An Efficient Dynamic Engineering Implementation Architecture for MIMO Radar System

Abstract

Multiple-input multiple-output (MIMO) technology can significantly improve radar system resolution and target detection ability and become the core technology in millimeter-wave radar systems. Each antenna of a MIMO radar system generates a separate data stream, resulting in an exponential increase in the amount of data. The MIMO radar system needs to process these data efficiently and quickly to ensure the real-time performance of the system. Conventional MIMO radar engineering systems have difficulty meeting the efficient processing requirements for large data volume processing, and an efficient dynamic engineering implementation architecture for a MIMO radar system is necessary. In such cases, we analyze the MIMO radar engineering system principle and propose an efficient dynamic MIMO radar engineering implementation architecture, which reduces the data storage step after the Doppler fast Fourier transform (FFT) and improves the storage efficiency and processing speed. In addition, we validate the effectiveness of the method by implementing it on a field-programmable gate array (FPGA) platform. The results show that the method is better than the conventional MIMO radar engineering implementation architecture regarding storage resource consumption and processing speed. Specifically, for the same antenna channel data volume, the MIMO radar system with the proposed architecture achieves a 50% reduction in storage resource consumption and a 15% improvement in processing speed compared to the conventional architecture. The proposed architecture demonstrates better antenna array compatibility across different antenna arrays. This work provides new ideas and methods for the efficient engineering implementation of MIMO radar systems. This work can support the practical application of MIMO radar, especially the engineering applications of large-scale array MIMO radar systems.

1. Introduction

With their multi-input and multi-output architectures, MIMO radar systems have emerged as a significant development direction in modern radar technology. By integrating MIMO technology with Frequency Modulated Continuous Wave (FMCW) signals, such systems not only enhance the resolution of the radar but also strengthen the target detection performance. The core advantage of MIMO radars lies in their ability to simultaneously detect targets from multiple angles, thereby acquiring more comprehensive target information, including the distance, velocity, and spatial location [1,2]. This capability endows MIMO radars with prominent application advantages in complex environments, such as urban canyons or dense air traffic.

In MIMO radar systems, each antenna serves as an independent data stream generator, which leads to exponential data growth. To handle these massive data volumes, radar systems must possess efficient data processing and storage capabilities [3]. Traditional radar systems might only utilize a single antenna, whereas MIMO radar systems need to process data from multiple antennas, which undoubtedly increases the complexity of the system. The radar system must process and analyze these data in real time or near-real time to achieve rapid target detection and tracking; therefore, how to effectively process and store these large amounts of data while ensuring system performance has become a crucial issue in designing and implementing MIMO radar systems.

To address this challenge, researchers have proposed various solutions spanning multiple aspects, including hardware design, algorithm optimization, and resource management strategies. In terms of hardware design, by employing high-performance hardware components, such as a Digital Signal Processor (DSP) [4], FPGA [5] and a Graphics Processing Unit (GPU) [6], the speed of signal processing can be significantly enhanced. This high-performance hardware can rapidly process vast amounts of data, meeting the system’s real-time performance requirements [7,8]. However, utilizing such hardware also introduces issues such as increased costs, heat dissipation problems, and difficulties in physical size management [9], all of which may limit the application and dissemination of the system. In terms of algorithm optimization, techniques such as compressive sensing, data sparsification, and improved FFT algorithms have been proposed to reduce the demand for computational resources. These techniques effectively reduce the amount of data to be processed, alleviating the burden on hardware and improving system operating efficiency. However, the optimization and implementation of these algorithms often increase the complexity of the system, requiring elaborate design and maintenance. Moreover, their performance usually depends on specific operating conditions and system configurations [10,11].

Modern technologies such as compressed sensing and deep learning have demonstrated higher efficiency or lower computational load in certain specific scenarios. However, the classic FFT architecture remains widely used due to its long-standing engineering application experience and well-established hardware implementations. Engineers have conducted extensive optimizations on this architecture, particularly in memory management and efficient utilization of storage bandwidth. These studies rely on a two-storage structure, where Range FFT data and Doppler FFT data are stored separately [12,13]. In this study, we refer to this architecture as the conventional engineering implementation architecture.

In this paper, we aim to enhance storage efficiency and processing speed for MIMO radar systems by modifying the engineering implementation architecture, without relying on expensive hardware upgrades or complex algorithmic tuning. We analyze the limitations of the conventional architecture and propose an efficient dynamic engineering implementation architecture. Our proposed method leverages the characteristics of angle estimation, which depends on target detection results and the selectivity and latency of input data. Instead of precomputing and storing the entire data set’s Doppler FFT in advance, this method dynamically extracts data corresponding to specific target ranges and performs real-time Doppler FFT computation after target detection. Subsequently, a secondary filtering process is applied based on velocity information, reducing storage resource consumption while enhancing flexibility in data storage and management. Additionally, a data retrieval and reordering mechanism is introduced to improve adaptability to different antenna array configurations. Experimental results demonstrate that the efficient dynamic MIMO radar engineering system outperforms conventional systems in storage resource consumption and real-time processing performance. Specifically, under the same antenna channel data volume, the proposed efficient dynamic engineering implementation architecture achieves a 50% reduction in storage resource consumption and a 15% improvement in processing speed. Furthermore, it exhibits enhanced MIMO antenna array compatibility, where, given a fixed number of virtual antenna channels, the system can adapt to changes in antenna array configurations simply by reconfiguring the array information. Notably, the overall processing time remains stable regardless of variations in the antenna array layout, ensuring system robustness and efficiency.

The structure of this paper is organized as follows: Section 2 discusses the fundamental principles of MIMO radar and analyzes the architecture of traditional MIMO radar engineering implementation systems and their limitations. Section 3 proposes a new efficient dynamic MIMO radar engineering implementation architecture and describes the process implemented on an FPGA platform for this efficient dynamic MIMO radar engineering system. Section 4 constructs two types of radio frequency (RF) boards, each equipped with different antenna arrays, to assess the effectiveness of the dynamic MIMO radar engineering implementation system. Section 5 analyzes the performance of the efficient dynamic MIMO radar engineering implementation system through practical experiments. Section 6 concludes the work presented in this paper.

2. MIMO FMCW Radar Principle and System Analysis

2.1. Fundamentals of FMCW Radar

FMCW radar has become the most widely used system in MIMO radar applications due to its ability to simultaneously measure target distance and velocity, its high resolution and measurement accuracy, and its relatively simple and cost-effective hardware implementation with high energy efficiency. Moreover, the continuous wave signal characteristics of FMCW radar align seamlessly with the multi-channel transmission and reception framework of MIMO technology, enabling significant improvements in spatial resolution and multi-target detection capabilities. Accordingly, this paper introduces the fundamental principles of radar using FMCW as a representative example. The following sections provide a detailed explanation of the working principles of FMCW radar, including the methodologies for measuring target distance, velocity, and angle. FMCW radars acquire target information by transmitting FMCW signals and capturing the reflected signals from the target. The radar system exhibits a time delay, denoted as $\tau$, between the transmitted and received signals, and this delay represents the time required for the signal to propagate between the radar and the target. In the radar system, the mixer combines the received signal with the contemporaneously transmitted signal to generate an intermediate frequency (IF) signal, commonly called a beat frequency signal [14]. The frequency of this beat frequency signal serves as a basis for calculating the distance between the radar and the target. The correlation between the distance to the target and the IF signal can be mathematically represented as follows:

(1)$R={\displaystyle \frac{c{f}_b{T}_c}{2B}},$

where R denotes the distance between the radar and the target, c is the speed of light, $T_c$ represents the chirp cycle, B refers to the bandwidth, and $f_b$ signifies the frequency of the beat frequency signal. To measure the velocity of a target, the radar system must transmit at least two chirp pulses separated by a time interval $T_c$ [15,16]. Given that the signal frequency substantially exceeds the bandwidth, the phase change $\Delta \varphi$ induced by the target velocity can be articulated as

(2)$\Delta \varphi =2\pi f_{0}t={\displaystyle \frac{4\pi f_0\Delta R}{c}}={\displaystyle \frac{4\pi v{T}_c}{\lambda }},$

where $\lambda$ represents the wavelength of the signal, and v is the velocity of the target. The relationship between the phase difference and the Doppler shift within the chirp period $T_c$ is formulated as

(3)$\Delta \phi =2\pi f_d{T}_c,$

where $f_d$ corresponds to the Doppler shift resulting from the target velocity v. The relationship between the velocity and the Doppler shift is described by

(4)$f_d={\displaystyle \frac{2v}{\lambda }},$

To estimate the azimuth of a target, the radar system must be equipped with at least two receiving antennas [17]. It is assumed that these antennas are uniformly spaced, with a separation distance d between neighboring antennas, and that the angle between the target and the radar is $\theta$. Geometrically, the difference in the propagation distance of the signal between the receiving antennas is $dsin\left(\theta \right)$, resulting in a phase difference $\Delta \omega$ between the signals received at neighboring antennas, expressed as

(5)$\Delta \omega ={\displaystyle \frac{2\pi }{\lambda }}dsin\left(\theta \right),$

Therefore, the angle of arrival $\theta$ of the target can be calculated using the following formula [18]:

(6)$\theta =arcsin\left({\displaystyle \frac{\Delta \omega \lambda }{2\pi d}}\right),$

In practical applications, critical parameters such as the beat signal frequency $f_b$, Doppler shift $f_d$, and phase difference $\Delta \omega$ can be accurately extracted through the implementation of the FFT and signal detection techniques. Angular resolution is the ability of a radar system to distinguish between two targets with the same distance and velocity but different azimuths [19]. Generally, in radar systems, a larger antenna aperture correlates with enhanced angular resolution [20]. The angular resolution of a radar can be estimated by the following equation [21]:

(7)$\Delta \theta ={\displaystyle \frac{\lambda }{D}},$

where $\Delta \theta$ represents the angular resolution in radians, and D denotes the aperture size of the antenna array.

2.2. MIMO Principle Analysis

MIMO radar systems form a virtual aperture by combining multiple transmitting and receiving antennas, which is equivalent to a more extensive physical antenna aperture [22]. The fundamental principle behind utilizing MIMO techniques to enhance the radar antenna aperture is depicted in Figure 1. This figure illustrates a radar configuration with two transmitting (TX) antennas and four receiving (RX) antennas. Using the first RX antenna as a reference, the signal from TX1 generates a phase sequence $[0,\omega ,2\omega ,3\omega ]$ across the four RX antennas. Given that the receiving antennas at TX2 are spaced $4d$ apart from TX1, the signal from TX2 travels an additional path length of $4dsin\left(\theta \right)$ relative to that from TX1. Consequently, the signals at each RX antenna exhibit an incremental phase shift of $4\omega$ compared to the signal from TX1 [23]. Therefore, the phase sequence at the four RX antennas becomes $[4\omega ,5\omega ,6\omega ,7\omega ]$. By amalgamating the phase sequences from the two transmissions across the four RX antennas, the complete sequence becomes $[0,\omega ,2\omega ,3\omega ,4\omega ,5\omega ,6\omega ,7\omega ]$. This configuration of 2TX-4RX antennas effectively synthesizes a virtual array comprising eight RX antennas [24].

In MIMO radar systems, the angular resolution is intrinsically linked to the configuration of the antenna array. The angular resolution can be further enhanced by optimizing the aperture of the virtual array, which typically exhibits a larger effective aperture than the actual physical antenna array [25]. The aperture of the virtual array is defined by the following equation:

(8)$D_{\mathrm{virtua}1}=(N_t·N_r-1)·d,$

where $N_t$ is the number of transmitting antennas, $N_r$ is the number of receiving antennas, and d is the spacing between antennas.

Therefore, the angular resolution of the MIMO radar can be calculated by using the aperture of the virtual array as follows:

(9)$\Delta {\theta }_{\mathrm{MIMO}}={\displaystyle \frac{\lambda }{D_{\mathrm{virtual}}}}$

While MIMO technology significantly enhances the capabilities of radar systems, it also introduces complexities. MIMO radar systems increase the number of virtual antennas by transmitting and receiving multiple times, leading to a substantial increase in data volume. This surge in data volume exerts considerable pressure on the storage resources of the MIMO radar system. Moreover, to fully leverage the benefits of MIMO technology, the system must perform complex signal processing on a large data set within a minimal timeframe, challenging the real-time performance of the system [26]. Exploring more effective engineering implementation strategies is imperative to ensure the efficient processing of large data volumes in MIMO radar systems.

2.3. Double Storage Engineering Implementation Architecture for MIMO Radar Systems

As shown in Figure 2, the widely used double storage MIMO radar system engineering implementation architecture in engineering applications consists of several key components: Range FFT, Range FFT data storage, Doppler FFT, Doppler data storage, target detection, and angle estimation. The specific process is as follows:

Step 1: The Range FFT module obtains the digital IF signal from the analog-to-digital converter (ADC) sampler and performs Range FFT processing.

Step 2: The data are stored in mass memory I. Since the data structure obtained from Range FFT is different from the input format for the subsequent Doppler FFT, the data need to be temporarily stored for dimension conversion and Doppler processing.

Step 3: The Range FFT data stored in mass memory I are managed by the memory read/write control unit, which retrieves the data as needed and inputs them into the Doppler FFT module for processing, forming the complete Range–Doppler Matrix (RDM).

Step 4: The RDM data are fully stored in mass memory II.

Step 5: The target detection algorithm is executed to obtain the target’s range and velocity information.

Step 6: Based on the detected target’s location (range, velocity), the system indexes and retrieves the corresponding antenna channel data from mass memory II, forming the target antenna data vector, which is then input into the angle estimation module [27].

Step 7: Finally, the target’s angle is estimated by executing the angle estimation algorithm.

In the conventional double storage MIMO radar engineering implementation system, the radar data undergo two complete full-frame storage operations between two FFT processes. Assuming that the bit width of the radar data, denoted as $W_d$, remains constant throughout the signal processing flow, the size of the storage resource $S_m$ required for the two data storage operations can be expressed as follows:

(10)$S_m=2W_{d}MK{N}_a,$

where M denotes the number of range sampling points, K signifies the number of velocity sampling points, and $N_a$ represents the volume of virtual channel data.

The total number of data read and write operations, $A_{m1}$, for the two data stores is calculated as follows:

(11)$A_{m1}=3MK{N}_a,$

During the angle estimation phase, retrieving the data corresponding to each detected target across all virtual antenna channels from memory II is imperative. Let C represent the number of detected targets. The total number of data retrieval operations, denoted as $A_{m2}$, can be calculated using the following equation:

(12)$A_{m2}=C{N}_a,$

where $N_a$ denotes the amount of virtual channel data.

If the system uses a processing time unit of T, where T represents the time required by the radar system to process a single data point, the processing time $T_s$ for the entire system can be formulated as follows:

(13)$T_s=(A_{m1}+A_{m2})T+T_o,$

where $T_o$ is the algorithmic elapsed time outside the data stream.

By combining Formulas (11) and (12), the processing time of the entire system can be derived, as presented in the following equation:

(14)$T_s=(3MK{N}_a+C{N}_a)T+T_o,$

3. Efficient Dynamic Engineering Implementation Architecture for MIMO Radar Systems

3.1. Principles of Efficient Dynamic Engineering Implementation Architecture for MIMO Radar Systems

In the conventional system, storing the data after Doppler FFT is necessary for the angle estimation process following target detection. The distance and velocity information of the target within the radar signal cannot be anticipated until the target detection process is concluded. Consequently, the data stored after the Doppler FFT must encompass full-frame data of all virtual antenna channels, thereby imposing high demands on memory resources for the MIMO radar system. The angle estimation algorithm is typically executed after target detection, as it necessitates further processing using the target distance and velocity furnished by the target detection [28]. This implies that the angle estimation module has no requirement for data until the completion of target detection. Moreover, since the objective of angle estimation is to ascertain the azimuth and pitch angles of the detected target, the angle estimation algorithm only needs to process the data associated with a specific target rather than the entire data set. Such a lag and selectivity characteristic enables the system to be optimized concerning data storage and processing.

Therefore, we propose an efficient dynamic engineering implementation architecture for the MIMO radar system. The system stores only the data after the Range FFT and incorporates a state judgment mechanism during the subsequent Doppler FFT processing. If the next operation to be carried out is target detection, all the data are read to perform the Doppler FFT. Conversely, if the next operation is angle estimation, the data after the Range FFT are selectively extracted based on the target detection result to conduct the Doppler FFT. The advantage of this approach lies in its full utilization of the lag characteristic of the angle estimation module. The system selectively reads the data after the Range FFT to perform the Doppler FFT according to the subsequent processing operation rather than immediately storing all the data after the Doppler FFT is executed. This omission of directly storing the data after the Doppler FFT is made possible. Suppose that the next step of the system is angle estimation. In that case, only the data corresponding to the target must be read to perform the Doppler FFT, thereby obtaining the data of all the virtual antenna channels associated with the target. The amount of data required for performing the Doppler FFT, in this case, is substantially less than that of the entire frame of data used for target detection.

Based on this feature, an antenna data reordering and retrieval module is added before angle estimation. This module can sort and filter all the virtual antenna channels corresponding to the target with minimal hardware resources, eliminating redundant antenna channels and ensuring that the data arrangement order aligns with the physical layout of the antenna array. A conventional system performs the Doppler FFT on all the data and stores them. The data required for angle estimation must then be retrieved from the extensive antenna channel data, a process which is inefficient and incurs a high cost when modifying the system in response to changes in the antenna physical arrangement. The proposed system architecture, leveraging the selective Doppler FFT and the data reordering and retrieval module, significantly enhances the data sorting efficiency, reduces the complexity of data antenna array adaptability sorting, and exhibits enhanced antenna array compatibility.

As illustrated in Figure 3, the proposed architecture for the MIMO radar system follows a specific sequence of operations:

• (a). Firstly, the Range FFT module receives the IF signal from the ADC sampler and performs Range FFT, and the data are stored in memory I.

• (b). The system reads the data of all virtual antenna channels sequentially from memory I and inputs them to the Doppler FFT module for processing.

• (c). The 2D-FFT processed data are input into the real-time target detection algorithm for target detection to obtain the distance R and velocity V of the target.

• (d). Based on the distance information of the target, the Doppler FFT is performed by reading the velocity dimension data of all virtual antenna channels at the target distance R from memory I. $N_a$ Doppler data vectors $V_d$ are finally obtained, and the number of $V_d$ is equal to the number of velocity samples K.

• (e). Based on the velocity information V of the target, the antenna data corresponding to the target are extracted from $N_a$ Doppler data vectors. The set $S_a$ of all virtual antenna channel data associated with this target is obtained, and the data set size is $N_a$.

• (f). Configure the antenna array parameters, reorder the data in the data set $S_a$ based on the antenna array arrangement information, and obtain the antenna data vector $V_a$.

• (g). Finally, the antenna data vector $V_a$ is output for angle estimation.

The differences compared with the conventional architecture are as follows:

Firstly, the proposed system solely stores the whole frame data after the Range FFT, eliminating the data storage step following the Doppler FFT. Upon target detection, the velocity dimension data from all virtual antenna channels at the target distance R are retrieved from memory I based on the target distance information, and a real-time target distance-selective Doppler FFT is performed to acquire Doppler data vectors $V_d$. Subsequently, the antenna data corresponding to the target are extracted from these $N_a$ Doppler data vectors based on the velocity information V, resulting in a set of all virtual antenna channel data corresponding to the target $S_a$. This modification leverages the latency in the demand for data after Doppler FFT by the angle estimation module, obviating the need for pre-storage of the entire frame of data after Doppler FFT and target detection, and the Doppler dimensional data are extracted based on the distance information to execute the real-time Doppler FFT. This significantly reduces the volume of data post-distance selection, facilitating subsequent Doppler processing and data storage.

Secondly, the conventional architecture involves storing the entire frame data after 2D FFT, necessitating extensive filtering and sorting from a large data set based on antenna array information when accessing the antenna data for a specific target. This increases the complexity of system design and diminishes antenna array compatibility. Conversely, in the high-efficiency dynamic MIMO radar system, the total data volume after 2D FFT is $C{N}_a$, where C represents the number of targets and $N_a$ denotes the total number of virtual antenna channels. For various antenna arrays, this configuration requires only minimal reordering and retrieval processing based on the antenna array information, with the number of antenna data being $N_a$. This obviates the need to modify other system components, substantially enhancing the compatibility.

In this proposed architecture, only the whole frame of radar data after the Range FFT is retained between two FFT processing steps. Assuming a constant radar data bit width $W_d$ throughout the processing flow, the size of storage resource ${\tilde{S}}_m$ utilized for data storage after Range FFT can be outlined as follows:

(15)${\tilde{S}}_m=W_{d}MK{N}_a,$

where M denotes the number of distance sampling points, K denotes the number of velocity sampling points, and $N_a$ denotes the amount of virtual channel data.

The total number of data reads and writes ${\tilde{A}}_{m1}$ for the data store is as follows:

(16)${\tilde{A}}_{m1}=2MK{N}_a,$

In the angle estimation data reading step, it is necessary to read the velocity dimensional data of all virtual antenna channels corresponding to each target distance from memory I. Assuming that the number of detected targets is C, the number of data readings ${\tilde{A}}_{m2}$ is calculated as follows:

(17)${\tilde{A}}_{m2}=C{N}_{a}K,$

If the processing time unit utilized by the system is T, then the processing time ${\tilde{T}}_s$ of the system can be expressed as follows:

(18)${\tilde{T}}_s=({\tilde{A}}_{m1}+{\tilde{A}}_{m2})T+T_o,$

where $T_o$ is the algorithmic elapsed time outside the data stream.

Integrating Formulas (16)–(18) yields the following derived equation:

(19)${\tilde{T}}_s=(2MK{N}_a+C{N}_{a}K)T+T_o,$

where $T_o$ is the algorithm time consumption outside the data stream.

Given the limited number of targets, where $C\ll M$, the proposed system substantially decreases the usage of storage resources. Consequently, the system processing time ${\tilde{\mathit{T}}}_s$ is notably reduced compared to ${\mathit{T}}_s$, thereby significantly enhancing the real-time performance.

3.2. Engineering Implementation Process

We have deployed both architectures on an FPGA-based platform to evaluate the performance of the efficient dynamic engineering implementation architecture for a MIMO radar system. This section outlines the implementation flow on the FPGA platform and elaborates on the novel key steps involved in data management.

Significantly, to facilitate rapid prototyping and expedite the testing process for validating the effectiveness of the new architecture, we selected the well-acquainted FPGA platform as the core processing unit. Although the proposed architecture presented in this paper is experimentally validated on an FPGA platform, the architecture is optimized by refining the signal processing workflow and employing a modular design approach. Significantly, this optimization is not constrained to any specific hardware platform; hence, the foundational concepts and methodologies are universally applicable. They can be adapted and optimized for different processing platforms according to specific requirements.

3.2.1. FPGA-Based Engineering Implementation Flow

To validate the performance of the proposed system, this paper presents an engineering implementation architecture based on the FPGA platform. The process is depicted in Figure 4 and is outlined as follows:

Step 1: First, initialize the antenna array parameter configuration according to the actual arrangement of the antenna array and configure the system mode to state I.

Step 2: Acquire the ADC data, perform the Range FFT, and store the data in the DDR mass memory.

Step 3: Determine the system mode and perform Doppler FFT. If it is mode state I, read the whole frame and perform a full range dimension Doppler FFT, that is, perform Range FFT on all data.

Step 4: Perform target detection processing on the data after Doppler FFT, obtain the distance and speed of the target, and switch the system mode to state II.

Step 5: Based on the distance information of the target, selectively read the Doppler dimension data at the target distance from the DDR memory for Doppler FFT, and obtain the Doppler data vectors of all the virtual antenna channels with the target distance index.

Step 6: Select the data at the target of all virtual antenna channels from the Doppler data vector based on the velocity information of the target detection result. Obtain the target point data set for all virtual antennas.

Step 7: Using the antenna array information, reorder the data for all virtual antenna channels and filter out redundant channel data to ensure that the data arrangement adheres to the predefined array configuration rules.

Step 8: Conduct angle estimation processing in horizontal and vertical dimensions.

Step 9: Complete the processing for the current data frame and proceed to the next frame.

3.2.2. Efficient Dynamic Data Frame Storage Management

In the engineering implementation process outlined in Section 3.2.1, an efficient and dynamic data frame storage management mechanism is employed to effectively manage the data interchange between the memory and the algorithmic process. The angle estimation step is initiated after target detection, and it is crucial to note that the angle estimation algorithm requires processing only the data pertinent to a specific target rather than the entire data frame. This indicates that the data requirements for the angle estimation algorithm are both delayed and selective. Consequently, rather than storing and processing all data immediately following the Doppler FFT, the system is designed to selectively extract and process data based on the outcomes of the target detection, thus optimizing both storage and processing resources. As shown in Figure 5, in the engineering implementation process, we divide the state of the system after the Range FFT into two types:

When the system is in state I, the data frame storage management module retrieves and outputs radar data across the whole range dimension. This subset of data subsequently forms the RDM after Doppler FFT processing, fulfilling the data requirements for the target detection algorithm.

Conversely, when the system is in state II, the data frame storage management module retrieves data according to the target distance dimension information provided by the target detection algorithm and outputs it. This subset of data forms a Doppler data set after Doppler FFT processing. Following the data reordering process of the antenna array, this data set meets the requirements of the angle estimation algorithm.

Consequently, integrating a state judgment mechanism within the engineering implementation process, in conjunction with a real-time Doppler FFT module, facilitates the fulfillment of data requirements for both target detection and angle estimation in a time-shared approach. Thus, by implementing the state judgment mechanism in the engineering process and pairing it with the real-time Doppler FFT module, we can efficiently meet the data needs for target detection and angle estimation in a time-efficient manner. This strategy significantly optimizes the storage demands for data sets following Doppler FFT processing.

3.2.3. Antenna Data Reordering and Retrieval

As discussed in Section 3.2.2, the data output from the data frame storage management module in State II are typically organized according to the sequence of the transmitting antennas. However, in MIMO radar systems, the precision of angle estimation algorithms critically depends on the exact positioning and the relative configuration of the receiving antennas. This precise arrangement is essential for accurately determining the angles of arrival of incoming signals. If the data are not structured in alignment with the configuration of the receiving antennas, it may result in significant inaccuracies in angle estimation, consequently impairing the overall performance and accuracy of the radar system. Therefore, despite the initial organization of the data according to the sequence of the transmitting antennas, it is imperative to reorganize the data’s suitably to fulfill the specific requirements of the angle estimation algorithms. This reorganization ensures that the radar system operates with optimal accuracy and efficiency. The reordering and retrieval module is tasked with transforming the data from the transmission sequence to the physical arrangement sequence and eliminating duplicate data channels, which may arise due to redundancy in the array design or may be intentionally introduced to enhance signal strength. By reducing the volume of data that require processing in subsequent steps, this module helps to prevent the inclusion of unnecessary, redundant information in the angle estimation process.

The operational principle of the antenna data reordering and retrieval module is depicted in Figure 6. The inputs to this module consist of data from all virtual antenna channels that have undergone velocity-selective filtering, with the antenna array parameters serving as the critical configuration information for this module.

As illustrated in Figure 6, the core implementation of the antenna data reordering and retrieval module is outlined as follows:

Step 1: Firstly, all virtual antenna channel data are stored in the antenna data set random-access memory (RAM). It is important to note that the virtual antenna reception data are arranged sequentially according to the transmission order of the transmitting antennas and are not aligned with the actual physical arrangement of the antennas. Consequently, these data cannot be directly utilized in the angle estimation algorithm.

Step 2: Next, the read data address is calculated based on the antenna array layout information. Specific virtual antenna target point data for angle estimation are then selectively retrieved from the antenna data set RAM in a jumpwise manner to form the target antenna data vector. This target antenna data vector is reorganized to fulfill the data requirements of the angle estimation algorithm.

The target antenna data vector is systematically reordered to comply with the data specifications necessary for the angle estimation algorithm. Given that various radar systems may employ distinct antenna array configurations, the requisite data processing for alternative antenna arrays can be achieved by simply reconfiguring the antenna array information parameters stored in the antenna array parameter storage memory. By dynamically selecting and processing pertinent channels, the system is engineered to automatically adjust to diverse array structures and configurations, enhancing its adaptability and efficiency in different operational contexts.

4. Antenna Array Design and Scenario Analysis

4.1. MIMO Radar Antenna Array Design

We constructed two RF boards, each equipped with distinct antenna arrays, to evaluate the proposed system. Each RF board incorporates four cascaded radar chips, supporting 12 transmitting antennas and 16 receiving antennas. These antennas are utilized to emit millimeter-wave signals and capture the return signals from targets. Every receiver channel on the radar chips is fitted with a high-performance digital sampler, which is responsible for converting the received analog signals into digital format. It is essential to highlight that all RF boards share identical hardware configurations, with the only variation being the arrangement of the antenna array.

Figure 7a,b depict the physical diagram and the arrangement structure of the first antenna array, respectively. This configuration organizes the twelve transmitting antennas and the sixteen receiving antennas in a one-dimensional linear array. As illustrated in Figure 7c, the 192 virtual antenna receive channels are arranged in a one-dimensional linear configuration along the horizontal axis, with an equivalent virtual antenna aperture of 64 times the signal wavelength, which does not facilitate vertical detection capabilities. Taking the 77 GHz signal as an example, the signal wavelength is approximately 3.89 mm, which means the size of the virtual antenna aperture is 248.96 mm. In Figure 7c, receive channels with the same color represent those corresponding to the same transmitting antenna channels. A total of 12 colors are used, indicating that all virtual receiving antenna channels correspond to 12 transmitting antenna channels.

Figure 8a,b illustrate the physical diagram of the second antenna array and the structure of the antenna physical arrangement, respectively. In this configuration, the transmitting and receiving antennas are organized in a two-dimensional planar layout, facilitating target detection across both vertical and horizontal dimensions. This arrangement enhances the system’s capability to capture a more comprehensive spatial representation of the target area. As shown in Figure 8c, the 192 virtual antenna receive channels are arranged in a two-dimensional planar configuration along both the horizontal and vertical directions. The equivalent virtual antenna horizontal aperture is 70 times the signal wavelength, while the vertical aperture is 23.25 times the signal wavelength. Taking the 77 GHz signal as an example, the signal wavelength is approximately 3.89 mm, resulting in a virtual antenna horizontal aperture size of 272.3 mm and a vertical aperture size of 90.44 mm.

In the figure, receive channels represented by the same color correspond to the same transmitting antenna channel. A total of 12 colors are used, indicating that all virtual receiving antenna channels correspond, respectively, to 12 transmitting antenna channels.

The specific parameters of the two antenna arrays are detailed in Table 1. By utilizing these distinct antenna array configurations, we can comprehensively assess the performance of the efficient dynamic MIMO radar engineering implementation system. This approach enables a thorough evaluation of system adaptability and effectiveness across different spatial configurations.

4.2. Test Scenario and Parameter Configuration

We validate and compare the performance of the proposed system by testing it on a moving vehicle on an actual roadway. Figure 9a illustrates how the radar is installed. As shown in Figure 9b, the RF panel antenna is oriented to face the direction of vehicle travel, enabling the radar to perform the spatial detection of other vehicles on the road from a side-view perspective.

The radar data sampling information used in the experiment is shown in Table 2, where the number of distance dimension sampling points is 1024, the number of Doppler sampling points is 256, and the data bit width is 32 bits.

5. Results

5.1. Resource Consumption and Processing Time

As discussed previously, the proposed system demonstrates significant advantages in terms of reduced storage resource consumption and enhanced real-time performance. To quantitatively assess the performance of this system, we measure the storage resource usage and the time elapsed to process a single data frame. Given that the data volume in the MIMO radar system is influenced by various factors, such as the number of sampling points and antennas, we manipulate the number of transmitting channels to vary the data volume of the system for a comprehensive performance evaluation. Adhering to the principle of univariate experimentation, we fix the number of distance sampling points and speed sampling points at 1024 and 256, respectively. Subsequently, we progressively increase the number of active transmitting antenna channels from 1 to 12, incrementally increasing the number of synthesized virtual antenna channels. We then record and compare the storage resource consumption and system time consumption for configurations with varying numbers of virtual antenna channels.

In the experimental outcomes, depicted in Figure 10, the horizontal axis represents the number of activated transmitting antennas, while the vertical axis indicates the size of the storage resources required by the system. The blue line illustrates the storage resource consumption of a MIMO radar system employing a double storage architecture (DS-MRS), whereas the orange line represents the storage resource consumption for a system utilizing an efficient dynamic storage architecture (ED-MRS). It is evident from the graph that as the number of activated transmitting antenna channels increases, the storage resource requirements for both signal processing systems increase accordingly.

Notably, at equivalent numbers of transmitting antenna channels, the proposed system utilizes approximately 50% less storage resources than the conventional system. Furthermore, as the number of transmitting antenna channels increases, the growth rate of storage resource consumption in the efficient dynamic MIMO radar system is more gradual. This finding holds substantial relevance for implementing large-scale array MIMO radar projects.

As depicted in Figure 11, the horizontal axis represents the number of activated transmitting antennas, while the vertical axis indicates the system time consumption for processing a frame of data. The blue, black, and orange lines in the graph represent the time consumption of the proposed systems under different system clock speeds of 100 MHz, 200 MHz, and 300 MHz, respectively. As shown in Figure 11, the elapsed time required for processing a frame of data in both the conventional system and the proposed system increases as the number of virtual antennas grows. However, with an equivalent number of transmitting antenna channels, the proposed system demonstrates a processing speed enhancement of approximately 15% compared to the conventional system. Furthermore, we conducted experiments at three distinct system clock speeds. As illustrated in Figure 11, different colors represent the three clock speeds. The results indicate that the proposed system exhibits more pronounced real-time advantages at lower signal processing clock rates, achieving significant reductions in processing time. This performance enhancement underscores the system suitability for applications requiring high efficiency and speed in real-time signal processing environments.

5.2. Compatibility of Different Arrays

Utilizing a one-dimensional antenna array I and a two-dimensional planar antenna array II, the performance of the proposed system is validated for both planar detection and three-dimensional spatial detection. We transfer 10 frames of data from the processing results to a computer via a network port. We analyze the detection performance of the real-time radar signal processing system using MATLAB R2021bTo validate the system’s output, we simultaneously use real-time kinematic(RTK) system to measure the target’s position in real-time and synchronize these measurements with the radar data based on timestamps. As shown in Figure 12a, the black circles represent the RTK-derived positions corresponding to each radar data point, serving as the reference. The colored trajectories represent radar outputs from different data frames. Due to the large reflective surface of the target, multiple points are detected in each frame.

Figure 12b illustrates the error curve between the radar measurement points and the reference positions. For each frame, the error is calculated as the average distance between all radar measurement points and the reference position. As shown in the figure, the radar-derived trajectory closely matches the reference trajectory, with the average positional error of the radar output points in each frame not exceeding 1.2 m. Therefore, the radar output points accurately represent the actual motion trajectory of the target.

Figure 13a,b display the output results after signal acquisition and real-time processing for the conventional system and the proposed system, respectively. The horizontal axis represents the horizontal position of the target, the vertical axis represents the longitudinal distance of the target, and target traces from the same data frame are denoted by identical colors. Notably, when employing RF board I, while the motion of the target introduces variations across different data frames, the overall results exhibit a high degree of similarity for the same radar system. The signal processing system can detect and output 3–5 point cloud data values per frame for a single target vehicle traveling on an actual road. By aggregating data from multiple frames, the output results of the radar system consistently align with the exact motion trajectory of the target.

Figure 14a,b present the output results from the conventional system (DS-MRS) and the proposed system (ED-MRS), respectively. The horizontal axis represents the horizontal position of the target, the vertical axis represents the longitudinal distance of the target, and the third axis (z-axis) represents the height information of the target, while target traces from the same data frame are denoted by identical colors. When utilizing RF board II, in contrast to the results obtained with the RF board I, the system can acquire information about the vertical orientation of the target. A comparative analysis of the results from the different systems indicates that the proposed system can produce detection results that are consistent with the actual motion of the target. Moreover, it achieves functionality comparable to that of a conventional radar system in one-dimensional planar and two-dimensional spatial detection scenarios.

To comprehensively assess the compatibility and stability performance of the proposed engineering system across different antenna arrays, we processed 10 consecutive frames of data from both antenna array I and antenna array II. We recorded the time that the system took to process each frame and conducted a detailed analysis. Figure 15 depicts the time consumption of the conventional MIMO radar engineering system (DS-MRS), while Figure 16 illustrates the time consumption of the proposed system (ED-MRS). In these figures, the horizontal axis represents the frame number of the data, and the vertical axis represents the system time consumption for processing one frame of data. The blue color indicates the time consumption for processing data from antenna array I, and the orange color represents the time consumption for processing data from antenna array II.

By comparing Figure 15 and Figure 16, it can be concluded that, compared to the proposed system, the conventional system exhibits relatively larger fluctuations in time consumption between antenna array I and antenna array II. The time consumption difference between the two arrays is in the order of several hundred microseconds. This discrepancy arises primarily because the conventional system involves extensive data sorting and filtering within a large volume of stored RDM data. Antenna array I, being a one-dimensional linear array, exhibits lower angular complexity and fewer effective antenna channels, simplifying the data retrieval process for the angle estimation module. Consequently, this results in a reduced processing time for each data frame.

Conversely, in Figure 16, the time consumption for processing data from both antenna arrays is notably similar to the proposed system. This uniformity is attributed to the enhanced system architecture, which facilitates more flexible and compatible data sorting and screening. In this system, the extensive data processing tasks before the antenna data reordering and retrieval module do not necessitate specific adjustments for different antenna arrays, thus minimizing variations in processing time across different data volumes. Furthermore, during the antenna data reorganization stage, the input data have already been filtered through the Doppler FFT based on target distance and velocity information. Consequently, the volume of data requiring processing in the antenna data reorganization retrieval module is significantly reduced, equivalent to the total number of virtual antenna channels. Therefore, changes in the antenna array have a minimal impact on the array matchability data screening and sorting processes. This indicates that the proposed system exhibits superior compatibility with various antenna arrays.

6. Conclusions

In this work, we introduce an efficient dynamic MIMO radar engineering implementation architecture designed for practical applications in large-scale array MIMO radar systems. This architecture offers several advantages: (i) The data after the Range FFT aew stored only once, instead of the traditional twice FFT data storage, which reduces the storage resource consumption of the system. (ii) Data processing after the Range FFT is divided into two modes: target detection and angle estimation. The dynamic Doppler FFTs for the two cases are performed in different time slots, reducing the number of data readings and writings and improving data processing efficiency. (iii) The compatibility and flexibility of the system antenna array are improved by improving the realization architecture of the system. This is of great significance to improve the real-time performances of MIMO radar systems and can support the future trend of MIMO radar systems toward large-scale arrays.

We aim to explore further the integration of our proposed architecture into various hardware platforms, enhancing its adaptability and efficiency. We will further optimize the system architecture to better address application scenarios that require higher resolution and larger-scale arrays, while also increasing processing speeds. We hope the methodologies and insights derived from this work will stimulate innovative approaches in MIMO radar engineering applications and contribute to advancing the field.

Author Contributions

Conceptualization, R.W. and H.Z.; investigation, H.Z.; writing—original draft preparation, R.W.; writing—review and editing, R.W. and L.X. All authors have read and agreed to the published version of this manuscript.

Data Availability Statement

The original contributions presented in the study are included in the article, further inquiries can be directed to the corresponding author.

Conflicts of Interest

This manuscript has not been published or presented elsewhere in part or its entirety and is not under consideration by another journal. There are no conflicts of interest to declare.

Abbreviations

The following abbreviations are used in this manuscript:

MIMO	Multiple-input multiple-output
FFT	Fast Fourier transform
FPGA	Field-programmable gate array
FMCW	Frequency Modulated Continuous Wave
DSP	Digital Signal Processing
GPU	Graphics Processing Unit
RDM	Range–Doppler Matrix
IF	Intermediate frequency
ADC	Analog-to-digital converter
TX	Transmitting antennas
RX	Receiving antennas
RF	Radio frequency
DDR	Double data rate SDRAM
RAM	Random-access memory
RTK	Real-time kinematic

Footnotes

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Figures and Tables

Figure 1. MIMO radar virtual aperture expansion principle.

Figure 2. Double storage engineering implementation architecture for MIMO radar system.

Figure 3. Efficient dynamic engineering implementation architecture for MIMO radar system.

Figure 4. Efficient dynamic MIMO radar system engineering implementation process.

Figure 5. Efficient dynamic data frame storage management.

Figure 6. Antenna data reordering and retrieval.

Figure 7. Antenna array I. (a) RF board. (b) Antenna array layout. (c) Virtual antenna RX channels.

Figure 8. Antenna array II. (a) RF board. (b) Antenna array layout. (c) Virtual antenna RX channels’ arrangement.

Figure 9. Experimental test scenarios. (a) Radar set up real scenarios. (b) Radar position relative to the target.

Figure 10. Storage resource consumption.

Figure 11. System processing time.

Figure 12. Target trajectory. (a) Comparison of radar and RTK output results. (b) Difference between radar and RTK output results.

Figure 13. Antenna array I processing results. (a) DS-MRS antenna array I target trajectory. (b) ED-MRS antenna array I target trajectory.

Figure 14. Antenna array II processing results. (a) DS-MRS antenna array II target trajectory. (b) ED-MRS antenna array II target trajectory.

Figure 15. DS-MRS processing time.

Figure 16. ED-MRS processing time.

Table 1

Antenna array parameters.

Antenna Array Layout	Number of Effective Antennas	Antenna Aperture Size	Direction
I	94	248.96 mm	Horizontal
	-	-	Vertical
II	113	272.3 mm	Horizontal
	32	90.44 mm	Vertical

Table 2

Radar sampling parameters.

Parameter	Description	Value
$f_s$	ADC Sample Rate	8 MHz
$W_d$	Data Bit Width	32 bits
M	ADC Sample Rate	1024
K	Number of Chirps	256

References

1. Zhao, Y.; Liu, H. Overview of MIMO Radar Technology. J. Data Acquis. Process.; 2018; 33, pp. 389-399.

2. Li, X.; Dong, T.; Huang, G. Overview of MIMO Radar and Its Characteristics. Mod. Def. Technol.; 2015; 43, pp. 124–131+149.

3. Larsson, E.G.; Edfors, O.; Tufvesson, F.; Marzetta, T.L. Massive MIMO for next-generation wireless systems. IEEE Commun. Mag.; 2014; 52, pp. 186-195. [DOI: https://dx.doi.org/10.1109/MCOM.2014.6736761]

4. Pang, N.; Xu, P. Implementation of MIMO Radar Signal Processing Based on Multi-core DSP. Appl. Electron. Tech.; 2014; 40, pp. 4-6.

5. Meinl, F.; Schubert, E.; Kunert, M.; Blume, H. Realtime FPGA-based Processing Unit for a High-Resolution Automotive MIMO Radar Platform. Proceedings of the 12th European Radar Conference; Paris, France, 9–11 September 2015; pp. 213-216.

6. Liu, G.; Bao, Y.; Yue, N.; Wang, S.; Wu, H.; Liu, Q.; Yang, W. A GPU-based real-time processing system for frequency division multiple-input-multiple-output radar. IET Radar Sonar Navig.; 2023; 17, pp. 1524-1537. [DOI: https://dx.doi.org/10.1049/rsn2.12439]

7. Pitre, E.; Roberge, V.; Bray, J.; Hefnawi, M. MIMO Radar Hardware Acceleration with Enhanced Resolution. ITM Web Conf.; 2022; 48, 01013. [DOI: https://dx.doi.org/10.1051/itmconf/20224801013]

8. Suto, K.; Hara, Y.; Ohtsuki, T. Throughput maximization transmission control scheme for MIMO systems. Proceedings of the 2004 IEEE 59th Vehicular Technology Conference. VTC 2004-Spring (IEEE Cat. No.04CH37514); Milan, Italy, 17–19 May 2004; Volume 2, pp. 603-607.

9. Chang, Q.; Gong, B.; Lu, K.; Tao, D. A High-Performance Transmission Mechanism for Ground Phased Array Radar Data Stream Based on DPDK. J. Small Micro Comput. Syst.; 2023; 44, pp. 2619-2624.

10. Fromenteze, T.; Yurduseven, O.; Berland, F.; Decroze, C.; Smith, D.R.; Yarovoy, A.G. Accelerating short range MIMO imaging with optimized Fourier processing. Proceedings of the Passive and Active Millimeter-Wave Imaging XXII, SPIE Defense + Commercial Sensing; Baltimore, MD, USA, 18–19 April 2020; Volume 11411, pp. 1-6.

11. Liu, D.; Zhang, Y.; Liu, K. Compressed Sensing Three-Dimensional Imaging Based on FMCW MIMO Millimeter Wave Radar. J. China Univ. Metrol.; 2024; 35, pp. 496-508.

12. Guo, M.; Zhao, D.; Wu, Q.; Wu, J.; Li, D.; Zhang, P. An Integrated Real-Time FMCW Radar Baseband Processor in 40-nm CMOS. IEEE Access; 2023; 11, pp. 36041-36051. [DOI: https://dx.doi.org/10.1109/ACCESS.2023.3265814]

13. Damnjanović, D.; Petrović, M.L.; Milovanović, V.M. On Hardware Implementations of Two - Dimensional Fast Fourier Transform for Radar Signal Processing. Proceedings of the IEEE EUROCON 2023—20th International Conference on Smart Technologies; Torino, Italy, 6–8 July 2023; pp. 400-405.

14. Yang, B.; Zhang, H.; Du, M.; Wang, A.; Xiong, K. Urban Traffic Congestion Alleviation System Based on Millimeter Wave Radar and Improved Probabilistic Neural Network. IET Radar Sonar Navig.; 2024; 18, pp. 327-343. [DOI: https://dx.doi.org/10.1049/rsn2.12443]

15. Xiao, C. Design and Implementation of a Real-Time Processing System for Vehicle-Mounted Millimeter Wave Radar Based on a Domestic GPU. Ph.D. Dissertation; University of Electronic Science and Technology of China: Chengdu, China, 2024.

16. Kim, J.; Khang, S.; Choi, S.; Eo, M.; Jeon, J. Implementation of MIMO Radar-Based Point Cloud Images for Environmental Recognition of Unmanned Vehicles and Its Application. Remote Sens.; 2024; 16, 1733. [DOI: https://dx.doi.org/10.3390/rs16101733]

17. Yang, B.; Zhang, H.; Chen, Y.; Zhou, Y.; Peng, Y. Urban Traffic Imaging Using Millimeter-Wave Radar. Remote Sens.; 2022; 14, 5416. [DOI: https://dx.doi.org/10.3390/rs14215416]

18. Rai, P.K.; Kumar, A.; Khan, M.Z.A.; Soumya, J.; Cenkeramaddi, L.R. Angle and Height Estimation Technique for Aerial Vehicles using mmWave FMCW Radar. Proceedings of the 2021 International Conference on Communication Systems & Networks (COMSNETS); Bangalore, India, 5–9 January 2021; pp. 104-108.

19. Liao, B. Fast Angle Estimation for MIMO Radar With Nonorthogonal Waveforms. IEEE Trans. Aerosp. Electron. Syst.; 2018; 54, pp. 2091-2096. [DOI: https://dx.doi.org/10.1109/TAES.2018.2847958]

20. Ji, C.; Zhang, F. Dimensionality Reduction DOA Estimation Method for Non-Circular Signals in Coprime MIMO Radar. J. Microwaves; 2022; 38, pp. 95-100.

21. Skolnik, M.I. Introduction to Radar Systems; 3rd ed. McGraw-Hill: New York, NY, USA, 2001.

22. Liu, G.; Yang, W.; Li, P.; Qin, G.; Cai, J.; Wang, Y.; Wang, S.; Yue, N.; Huang, D. MIMO Radar Parallel Simulation System Based on CPU/GPU Architecture. Sensors; 2022; 22, 396. [DOI: https://dx.doi.org/10.3390/s22010396]

23. Yang, B.; Liu, S.; Zhang, H.; Zhou, Y. A Velocity Ambiguity Resolution Algorithm Based on Improved Hypothetical Phase Compensation for TDM-MIMO Radar Traffic Target Imaging. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens.; 2024; 17, pp. 3409-3424. [DOI: https://dx.doi.org/10.1109/JSTARS.2024.3352082]

24. Tan, X.; Roberts, W.; Li, J.; Stoica, P. Sparse Learning via Iterative Minimization With Application to MIMO Radar Imaging. IEEE Trans. Signal Process.; 2011; 59, pp. 1088-1100. [DOI: https://dx.doi.org/10.1109/TSP.2010.2096218]

25. Huang, L.; Liu, A.; Gao, C. Research Progress on Centralized MIMO Radar: Orthogonal Waveform Design and Signal Processing. Radar Sci. Technol.; 2023; 21, pp. 1–15+23.

26. Wei, Z. Research on Large-Scale MIMO Communication and Radar Signal Processing Technology. Ph.D. Dissertation; Beijing University of Posts and Telecommunications: Beijing, China, 2024.

27. Pitre, E.; Roberge, V.; Bray, J.; Hefnawi, M. Comparison of Massively Parallel Algorithms on Graphics Processing Unit for MIMO Radar. e-Prime-Adv. Electr. Eng. Electron. Energy; 2022; 2, 100063. [DOI: https://dx.doi.org/10.1016/j.prime.2022.100063]

28. He, Z.; Cheng, Z.; Li, J.; Zhang, W.; Shi, J.; Su, Y.; Deng, M. A Review of Centralized MIMO Radar Research. J. Radar; 2022; 11, pp. 805-829.