CN113259033A - High-speed wave beam control method in dynamic millimeter wave communication scene based on FPGA - Google Patents

Abstract

The invention discloses a high-speed beam control method in a dynamic millimeter wave communication scenario based on FPGA, comprising the following steps: at the receiving end of the communication system, the beam is initialized by scanning a full codebook to establish an initial communication beam; establishing a beam control module The data communication between the FPGA and the baseband processing FPGA is used to trigger the beam tracking process and perform beam measurement; establish a beam training state machine to realize the beam tracking algorithm model; transmit the antenna control information through the SPI high-speed serial port to realize the control of the antenna by the beam control module; After the beam training is completed, the beam quality is judged. If it is determined that the beam selection fails, beam recovery is required; after the beam training is completed, the best transmission beam is selected for data transmission, and the training trigger signal of the next cycle is waited. The invention realizes the beam switching interval of microsecond level by defining the SPI high-speed mode writing protocol to control the antenna.

Description

A high-speed beam steering method in dynamic millimeter-wave communication scenarios based on FPGA

technical field

The invention relates to a large-scale array antenna beam forming and tracking technology, and belongs to the field of wireless communication.

Background technique

With the continuous development of information technology, the amount of information has increased explosively, and new business scenarios have emerged one after another, which has put forward more stringent requirements for the bandwidth of the communication system. Traditional communication frequency bands are gradually difficult to meet this demand. The characteristics of large bandwidth, low latency, and abundant spectrum resources have attracted the attention of researchers. Communication research on the millimeter wave frequency band has begun to be widely carried out, and a very important research in the field of millimeter wave is about beamforming technology. Because the millimeter wave frequency band has various advantages, it has the disadvantage of easy attenuation, and the beamforming technology resists attenuation by concentrating the antenna beam gain in one direction.

Although beamforming greatly improves the transmit power gain, due to its strong directivity, when the user moves, the beam will be misaligned and the received signal power will drop sharply, making it difficult to maintain communication quality. So real-time alignment of beam directions when the user moves becomes an unavoidable problem. At present, there have been many achievements in the research on beam tracking, such as the scanning of the entire beam codebook and the hierarchical search method, which has a high alignment rate but a large training cost. In order to reduce the overhead, the method of using auxiliary information, such as the estimation of the angle of arrival, the estimation of the channel state information, etc., is only in the theoretical stage and it is difficult to realize. In order to reduce overhead while being easy to implement, beam tracking methods combined with machine learning are starting to gain more attention.

In order to ensure that the user's real-time communication is not interrupted, it is necessary to periodically conduct beam search to always control the beam point in a better direction, which requires continuous beam training process, and the range of each search must be controlled within a small The beam switching speed must be as fast as possible, so that the entire beam search process is fast, so as not to affect the normal communication use of the user. Because of its superior processing performance, FPGA can quickly complete the whole process, and the bandit algorithm combined with historical experience learning can quickly reduce the range of beam search and reduce overhead. Therefore, the present invention makes a reasonable recommendation based on the historical training experience of the bandit algorithm, and implements it on the hardware through FPGA. Combined with the characteristics of high-speed processing, the training consumption and training cycle are greatly reduced. The overhead of training beams never achieves the alignment of beam directions when the user is constantly moving, and the communication link is always stable and uninterrupted.

SUMMARY OF THE INVENTION

In the communication in the millimeter wave frequency band, beam alignment is required due to the application of beamforming technology. Since the user moves in real time, a fast and frequent beam tracking process is required, so that the user's communication link is always stable and available. This requires the design of a reliable beam steering module to achieve the purpose of high-speed training beams in the millimeter-wave hardware system, and to have high robustness and beam recovery mechanisms.

The purpose of the present invention is to provide a high-speed beam control method in a dynamic millimeter wave communication scenario based on FPGA, by defining the SPI high-speed mode writing protocol to control the antenna, and realizing the beam switching interval of microsecond level. According to the characteristics of fast beam training, the period of beam training is shortened, so that the scope of each beam training is greatly reduced. At the same time, the random selection optimization model method is used to continuously learn the training experience, and further reduce the scope of the training beam according to the training results.

To achieve the above object, the technical scheme adopted in the present invention is:

An FPGA-based high-speed beam steering method in a dynamic millimeter wave communication scenario, comprising the following steps:

Step 1, at the receiving end of the communication system, initialize the beam by scanning the full codebook, and establish an initial communication beam;

Step 2, establishing data communication between the beam control module and the baseband module equipped with the FPGA through the FIFO data structure and the PXI-trigger register, for triggering the beam tracking process and performing beam measurement;

Step 3, establishing a beam training state machine to implement a beam tracking algorithm model;

In step 4, the control of the antenna by the beam control module is realized by transmitting the antenna control information through the SPI module in FIG. 4;

Step 5: Judge the beam quality after each beam training, and if it is judged that the beam selection fails, beam recovery needs to be performed;

Step 6: After the beam training ends, select the best transmission beam for data transmission, and wait for the training trigger signal of the next cycle.

The step 1 is as follows:

其中M是码本的大小，f_i表示天线响应向量；根据天线格式将每个码字对应的控制比特流提前以字节形式写入memory中；通信系统接收端的波束控制模块首先对全码本进行一次全扫描，全扫描的实现通过建立初始化状态机来实现，状态转移过程为：先初始化波数序号i＝0，最佳波束序号i_max＝0，最佳波束对应的功率p_max＝0，然后进入SPI写入过程，根据天线需求，需要分多个阶段将波束控制信息，收发控制信息，使能信号写入天线；多个阶段总共写入n_b个字节的数据，其中每一阶段写入的数据均从memory内存模块中读取，起始地址为i×n_b+o_f，o_f为偏移地址；每一阶段写入的字节数为n_bt，比特数为n_bit；通过SPI_START信号通知SPI模块开始往天线写入当前阶段的数据；然后等待SPI模块反馈的SPI_DONE信号，当检测到该信号为高电平则说明本阶段数据写入完成；此时进行一个判断，多个阶段是否全部写入完成，若未完成则写入下一阶段数据，若完成则判断码本是否扫描完成，若完成则结束该过程进入波束跟踪的触发信号监测状态，等待进行波束跟踪的过程，若未完成扫描则更新最佳波束及最佳波束对应的功率；更新最佳波束信息后进行一次判断，是否所有波束方向都已搜索完毕，若未搜索完毕将波束序号加1重新进行SPI模块写入，若已经搜索完毕，则将最佳波束作为执行波束序号再执行一次SPI模块写入。The codebook space is defined as

Among them, M is the size of the codebook, and f _i represents the antenna response vector; according to the antenna format, the control bit stream corresponding to each codeword is written into the memory in advance in the form of bytes; the beam control module at the receiving end of the communication system A full scan is performed, and the realization of the full scan is realized by establishing an initialization state machine. The state transition process is: first initialize the wave number sequence number i=0, the optimal beam sequence number i _max =0, the power corresponding to the optimal beam p _max =0, Then enter the SPI writing process. According to the antenna requirements, it is necessary to write beam control information, send and receive control information, and enable signals into the antenna in multiple stages; a total of n _b bytes of data are written in multiple stages, and each stage The written data is read from the memory memory module, the starting address is _i ×n _b +of, and of _is the offset address; the number of bytes written in each stage is n _bt , and the number of bits is n _bit ; Inform the SPI module to start writing the data of the current stage to the antenna through the SPI_START signal; then wait for the SPI_DONE signal fed back by the SPI module, when it is detected that the signal is high, it means that the data writing at this stage is completed; At this time, a judgment is made, Whether the writing of multiple stages is completed, if not, write the data of the next stage, if completed, judge whether the codebook is scanned or not, if completed, end the process and enter the trigger signal monitoring state of beam tracking, waiting for beam tracking. Process, if the scanning is not completed, update the optimal beam and the power corresponding to the optimal beam; after updating the optimal beam information, make a judgment to determine whether all beam directions have been searched, if not, add 1 to the beam serial number and perform SPI again Module write, if the search has been completed, then execute the SPI module write again with the optimal beam as the execution beam serial number.

The step 2 is as follows:

In the process of beam initialization and beam tracking, beam performance measurement needs to be performed, and the received power is calculated by the baseband module at the receiving end of the communication system, which is used as an indicator to judge the performance of the beam direction selected by the current antenna. The performance of all beams within the range is measured, and the beam with the best measurement result is taken out for communication in the data transmission phase; at the same time, the beam tracking process is carried out periodically, each period is the time of a communication frame, and each frame contains Beam training field. When the beam training field starts, the baseband module needs to notify the beam control module of the start of the round of beam tracking. The PXI-trigger register is used to achieve low-latency information transmission, and the register is set to high level when the enable signal is transmitted. , and lasts a long enough period to ensure that it can be read by the beam control module; the performance of the scanning beam is obtained through the beam measurement process, and the optimal beam and corresponding power are obtained through performance comparison:

Among them, i _max represents the optimal beam, p _max represents the power corresponding to the optimal beam, and p _i represents the power corresponding to the beam i;

In the power acquisition of the current beam, in order to compensate for the delay caused by power calculation and PXIe bus transmission, the waiting time τμs before the current optimal beam is updated.

The step 3 is as follows:

当前时刻的最佳波束：The entire codebook space is obtained after beam initialization is performed

The best beam at the current moment:

After the initial process is completed, the beam training phase is entered. This phase is enabled by the training trigger signal and completed by establishing a beam training state machine. The state transition process is as follows: the beam tracking phase adopts a random selection optimization model, and presets multiple beam behaviors. Each behavior is represented as (b _last , o, c), b _last represents the best beam obtained from the last beam training, the initial value is i _max , o represents the offset of the first beam in the behavior relative to b _last , and c represents the behavior in the The number of beams, and the beams in the behavior are continuous; the beams included in (b _last ,o,c) are (b _last -o,b _last -o+1,b _last -o+2,…,b _last -o+c-1); after starting training, first initialize the parameters required for the training process, create multiple parameter arrays to store the offset value of the behavior, the number of beams, the average return value, the number of rounds selected, and the UCB value, the index of each array corresponds to the behavior number one-to-one; after initialization, select the behavior with the largest UCB value from the UCB array for training, and use the behavior as an index to obtain its corresponding value in the offset value array and the beam number array , to generate the candidate beam set of the behavior; the initial value of UCB should be set to a maximum value so that each behavior is trained at least once; the receiving end of the communication system measures the beam sets in the behavior respectively, and the antenna writing process of the beam Consistent with the beam initialization process, the baseband module calculates the energy of the current beam received signal in the time domain by the following formula:

where y _i is the time domain signal after channel estimation and equalization at the receiving end, and T represents the calculated length of the received signal;

After all the beams in the behavior are measured, the best beam is selected, and the process is the same as beam initialization; then the reward for this training needs to be determined:

Among them, r ₁ and r ₂ are fixed values, and p _thr is the power threshold of the receiver; according to the reward value, the average reward array is updated through the current behavior index, and the behavior selects the parameter values corresponding to the round number array and the UCB value array, and at the same time, the total training round is updated. count:

Among them, μ represents the average reward value, n _sel represents the number of rounds selected for this training behavior, n _to represents the total number of rounds, and u represents the UCB value.

The step 4 is as follows:

The SPI module obtains the corresponding byte data from the memory according to the beam index and data volume parameters passed in by the beam initialization and beam training state machine, and then writes the data into the antenna through the serial port through the custom high-speed SPI write protocol; custom SPI write The process of entering the protocol is: obtain the starting address of the data to be written _i ×n _b +of , the total number of bytes is n _bt , the number of bits is n _bit , and then write all the n _bt data into the FIFO storage structure; Set up a control to indicate the remaining bits to be written, and initialize it to the total number of bits n _bits , and write the clock signal to the clock interface at the same time; set the control v to store the data written to the antenna in each round, and then write the data bit by bit Enter the antenna, v is a data of 1-bit size; set the initial bit index in v, and write one by one from the bit data corresponding to the index; since the total number of bits to be written, n _bits , is usually not an integer multiple of 1, so write The first few bits of the first data entered are usually invalid bits. At this time, the initial bit index is set to the start of the valid bits, and the value is l-(l×n _bt -n _bit )-1; the storage structure from the FIFO The read data is stored in v, and it is judged whether the total number of bits is equal to the remaining number of bits. If they are equal, it means that it is at the beginning of writing. It is necessary to pull down the chip select signal and the antenna enable signal, and at the same time index the initial bit of v. Set as above; if they are not equal, it means that it is not the first time to read from the FIFO at this moment, just set the initial bit index in v to 1; write the clock signal to the antenna in each working cycle of the FPGA, and write v at the same time The bit data corresponding to the current index in the index is then decremented by 1, and the remaining number of bits to be written is decremented by 1; it is judged whether the index is 0, and if it is not 0, the process of writing bits and clock signals is repeated. If it is 0, judge whether the remaining number of bits to be written is 0. If it is not 0, continue to read data from the FIFO and perform bit writing. If it is 0, it means that n _bit data writing is completed, and reset at this time. Chip select signal and antenna enable signal, since the antenna needs response time, the enable signal should be reset later than the chip select signal according to the length of the response time.

The step 5 is as follows:

Set the minimum communication quality threshold p _lim , which represents the minimum performance threshold for maintaining normal communication quality; set up a control for the number of consecutive failures n _fail = 0, and judge whether the training result at this time meets the standard according to the optimal beam corresponding power value obtained from beam training. The minimum communication standard is n _fail =n _fail +1, and the beam used in the data transmission phase is still the training result that met the standard last time. If the standard is met, n _fail = 0; when n _fail reaches a certain threshold n _lim , execute again Steps of initial beam establishment to restore optimal beam.

The step 6 is as follows:

In the data transmission phase of each frame, the optimal beam obtained by training is used for communication. At the same time, the beam control module stays in the enable detection phase and continues to detect the PXI-trigger register signal. If it detects a high level, it enters the beam training process.

Beneficial effect: Compared with the prior art, the present invention has the following beneficial effects:

(1) The present invention implements an intelligent learning beam tracking algorithm based on an FPGA, and utilizes the high-speed processing characteristics of the FPGA to reduce the beam switching time to the microsecond level, thereby reducing the training duration and training period.

(2) The present invention makes full use of historical experience by combining the reinforcement learning method, further greatly reduces the training overhead, thereby realizing fast beam tracking, ensuring real-time beam alignment in the user's moving state, and always stable communication quality.

(3) The present invention realizes the coordinated work among the beam control module, the baseband module and the antenna module by designing a perfect beam control system, and has good robustness.

(4) The beam control module of the present invention has a flexible design, is easy to replace with different beam tracking algorithms and antennas, and can be extended to the mode of beam tracking at the same time of transmission and reception.

Description of drawings

Figure 1 is a beam initialization state transition diagram;

Fig. 2 is a beam tracking state transition diagram;

Fig. 3 is the flow chart of SPI module;

FIG. 4 is a flow chart of the beam control system;

5 is a schematic diagram of a frame structure in a communication system;

Figure 6 is a hardware platform diagram of an implementation case.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings.

The present invention provides a high-speed beam control method in an FPGA-based dynamic millimeter-wave communication scenario. Based on the NI millimeter-wave system, a single FPGA is used to control the beam switching and data interaction process, thereby realizing a complete beam control function. The FPGA receives the power information of the baseband for the performance index judgment of the beam, and reads the trigger signal from the baseband to periodically control the beam tracking process. By implementing an intelligent beam tracking method based on a random selection optimization model, among the defined multiple behaviors, the optimal behavior is selected through the UCB strategy to determine the scanning beam range for beam measurement and tracking. By defining the high-speed controllable transmission mode of the SPI serial port, the bit stream writing of the antenna phase shift information at a controllable rate is realized. Specifically include the following steps:

其中M是码本的大小，f_i表示天线响应向量；根据天线格式将每个码字对应的控制比特流提前以字节形式写入memory中；通信系统接收端的波束控制模块首先对全码本进行一次全扫描，全扫描的实现通过建立初始化状态机来实现，状态转移过程如图1所示：先初始化波数序号i＝0，最佳波束序号i_max＝0，最佳波束对应的功率p_max＝0，然后进入SPI写入过程，根据天线需求，需要分多个阶段将波束控制信息，收发控制信息，使能信号写入天线；多个阶段总共写入n_b个字节的数据，其中每一阶段写入的数据均从memory内存模块中读取，起始地址为i×n_b+o_f，o_f为偏移地址；每一阶段写入的字节数为n_bt，比特数为n_bit；通过SPI_START信号通知SPI模块开始往天线写入当前阶段的数据；然后等待SPI模块反馈的SPI_DONE信号，当检测到该信号为高电平则说明本阶段数据写入完成；此时进行一个判断，多个阶段是否全部写入完成，若未完成则写入下一阶段数据，若完成则判断码本是否扫描完成，若完成则结束该过程进入波束跟踪的触发信号监测状态，等待进行波束跟踪的过程，若未完成扫描则更新最佳波束及最佳波束对应的功率；更新最佳波束信息后进行一次判断，是否所有波束方向都已搜索完毕，若未搜索完毕将波束序号加1重新进行SPI模块写入，若已经搜索完毕，则将最佳波束作为执行波束序号再执行一次SPI模块写入。Step 1: Define the codebook space as

Among them, M is the size of the codebook, and f _i represents the antenna response vector; according to the antenna format, the control bit stream corresponding to each codeword is written into the memory in advance in the form of bytes; the beam control module at the receiving end of the communication system A full scan is performed, and the realization of the full scan is realized by establishing an initialization state machine. The state transition process is shown in Figure 1: first initialize the wave number number i=0, the optimal beam number i _max = 0, and the power p corresponding to the optimal beam _max = 0, and then enter the SPI writing process. According to the requirements of the antenna, the beam control information, the sending and receiving control information, and the enabling signal need to be written into the antenna in multiple stages; a total of n _b bytes of data are written in multiple stages, The data written in each stage is read from the memory memory module, the starting address is _i ×n _b +of, and of _is the offset address; the number of bytes written in each stage is n _bt , bit The number is n _bit ; inform the SPI module to start writing the data of the current stage to the antenna through the SPI_START signal; then wait for the SPI_DONE signal fed back by the SPI module, when it is detected that the signal is high, it means that the data writing at this stage is completed; at this time A judgment is made to determine whether all the writing of multiple stages is completed. If not, write the data of the next stage. If it is completed, judge whether the scanning of the codebook is completed. If it is completed, end the process and enter the trigger signal monitoring state of beam tracking, and wait. In the process of beam tracking, if the scanning is not completed, the optimal beam and the corresponding power of the optimal beam are updated; after updating the optimal beam information, a judgment is made to determine whether all beam directions have been searched. 1. Re-write the SPI module. If the search has been completed, use the optimal beam as the execution beam serial number and execute the SPI module write again.

Step 2: In the process of beam initialization and beam tracking, it is necessary to measure the performance of the beam, and calculate the received power through the baseband module at the receiving end of the communication system, and use this as an indicator to judge the performance of the beam direction selected by the current antenna; The performance of all beams within the sub-scanning beam range is measured, and the beam with the best measurement result is taken out for communication in the data transmission phase; at the same time, the beam tracking process is performed periodically, each period is the time of a communication frame, and each There is a beam training field in the frame. When the beam training field starts, the baseband module needs to notify the beam control module of the start of the round of beam tracking. Use the PXI-trigger register to achieve low-latency information transmission. When transmitting the enable signal, set the register to High level, and lasts for a long enough period to ensure that it can be read by the beam control module; the performance of the scanning beam is obtained through the beam measurement process, and the optimal beam and corresponding power are obtained through performance comparison:

Among them, i _max represents the optimal beam, p _max represents the power corresponding to the optimal beam, and p _i represents the power corresponding to the beam i;

In the power acquisition of the current beam, in order to compensate for the delay caused by power calculation and PXIe bus transmission, the waiting time τμs before the current optimal beam is updated.

当前时刻的最佳波束：Step 3: After the beam initialization is performed, the entire codebook space is obtained

The best beam at the current moment:

After the initial process is completed, the beam training phase is entered. This phase is enabled by the training trigger signal and is completed by establishing a beam training state machine. The state transition process is shown in Figure 2. The beam tracking phase adopts a random selection optimization model and presets multiple beam behaviors , each behavior is represented as (b _last , o, c), b _last represents the best beam obtained from the last beam training, the initial value is i _max , o represents the offset of the first beam in the behavior relative to b _last , c represents the number of beams in the behavior, and the beams in the behavior are continuous; the beams included in (b _last ,o,c) are (b _last -o,b _last -o+1,b _last -o+2 ,...,b _last -o+c-1); after starting the training, first initialize the parameters required for the training process, and establish multiple parameter arrays to store the offset value of the behavior, the number of beams, the average return value, and the selected round. number, and UCB (Upper Confidence Bound) value, the index of each array corresponds to the behavior number one-to-one; after initialization, select the behavior with the largest UCB value from the UCB array for training, and use the behavior as an index to obtain its offset value The corresponding value in the array and the number of beams array, generate the candidate beam set of the behavior; the initial value of UCB should be set to a maximum value so that each behavior is trained at least once; For measurement, the antenna writing process of the beam is consistent with the process of beam initialization. The baseband module calculates the energy of the current beam received signal in the time domain by the following formula:

where y _i is the time domain signal after channel estimation and equalization at the receiving end, and T represents the calculated length of the received signal;

After all the beams in the behavior are measured, the best beam is selected, and the process is the same as beam initialization; then the reward for this training needs to be determined:

Among them, r ₁ and r ₂ are fixed values, and p _thr is the power threshold of the receiver; according to the reward value, the average reward array is updated through the current behavior index, and the behavior selects the parameter values corresponding to the round number array and the UCB value array, and at the same time, the total training round is updated. count:

Among them, μ represents the average reward value, n _sel represents the number of rounds selected for this training behavior, n _to represents the total number of rounds, and u represents the UCB value.

Step 4: The SPI module obtains the corresponding byte data from the memory according to the beam index and data volume parameters passed in by the beam initialization and beam training state machine, and then writes the data into the antenna through the serial port through the custom high-speed SPI writing protocol; Define the SPI write protocol flow as shown in Figure 3, obtain the starting address of the data to be written _i ×n _b +of , the total number of bytes n _bt , the number of bits is n _bit , and then all n _bt data Write into the FIFO storage structure; establish a control to indicate the remaining number of bits to be written, and initialize it to the total number of bits n _bit , and write the clock signal to the clock interface at the same time; set the control v to store the data written to the antenna in each round, Then write the data into the antenna bit by bit, v is a data of 1 bit size; set the initial bit index in v, and start writing one by one from the bit data corresponding to the index; because the total number of bits to be written, n _bits are usually not Integer multiples of l, so the first few bits of the first data written are usually invalid bits. At this time, the initial bit index is set to the beginning of the valid bits, and the value is l-(l×n _bt -n _bit ) -1; read data from the FIFO storage structure and store it in v, and judge whether the total number of bits is equal to the remaining number of bits. If they are equal, it means that the writing is in the beginning stage, and the chip select signal and antenna enable signal need to be pulled down. At the same time, set the initial bit index of v as above; if it is not equal, it means that it is not the first time to read from the FIFO at this moment, and you only need to set the initial bit index in v to 1; write the clock to the antenna in each working cycle of the FPGA signal, and write the bit data corresponding to the current index in v at the same time, then reduce the index by 1, and reduce the number of remaining bits to be written by 1; judge whether the index is 0, and repeat the process of writing bits and clock signals if it is not 0. If it is 0, judge whether the remaining number of bits to be written is 0. If it is not 0, continue to read data from the FIFO and perform bit writing. If it is 0, it means that n _bit data writing is completed, and reset at this time. Chip select signal and antenna enable signal, since the antenna needs response time, the enable signal should be reset later than the chip select signal according to the length of the response time.

Step 5: Set the minimum communication quality threshold p _lim , which represents the minimum performance threshold for maintaining normal communication quality; set up a control for the number of consecutive failures n _fail =0, and judge whether the training result at this time meets the standard according to the optimal beam corresponding power value obtained by beam training, If the minimum communication standard is not met, n _fail =n _fail +1, and the beam used in the data transmission phase is still the last training result that met the standard. If the standard is met, n _fail =0; when n _fail reaches a certain threshold n _lim When re-executing the initial beam-building steps to restore the optimal beam.

Step 6: In the data transmission phase of each frame, the optimal beam obtained by training is used for communication. At the same time, the beam control module stays in the enable detection phase and continues to detect the PXI-trigger register signal. If a high level is detected, the beam training process is entered. .

Figure 5 shows a schematic diagram of a training period, each communication frame is a training period, including a beam training phase and a data transmission phase. Based on the NI millimeter wave prototype verification machine platform, a specific implementation case is described, in which the FPGA 7902 is used as the baseband processing part, the 7820 is used as the beam control part, the code environment is LabVIEW 2015, the training period is t _s , and each training time is _tr . The number of actions and the beam offset and number of beams for each action. Calculate the power value in the 7902:

By establishing the point-to-point communication FIFO transmission power value of 7902 and 7820, the trigger signal is established to transmit the beam training enable signal in real time. The array antenna adopts the phased array antenna ZMB28-64TRA. It needs to write n _bits of data in four stages when switching one direction, and supports SPI serial port high-speed mode writing. The beam direction is controlled by the phased matrix H, which contains a total of 64 antenna elements. The relationship between beam direction θ and phase shift is given by:

代表水平或垂直相移，k₀是自由空间中的波数，d是元素间距。通常，d取λ/2，取2π/λ。根据角度值计算出相移值，然后根据相移值生成控制帧数据存入7820的memory中。in,

represents the horizontal or vertical phase shift, _k0 is the wavenumber in free space, and d is the element spacing. Usually, d takes λ/2 and takes 2π/λ. Calculate the phase shift value according to the angle value, and then generate the control frame data according to the phase shift value and store it in the memory of the 7820.

After the FPGA works, the initialization process is carried out. The offsets of all options are initialized as _{ o ₁ , o ₂ ,...on }, the number of options training is {0,0,...,0}, the total number of training _nto =0, the current The beam sequence number i=0, the optimal beam sequence number i _max =0, and the optimal beam corresponding power p _max =0. The codebook is designed as

Switch all beams of the codebook in turn, write data into the antenna in four stages each time, and write n _b bytes of data in total in the four stages. The data written in each stage is read from the memory memory module. The starting address is i×n _b +of , and of _{f is} the offset address. The number of bytes written in each stage is n _bt , and the number of bits is n _bit . Through the SPI_START signal, the SPI module is notified to start writing the data of the current stage to the antenna. Then wait for the SPI_DONE signal fed back by the SPI. When it is detected that the signal is high, it means that the data writing at this stage is completed. At this time, a judgment is made to determine whether all the writing of multiple stages is completed. If not, write the data of the next stage. If it is completed, it is judged whether the scanning of the codebook is completed. If it is completed, the process is ended and the trigger signal monitoring state of beam tracking is entered. , wait for the beam tracking process, and update the optimal beam and the power corresponding to the optimal beam if the scanning is not completed. After updating the optimal beam information, a judgment is made to determine whether all beam directions have been searched. If the search is not completed, add 1 to the beam serial number and rewrite the SPI module. If the search has been completed, according to the obtained optimal beam i _max , The optimal beam is used as the execution beam serial number to execute the SPI module writing again, and the last optimal beam b _last = i _max is updated to enter the beam tracking stage and wait for the beam training trigger signal.

When the trigger signal is high, the beam training process is triggered, and it is triggered every t _s . After initializing the parameters, choose the best UCB option:

Get the start and end sequence numbers of the search beam range:

i=b _last +o _a ,i _final =i+c

The beams within the training options are switched for each beam direction by defining the SPI protocol. Select the beam and power, and update the parameters according to the training results. Set the reward for this training to:

the best

x ₁ and x ₂ are fixed values, and the optimization goal is to track the success rate, that is, the average reward value:

Update both the training count and the total training count for the option:

Calculate the UCB value of each option as follows:

Perform error monitoring on the training results to determine whether the selected beam direction meets the standard. By comparing the optimal beam power with the minimum standard power, if it fails to meet the standard, the number of consecutive misalignments will be increased by one, n _fail = n _fail +1, and at the same time the optimal beam power will be increased by one. The beam direction is set to the last best beam, and if it reaches the standard, n _fail =0. If the number of consecutive misalignments reaches a certain threshold _nfail > _nlim , then re-enter the initialization process to scan the codebook to restore the optimal beam, otherwise wait for the next training trigger signal.

In the data transmission phase of each frame, the optimal beam obtained by training is used for communication. At the same time, the beam control module stays in the enable detection phase and continues to detect the PXI-trigger register signal. If it detects a high level, it enters the beam training process.

When transmitting live video data through this platform, as shown in Figure 6, the receiving end is in a real-time moving state. By designing a beam control module to implement an FPGA-based intelligent beam tracking algorithm, the continuous alignment of the sending and receiving ends can be realized, and the live video playback of the receiving end can be realized. Always keep it smooth and clear, without the phenomenon of stuttering and blurring.

The above is only the preferred embodiment of the present invention, it should be pointed out that: for those skilled in the art, without departing from the principle of the present invention, several improvements and modifications can also be made, and these improvements and modifications are also It should be regarded as the protection scope of the present invention.

Claims (7)

1. a high-speed beam control method in a dynamic millimeter wave communication scene based on FPGA, is characterized in that: comprise the following steps: Step 1, at the receiving end of the communication system, initialize the beam by scanning the full codebook, and establish an initial communication beam; Step 2, establishing data communication between the beam control module and the baseband module equipped with the FPGA through the FIFO data structure and the PXI-trigger register, for triggering the beam tracking process and performing beam measurement; Step 3, establishing a beam training state machine to implement a beam tracking algorithm model; In step 4, the control of the antenna by the beam control module is realized by transmitting the antenna control information through the SPI module in FIG. 4; Step 5: Judge the beam quality after each beam training, and if it is judged that the beam selection fails, beam recovery needs to be performed; Step 6: After the beam training ends, select the best transmission beam for data transmission, and wait for the training trigger signal of the next cycle. 2. the high-speed beam control method in the FPGA-based dynamic millimeter wave communication scene according to claim 1, is characterized in that: described step 1 is as follows: The codebook space is defined as

Among them, M is the size of the codebook, and f _i represents the antenna response vector; according to the antenna format, the control bit stream corresponding to each codeword is written into the memory in advance in the form of bytes; the beam control module at the receiving end of the communication system A full scan is performed, and the realization of the full scan is realized by establishing an initialization state machine. The state transition process is: first initialize the wave number sequence number i=0, the optimal beam sequence number i _max =0, the power corresponding to the optimal beam p _max =0, Then enter the SPI writing process. According to the antenna requirements, it is necessary to write beam control information, send and receive control information, and enable signals into the antenna in multiple stages; a total of n _b bytes of data are written in multiple stages, and each stage The written data is read from the memory memory module, the starting address is _i ×n _b +of, and of _is the offset address; the number of bytes written in each stage is n _bt , and the number of bits is n _bit ; Inform the SPI module to start writing the data of the current stage to the antenna through the SPI_START signal; then wait for the SPI_DONE signal fed back by the SPI module, when it is detected that the signal is high, it means that the data writing at this stage is completed; At this time, a judgment is made, Whether the writing of multiple stages is completed, if not, write the data of the next stage, if completed, judge whether the codebook is scanned or not, if completed, end the process and enter the trigger signal monitoring state of beam tracking, waiting for beam tracking. Process, if the scanning is not completed, update the optimal beam and the power corresponding to the optimal beam; after updating the optimal beam information, make a judgment to determine whether all beam directions have been searched, if not, add 1 to the beam serial number and perform SPI again Module write, if the search has been completed, then execute the SPI module write again with the optimal beam as the execution beam serial number. 3. the high-speed beam control method in the dynamic millimeter wave communication scene based on FPGA according to claim 1, is characterized in that: described step 2 is as follows: In the process of beam initialization and beam tracking, beam performance measurement needs to be performed, and the received power is calculated by the baseband module at the receiving end of the communication system, which is used as an indicator to judge the performance of the beam direction selected by the current antenna. The performance of all beams within the range is measured, and the beam with the best measurement result is taken out for communication in the data transmission phase; at the same time, the beam tracking process is carried out periodically, each period is the time of a communication frame, and each frame contains Beam training field. When the beam training field starts, the baseband module needs to notify the beam control module of the start of the round of beam tracking. The PXI-trigger register is used to achieve low-latency information transmission, and the register is set to high level when the enable signal is transmitted. , and lasts a long enough period to ensure that it can be read by the beam control module; the performance of the scanning beam is obtained through the beam measurement process, and the optimal beam and corresponding power are obtained through performance comparison:

Among them, i _max represents the optimal beam, p _max represents the power corresponding to the optimal beam, and p _i represents the power corresponding to the beam i; In the power acquisition of the current beam, in order to compensate for the delay caused by power calculation and PXIe bus transmission, the waiting time τμs before the current optimal beam is updated. 4. the high-speed beam control method in the FPGA-based dynamic millimeter wave communication scene according to claim 1, is characterized in that: described step 3 is as follows: The entire codebook space is obtained after beam initialization is performed

The best beam at the current moment:

After the initial process is completed, the beam training phase is entered. This phase is enabled by the training trigger signal and completed by establishing a beam training state machine. The state transition process is as follows: the beam tracking phase adopts a random selection optimization model, and presets multiple beam behaviors. Each behavior is represented as (b _last , o, c), b _last represents the best beam obtained from the last beam training, the initial value is i _max , o represents the offset of the first beam in the behavior relative to b _last , and c represents the behavior in the The number of beams, and the beams in the behavior are continuous; the beams included in (b _last ,o,c) are (b _last -o,b _last -o+1,b _last -o+2,…,b _last -o+c-1); after starting training, first initialize the parameters required for the training process, create multiple parameter arrays to store the offset value of the behavior, the number of beams, the average return value, the number of rounds selected, and the UCB value, the index of each array corresponds to the behavior number one-to-one; after initialization, select the behavior with the largest UCB value from the UCB array for training, and use the behavior as an index to obtain its corresponding value in the offset value array and the beam number array , to generate the candidate beam set of the behavior; the initial value of UCB should be set to a maximum value so that each behavior is trained at least once; the receiving end of the communication system measures the beam sets in the behavior respectively, and the antenna writing process of the beam Consistent with the beam initialization process, the baseband module calculates the energy of the current beam received signal in the time domain by the following formula:

where y _i is the time domain signal after channel estimation and equalization at the receiving end, and T represents the calculated length of the received signal; After all the beams in the behavior are measured, the best beam is selected, and the process is the same as beam initialization; then the reward for this training needs to be determined:

Among them, r ₁ and r ₂ are fixed values, and p _thr is the power threshold of the receiver; according to the reward value, the average reward array is updated through the current behavior index, and the behavior selects the parameter values corresponding to the round number array and the UCB value array, and at the same time, the total training round is updated. count:

Among them, μ represents the average reward value, n _sel represents the number of rounds selected for this training behavior, n _to represents the total number of rounds, and u represents the UCB value. 5. the high-speed beam control method in the FPGA-based dynamic millimeter wave communication scene according to claim 1, is characterized in that: described step 4 is as follows: The SPI module obtains the corresponding byte data from the memory according to the beam index and data volume parameters passed in by the beam initialization and beam training state machine, and then writes the data into the antenna through the serial port through the custom high-speed SPI write protocol; custom SPI write The process of entering the protocol is: obtain the starting address of the data to be written _i ×n _b +of , the total number of bytes is n _bt , the number of bits is n _bit , and then write all the n _bt data into the FIFO storage structure; Set up a control to indicate the remaining bits to be written, and initialize it to the total number of bits n _bits , and write the clock signal to the clock interface at the same time; set the control v to store the data written to the antenna in each round, and then write the data bit by bit Enter the antenna, v is a data of 1-bit size; set the initial bit index in v, and write one by one from the bit data corresponding to the index; since the total number of bits to be written, n _bits , is usually not an integer multiple of 1, so write The first few bits of the first data entered are usually invalid bits. At this time, the initial bit index is set to the start of the valid bits, and the value is l-(l×n _bt -n _bit )-1; the storage structure from the FIFO The read data is stored in v, and it is judged whether the total number of bits is equal to the remaining number of bits. If they are equal, it means that it is at the beginning of writing. It is necessary to pull down the chip select signal and the antenna enable signal, and at the same time index the initial bit of v. Set as above; if they are not equal, it means that it is not the first time to read from the FIFO at this moment, just set the initial bit index in v to 1; write the clock signal to the antenna in each working cycle of the FPGA, and write v at the same time The bit data corresponding to the current index in the index is then decremented by 1, and the remaining number of bits to be written is decremented by 1; it is judged whether the index is 0, and if it is not 0, the process of writing bits and clock signals is repeated. If it is 0, judge whether the remaining number of bits to be written is 0. If it is not 0, continue to read data from the FIFO and perform bit writing. If it is 0, it means that n _bit data writing is completed, and reset at this time. Chip select signal and antenna enable signal, since the antenna needs response time, the enable signal should be reset later than the chip select signal according to the length of the response time. 6. The high-speed beam control method in the FPGA-based dynamic millimeter-wave communication scenario according to claim 1, wherein the step 5 is as follows: Set the minimum communication quality threshold p _lim , which represents the minimum performance threshold for maintaining normal communication quality; set up a control for the number of consecutive failures n _fail = 0, and judge whether the training result at this time meets the standard according to the optimal beam corresponding power value obtained from beam training. The minimum communication standard is n _fail =n _fail +1, and the beam used in the data transmission phase is still the training result that met the standard last time. If the standard is met, n _fail = 0; when n _fail reaches a certain threshold n _lim , execute again Steps of initial beam establishment to restore optimal beam. 7. the high-speed beam control method in the dynamic millimeter wave communication scene based on FPGA according to claim 1, is characterized in that: described step 6 is as follows: In the data transmission phase of each frame, the optimal beam obtained by training is used for communication. At the same time, the beam control module stays in the enable detection phase and continues to detect the PXI-trigger register signal. If it detects a high level, it enters the beam training process.

Images