A massive MIMO beam training method and apparatus

Abstract

This application discloses a large-scale MIMO beam training method and apparatus. By constructing a shared training system consisting of a basic DFT codebook, a constant-multiple-code target response matrix, and its corresponding combined coding codebook, and performing maximum a posteriori decoding based on the energy observation sequence, it is possible to accurately select the beamforming vector with maximum gain with high probability even under time-delay constraints with a limited number of training rounds. Compared with traditional exhaustive search methods, this significantly reduces the number of queries in beam training; compared with binary search and convolutional coding search schemes, it improves decoding accuracy with the same number of training rounds, thus achieving a beam training performance improvement that balances low latency and high accuracy.

Description

Technical Field

[0001] This application relates to, but is not limited to, the field of communication technology, and in particular to a large-scale MIMO beam training method and apparatus. Background Technology

[0002] In ultra-large-scale multiple-input multiple-output (XL-MIMO) systems for future sixth-generation (6G) communication, the electromagnetic field environment and time delay constraints caused by changes in antenna aperture and number of antennas are different, making the traditional beam training method in MIMO no longer applicable. It is necessary to design a beam training scheme with acceptable complexity in a suitable modeling environment combined with new time delay constraints. Summary of the Invention

[0003] This application provides a method and apparatus for large-scale MIMO beam training, which can achieve beam training with low latency and high accuracy, and can meet the needs of future wireless communication systems.

[0004] This invention provides a method for large-scale MIMO beam training, comprising:

[0005] The sending and receiving ends generate and share a training codebook system, which includes the basic DFT codebook. The target response matrix constructed using constant multiple codes and the combined codebook corresponding to the target response matrix. ;

[0006] The transmitting end sends training signals round by round according to the combined coding codebook;

[0007] The receiver receives the training signal and records the energy, accumulating observation data for decoding.

[0008] The receiver uses the accumulated observation data and target response matrix to perform maximum a posteriori decoding and select the optimal beam direction;

[0009] The receiver feeds back the optimal beam direction to the transmitter, which then locks the beam vector based on the optimal beam direction for subsequent communication.

[0010] In one exemplary instance, the target response matrix constructed using constant multiple codes... Satisfy: In each row The number of vectors should be kept as small as possible to concentrate the energy of the training beam and make the signal easy to distinguish; the minimum Hamming distance between different column vectors should be large to make the response modes in different directions significantly different.

[0011] In one exemplary instance, the target response matrix The design adopts a column constant code structure, including:

[0012] The target response matrix Each column includes One element has a value of 1, and the rest are 0. It is a pre-set value;

[0013] The target response matrix Each column can be considered as a column of length . of Constant duplicate codes; from all candidates Selection of constant multiple codes from the set The target response matrix is ​​composed of codewords. .

[0014] In one exemplary instance, the combined codebook is based on the target response matrix. Generation, including:

[0015] For the basic DFT codebook The combination is performed to make the receiver's response approximate the ideal response, so that the shared combined codebook satisfies... At this time, the first The vector sent in the round-robin query is The Transpose of a line.

[0016] In one exemplary instance, the receiver receives training signals and records energy to accumulate observation data for decoding, including:

[0017] The receiving end uses a receiving vector. Perform reception and analyze the received codewords. The energy is recorded, and the signal energy is recorded as... ;go through Round of queries, the recorded energy sequence is The energy sequence As the observation data accumulated for decoding;

[0018] in, For the first The codewords received in the round.

[0019] In one exemplary instance, selecting the optimal beam direction includes:

[0020] Using the received signal energy sequence and threshold parameters Calculate the likelihood ratio (LLR);

[0021] Accumulate and compare the LLRs of all rounds across each candidate direction, and then use maximum a posteriori decoding on the target response matrix. All Compare the column codewords corresponding to each candidate direction, and select the beam direction corresponding to the column codeword with the largest cumulative likelihood value. And take it as the optimal beam direction.

[0022] In one exemplary instance, the receiver feeds back the optimal beam direction to the transmitter, including:

[0023] The receiver selects the optimal beam direction to obtain the transmission vector in the DFT codebook. The signal is then returned to the receiving end as the beamforming vector used for communication.

[0024] This application also provides a computer-readable storage medium storing computer-executable instructions for executing any of the above-described large-scale MIMO beam training methods.

[0025] This application embodiment further provides a computer device, including a memory and a processor, wherein the memory stores the following instructions executable by the processor: steps for performing the large-scale MIMO beam training method described in any of the above claims.

[0026] This application also provides a large-scale MIMO beam training device, applied at the transmitting end, including:

[0027] The negotiation module is used to generate and share the training codebook system with the receiving end. The training codebook system includes the basic DFT codebook, the target response matrix constructed using constant heavy codes, and the combined coding codebook corresponding to the target response matrix.

[0028] The polling processing module is used to send training signals round by round according to the combined encoding codebook;

[0029] The first processing module is used to receive the optimal beam direction from the transmitter and lock the beam vector according to the optimal beam direction for subsequent communication.

[0030] This application also provides a large-scale MIMO beam training device for use at a receiver, comprising:

[0031] The second processing module is used to receive training signals from the transmitter and record energy, accumulating observation data for decoding;

[0032] The determination module is used to perform maximum a posteriori decoding using accumulated observation data and the target response matrix to select the optimal beam direction;

[0033] The feedback module is used to feed back the optimal beam direction to the transmitter.

[0034] The large-scale MIMO beamforming method provided in this application constructs a shared training system consisting of a basic DFT codebook, a constant-multiple-code target response matrix, and its corresponding combined coding codebook. It then performs maximum a posteriori decoding based on energy observation sequences. This allows for the accurate selection of the beamforming vector with maximum gain with high probability, even under time constraints with limited training rounds. Compared to traditional exhaustive search methods, this significantly reduces the number of queries during beam training. Compared to binary search and convolutional coding search schemes, it improves decoding accuracy with the same number of training rounds, thus achieving a beamforming performance improvement that balances low latency and high accuracy.

[0035] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the description, claims, and drawings. Attached Figure Description

[0036] The accompanying drawings are used to provide a further understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0037] Figure 1 This is a flowchart illustrating the large-scale MIMO beam training method in an embodiment of this application.

[0038] Figure 2 This is a schematic diagram of the structural composition of one embodiment of the large-scale MIMO beam training device in this application.

[0039] Figure 3 This is a schematic diagram of the composition structure of another embodiment of the large-scale MIMO beam training device in this application.

[0040] Figure 4 This is a schematic diagram comparing the misselection rate of different beam training methods under different signal-to-noise ratio conditions in the embodiments of this application. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be arbitrarily combined with each other.

[0042] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings, which illustrate embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of this application will be thorough and complete.

[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0044] It is understood that the terms "first" and "second" used in this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0045] It is understood that the term "connection" in the following embodiments should be understood as "electrical connection," "communication connection," etc., if the connected circuits, modules, units, etc., have electrical signal or data transmission with each other.

[0046] When used herein, the singular forms of “a,” “an,” and “the” may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “comprising / including” or “having,” etc., specify the presence of the stated features, wholes, steps, operations, components, parts, or combinations thereof, but do not preclude the possibility of the presence or addition of one or more other features, wholes, steps, operations, components, parts, or combinations thereof. Meanwhile, the term “and / or” as used in this specification includes any and all combinations of the associated listed items.

[0047] The steps illustrated in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases the steps shown or described may be performed in a different order than that presented here.

[0048] In traditional MIMO systems, dual transmission and reception require the design of the antenna array's amplitude and phase to achieve maximum beam gain. The main steps include: the transmitter using beamforming vectors... For transmission symbols Shaping is performed to obtain the transmitted signal. In the channel The transmission is performed at the upper end, and the receiving end uses a receiving shaped vector. The received signal is obtained by shaping the received signal. To achieve optimal communication performance, the normalized transmit beamforming vector needs to be optimized. and receiving the shaping vector Specific design is carried out to improve beam gain Maximizing this process is called beamforming, where the beamforming vector at the transmitting end... and the receiving shaping vector at the receiving end The selection range is based on the traditional DFT codebook. In far-field scenarios using ULA antennas, the DFT codebook can be given according to the resolution corresponding to the number of antennas at both the transmitting and receiving ends.

[0049] According to the IEEE 802.11ad standard, current search methods include exhaustive search and interactive search, with the base station having... One antenna, with one on the user side. Taking a single antenna as an example: exhaustive search involves sending and testing all codewords in the codebook. The base station needs to repeat this process. Each test This code, the user side faces this In the interactive search, the user side activates only one antenna for base station training, and the user side is trained only after the base station's optimal antenna is fixed. Both specifications assume that the user-side beam is fixed, so the optimization problem can be simplified to determining the optimal codeword for the base station, i.e., assuming... The optimal value (in other words, the receiver beamforming vector is considered to be optimal) (Always in an ideal state), for example, using the example Select Make Maximum. Grouping or encoding design based on the basic codebook can increase the number of directions detected per burst, thereby reducing the number of query rounds and meeting practical needs. Common beam training schemes include gauge-based exhaustive search, the subsequently proposed binary search, group search, and the recently proposed encoding search scheme. Increasing the number of directions detected per burst leads to a decrease in energy in each direction, thus reducing the accuracy of resolution. The main purpose of beam training is to improve the accuracy of selecting beamforming vectors while satisfying the time delay (i.e., the number of query rounds) constraint.

[0050] In theory, if the channel matrix Singular value decomposition (SVD) can yield the optimal transmit and receive beam vectors. However, this method is computationally extremely expensive and difficult to implement in practical systems. Therefore, existing beam training methods typically rely on DFT codebooks, mainly including three categories: exhaustive search, hierarchical search, and encoding search.

[0051] The basic approach of exhaustive search is to traverse all codewords in the DFT codebook. Its advantage is extremely high accuracy, and theoretically, it can find the globally optimal beam. However, its drawbacks are also quite obvious: the number of query rounds is exactly the same as the codebook size, which equals the number of antennas. In XL-MIMO systems, the number of antennas increases dramatically from 64-128 in 5G base stations to 512 or even thousands. At this point, the codebook capacity increases drastically, and exhaustive search leads to unacceptable training latency.

[0052] Typical methods for hierarchical search include binary search and group search. Binary search can eliminate half of the remaining directions each time, which is highly efficient. However, in the early stages of training, due to the low energy of a single direction, it is prone to misclassification, thus reducing overall accuracy. In addition, its process relies on frequent uplink feedback, which introduces additional latency when uplink bandwidth is limited. Group search first determines the group and then gradually subdivides within the group. It has high overall accuracy, but the number of queries is still linearly related to the number of antennas, and the latency is still too high in large-scale arrays.

[0053] Encoding search methods attempt to enhance misjudgment correction capabilities by introducing redundant encoding (such as convolutional codes), thereby maintaining high efficiency while ensuring a certain level of accuracy. Although its performance is improved compared to hierarchical search, the number of training rounds is still related to the encoding method, and generally remains at a linear level similar to binary search. While its performance is better than binary search, when using traditional encoding methods (such as Hamming codes, convolutional codes, etc.), its decision threshold is still low (probing half the direction in a single attempt leads to energy dispersion), and accuracy issues remain.

[0054] In XL-MIMO scenarios, these problems are amplified. For example, in a far-field ULA system with 1024 antennas, the DFT codebook contains 1024 candidate vectors, and the training overhead is unacceptable whether exhaustive search or group search is used. At the same time, high-level codebooks have poor decision thresholds under low signal-to-noise ratio conditions, leading to a decrease in selection accuracy; in multicast user scenarios, existing solutions also struggle to meet the needs of simultaneous training for multiple users.

[0055] In other words, while beam training methods in related technologies can be applied to small-scale MIMO systems, they face a significant contradiction between latency and accuracy in XL-MIMO. To maintain high accuracy while meeting low latency constraints, this application provides a large-scale MIMO beam training method that can be used for far-field beam training in XL-MIMO systems, meeting the needs of future wireless communication systems.

[0056] Figure 1 This is a flowchart illustrating the large-scale MIMO beam training method in an embodiment of this application, as shown below. Figure 1 As shown, it may include:

[0057] Step 100: The transmitting end and the receiving end generate and share a training codebook system. The training codebook system includes a basic DFT codebook, a target response matrix constructed using constant multiple codes, and a combined coding codebook corresponding to the target response matrix.

[0058] In one exemplary instance, step 100 may include:

[0059] The transmitting and receiving ends (e.g., base station and user terminal) negotiate and agree on the basic codebook and target response codebook required for training. The basic codebook is a traditional Discrete Fourier Transform (DFT) codebook. Its size is ,in, This refers to the number of antennas at the transmitting end (e.g., a base station). DFT codebook Its purpose is to provide optional beamforming vectors for subsequent beam training.

[0060] Based on this, the sending end designs the target response matrix. Its size is ,in, This represents the number of query rounds during training. Target response matrix. By employing constant-multiple-code construction, we ensure that each column has the same number of 1s (column constant-multiple-code property) while maintaining a large Hamming distance between different column vectors, thereby improving decoding accuracy with a limited number of query rounds.

[0061] Subsequently, based on the target response matrix Generate a combined codebook Combined encoding codebook The size is also Combined encoding codebook By linearly combining the columns of the DFT codebook, the receiver's response is made to approximate the target response matrix. The defined ideal energy distribution. At this point, the sender and receiver have completed the codebook sharing, laying the foundation for subsequent training.

[0062] In one embodiment, the basic DFT codebook is as shown in formula (1): (1)

[0063] As can be seen from formula (1), the codebook consists of multiple vectors. Composition, each vector corresponds to an angle. These angles are not chosen arbitrarily, but calculated based on uniformly distributed cosine values. Formula (1), single beam vector , The signals output by each antenna are sequentially separated by one phase. In this way, the signal will be amplified in a certain direction to form a beam; This indicates that normalization is being performed to ensure that the transmission power remains constant. This means that all these beam vectors are pairwise orthogonal and do not interfere with each other.

[0064] In this embodiment of the application, the basic DFT codebook is derived from... The antenna array consists of several standard beam vectors, each generated by applying equal-amplitude, phase-by-phase complex exponential weights to the antenna array, ensuring that all beams are orthogonal to each other and do not interfere with each other.

[0065] In one exemplary instance, the target response matrix is: , The bank representative The ideal response at each angle defines which beam directions participate in signal superposition during each round of training, thereby controlling the energy distribution pattern at the receiver. Indicates the number of query rounds during training. This represents the total number of candidate beam directions. This is important for training a limited number of rounds. It still maintains high decoding accuracy, and the target response matrix is It needs to satisfy: In each row The number of columns represents the number of energy distributions. A smaller number (i.e., concentrated training beam energy, easily distinguishable signals) and a larger minimum Hamming distance between different column vectors (i.e., significant differences in response patterns in different directions, facilitating decoding and discrimination) improve detection accuracy. To meet these requirements, in one embodiment, the target response matrix... A column-constant weight code structure can be used, where each column contains exactly one column of constant weight code. One element has a value of 1, and the rest are 0. These are pre-set values; the number of 1s in each column remains the same to ensure a balanced number of activations in each beam direction during training; each column of the matrix can be considered as a column of length... of Constant duplicate codes; from all candidates Selection of constant multiple codes from the set The target response matrix is ​​composed of codewords. The methods of construction may include:

[0066] First, a set of candidate constant codes is generated in a structured manner.

[0067] Structured generation indivual The codeword consists of two parts, one part being... The frequently repeated codewords, part of which are The full permutation of the codewords is combined into a total of 100. indivual The minimum Hamming distance for constant-multiplexed codewords is .in, The generation of constant codewords includes: selecting Simplex code (i.e., the dual code of Hamming code). , for of A set consisting of cyclic shifts, Each of the digits corresponds to Each The cyclic shift generates The frequently repeated code words.

[0068] Then, the optimization and screening process forms the final target response matrix. .

[0069] Using an optimized filtering method, firstly in Select from the constant homocode characters Then, based on the minimum Hamming distance not decreasing, the codewords in the codebook are gradually replaced iteratively. The end of the round is used as the target response matrix. .

[0070] In one embodiment, based on the target response matrix Generate a combined codebook This can include: combining the basic DFT codebook to make the receiver's response approximate the ideal response, so that the shared combined coded codebook satisfies At this time, the first The vector sent in the round-robin query is The Transpose of a line.

[0071] For the target response matrix The Okay, if the first one If the nth element is equal to 1, then the nth element in the basic DFT codebook will be... The column codewords participate in the construction of the training beam in this round, that is, the training beam is formed by linearly superimposing the DFT column vectors, so that the observation energy obtained by the receiver in the direction of the beam is similar to the target response matrix. The expected response distribution is consistent.

[0072] In one embodiment, The target response matrix is ​​a pre-defined constant. The number of elements with a value of 1 in each column represents the number of times each beam direction is activated during the entire training process. The value of and the number of training rounds Number of antennas The system's requirements for misjudgment probability and training latency are related, and the appropriate option can be selected based on the actual application scenario and the above constraints.

[0073] Step 101: The transmitting end sends training signals round by round according to the combined coding codebook.

[0074] In one exemplary instance, during the training phase, the sending end performs a total of Downlink transmission in round 1. In the first... During rounds, the base station retrieves the combined coding codebook. Selected from The line is then taken, and its transpose is normalized and used as the transmission vector. In this way, the signal transmitted by the base station in each round corresponds to a specific combination of training directions.

[0075] The combined codebook formed in step 100 The training direction for each round has been determined, and step 101 is to specifically transmit it into the channel.

[0076] Step 102: The receiver receives the training signal and records the energy to accumulate observation data for decoding.

[0077] In one exemplary instance, the sender transmits a vector. via channel Transmitted to the receiving end (such as the user end), receiving the signal (i.e., the first) The codeword received in the round is: ,in, This is noise. Because the receiver assumes only a single antenna is used (i.e., the receive vector...). Therefore, the receiving end only needs to perform statistical analysis on the received signal energy to obtain: ,Finish After one round of training, the receiver obtains the energy sequence: The energy sequence The observational data accumulated for decoding will serve as the basis for decoding decisions in subsequent steps.

[0078] In one embodiment, the receiver employs a receive vector. Perform reception and analyze the received codewords. The energy is recorded, that is Specifically, this can include:

[0079] Analog channel is ,in, for Path, the rest are indivual path, Represents the multi-link gain. , As the starting angle, Angle of arrival. (The first...) The codeword received in the round Energy record as .go through Round of queries, the recorded energy sequence is .

[0080] Through steps 101 and 102, the training signal from the transmitting end is transmitted in the channel, and the receiving end retains the training information through energy recording, providing observation data for decoding.

[0081] Step 103: The receiver uses the accumulated observation data and target response matrix to perform maximum a posteriori decoding and select the optimal beam direction.

[0082] In one exemplary instance, the user terminal uses a shared target response matrix. and the received energy sequence The maximum a posteriori probability (MAP) criterion is used for decoding to determine the optimal beam direction. In one embodiment, it may include:

[0083] First, the likelihood ratio (LLR) is calculated based on the energy decision, using the received signal energy sequence. and threshold parameters Calculate LLR to obtain: ,in, For zero-order modified Bessel functions, For noise variance, For the first The energy recorded by the wheel.

[0084] Then, the LLRs from all rounds are accumulated and compared across all candidate directions to select the most likely beam direction. This can include: employing maximum a posteriori decoding on the target response matrix. All Compare the column codewords corresponding to each candidate direction, and select the beam direction corresponding to the column codeword with the largest cumulative likelihood value. And use it as the optimal beam direction.

[0085] For the target response matrix Each column ,Right now It is the first one List the codewords and accumulate their corresponding energy decision values: , Represents the target response matrix The line, number Column elements When; maximize selection column index That is, to obtain the angle corresponding to the optimal beam direction. .

[0086] Analysis of the above maximum a posteriori decoding reveals the relationship between the misclassification probability and the energy threshold. And related to the Hamming distance between different columns.

[0087] In this step, the receiver uses the recorded energy sequence to complete the decoding, and the maximum a posteriori criterion ensures that high accuracy can still be maintained under a limited number of interrogation rounds.

[0088] Step 104: The receiver feeds back the optimal beam direction to the transmitter, and the transmitter locks the beam vector according to the optimal beam direction for subsequent communication.

[0089] In one exemplary instance, the receiver selects the beam direction. The corresponding transmission vector in the DFT codebook is obtained. It is then returned to the receiving end as the beamforming vector used for communication.

[0090] In this step, the user terminal determines the optimal beamforming vector based on the decoding results. That is, the angle in the DFT codebook corresponding to the optimal beam direction. The corresponding vector. The receiver returns this vector index to the transmitter, which then selects the beamforming vector for subsequent communication.

[0091] Through this step, the angle obtained by decoding directly corresponds to the codeword in the DFT codebook. After being returned to the receiving end, the beam training process ends, and communication enters the normal data transmission stage.

[0092] The large-scale MIMO beam training method provided in this application constructs a shared training system consisting of a basic DFT codebook, a constant-weighted target response matrix, and its corresponding combined coding codebook, and performs maximum a posteriori decoding based on the energy observation sequence, thereby enabling training with a limited number of rounds. Under time delay constraints, it can still accurately select the beamforming vector with maximum gain with high probability. Compared with traditional exhaustive search methods, it significantly reduces the number of queries for beam training; compared with binary search and convolutional coding search schemes, it improves decoding accuracy with the same number of training rounds, thus achieving a beam training performance improvement that balances low latency and high accuracy.

[0093] This application also provides a computer-readable storage medium storing computer-executable instructions for performing any of the above-described large-scale MIMO beam training methods.

[0094] This application further provides a computer device, including a memory and a processor, wherein the memory stores the following instructions executable by the processor: steps for performing the large-scale MIMO beam training method described in any of the preceding claims.

[0095] Figure 2 This is a schematic diagram of the structural composition of an embodiment of a large-scale MIMO beam training device in this application, applied to the transmitting end, such as... Figure 2 As shown, it may include:

[0096] The negotiation module is used to generate and share the training codebook system with the receiving end. The training codebook system includes the basic DFT codebook, the target response matrix constructed using constant heavy codes, and the combined coding codebook corresponding to the target response matrix.

[0097] The polling processing module is used to send training signals round by round according to the combined encoding codebook;

[0098] The first processing module is used to receive the optimal beam direction from the transmitter and lock the beam vector according to the optimal beam direction for subsequent communication.

[0099] In one exemplary instance, the sender is a base station.

[0100] Figure 3 This is a schematic diagram of the structural composition of another embodiment of the large-scale MIMO beam training device in this application, applied to the receiving end, such as... Figure 3 As shown, it may include:

[0101] The second processing module is used to receive training signals from the transmitter and record energy, accumulating observation data for decoding;

[0102] The determination module is used to perform maximum a posteriori decoding using accumulated observation data and the target response matrix to select the optimal beam direction;

[0103] The feedback module is used to feed back the optimal beam direction to the transmitter.

[0104] It should be noted that, in one embodiment, at least a portion of the basic DFT codebook, the target response matrix, and the combined encoded transmission codebook can be pre-generated by the transmitter and sent to the receiver via control signaling; or it can be locally generated by the transmitter and receiver respectively according to pre-agreed construction rules, thereby forming a consistent shared training codebook system.

[0105] In one exemplary instance, the receiving end is the user end.

[0106] The large-scale MIMO beam training device provided in this application constructs a shared training system consisting of a basic DFT codebook, a constant-weighted target response matrix, and its corresponding combined coding codebook, and performs maximum a posteriori decoding based on energy observation sequences, thereby enabling training rounds with a limited number of training rounds. Under time delay constraints, it can still accurately select the beamforming vector with maximum gain with high probability. Compared with traditional exhaustive search methods, it significantly reduces the number of queries for beam training; compared with binary search and convolutional coding search schemes, it improves decoding accuracy with the same number of training rounds, thus achieving a beam training performance improvement that balances low latency and high accuracy.

[0107] The following describes in detail the large-scale MIMO beam training method under low signal-to-noise ratio conditions provided in this application embodiment, with reference to an example.

[0108] The number of antennas corresponding to massive MIMO Taking beam training simulation testing as an example. Assume that the base station and 1000 users are each undergoing beam training, and the channel signal-to-noise ratio is designed to be... dB.

[0109] First, channel initialization is performed. For users... ,exist Randomly select the starting angle within the range Analog Channel .

[0110] Next, the codebook is initialized. Before training begins, the two communicating parties negotiate and share the basic DFT codebook. Target response matrix and combined codebook .in,

[0111] Basic DFT codebook ,in, .

[0112] The design of a constant target response matrix includes: taking , , , A total of 1456 (20,6) constant multiple codes were generated, and 1024 of them were optimized and selected as the ideal target response matrix. Considering the low total sample size, the number of screening rounds is set to 2;

[0113] Take the same , , , A total of 23,936 (40,6) constant multiple codes were generated, and 1,024 of them were optimized and selected as the ideal target response matrix. The filter rounds are set to 20.

[0114] Encoding and sending codebook The generation includes: based on , respectively obtained and As a combined codebook.

[0115] Then, send Round of test vectors, the first During round-robin queries, the base station sends information vectors. for The The normalized vector of the transpose of a row.

[0116] After that, the user client's first The round receive codeword is Record the energy sequence as .

[0117] Then, maximum a posteriori decoding is used to select the transmission shaping vector with the largest gain:

[0118] calculate The List corresponding value;

[0119] The largest of the 1024 calculated values ​​is selected as the most likely angle. Corresponding vector .

[0120] Finally, the selected send shaping vector is returned. As the transmission shaping vector in the communication process,

[0121] In this embodiment, the returned vector is compared with the user's starting angle range during the test initialization phase to accumulate whether the correct maximum gain shaping vector has been selected. The selected error rate is compared with exhaustive search, binary search, and convolutional coding search. The results of this embodiment are shown in Table 1. Through software simulation based on Matlab, the simulation results are as follows: Figure 4 As shown.

[0122] The latency overhead (in terms of the number of query rounds) is shown in Table 1:

[0123]

[0124] Table 1

[0125] As shown in Table 1, this invention maintains a certain number of query rounds. Level (e.g., (20,6) constant key corresponds to) The constant multiple code (40,6) corresponds to: Under these conditions, beam selection accuracy close to exhaustive search can be obtained at SNR of 5dB and 0dB respectively, and the bit error rate is lower than that of binary codebook and convolutional code schemes at the same number of rounds.

[0126] Figure 4 This is a schematic diagram comparing the misselection rate of different beam training methods under different signal-to-noise ratio conditions in the embodiments of this application, as shown below. Figure 4 As shown, the (20,6) constant multiple code scheme and the (40,6) constant multiple code scheme proposed in the embodiments of this application (such as...) Figure 4 The LWC-20 and LWC-40 configurations significantly outperformed both binary search (BIN) and convolutional code search (CONV) across the entire SNR range, and maintained a low misselection rate even in the low-to-medium signal-to-noise ratio region (SNR ≤ 0 dB). Among them, the (40,6) configuration, due to its higher number of training rounds, had an overall performance closer to the exhaustive search (ES) benchmark. Figure 4 The horizontal axis represents the signal-to-noise ratio (SNR), and the vertical axis represents the misselection rate (misrate).

[0127] As can be seen, when the required number of query rounds reaches... Under the time delay constraint of the same level, the response design of this scheme can effectively reduce the error rate of maximum gain beam selection when the number of query rounds is the same, and increasing the number of query rounds can also significantly improve the accuracy of training design based on constant multiple codes.

[0128] The large-scale MIMO beamforming method provided in this application can more accurately select the beamforming vector with maximum gain under time delay constraints. Specifically, the constant-multiplexity code-based coded beamforming scheme, compared to exhaustive search, reduces the number of query rounds from... Reduced to linear The algorithm achieves a certain level of accuracy while maintaining a certain level of precision. Compared to binary search and convolutional coding search, it improves accuracy with the same number of query rounds. In the design of the response matrix based on constant-multiple-code, the trade-off between the number of query rounds and the accuracy of selecting the maximum-gain beamforming vector is characterized. It presents that, under the constraint of time delay (query discussion) in large-scale MIMO beam training, the optimization direction of the response matrix design is to minimize code weight and maximize code distance. Finally, a practical use case is given using constant-multiple-code encoding.

[0129] Although the embodiments disclosed in this application are as described above, the content described is merely for the purpose of understanding this application and is not intended to limit this application. Any person skilled in the art to which this application pertains may make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in this application; however, the scope of patent protection of this application shall still be determined by the scope defined in the appended claims.

Claims

1. A method for training large-scale MIMO beams, characterized in that, include: The sending and receiving ends generate and share a training codebook system, which includes the basic DFT codebook. The target response matrix constructed using constant multiple codes and the combined codebook corresponding to the target response matrix. ; The transmitting end sends training signals round by round according to the combined coding codebook; The receiver receives the training signal and records the energy, accumulating observation data for decoding. The receiver uses the accumulated observation data and target response matrix to perform maximum a posteriori decoding and select the optimal beam direction; The receiver feeds back the optimal beam direction to the transmitter, and the transmitter locks the beam vector based on the optimal beam direction for subsequent communication. The target response matrix constructed using constant multiple codes is described above. Satisfy: In each row The number of vectors should be kept as small as possible to concentrate the energy of the training beam and make the signal easy to distinguish; the minimum Hamming distance between different column vectors should be large to make the response patterns in different directions significantly different. The combined codebook is generated according to a target response matrix generating, comprising: For the basic DFT codebook The combination is performed to make the receiver's response approximate the ideal response, so that the shared combined codebook satisfies... At this time, the first The vector sent in the round-robin query is The Transpose of a line.

2. The massive MIMO beam training method of claim 1, wherein, The target response matrix The design adopts a column constant code structure, including: The target response matrix Each column includes One element has a value of 1, and the rest are 0. It is a pre-set value; The target response matrix Each column can be considered as a column of length . of Constant duplicate codes; from all candidates Selection of constant multiple codes from the set The target response matrix is ​​composed of codewords. .

3. The massive MIMO beam training method of claim 1, wherein, The receiving end receives the training signal and records the energy, accumulating observation data for decoding, including: The receiving end uses a receiving vector. Perform reception and process the received codewords. The energy is recorded, and the signal energy is recorded as... ;go through Round of queries, the recorded energy sequence is The energy sequence This serves as the observation data accumulated for decoding; in, For the first The codewords received in the round.

4. The large-scale MIMO beam training method according to claim 3, wherein, The selection of the optimal beam direction includes: Using the received signal energy sequence and threshold parameters Calculate the likelihood ratio (LLR); Accumulate and compare the LLRs of all rounds across each candidate direction, and then use maximum a posteriori decoding on the target response matrix. All Compare the column codewords corresponding to each candidate direction, and select the beam direction corresponding to the column codeword with the largest cumulative likelihood value. And take it as the optimal beam direction.

5. The massive MIMO beam training method of claim 1, wherein, The receiver feeds back the optimal beam direction to the transmitter, including: The receiver selects the optimal beam direction to obtain the transmission vector in the DFT codebook. The signal is then returned to the receiving end as the beamforming vector used for communication.

6. A computer-readable storage medium storing computer-executable instructions for performing the large-scale MIMO beam training method according to any one of claims 1-5.

7. A large-scale MIMO beam training device, characterized in that, Applied to the sending end, including: The negotiation module is used to generate and share a training codebook system with the receiving end. The training codebook system includes a basic DFT codebook, a target response matrix constructed using constant-multiple codes, and a combined encoding codebook corresponding to the target response matrix; wherein, the target response matrix constructed using constant-multiple codes... Satisfy: In each row The number of columns should be kept as small as possible to concentrate the training beam energy and make the signals easy to distinguish; the minimum Hamming distance between different column vectors should be large to make the response patterns in different directions significantly different; the combined coding codebook is based on the target response matrix. Generation, including: the underlying DFT codebook The combination is performed to make the receiver's response approximate the ideal response, so that the shared combined codebook satisfies... At this time, the first The vector sent in the round-robin query is The transpose of a line; The polling processing module is used to send training signals round by round according to the combined encoding codebook; The first processing module is used to receive the optimal beam direction from the transmitter and lock the beam vector according to the optimal beam direction for subsequent communication.

8. A large-scale MIMO beam training device, characterized in that, Applied to the receiving end, including: The second processing module is used to receive training signals sent round by round from the transmitter according to the combined coding codebook and record the energy, so as to accumulate observation data for decoding; The determination module is used to perform maximum a posteriori decoding using accumulated observation data and the target response matrix to select the optimal beam direction; wherein, the target response matrix... It is constructed using constant multiple codes, satisfying: in each row The number of columns should be kept as small as possible to concentrate the training beam energy and make the signals easy to distinguish; the minimum Hamming distance between different column vectors should be large to make the response patterns in different directions significantly different; the combined coding codebook is based on the target response matrix. Generation, including: the underlying DFT codebook The combination is performed to make the receiver's response approximate the ideal response, so that the shared combined codebook satisfies... At this time, the first The vector sent in the round-robin query is The transpose of a line; The feedback module is used to feed back the optimal beam direction to the transmitter.

Images