Signal detection method and system based on efficient diagonal random conjugate gradient
By combining an efficient diagonal stochastic conjugate gradient algorithm with stochastic gradient estimation and adaptive step size, the problems of high computational complexity and slow convergence in large-scale MIMO systems are solved, and efficient and accurate signal detection is achieved.
Patent Information
- Application Number
- CN202510921078.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-04
- Publication Date
- 2025-09-05
AI Technical Summary
The traditional conjugate gradient method has high computational complexity and slow convergence speed in large-scale MIMO systems, and is difficult to adapt to dynamic differences in channel gain, affecting signal detection efficiency.
An efficient diagonal stochastic conjugate gradient algorithm is adopted, combined with stochastic gradient estimation and adaptive diagonal step size, to iteratively approximate the optimal solution through internal and external loop structures, and dynamically adjust the step size to adapt to channel changes.
Significantly reduce computational complexity, improve signal detection efficiency and accuracy, adapt to complex channel environments, and enhance the practicality of large-scale MIMO systems.
Smart Images
Figure CN120602050A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of wireless communications, and more particularly to a signal detection method and system based on efficient diagonal random conjugate gradient. Background Art
[0002] The contradiction between the scarcity of spectrum resources and the exponential growth of data transmission demand has become increasingly prominent in the development of wireless communication technology. Traditional single-antenna systems, constrained by Shannon's channel capacity theorem, struggle to simultaneously guarantee high transmission rates and communication reliability within limited frequency bands. Multiple-Input, Multiple-Output (MIMO) technology, through the collaborative operation of multiple antenna arrays at the transmit and receive ends, innovatively transforms the multipath propagation characteristics of wireless channels into spatial resources. Without increasing bandwidth or transmit power, it achieves three breakthroughs: exponentially increasing spectrum efficiency and system capacity through spatial multiplexing; enhancing communication reliability through spatial diversity; and optimizing transmission rates by combining precoding techniques. This has made MIMO a core technology pillar of 5G and future communication systems. However, as massive MIMO antennas scale to tens or even hundreds of elements, the system faces challenges such as a surge in channel estimation complexity and an exponential increase in signal detection computational complexity. Therefore, designing receiving algorithms that combine low computational complexity with near-optimal detection performance has become a key research direction for advancing the practical application of MIMO systems.
[0003] In massive multiple-input, multiple-output (MIMO) systems, signal detection is a crucial step, directly impacting system performance and communication quality. Traditional linear detectors, such as the minimum mean square error (MMSE) detector, face significant challenges in practical applications. The MMSE detector requires matrix inversion, and its computational complexity increases dramatically as the number of antennas increases. This results in an extremely large computational load for the MMSE detector, significantly reducing detection efficiency. To reduce computational complexity, the conjugate gradient method (CGM) has been introduced for signal detection in massive MIMO systems. The CGM gradually approaches the optimal solution through iteration, avoiding direct matrix inversion. However, the CGM uses a fixed step size for iteration, which limits its convergence rate.
[0004] The traditional conjugate gradient method (CG) uses a fixed step size and deterministic gradient updates to iterate signal estimates, residuals, and conjugate directions. However, its reliance on global gradient calculations makes it difficult to adapt to the dynamic differences in channel gains across dimensions in massive MIMO systems. Therefore, further research is needed to address these technical issues. Summary of the Invention
[0005] In view of this, the present invention provides a signal detection method and system based on efficient diagonal random conjugate gradient, designs an efficient diagonal random conjugate gradient algorithm (DSCG) to replace the traditional conjugate gradient algorithm, and cleverly integrates stochastic gradient estimation with adaptive diagonal step size. Through stochastic gradient estimation, the characteristic information of the signal can be captured more accurately and the influence of gradient noise can be reduced; the adaptive diagonal step size can dynamically adjust the step size according to the actual situation during the iteration process, significantly improving the convergence speed. This combination not only effectively reduces the computational complexity, but also greatly improves the detection efficiency, providing a more efficient and practical method for signal detection in large-scale MIMO systems.
[0006] In order to achieve the above object, the present invention adopts the following technical solutions:
[0007] A signal detection method based on efficient diagonal random conjugate gradient includes the following steps:
[0008] S1. Receive the signal to be detected and model it as the signal model, determine the objective function of minimum mean square error detection, and introduce a quadratic regularization term;
[0009] S2, building a signal detection model based on an efficient diagonal stochastic conjugate gradient algorithm;
[0010] S3. Generate multi-SNR training data based on the signal model and actual communication scenario, and preprocess the data;
[0011] S4. Use multiple signal-to-noise ratio training data to perform offline training on the signal detection model and optimize the model parameters;
[0012] S5. Deploy the trained signal detection model to the actual communication system to achieve real-time detection of massive MIMO signals.
[0013] Optionally, S1 specifically includes: the received signal is modeled as y=Hx+n, where H is the channel matrix, x is the transmitted signal vector, and n is the noise; the objective function is obtained by minimizing the regularized mean square error: Deriving the normal equations in is the regularization parameter related to the noise variance, H H represents the conjugate transpose of the channel matrix, and I is a unit vector.
[0014] Optionally, in S2, the efficient diagonal stochastic conjugate gradient algorithm integrates stochastic recursive gradient estimation and an adaptive diagonal step size adjustment mechanism to iteratively approximate the optimal solution through an inner and outer loop structure.
[0015] Optionally, in S2, the specific implementation process of the efficient diagonal stochastic conjugate gradient algorithm is as follows:
[0016] The variance is controlled by using a stochastic recursive gradient algorithm framework under small batches, which includes an outer loop and an inner loop. The outer loop periodically calculates the full gradient as the baseline gradient. The inner loop randomly samples small batches of samples to calculate the stochastic gradient.
[0017] The diagonal Barzilai-Borwein step size rule is used to dynamically adjust the step size of each dimension to replace the fixed step size in the conjugate gradient method, combined with the long BB step size Short BB step length And the weighted update of historical step information is achieved through the following formula:
[0018] in Among them, s k =x k -x k-1 , is the parameter difference between adjacent iteration points; is the gradient difference between adjacent iteration points.
[0019] Optionally, in S3, generating multi-SNR training data specifically includes: simulating multiple fading scenarios to construct a channel matrix H based on the signal model y=Hx+n and actual scene characteristics, using high-order modulation to generate a transmission signal x, injecting additive white Gaussian noise and controlling the signal-to-noise ratio to generate a large amount of training data.
[0020] Optionally, in S3, data preprocessing includes normalizing the received signal, normalizing the channel matrix, and processing the transmitted signal labels to unify the data scale and ensure the balance of features in each dimension.
[0021] Optionally, in S4, offline training includes: initializing the parameters of the efficient diagonal random conjugate gradient algorithm and the initial values of the signal estimation based on the preprocessed actual communication scenario data, dynamically adjusting the step size of each dimension using the diagonal matrix through iterative optimization, combining small-batch debiasing gradient and search direction optimization to approach the optimal estimate; terminating the training with the residual threshold or the maximum number of iterations, and saving the optimal model parameters adapted to different channel scenarios.
[0022] Optionally, in S5, real-time detection includes: embedding the trained signal detection model into the communication system, receiving the noisy signal in real time and performing channel estimation and standardization preprocessing, loading the optimal parameters and iteratively updating the signal estimation value, and outputting the final detection result when the residual convergence condition is met.
[0023] A signal detection system based on efficient diagonal random conjugate gradient includes the following steps:
[0024] Signal model building module: used to receive the signal to be detected and model it as the signal model, determine the objective function of minimum mean square error detection, and introduce a quadratic regularization term;
[0025] Signal detection model building module: used to build a signal detection model based on the efficient diagonal stochastic conjugate gradient algorithm;
[0026] Data generation and preprocessing module: used to generate multi-SNR training data based on the signal model and actual communication scenario, and preprocess the data;
[0027] Model parameter optimization module: used to perform offline training of the signal detection model using multiple signal-to-noise ratio training data and optimize model parameters;
[0028] Signal detection module: used to deploy the trained signal detection model to the actual communication system to achieve real-time detection of massive MIMO signals.
[0029] It can be seen from the above technical solution that compared with the existing technology, the present invention provides a signal detection method and system based on efficient diagonal random conjugate gradient, which deeply integrates random recursive gradient estimation with diagonal adaptive step size, and realizes precise adaptation of the characteristics of various dimensions of large-scale MIMO signals while suppressing the influence of noise. It effectively solves the problems of complex calculation and low convergence efficiency of traditional methods in complex channel environments, significantly improves the efficiency and accuracy of signal detection, and provides solid technical support for the efficient operation of large-scale MIMO systems. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.
[0031] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION
[0032] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0033] The embodiment of the present invention discloses a signal detection method based on efficient diagonal random conjugate gradient, such as Figure 1 As shown, the following steps are included:
[0034] S1. Receive the signal to be detected and model it as the signal model, determine the objective function of minimum mean square error detection, and introduce a quadratic regularization term;
[0035] S2, building a signal detection model based on an efficient diagonal stochastic conjugate gradient algorithm;
[0036] S3. Generate multi-SNR training data based on the signal model and actual communication scenario, and preprocess the data;
[0037] S4. Use multiple signal-to-noise ratio training data to perform offline training on the signal detection model and optimize the model parameters;
[0038] S5. Deploy the trained signal detection model to the actual communication system to achieve real-time detection of massive MIMO signals.
[0039] In S1, the following steps are specifically included:
[0040] In the signal detection scenario, the received signal can be modeled as y = Hx + n, H∈ N×K is the channel matrix (N is the number of receiving antennas, K is the number of transmitting antennas), x∈ K×1 is the transmitted signal vector, and n is the noise.
[0041] The core of minimum mean square error (MMSE) detection is to estimate the signal by optimizing Minimizing the regularized mean square error is equivalent to solving the regularized least squares problem:
[0042]
[0043] This problem is solved by introducing a quadratic regularization term Suppress noise amplification. By taking the derivative of the objective function and setting the gradient to zero, the normal equation Ax = b can be derived. is a symmetric positive definite matrix (composed of the conjugate transpose of the channel matrix and the superposition of the noise variance matrix), b = H H y is the matched filter output vector, and solving this equation is the core computational task of MMSE detection.
[0044] In S2, the specific process is as follows:
[0045] The present invention adopts the Stochastic Recursive Gradient Algorithm (SARAH) framework under small batches to control variance. The internal and external loop structures are as follows: the outer loop periodically calculates the full gradient as the baseline gradient; the inner loop randomly samples small batches of samples to calculate the stochastic gradient to ensure that the variance is controllable.
[0046] The present invention uses the diagonal Barzilai-Borwein (DBB) step size rule to replace the fixed step size in the conjugate gradient method. First, in order to achieve the step size in the long BB step size Short BB step length Adaptive adjustment between the solution and the weighted average solution is introduced by formula:
[0047]
[0048] These formulas take the intermediate value strategy and combine the parameter update amount of the current dimension Gradient difference and historical step information or Dynamically select the step size mode. In addition, for random environments, the following formula is used as a variant:
[0049]
[0050] The parameters θ and m are introduced to further adapt the step size requirements under random noise. These formulas together enrich the implementation of the DBB step size rule, making it better able to cope with the time-varying and random nature of the channel and improving the algorithm's adaptability in complex scenarios.
[0051] Specifically, the iterative process of the efficient diagonal stochastic conjugate gradient method proposed in the present invention is:
[0052] First, perform the initialization steps:
[0053] Set the update frequency m, the mini-batch sample size b, and determine the initial point Introduce parameters α and θ and define the initial diagonal matrix U0=η0I d (η0 is the initial step length, I d represents the d-dimensional identity matrix), and given the probability distribution Q = {q1,q2,...,q n}, used for random sampling of subsequent small batch samples.
[0054] Then the outer loop:
[0055] make The objective function is Calculate full gradient It is the gradient of the objective function and provides benchmark gradient information for subsequent iterations.
[0056] Then determine the initial search direction:
[0057] If s=1, then d0=-v0; if s>1, then d0=-p s-1 , and set
[0058] Furthermore, the inner loop iteration begins:
[0059] pass Update the parameters, where It is a diagonal matrix that realizes independent adjustment of the step size of each dimension, making the update more suitable for the channel characteristics of each dimension.
[0060] Select a mini-batch of size b Each i∈I k Random sampling is performed according to distribution Q to ensure the randomness and unbiasedness of gradient estimation.
[0061] calculate The fixed noise is offset by the gradient difference between the previous and subsequent gradients, which reduces the interference of noise on gradient estimation and improves the robustness of the algorithm in low signal-to-noise ratio scenarios.
[0062] use Compute the conjugate parameter, which is automatically bounded Ensure the stability and rationality of the search direction.
[0063] pass Update the search direction, combine the current stochastic gradient with the historical search direction, generate a new conjugate direction, and improve search efficiency.
[0064] Finally, the loop ends:
[0065] After the inner loop is completed, p is obtained s =v m There are two options to determine Option I is to randomly select t from {0,1,…,m}, and let Option II is a direct command Finally, according to the above, The formula calculates the step size, and then uses the formula or renew Prepare for the next iteration.
[0066] In S3, the specific process is as follows:
[0067] Based on the signal model y = Hx + n and actual scene characteristics, the channel matrix H is constructed by simulating various fading scenarios (such as Rayleigh, Ricean, and correlated fading). High-order modulation is used to generate the transmitted signal x. Additive white Gaussian noise is injected and the signal-to-noise ratio is controlled to generate a large amount of training data. The data is then preprocessed, including normalization of the received signal y, standardization of the channel matrix H, and labeling of the transmitted signal x. This unifies the data scale and ensures balanced features across all dimensions, providing high-quality and diverse data support for subsequent model training.
[0068] In S4, the specific process is as follows:
[0069] First, the key parameters and initial signal estimation values of the DSCG algorithm are initialized based on preprocessed multi-scenario data. Subsequently, through an iterative optimization process, the step sizes of each dimension are dynamically adjusted using a diagonal matrix. This process combines debiased gradient calculations and search direction optimization on small batches of samples to gradually approach the optimal signal estimation. Training is criterioned by residual accuracy, terminating when a preset error threshold or the maximum number of iterations is reached. Ultimately, the optimal model parameters for different channel fading scenarios are saved, forming an efficient model for massive MIMO signal detection.
[0070] In S5, the specific process is as follows:
[0071] During online deployment, the offline trained DSCG model is embedded in the actual communication system, and noisy MIMO signals are received in real time and preprocessed, including channel estimation, noise variance calculation, and signal normalization. The optimal parameters obtained through training are then loaded, and the step size mode is dynamically matched according to the real-time channel scenario. The signal estimation value is gradually updated through iterative detection. When the residual convergence conditions or real-time processing time requirements are met, the final signal estimation result is output, realizing efficient real-time detection of large-scale MIMO signals in complex scenarios.
[0072] This embodiment also discloses a signal detection system based on efficient diagonal random conjugate gradient, comprising the following steps:
[0073] Signal model building module: used to receive the signal to be detected and model it as the signal model, determine the objective function of minimum mean square error detection, and introduce a quadratic regularization term;
[0074] Signal detection model building module: used to build a signal detection model based on the efficient diagonal stochastic conjugate gradient algorithm;
[0075] Data generation and preprocessing module: used to generate multi-SNR training data based on the signal model and actual communication scenario, and preprocess the data;
[0076] Model parameter optimization module: used to perform offline training of the signal detection model using multiple signal-to-noise ratio training data and optimize model parameters;
[0077] Signal detection module: used to deploy the trained signal detection model to the actual communication system to achieve real-time detection of massive MIMO signals.
[0078] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.
[0079] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A signal detection method based on efficient diagonal random conjugate gradient, characterized in that: The following steps are involved: S1. Receive the signal to be detected and model it as the signal model, determine the objective function of minimum mean square error detection, and introduce a quadratic regularization term; S2, building a signal detection model based on an efficient diagonal stochastic conjugate gradient algorithm; S3. Generate multi-SNR training data based on the signal model and actual communication scenario, and preprocess the data; S4. Use multiple signal-to-noise ratio training data to perform offline training on the signal detection model and optimize the model parameters; S5. Deploy the trained signal detection model to the actual communication system to achieve real-time detection of massive MIMO signals.
2. The signal detection method based on efficient diagonal random conjugate gradient according to claim 1, characterized in that: Specifically, S1 includes: the received signal is modeled as y=Hx+n, where H is the channel matrix, x is the transmitted signal vector, and n is the noise; the objective function is to minimize the regularized mean square error: Deriving the normal equations in is the regularization parameter related to the noise variance, H H represents the conjugate transpose of the channel matrix, and I is a unit vector.
3. The signal detection method based on efficient diagonal random conjugate gradient according to claim 1, characterized in that: In S2, the efficient diagonal stochastic conjugate gradient algorithm combines stochastic recursive gradient estimation with an adaptive diagonal step size adjustment mechanism to iteratively approach the optimal solution through an inner and outer loop structure.
4. The signal detection method based on efficient diagonal random conjugate gradient according to claim 1, characterized in that: In S2, the specific implementation process of the efficient diagonal stochastic conjugate gradient algorithm is as follows: The variance is controlled by using a stochastic recursive gradient algorithm framework under small batches, which includes an outer loop and an inner loop. The outer loop periodically calculates the full gradient as the baseline gradient. The inner loop randomly samples small batches of samples to calculate the stochastic gradient. The diagonal Barzilai-Borwein step size rule is used to dynamically adjust the step size of each dimension to replace the fixed step size in the conjugate gradient method, combined with the long BB step size Short BB step length And the weighted update of historical step information is achieved through the following formula: Among them, s k =x k -x k-1 , is the parameter difference between adjacent iteration points; is the gradient difference between adjacent iteration points.
5. The signal detection method based on efficient diagonal random conjugate gradient according to claim 1, characterized in that: In S3, generating multi-SNR training data specifically includes: constructing the channel matrix H based on the signal model y=Hx+n and actual scene characteristics by simulating multiple fading scenarios, generating the transmission signal x using high-order modulation, injecting additive white Gaussian noise and controlling the SNR to generate a large amount of training data.
6. The signal detection method based on efficient diagonal random conjugate gradient according to claim 1, characterized in that: In S3, data preprocessing includes normalization of the received signal, standardization of the channel matrix, and label processing of the transmitted signal to unify the data scale and ensure the balance of features in each dimension.
7. The signal detection method based on efficient diagonal random conjugate gradient according to claim 1, characterized in that: In S4, offline training includes: initializing the parameters of the efficient diagonal random conjugate gradient algorithm and the initial values of the signal estimation based on pre-processed actual communication scenario data, dynamically adjusting the step size of each dimension through iterative optimization using the diagonal matrix, and combining small-batch debiasing gradients with search direction optimization to approximate the optimal estimate; terminating training with a residual threshold or the maximum number of iterations, and saving the optimal model parameters adapted to different channel scenarios.
8. The signal detection method based on efficient diagonal random conjugate gradient according to claim 1, characterized in that: In S5, real-time detection includes: embedding the trained signal detection model into the communication system, receiving the noisy signal in real time and performing channel estimation and normalization preprocessing, loading the optimal parameters and iteratively updating the signal estimation value, and outputting the final detection result when the residual convergence condition is met.
9. A signal detection system based on efficient diagonal random conjugate gradient, characterized in that: The following steps are involved: Signal model building module: used to receive the signal to be detected and model it as the signal model, determine the objective function of minimum mean square error detection, and introduce a quadratic regularization term; Signal detection model building module: used to build a signal detection model based on the efficient diagonal stochastic conjugate gradient algorithm; Data generation and preprocessing module: used to generate multi-SNR training data based on the signal model and actual communication scenario, and preprocess the data; Model parameter optimization module: used to perform offline training of the signal detection model using multiple signal-to-noise ratio training data and optimize model parameters; Signal detection module: used to deploy the trained signal detection model to the actual communication system to achieve real-time detection of massive MIMO signals.