal transmission. The detection output The calculation formula is: ， wherein denotes the rounding to the nearest constellation point, denotes a set of transmit signals subject to the constellation point mapping, the dimension of each transmit signal in the set of transmit signals being . 9.A device for signal transmission of a MIMO system based on penalized gradient, applied to a receiving end, and having the characteristics that, The method comprises: a first data acquisition module, configured to acquire the quantized received signal and a quantization upper limit and a quantization lower limit ; a second data acquisition module, configured to acquire a channel matrix by using a pilot for channel estimation ; an initializing module, configured to initialize parameters of a quantization-penalized gradient algorithm, wherein a gradient descent strategy in the quantization-penalized gradient algorithm is gradient descent, proximal projection and extrapolation interaction calculation; an estimation module for updating the transmitted signal estimation vector based on the quantized penalized gradient algorithm , the quantized received signal , the quantization upper bound , and the quantization lower bound to obtain a final transmitted signal estimation vector ; a decoding decision module for performing a decoding decision on the transmit signal estimate vector performing a decoding decision process to obtain a final detection output .

10. A penalized gradient-based MIMO system signal transmission system applied to a receiving end, characterized in that, comprising a storage medium and a processor; the storage medium is configured to store instructions; the processor is configured to operate according to the instructions to perform the method according to any one of claims 1-8.

Images