Differential coding assisted machine learning channel state information feedback method and device

The differential coding-assisted machine learning method improves CSI feedback accuracy and reduces bandwidth and energy consumption by encoding CSI signals differentially and using a lightweight network, addressing the limitations of existing methods.

CN115189742BActive Publication Date: 2024-12-03XIHUA UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202210805014.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-07-08
Publication Date
2024-12-03
Estimated Expiration
2042-07-08

AI Technical Summary

Technical Problem

Existing CSI feedback methods face challenges in reducing uplink bandwidth resource occupancy and user end energy consumption while maintaining accurate reconstruction of downlink CSI, particularly due to superposition interference and the need for uplink channel estimation guidance.

Method used

A differential coding-assisted machine learning method that differentially encodes and modulates CSI feedback signals using OFDM, combined with a lightweight enhanced CSI network, to improve reconstruction accuracy and reduce bandwidth and energy consumption.

Benefits of technology

Saves uplink bandwidth resources and reduces user end energy consumption while enhancing the reconstruction accuracy of downlink CSI through differential coding and machine learning, avoiding the need for uplink channel estimation guidance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115189742B_ABST
    Figure CN115189742B_ABST
Patent Text Reader

Abstract

The present invention discloses a differential coding-assisted machine learning channel state information feedback method and device, including: user end: the user end performs differential coding after real-valued, quantized and digitally modulated downlink CSI vector into modulated signal, obtains differential coded signal and then feeds it back to base station; base station end: adopts character-by-character differential decoding mode to obtain differential decoded signal for received signal; performs digital demodulation and dequantization processing on differential decoded signal to obtain downlink CSI recovery information, and constructs enhanced CSI network Enh‑CsiNet to enhance the downlink CSI recovery information to obtain downlink CSI with enhanced final recovery accuracy. The present invention avoids the user end sending guidance, thereby saving bandwidth resources, reducing the energy consumption of the user end, and improving the reconstruction accuracy of downlink CSI.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of channel state information feedback of a large-scale multiple-input multiple-output system in frequency division duplex mode, and in particular to a differential coding-assisted machine learning channel state information (CSI) feedback method and device. Background Art

[0002] In order to reduce feedback overhead and thus reduce uplink bandwidth resource occupancy and user (UE, User Equipment) end energy consumption, existing methods mainly focus on compressed feedback of downlink CSI. The CSI feedback method based on compressed sensing (CS, compressed sensing) reduces feedback overhead by developing a signal sparse structure. However, the CSI feedback based on CS is based on the assumption of sparse downlink CSI under a certain sparse basis, which is sometimes inconsistent with the actual situation, resulting in a significant decrease in the sparse reconstruction performance of downlink CSI. Applying deep learning (DL, Deep Learning) to CSI feedback effectively reduces the CSI feedback overhead while improving the CSI feedback accuracy. The DL-based superposition CSI feedback method superimposes the downlink CSI on the uplink user data sequence (UL-US, Uplink User Data Sequences) and then feeds it back to the base station (BS, Base Station) end, avoiding the occupation of additional uplink bandwidth resources due to feedback of downlink CSI and improving spectrum efficiency. Nevertheless, superposition CSI feedback has superposition interference, which reduces the CSI feedback accuracy. Moreover, in order for the BS to demodulate the UL-US, the UE usually needs to send guidance for uplink channel estimation to the BS. This inevitably occupies uplink bandwidth resources and consumes UE energy. Fortunately, the differential coding-assisted machine learning CSI feedback method avoids the guidance overhead for uplink channel estimation by differential means, thereby reducing uplink bandwidth resource occupancy and saving UE energy consumption. The introduction of neural networks further improves the reconstruction accuracy of downlink CSI. Summary of the invention

[0003] The purpose of the present invention is to provide a differential coding assisted machine learning channel state information feedback method and device, which, compared with the existing uplink guided channel estimated CSI feedback, saves uplink bandwidth resources and reduces user end energy consumption while improving the reconstruction accuracy of downlink CSI.

[0004] The differential coding-assisted machine learning channel state information feedback method includes:

[0005] User-side processing:

[0006] S1. Downlink CSI vector The real-valued, quantized and digitally modulated signal s is differentially encoded to obtain a differentially coded signal z, and the differentially coded signal z is OFDM modulated and fed back to the base station.

[0007] The real-value conversion is to convert the downlink CSI vector h into real-valued CSI information

[0008] The quantization is to measure the real-valued CSI information Perform a-bit quantization to obtain the bit stream information of the real and imaginary parts

[0009] The digital modulation is to convert the bit stream information Perform digital modulation to form a modulated signal

[0010] in, is the modulation order;

[0011] The differential encoding is to modulate the signal s = [s 1 ,s 2 ,...,s N ] T According to the rules Perform differential encoding to obtain the differential encoding signal z=[z 1 ,z 2 ,...z N ] T ;

[0012] Where N = 2aMγ; UE and BS agree on z 0 = 1, then there is no need to send z 0 , the differential encoding signal z is discarded z 0 The differential encoded signal.

[0013] Base station processing:

[0014] S2. The base station performs orthogonal frequency division multiplexing (OFDM) demodulation to obtain the received signal The received signal r is decoded character by character to obtain a differential decoded signal

[0015] The differential decoding is achieved by solving get;

[0016] in, represents the i-th decision value; r i 、r i-1 Respectively represent the i-th and i-1-th elements in r; st Indicated in the constellation set The superscript * indicates the conjugate operation, Re[·] indicates the real part, and max[·] indicates the maximum value operation.

[0017] S3. Decode the differential signal Perform digital demodulation and dequantization to obtain downlink CSI recovery information And build an enhanced CSI network Enh-CsiNet to recover downlink CSI information Perform enhancement processing to obtain the downlink with enhanced recovery accuracy

[0018] The digital demodulation is to decode the differential signal Perform demodulation to obtain bit stream information

[0019] The dequantization process is based on the bit stream information obtained Perform the corresponding a-bit dequantization operation to obtain the downlink CSI recovery information

[0020] The enhanced CSI network Enh-CsiNEt described in step S3 further includes:

[0021] An input layer, a hidden layer with a Leaky Relu activation function, and an output layer with a linear activation function. The number of nodes in the input layer, hidden layer, and output layer are 2M, nM, and 2M respectively, and n represents the hidden layer node coefficient determined according to the engineering preset;

[0022] Building a training dataset The enhanced CSI network Enh-CsiNet is trained to obtain network parameters of the enhanced CSI network Enh-CsiNet;

[0023] The training labels It is obtained by converting the downlink CSI vector h into a real value, that is:

[0024] h label =[Re(h 1 ),Im(h 1 ),…,Re(h M ),Im(h M )] T

[0025] Among them, Im(·) means taking the imaginary part, [·] T Indicates the transpose operation;

[0026] When running online, the recovery information of the downlink CSI Enter the enhanced CSI network Enh-CsiNet to obtain downlink with enhanced recovery accuracy

[0027] The differential coding-assisted machine learning channel state information feedback device provided by the present invention, the user end device comprises:

[0028] A modulation module, which obtains a modulated signal s according to the design described in S1;

[0029] A differential encoding module, which obtains a differential encoding signal z according to the design described in S1;

[0030] The differential encoding module is connected after the modulation signal acquisition module.

[0031] The base station device includes:

[0032] The differential decoding module obtains the differential decoding signal according to the design described in S2

[0033] The precision enhancement module is designed according to S3 to obtain the downlink of the restored precision enhancement

[0034] The modulation module comprises:

[0035] Quantization unit, real-valued CSI information Perform a-bit quantization to obtain the bit stream information of the real and imaginary parts

[0036] Modulation unit, which converts the bit stream information Perform digital modulation to form a modulated signal

[0037] The precision enhancement module comprises:

[0038] Demodulation unit, which decodes the differential signal Perform demodulation to obtain bit stream information

[0039] The dequantization unit obtains the bit stream information Perform the corresponding a-bit dequantization operation to obtain the downlink CSI recovery information

[0040] The network enhancement unit recovers the downlink CSI information Input the enhanced CSI network Enh-CsiNet to obtain the downlink with enhanced recovery accuracy

[0041] The present invention utilizes differential coding to assist in CSI feedback, and further improves the reconstruction accuracy of downlink CSI through a lightweight enhanced CSI network Enh-CsiNet. In combination with the advantages of quantization coding, digital modulation and DL technology, the differentially coded signal after differential coding is fed back to the base station through OFDM modulation technology, and downlink CSI recovery information is obtained by differential decoding, digital demodulation and dequantization processing at the base station end. Then, the reconstruction accuracy of downlink CSI is further improved through a lightweight network enhanced Enh-CsiNet. While avoiding the UE end sending a guide for the BS end to perform channel estimation, uplink bandwidth resources and UE end energy are greatly saved, and the reconstruction accuracy of CSI is guaranteed.

[0042] Compared with the existing CSI feedback of uplink pilot channel estimation, the present invention saves uplink bandwidth resources and reduces user-side energy consumption by using differential coding technology, and improves the reconstruction accuracy of downlink CSI by using a lightweight neural network. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0044] Figure 2 It is a user end process design diagram of the present invention;

[0045] Figure 3 This is a flow chart of the base station end of the present invention. DETAILED DESCRIPTION

[0046] The present invention is described in detail below in conjunction with the embodiments and drawings, but it should be understood that the embodiments and drawings are only used to exemplify the present invention and do not constitute any limitation on the protection scope of the present invention. All reasonable changes and combinations within the scope of the inventive concept of the present invention fall within the protection scope of the present invention.

[0047] Reference Figure 1 , the specific differential coding assisted machine learning channel state information feedback method includes:

[0048] a1) Reference Figure 2 , the downlink CSI vector h is real-valued, quantized and digitally modulated into a modulated signal s, and then differentially encoded to obtain a differentially coded signal z, and the differentially coded signal z is OFDM modulated and fed back to the base station.

[0049] Among them, some more specific implementation methods are as follows:

[0050] The real-value conversion is to convert the downlink CSI vector h into real-valued CSI information Right now:

[0051]

[0052] Among them, Re(·) means taking the real part, Im(·) means taking the imaginary part, [·] T Indicates the transpose operation;

[0053] The quantization is to quantize the real-valued CSI information Perform a-bit quantization to obtain the real and imaginary bit stream information of the downlink CSI vector h

[0054] The digital modulation is to convert the bit stream information Perform digital modulation to form a modulated signal in is the modulation order (positive integer);

[0055] The differential encoding is to modulate the signal s = [s 1 ,s 2 ,...,s N ] T According to the rules Perform differential encoding to obtain the differential encoding signal z=[z 1 ,z 2 ,...z N ] T ;

[0056] Wherein, N = 2aMγ; the differential coding signal z is z agreed upon by the UE and the BS at the same time 0 =1, discard z 0 =1 after the differential encoding signal.

[0057] a2) Reference Figure 3 , the base station performs OFDM demodulation to obtain the received signal The received signal r is decoded character by character to obtain a differential decoded signal

[0058] The differential decoded signal is obtained by solving obtained;

[0059] in, represents the i-th decision value; r i 、r i-1 Respectively represent the i-th and i-1-th elements in r; s t Indicated in the constellation set The superscript * indicates the conjugate operation, Re[·] indicates the real part, and max[·] indicates the maximum value operation.

[0060] a3) Decode the differential signal Perform digital demodulation and dequantization to obtain downlink CSI recovery information And build an enhanced CSI network Enh-CsiNet to recover downlink CSI information Perform enhancement processing to obtain the downlink with enhanced recovery accuracy

[0061] The digital demodulation is to decode the differential signal Perform demodulation to obtain bit stream information

[0062] The dequantization process is based on the bit stream information obtained Perform the corresponding a-bit dequantization operation to restore the bit stream information to a real value and obtain the restored information of the downlink CSI

[0063] The enhanced CSI network Enh-CsiNEt further includes:

[0064] An input layer, a hidden layer with a Leaky ReLU activation function, and an output layer with a linear activation function. The number of nodes in the input layer, hidden layer, and output layer are 2M, nM, and 2M respectively, and n represents the hidden layer node coefficient determined according to the engineering preset;

[0065] Building a training dataset The enhanced CSI network Enh-CsiNet is trained to obtain network parameters of the enhanced CSI network Enh-CsiNet;

[0066] The training labels It is obtained by converting the downlink CSI vector h into a real value, that is:

[0067] h label =[Re(h 1 ),Im(h 1 ),…,Re(h M ),Im(h M )] T

[0068] The training loss function of the enhanced CSI network Enh-CsiNet adopts a mean square error loss function.

[0069] When running online, the recovery information of the downlink CSI Input the enhanced CSI network Enh-CsiNet, and finally obtain the downlink with enhanced recovery accuracy

[0070] The differential coding-assisted machine learning channel state information feedback device provided by the present invention, the user end device comprises:

[0071] A modulation module, which obtains a modulated signal s according to the design described in S1;

[0072] A differential encoding module, which obtains a differential encoding signal z according to the design described in S1;

[0073] The differential encoding module is connected after the modulation signal acquisition module.

[0074] The base station device includes:

[0075] The differential decoding module obtains the differential decoding signal according to the design described in S2

[0076] The precision enhancement module is designed according to S3 to obtain the downlink of the restored precision enhancement

[0077] The modulation module comprises:

[0078] Quantization unit, real-valued CSI information Perform a-bit quantization to obtain the bit stream information of the real and imaginary parts

[0079] Modulation unit, which converts the bit stream information Perform digital modulation to form a modulated signal

[0080] The precision enhancement module comprises:

[0081] Demodulation unit, which decodes the differential signal Perform demodulation to obtain bit stream information

[0082] The dequantization unit obtains the bit stream information Perform the corresponding a-bit dequantization operation to obtain the downlink CSI recovery information

[0083] The network enhancement unit recovers the downlink CSI information Input the enhanced CSI network Enh-CsiNet to obtain the downlink with enhanced recovery accuracy

[0084] Embodiment 1:

[0085] In step a1), a specific embodiment of differentially encoding the modulated signal s to obtain the differentially encoded signal z is as follows:

[0086] Taking QPSK modulation (modulation order γ = 2) and 4-bit quantization (a = 4) as an example, assuming that the length of the downlink CSI vector h is M = 2,

[0087] Then the length of the modulated signal s is N=8, and s is:

[0088]

[0089] According to the differential encoding rule The differential encoding signal z can be calculated as:

[0090]

[0091] Embodiment 2:

[0092] In step a2), the base station uses character-by-character differential decoding to obtain a differential decoded signal r. A specific embodiment is as follows:

[0093] Assume that the received signal According to the character-by-character differential decoding rules Perform differential decoding on the received signal r to obtain the differential decoded signal by The first element in Take this as an example, the specific solution process is as follows:

[0094] Constellation Set The CCP includes There are four constellation points. First, the first constellation point Substitute Re[s t ·r i * r i-1 ], the result is 1;

[0095] Then substitute the second to fourth constellation points in turn, and the results are 0, 0, and -1 respectively;

[0096] The result when substituting the first constellation point is the largest, and we get

[0097] Follow the same method to solve the remaining elements in r one by one, and the differential decoding signal is Finally, it can be expressed as:

[0098]

[0099] Embodiment 3:

[0100] In step a3), the downlink CSI recovery information Input into the enhanced CSI network Enh-CsiNet to obtain the downlink with enhanced recovery accuracy A specific embodiment is as follows:

[0101] Assume that downlink CSI recovery information for:

[0102]

[0103] The downlink CSI recovery information Input the enhanced CSI network Enh-CsiNet to obtain the real-valued downlink with enhanced accuracy for:

[0104]

[0105] It should be noted that those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the implementation methods of the present invention, and it should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed in the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. Differential coding assisted machine learning channel state information feedback method, It is characterized in that The following steps are involved: User-side processing: S1. Downlink CSI vector After real-valued, quantized and digitally modulated into a modulated signal s, differential encoding is performed to obtain a differentially coded signal z, and the differentially coded signal z is OFDM modulated and fed back to the base station; The real-value conversion is to convert the downlink CSI vector h into real-valued CSI information The quantization is to measure the real-valued CSI information Perform a-bit quantization to obtain the bit stream information of the real and imaginary parts The digital modulation is to convert the bit stream information Perform digital modulation to form a modulated signal in, is the modulation order, which is a positive integer; The differential encoding is to modulate the signal s = [s 1 ,s 2 ,...,s N ] T According to the rules Perform differential encoding to obtain the differential encoding signal z=[z 1 ,z 2 ,...z N ] T ; Wherein, N = 2aM / γ, the differential encoding signal z is the discarded z 0 =1 after the differential encoding signal; Base station processing: S2. The base station performs OFDM demodulation to obtain the received signal The received signal r is decoded character by character to obtain a differential decoded signal The differential decoding is achieved by solving get; in, represents the i-th decision value; r i 、r i-1 Respectively represent the i-th and i-1-th elements in r; s t Indicated in the constellation set The trial constellation point values ​​in , the superscript * indicates the conjugate operation, Re[·] indicates the real part, and max[·] indicates the maximum value operation; S3. Decode the differential signal Perform digital demodulation and dequantization to obtain downlink CSI recovery information And build an enhanced CSI network Enh-CsiNet to recover downlink CSI information Perform enhancement processing to obtain downlink CSI with enhanced recovery accuracy The digital demodulation is to decode the differential signal Perform demodulation to obtain bit stream information The dequantization process is based on the bit stream information obtained Perform the corresponding a-bit dequantization operation to obtain the downlink CSI recovery information 2. The differential coding-assisted machine learning channel state information feedback method according to claim 1, It is characterized in that The enhanced CSI network Enh-CsiNEt described in step S3 further includes: An input layer, a hidden layer with a Leaky Relu activation function, and an output layer with a linear activation function; the number of nodes in the input layer, hidden layer, and output layer are 2M, nM, and 2M respectively, and n represents the hidden layer node coefficient determined according to the engineering preset; Building a training dataset The enhanced CSI network Enh-CsiNet is trained to obtain network parameters of the enhanced CSI network Enh-CsiNet; The training labels It is obtained by converting the downlink CSI vector h into a real value, that is: h label =[Re(h 1 ),Im(h 1 ),…,Re(h M ),Im(h M )] T Among them, Im(·) means taking the imaginary part, [·] T Indicates the transpose operation; When running online, the recovery information of the downlink CSI Input the enhanced CSI network Enh-CsiNet to obtain downlink CSI with enhanced recovery accuracy 3. Differential coding assisted machine learning channel state information feedback device, It is characterized in that Used to implement the differential coding assisted machine learning channel state information feedback method according to claim 1 or 2; The client devices include: A modulation module, which obtains a modulated signal s according to the design of S1 in claim 1; The differential encoding module obtains the differential encoding signal z according to the design described in S1 of claim 1.

4. The differential coding-assisted machine learning channel state information feedback device according to claim 3, It is characterized in that The modulation module comprises: Quantization unit, real-valued CSI information Perform a-bit quantization to obtain the bit stream information of the real and imaginary parts Modulation unit, which converts the bit stream information Perform digital modulation to form a modulated signal 5. The differential coding-assisted machine learning channel state information feedback device according to claim 3, It is characterized in that The base station device includes: A differential decoding module, which obtains a differential decoding signal according to the design described in S2 of claim 1 The precision enhancement module is configured to obtain a downlink CSI with enhanced recovery precision according to the design of S3 in claim 1.

6. The differential coding-assisted machine learning channel state information feedback device according to claim 5, It is characterized in that The precision enhancement module comprises: Demodulation unit, which decodes the differential signal Perform demodulation to obtain bit stream information The dequantization unit obtains the bit stream information Perform the corresponding a-bit dequantization operation to obtain the downlink CSI recovery information The network enhancement unit recovers the downlink CSI information Input the enhanced CSI network Enh-CsiNet to obtain the downlink CSI with enhanced recovery accuracy

Citation Information

Patent Citations

  • Superposition coding CSI feedback method based on deep learning

    CN110166089A

  • CSI feedback and receiving method and device, and storage medium

    CN114070374A