Alternating Training Design Method for Large-Scale MIMO Multi-User Zero-Forcing Precoding

Abstract

The present invention provides an alternating training design method for large-scale MIMO multi-user zero-forcing precoding, including system modeling, antenna grouping, alternating training, and data transmission phases. Among them, the base station sends pilots on the grouped antennas for training, so that users can obtain their channels and feedback them to the base station until the obtained channels can support all users to reach the desired signal-to-interference-plus-noise ratio threshold, or all channels are obtained. Compared with full antenna training and partial training, the present invention improves the system transmission success rate and reduces the pilot training overhead; at the same time, compared with single-antenna alternating training, it can further reduce the additional time overhead and correspondingly improve the system transmission success rate.

Description

Technical Field

[0001] The present invention belongs to the technical field of wireless communication, and in particular relates to an alternating training design method for large-scale MIMO multi-user zero-forcing precoding. Background Art

[0002] In a frequency-division duplexing (FDD) large-scale MIMO system, downlink precoding and uplink data detection correspond to different operating frequency bands, and the fading experienced by the uplink and downlink channels usually has no correlation. Therefore, the base station needs to obtain the downlink channel state information (CSI) through channel feedback. The user terminal receives the pilot symbols from the base station, then performs channel estimation, and feeds back part or all of the estimated values to the base station. In the case of statistically independent channels, to ensure reliable channel estimation, the number of training time slots needs to be no less than the number of transmit antennas, and the consumed feedback resources are proportional to the number of antennas. Therefore, obtaining accurate CSI with low training overhead and feedback overhead has become an urgent problem to be solved.

[0003] The existing channel estimation and training schemes currently have three common points: 1) estimating all channel coefficients, 2) performing estimation using a fixed-length training interval, and 3) not considering the quality of service (QoS) that the already obtained CSI can provide during the training process. These aspects may limit the trade-off between training and data transmission, so there is still room for further optimization of the system performance.

[0004] Currently, existing research has proposed the concept of alternating training, which has the characteristics that the training length adapts to the instantaneous CSI and user QoS requirements. However, the existing alternating training schemes mainly consider the single-user MIMO scenario, so the alternating training in the multi-user scenario is worthy of further exploration. Summary of the Invention

[0005] The purpose of the present invention is to provide an alternating training design method for large-scale MIMO multi-user zero-forcing precoding to solve the problem that the existing scheme cannot obtain accurate CSI with low training overhead and feedback overhead.

[0006] To solve the above technical problems, the specific technical solution of the present invention is as follows:

[0007] An alternating training design method for large-scale multiple-input and multiple-output (MIMO) multi-user zero-forcing precoding includes the following steps:

[0008] Step 1. System Modeling Phase: Build a multi-user large-scale MIMO downlink channel model, propose an alternating training transmission model, and set the condition for the start of data transmission as the SINR of all users reaching the threshold;

[0009] Step 2. Antenna Grouping Phase: Considering the additional delay brought by the training feedback phase, correct the expected Signal to Interference plus Noise Ratio (SINR) threshold; Based on Zero-forcing (ZF) precoding, derive the transmission success rate index of alternating training. Considering the additional delay brought by the training feedback phase, determine the optimal number of antenna groups k according to the transmission success rate, and perform alternating training of k antennas;

[0010] Step 3. Alternating Training Phase: The base station first uses the first U antennas to send pilots to U users, and the users estimate the corresponding channel vectors and feedback them to the base station. The base station calculates the SINR of each user to determine whether the transmission requirement is met at this time. If not, continue with alternating training. Next, each time k antennas are trained. After the channel estimation and feedback of the users, the base station calculates the SINR of the users again. If the transmission requirement is reached, end the alternating training process with k antennas as a group;

[0011] Step 4. Data Transmission Phase: The base station designs ZF precoding according to the channel matrix fed back by the users and performs data transmission.

[0012] Furthermore, in the system modeling phase of Step 1, it includes the following specific steps:

[0013] Step 1.1. Build a multi-user large-scale MIMO downlink channel model as follows: Consider a single-cell multi-user large-scale MIMO downlink system, where a large antenna array is equipped at the base station to send signals to multiple users using the same time-frequency bandwidth. The base station is configured with M transmit antennas, assuming its antenna array has full radio frequency chains. The number of users is U, and each user is configured with a single antenna. Assume the channel is frequency-flat and block-fading, where the coherence interval is denoted as τ. In each time block, the channel vector from the base station to user u is denoted as:

[0014]

[0015] where, β u represents the large-scale channel coefficient related to distance, h u =[h u,1 ,…,h u,M represents the small-scale channel vector, h u,m is the small-scale channel from the m-th antenna to user u, I Mis an M×M identity matrix, and it is assumed that the small-scale channels between users are independent of each other. Define Then the overall channel matrix of the system can be expressed as:

[0016] G = BH

[0017] In an FDD system, channel reciprocity no longer holds. The base station obtains CSI through downlink training and CSI feedback from users. Each coherence interval τ is divided into two parts: a training interval and a data transmission interval. In the training interval, the base station sends training pilots, and users estimate CSI based on the received signals. Then, the users feedback the CSI to the base station; the data transmission interval comes after the training interval. In the data transmission interval, the base station uses a transmission scheme designed based on the CSI obtained in the training interval to send data. Let t denote the length of the training interval, then the length of the data transmission interval is τ - t;

[0018] Step 1.2. Propose an alternating training transmission model as follows: To save antenna training time, only part of the antennas are trained each time, and then the user side estimates and feedbacks CSI. The base station side determines whether to continue training until the conditions for starting transmission are met, or all M antennas are trained, then the antenna training stops. The base station obtains the first t columns of the downlink channel H through antenna training and CSI feedback, where t = U,…,M. After a training interval of length t, use to represent the channel vector between user u and the first t antennas, satisfying:

[0019]

[0020] where I t represents an identity matrix of dimension t×t, and 0 M-t,t represents an (M - t)×t matrix with all elements being 0. Then the partial channel matrix obtained by the base station side is represented by , represents the transpose of. The precoding matrix W(t) = [w1(t),…,w U (t)], and w u (t) is the u-th column of W(t), satisfying ||w u (t)|| = 1. Let s denote the data vector and n denote the Gaussian white noise vector, whose elements are independent of each other, and the received signal can be expressed as:

[0021]

[0022] where represents the power allocation matrix, P uDenote the power allocated to user u, at this time the received signal of user u is

[0023]

[0024] Step 1.3: Define that the SINR of all users reaches the threshold as the transmission requirement, that is, the condition for the start of transmission, specifically as follows: The SINR of the user can be expressed as

[0025]

[0026] Define γ u as the SINR threshold that user u expects to meet, and the transmission condition is

[0027]

[0028] Furthermore, the antenna grouping phase of step 2 includes the following specific steps:

[0029] Step 2.1: Considering the additional delay brought by the training feedback phase, correct the SINR threshold. Since the alternating training divides the training and feedback into multiple phases, each alternating feedback phase introduces additional delay, which accounts for a part of each training time slot, and the proportion is ε. Group the antennas from the (U + 1)-th to the M-th into k groups for alternating training, so as to reduce the additional time cost. Assume that the transmission start condition is met when training reaches the M-th antenna. Ignoring the additional delay, the transmission time interval is τ - M. When considering the additional delay The rate of user u is reduced to

[0030]

[0031] Define the corrected SINR threshold as That is, when the SINR of all users satisfies:

[0032]

[0033] it is considered that the antenna alternating training can be stopped and data transmission can be carried out;

[0034] Step 2.2: Determine the optimal number of antenna groups k according to the transmission success rate r su The base station side can be based on perform ZF precoding to obtain Use v u (t) to represent the u-th column of V(t), represent the transmit beamforming vector, and the SINR of user u can be expressed as

[0035]

[0036] Define X u = SINRu (t), there is X u ~Gamma(t - U + 1, P u β u ), and its probability density function (PDF) is expressed as:

[0037] Therefore, when the training length is t, the SINR of user u is greater than the threshold The probability is

[0038]

[0039] Due to the weak correlation of SINR among users, it is considered that the probability that the SINR of all users satisfies the threshold is

[0040]

[0041] There is a trade - off between the training interval t and the data transmission interval τ - t. A longer training interval can obtain better CSI, thus achieving better performance for each data transmission interval. However, at the same time, the decrease in the transmission interval brings a reduction in the system rate. Introduce the transmission success rate r su Evaluate the performance of the scheme. Given the channel coherence time τ, the total number of time slots is τ. The transmission success rate is defined as the ratio of the number of successfully transmitted time slots to the total number of time slots, where the successfully transmitted data time slots refer to the time slots when the SINR of each user reaches the SINR threshold. Let T denote the random variable of the training length, and define the variable Then r su The calculation method is as follows:

[0042]

[0043]

[0044] By maximizing the transmission success rate r su , the optimal number of antenna groups k can be found;! refers to the factorial

[0045] Furthermore, in the alternating training phase of step 3, including antenna training, CSI feedback, and SINR verification, the following specific steps are included:

[0046] Step 3.1: The base station sends U pilots to each user using the first U antennas. Each user estimates the corresponding U×1 channel vector and feeds back The base station side obtains the channel matrix At this time, the training length t = U;

[0047] Step 3.2: The base station side calculates SINR1(t), …, SINR according to the SINR formula in step 2.2U (t), if all users reach the SINR threshold, that is then end the antenna training phase, go to step 4, otherwise go to step 3.3;

[0048] Step 3.3: If the number of trained antennas t = M, then end step 3 and go to step 4; otherwise when t ≤ M, the base station sends pilots using the (t + 1)-th to the (t + 1 + k)-th antennas, and then user u estimates the corresponding channel coefficients and feeds them back to the base station side. The base station side obtains Let the training length t = t + k, and go to step 3.2.

[0049] An alternating training design method for large-scale MIMO multi-user zero-forcing precoding according to the present invention has the following advantages:

[0050] The present invention includes system modeling, antenna grouping, alternating training, and data transmission phases. To solve the problem of low-overhead acquisition of large-scale MIMO channels in FDD systems, the present invention proposes an alternating training scheme based on antenna grouping for zero-forcing precoding. In each training time slot, the base station sends pilots on the grouped antennas for users to obtain their channels and feed them back to the base station. Repeat the training of antenna grouping multiple times, based on the instantaneous channels obtained from the current and all previous training time slots, until all users reach the SINR threshold, or the channels of all base station antennas are obtained, then stop training and start data transmission. Compared with full antenna training and partial training, the present invention improves the system transmission success rate and reduces the pilot training overhead; at the same time, compared with single-antenna alternating training, the alternating training based on antenna grouping can further reduce the time overhead generated by the additional interaction between the base station and users in alternating training. And through the optimization of the number of grouped antennas, the system transmission success rate can be correspondingly improved. Description of the Drawings

[0051] Figure 1 is the flowchart of the present invention;

[0052] Figure 2(a) is the flowchart of the alternating training and data transmission of the scheme when the antenna grouping of the present invention is k > 1;

[0053] Figure 2(b) is the flowchart of the alternating training and data transmission of the scheme when training antennas one by one according to the present invention;

[0054] Figure 3 is γ of the present invention u = [-3dB, -5dB], ε = [0.1, 0.2], the comparison diagram of the transmission success rate of the alternating training scheme under different antenna grouping numbers;

[0055] Figure 4 is γ of the present invention u=-3dB, ε = 0.2, the transmission success rate comparison of the alternating training scheme, antenna-by-antenna training scheme, full training scheme and fixed-length partial training scheme under different transmit powers under the optimal antenna grouping k = 5. DETAILED DESCRIPTION

[0056] In order to better understand the purpose, structure and function of the present invention, the following further describes in detail an alternating training design method for massive MIMO multi-user zero-forcing precoding in conjunction with the accompanying drawings.

[0057] like Figure 1 As shown in FIG, the alternating training method for massive MIMO multi-user zero-forcing precoding proposed by the present invention is implemented in the following specific steps:

[0058] Step 1: System Modeling: Construct a multi-user massive MIMO downlink channel model, propose an alternating training transmission model, and set the conditions for starting data transmission.

[0059] The system modeling phase includes the following specific steps:

[0060] Step S101: Construct a multi-user massive MIMO downlink channel model as follows: Consider a single-cell multi-user massive MIMO downlink system, where the base station is equipped with a massive antenna array to transmit signals to multiple users using the same time-frequency bandwidth. The base station is configured with M transmit antennas, and its antenna array is assumed to have a full RF chain. The number of users is U, and each user is configured with a single antenna. The channel is assumed to be frequency-flat and block-fading, where the coherence interval is denoted by τ. Use β u Represents the large-scale channel coefficient related to distance, modeling Its species u Indicates shadow fading, and there is d u is the distance from the base station to user u, and v represents the path loss coefficient. u =[h u,1 ,…,h u,M ] represents the small-scale channel vector, h u,m is the small-scale channel from the mth antenna to user u, And it is assumed that the small-scale channels between users are independent of each other. In each time block, the channel vector from the base station to user u is Expressed as:

[0061] Define the large-scale fading coefficient matrix Small-scale channel matrix Then the entire channel matrix of the system It can be expressed as:

[0062] G = BH

[0063] In the FDD system, since channel reciprocity no longer holds, the base station obtains CSI through downlink training and the CSI feedback of users. Each coherence interval τ is divided into two parts: a training interval and a data transmission interval. In the training interval, the base station sends training pilots, and the user estimates CSI based on the received signals. Then, the user feeds back the CSI to the base station side; the data transmission interval comes after the training interval. In the data transmission interval, the base station side uses a transmission scheme designed based on the CSI obtained in the training interval to send data. Let t denote the length of the training interval, then the length of the data transmission interval is τ - t.

[0064] Step S102: Propose an alternating training transmission model. To save antenna training time, only part of the antennas are trained each time, and then the user side performs CSI estimation and feedback, and the base station side conducts a user SINR test to determine whether to continue training until the condition for starting transmission is met, or all M antennas are trained, then the antenna training stops. The specific transmission modeling is as follows: The base station obtains the first t columns of the downlink channel H through antenna training and CSI feedback, where t = U,..., M. After a training interval of length t, let denote the channel vector between user u and the first t antennas, satisfying:

[0065]

[0066] where I t denotes the identity matrix of dimension t×t, and 0 M-t,t denotes the (M - t)×t matrix with all elements being 0. Then the partial channel matrix obtained by the base station side is denoted by . The precoding matrix Let s denote the data vector and n denote the Gaussian white noise vector, whose elements are independent of each other, and the received signal can be expressed as:

[0067]

[0068] where denotes the power allocation matrix, P u denotes the power allocated to user u. At this time, the received signal of user u is

[0069]

[0070] Step S103: Define that the SINR of all users reaching the threshold is the transmission requirement, that is, the condition for starting transmission, as follows: The SINR of the user can be expressed as

[0071]

[0072] Define γ u as the SINR threshold that user u expects to satisfy, and the transmission condition is

[0073]

[0074] Step 2: In the antenna grouping stage: Considering the additional delay brought by the training feedback stage, training the antennas one by one alternately may be too costly. Therefore, alternating training with antenna grouping is considered.

[0075] The antenna grouping stage includes the following specific steps:

[0076] Step S201: Considering the additional delay brought by the training feedback stage, correct the SINR threshold. Since alternating training divides training and feedback into multiple stages, assuming that each alternating feedback stage introduces an additional delay, which accounts for a part of each training time slot, with a proportion of ε. Group the antennas from the (U + 1)-th to the M-th into k groups for alternating training, so as to reduce the additional time cost. Assume that the transmission start condition is met when training the M-th antenna. Ignoring the additional delay, the transmission time interval is τ - M. And when considering the additional delay At this time, the rate of user u is reduced to

[0077]

[0078] Define the corrected threshold as That is, when the following is satisfied:

[0079]

[0080] It is considered that the antenna alternating training can be stopped and data transmission can be carried out.

[0081] Step S202: Determine the optimal number of antenna groups k based on the transmission success rate. The base station side can perform ZF precoding based on to obtain Use v u (t) to represent the u-th column of V(t), represent the transmit beamforming vector, and the precoding matrix W(t) = [w1(t), …, w U (t)]. The SINR of user u can be expressed as

[0082]

[0083] Define X u = SINR u (t), and there is X u ~ Gamma(t - U + 1, P u β u), and its probability density function (PDF) is expressed as:

[0084] Therefore, when the training length is t, the probability that the SINR of user u is greater than the threshold is

[0085]

[0086] Due to the weak correlation of SINR among users, the probability that the SINRs of all users satisfy the threshold is considered to be

[0087]

[0088] There is a trade-off between the training interval t and the data transmission interval τ - t. A longer training interval can obtain better CSI, thus achieving better performance for each data transmission interval. However, at the same time, the decrease in the transmission interval leads to a reduction in the system rate. The transmission success rate r is introduced su to evaluate the performance of the scheme. Given the channel coherence time τ, the total number of time slots is τ. The transmission success rate is defined as the ratio of the number of successfully transmitted time slots to the total number of time slots, where the successfully transmitted data time slots refer to the time slots when the SINR of each user reaches the SINR threshold. Let T denote the random variable of the training length, and define Then r su is calculated as follows:

[0089]

[0090] Maximizing the transmission success rate r su can find the optimal number of antenna groups k.

[0091] Step 3: In the alternating training phase: The three steps of antenna training, CSI feedback, and SINR verification are carried out alternately until the condition for starting transmission is met, or all M antennas are trained, then the antenna training stops.

[0092] The alternating training phase includes the following specific steps:

[0093] Step S301: The base station sends U pilots to each user using the first U antennas, and each user estimates the corresponding U×1 channel vector and feeds back The base station obtains the channel matrix At this time, the training length t = U.

[0094] Step S302: The base station calculates SINR1(t), …, SINR U (t) according to the SINR formula in step 2.2. If all users reach the SINR threshold, that is Then end the antenna training phase and go to step S401; otherwise, go to step S303.

[0095] Step S303: If the number of trained antennas t = M, end step S401 and go to step 4; if t ≤ M, the base station sends pilots using the (t + 1)-th to the (t + 1 + k)-th antennas, and user u estimates the corresponding channel coefficients and feeds back. The base station obtains the training length t = t + k, and go to step S302.

[0096] Step 4: In the data transmission phase: The base station performs data transmission according to the CSI fed back by the user.

[0097] The data transmission phase includes the following specific steps:

[0098] Step S401: The base station performs ZF precoding according to the channel matrix fed back by the user to obtain the precoding matrix W(t) and perform data transmission.

[0099] The specific steps of the alternating training scheme for large-scale MIMO multi-user zero-forcing precoding include:

[0100] Step S501: For the alternating training phase and the data transmission phase in the present invention, the flowchart in Fig. 2(a) is given. First, the base station sends U pilots to each user using the first U antennas in the first U training time slots, and each user performs CSI feedback. Then, the base station performs SINR judgment and verification for the users. If the transmission condition is not satisfied, antenna group alternating training with k as the unit starts. In the next k time slots, the remaining k antennas are trained, and the CSI feedback of the users and the SINR verification at the base station end continue until the transmission start condition is satisfied. It can be calculated that within the coherence time τ, the training interval is t, and the CSI feedback and SINR verification delay is

[0101] Step S502: The flowcharts of the alternating training phase and the data transmission phase with one-by-one antenna training are shown in Fig. 2(b). First, the base station sends U pilots to each user using the first U antennas in the first U training time slots, and each user performs CSI feedback. Then, the base station performs SINR judgment and verification for the users. If the transmission condition is not satisfied, the remaining antennas are trained one by one, and the CSI feedback of the users and the SINR verification at the base station end are performed until the transmission start condition is satisfied. It can be calculated that within the coherence time τ, the training interval is t, and the CSI feedback and SINR verification delay is (1 + M - U)ε.

[0102] The specific steps for verifying the effect of the alternating training scheme for large-scale MIMO multi-user zero-forcing precoding include:

[0103] Step S601: At the coherence time τ = 200 and the path fading exponent v = 2. Set the number of base station antennas M = 100, the number of users U = 8, and the normalized noise power. Set the total transmit power to 40 dB, and the large-scale fading of users ranges from [-33 dB, -20 dB]. Set the unified SINR threshold γ u = -3 dB, -5 dB, ε = 0.1, 0.2, and use the proposed grouped alternating training scheme for training to obtain the transmission success rate in the corresponding cases. Considering that as small antenna groups are preferably used at low SINR thresholds to reduce the average training length, take k = 1, 2, 3, 4, 5, 6, 7, 8, 9, 10.

[0104] Step S602: In the case of γ u = -3 dB and ε = 0.2, respectively use the alternating training with the optimal group k = 5, the per-antenna training scheme, the full training scheme, and the fixed-length partial training scheme t = 80 to train the antennas, and obtain the transmission success rates of several schemes at different transmit powers. Set the transmit powers between users to be equal. Considering the value range of large-scale fading, the total transmit power value range is P = [40 dB, 50 dB].

[0105] Experimental results: As Figure 3 shown, for the additional delay ε introduced in the alternating feedback phase, at the same SINR threshold, the larger ε is, the larger the optimal antenna group number is. At the same additional delay ε, the higher the SINR threshold, the larger the corresponding optimal antenna group number will be. For a given SINR threshold and additional delay ε, the optimal antenna group number can be determined based on the maximization of the transmission success rate, and the conclusions of the simulation values and the theoretical values are consistent. For example, in γ u = -3 dB and ε = 0.2, selecting the antenna group number k = 5 can achieve the optimal transmission success rate performance.

[0106] As Figure 4 shown, taking the per-antenna training, the full training scheme, and the fixed-length partial training scheme t = 80 as the comparison methods, this example compares the theoretical values and the simulation values of the transmission success rate of the system under the total transmit power values.

[0107] It can be seen that:

[0108] For the derivation of the transmission success rate, the theoretical values and the simulation values are consistent. It can be seen that in γ u=-3dB. At medium and low signal-to-noise ratios, using only part of the training with t = 80 can achieve a better transmission success rate. Whether using the alternating training method of antenna grouping or training each antenna one by one, a better transmission success rate can be obtained compared to the full training scheme and the fixed-length partial training scheme. Moreover, the alternating training with the optimal number of antenna groups k = 5 can further reduce the additional delay generated by the alternating training in the system while improving the system performance compared to training each antenna one by one.

[0109] Therefore, the alternating training method for large-scale MIMO multi-user zero-forcing precoding proposed in the present invention takes into account the additional delay introduced by the alternating training in terms of feedback and SINR checking, and uses antenna grouping for alternating training, which further reduces the training overhead of the system while maximizing the system transmission success rate.

[0110] It can be understood that the present invention is described through some embodiments. Those skilled in the art know that, without departing from the spirit and scope of the present invention, various changes or equivalent replacements can be made to these features and embodiments. Additionally, under the teaching of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application belong to the scope protected by the present invention.

Claims

1. An alternating training design method for large-scale MIMO multi-user zero-forcing precoding, characterized in that It includes the following steps: Step 1, System Modeling Phase: Construct a multi-user large-scale MIMO downlink channel model, propose an alternating training transmission model, and define that the SINR of all users reaches a threshold as the transmission condition; Step 2, Antenna Grouping Phase: Considering the additional delay brought by the training feedback phase, correct the SINR threshold; Based on the zero-forcing ZF precoding, derive the transmission success rate index of alternating training, and determine the optimal number of antenna groups k according to the transmission success rate; Step 3, Alternating Training Phase: The base station first sends pilots to U users with the first U antennas, and the users estimate the corresponding channel vectors and feedback them to the base station; The base station calculates the SINR of each user to determine whether the transmission condition is met at this time. If not, continue with the alternating training. Next, each time k antennas are trained. After the channel estimation and feedback of the users, the base station calculates the user SINR again to determine whether the transmission condition is met. If it is met, end the alternating training process with k antennas as a group; Step 4, Data Transmission Phase: The base station performs ZF precoding according to the channel matrix fed back by the users and conducts data transmission.

2. The alternating training design method for large-scale MIMO multi-user zero-forcing precoding according to claim 1, characterized in that The system modeling phase described in Step 1 includes the following specific steps: Step 1.

1. Construct a multi-user large-scale MIMO downlink channel model as follows: Consider a single-cell multi-user large-scale MIMO downlink system, where the base station is equipped with M transmit antennas, assuming that its antenna array has full radio frequency chains; the number of users is U, and each user is equipped with a single antenna; assume that the channel is frequency-flat and block-fading, where the coherence interval is denoted as τ; within each time block, the downlink channel vector from the base station to user u is expressed as: where, β u represents the large-scale channel coefficient related to distance, h u = [h u,1 , …, h u,M represents the small-scale channel vector, h u,m is the small-scale channel from the m-th antenna to user u, I M is the M×M identity matrix, and it is assumed that the small-scale channels between users are independent of each other; the large-scale coefficient matrix is defined as represents the transpose of the channel vector h u , then the entire channel matrix of the system is expressed as: G = BH For the FDD system, the base station obtains the downlink channel through downlink training and user terminal feedback; each coherence time interval τ is divided into two parts: the training interval and the data transmission interval; let t represent the length of the training interval, then the length of the data transmission interval is τ - t; Step 1.

2. Propose an alternating training transmission model as follows: Only train some antennas each time, then the user side performs channel estimation and feedback, and the base station side determines whether to continue training until the condition for starting transmission is met, or all M antennas have been trained, then stop antenna training; The base station obtains the first t columns of the downlink channel H through antenna training and channel feedback, where t = U,…,M. After a training interval of length t, use to represent the channel vector between user u and the first t antennas, satisfying: where I t represents the identity matrix of dimension t×t, and 0 M-t,t represents the (M - t)×t matrix with all elements being 0; then the partial channel matrix obtained at the base station side is represented by denotes, denotes the transpose of; the precoding matrix and there is W(t) = [w1(t), …, w U (t)], w u (t) is the u-th column of W(t), satisfying ||w u (t)|| = 1; let s denote the data vector and n denote the noise vector, whose elements are independent of each other, and the received signal is expressed as: wherein represents a power allocation matrix, and P u represents the power allocated to user u; Step 1.3, Define that the SINR of all users reaches a threshold as the transmission requirement, that is, the condition for the start of transmission, specifically as follows: The SINR of the user can be expressed as Define γ u as the SINR threshold that user u expects to satisfy, and the transmission condition is 3. The alternating training design method for large-scale MIMO multi-user zero-forcing precoding according to claim 2, characterized in that The antenna grouping phase described in Step 2 specifically includes: Step 2.

1. Considering the additional delay caused by the training feedback phase, correct the SINR threshold: Since the alternating training divides the training and feedback into multiple phases, each alternating feedback phase introduces additional delay, which accounts for a part of each training time slot, with a proportion of ε; group the antennas from the (U + 1)-th to the M-th antenna for alternating training, with every k antennas as a group, so as to reduce the additional time cost; assume that the condition for starting transmission is met only when training reaches the M-th antenna, ignoring the additional delay, the transmission time interval is τ - M; and when considering the introduced additional delay At this time, the rate of user u is reduced to Define the corrected SINR threshold as When the SINR of all users satisfies it is considered to stop the antenna alternating training and conduct data transmission; Step 2.

2. According to the defined transmission success rate r su Determine the optimal number of antenna groups k, specifically as follows: Based on the base station side Perform ZF precoding, and the SINR of user u can be further expressed as Given the channel coherence time τ, the total number of time slots is τ; the transmission success rate is defined as the ratio of the number of successfully transmitted time slots to the total number of time slots, where the successfully transmitted data time slots refer to the time slots when the SINR of each user reaches the SINR threshold; define the variable r su The calculation method is as follows: By maximizing the transmission rate success rate r su , the optimal number of antenna groups k can be found.

4. The alternating training design method for large-scale MIMO multi-user zero-forcing precoding according to claim 3, characterized in that The alternating training phase described in Step 3 includes the following specific steps: Step 3.1: The base station sends U pilots to each user using the first U antennas, and each user estimates the corresponding U×1 channel vector and feeds back The base station obtains the channel matrix At this time, the training length t = U; Step 3.2: The base station calculates SINR1(t), …, SINR U (t) according to the SINR formula in Step 2.

2. If all users reach the SINR threshold, that is then end the antenna training phase and go to Step 4; otherwise, go to Step 3.

3. Step 3.

3. If the number of training antennas \(t = M\), then end Step 3 and go to Step 4; otherwise, when \(t\leq M\), the base station sends pilots using the \((t + 1)\)-th to the \((t + 1 + k)\)-th antennas, and user \(u\) estimates the corresponding channel coefficients and feeds back; the base station side obtains Let the training length \(t=t + k\), and go to Step 3.2.

Images