Beamforming hardware accelerator for radar systems

By optimizing beamforming analysis of the radar system using hardware accelerators and by iteratively storing and processing data slices, the resource-intensive problem was solved, and the computational efficiency and response speed of the radar system were improved.

CN116348786BActive Publication Date: 2026-11-17TEXAS INSTRUMENTS INC
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202180073167.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2020-11-10
Filing Date
2021-11-01
Publication Date
2026-11-17
Estimated Expiration
2041-11-01

AI Technical Summary

Technical Problem

Existing radar systems are resource-intensive when performing beamforming analysis, resulting in high computational complexity and affecting system response time and angular resolution.

Method used

Hardware accelerators (HWA) are used for iterative storage and processing of data slices. By using components such as address generators, selector logic circuits, multipliers and accumulators, data extraction and transposition operations are reduced, and beamforming operations are optimized.

Benefits of technology

Without reducing angular resolution or update rate, computational complexity is significantly reduced, and the processing efficiency and response speed of the radar system are improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116348786B_ABST
    Figure CN116348786B_ABST
Patent Text Reader

Abstract

A non-transitory computer-readable medium storage causes a processor to obtain instructions for an N×M range matrix and an M×S manipulation matrix, the N×M range matrix comprising radar data indexed by rate and antenna, and the M×S manipulation matrix comprising the expected phase indexed by antenna and assumed angle. For each unique X×Y range slice corresponding to a specific set of X rates, the processor stores the specific range slice in a first buffer (720). For each unique Y×Z manipulation slice corresponding to a specific set of Y antennas, the processor stores the specific manipulation slice in a second buffer (735). The processor performs beamforming operations on the range slices, manipulation slices, and intermediate slices, thereby storing the results as intermediate slices in a third buffer (780). After each manipulation and range slice of a specific set of X rates has been iterated, the processor stores the intermediate slices as beamforming slices of a specific set of X rates and assumed angles.
Need to check novelty before this filing date? Find Prior Art

Description

Background Technology

[0001] Some radar systems use beamforming to determine the angle of arrival of objects in the field of view. A radar system implementing beamforming first performs a range Fast Fourier Transform (FFT) and a Doppler FFT, resulting in a three-dimensional (3D) matrix consisting of a set of two-dimensional (2D) range matrices. Each 2D range matrix corresponds to a specific range of distances between the target and the radar system, called a range bin, and contains the rate range of the target through the antenna array, called a rate bin. The radar system performs beamforming analysis to determine which of a set of hypothetical angles, represented by manipulation vectors, most closely aligns with the target's angle of arrival. To test this set of hypothetical angles, each row of each 2D range matrix in the 3D data matrix is ​​multiplied by each manipulation vector, and the resulting spectra are summed. If a particular hypothetical angle corresponds to the actual angle of arrival of the target, the resulting spectra are constructively summed; otherwise, the resulting spectra are destructively summed.

[0002] Multiplying each row of each 2D extent matrix by each control vector results in each row and each control vector being extracted and re-extracted multiple times during beamforming analysis. These vector multiplications become more complex due to the combination of row-by-row and column-by-column operations, causing the 2D extent matrix to be transposed multiple times during beamforming analysis. Therefore, the total number of operations involved in beamforming analysis is quite high, making the radar system resource-intensive to execute.

[0003] For illustration, the example radar system calculates 256 range bins, 256 rate bins, and 16 hypothetical angles. The radar system performs 12 operations to test each hypothetical angle. In total, the radar system performs over 12 million operations during beamforming analysis to determine the angle of arrival of objects within the radar system's field of view. The signal processors capable of performing these operations are resource-intensive in terms of semiconductor die area, computation time, memory space, and so on. Some radar systems reduce the number of hypothetical angles to decrease the number of beamforming operations, which also reduces the system's angular resolution. Some radar systems reduce their update rate to allow for the additional processing time used for beamforming operations, which reduces the system's reaction time to objects in the field of view. Summary of the Invention

[0004] An apparatus includes one or more processor cores and one or more non-transitory computer-readable media storing machine instructions, which, when executed by the one or more processor cores, cause the one or more processor cores to obtain a data matrix and iteratively store data slices from the data matrix in a data buffer. Each data slice includes a unique portion of the data matrix. The processor core performs related operations on the data slices in the data buffer to obtain computed data slices of the data slices and stores the computed data slices. In some embodiments, the related operations cause the processor core to multiply and sum the data slices, a reference matrix, and intermediate slices. The processor core iteratively stores reference slices from the reference matrix in a second data buffer. Each reference slice includes a unique portion of the reference matrix corresponding to an index range contained in the data slice. The processor multiplies the reference slices with the data slices, sums the result with the intermediate slices, and stores the result as the intermediate slices.

[0005] The device further includes a hardware accelerator (HWA) comprising a first address generator, the data buffer, a first selector logic circuit (SLC), a multiplier and accumulator (MAC), a second SLC, a second data buffer, a third SLC, and a second address generator. The first address generator generates addresses for the data slices such that the data slices include a unique portion of the data matrix and provides the addresses to the memory interface system. The data buffer stores the data slices. The first SLC selectively outputs transposed data slices. The MAC uses the transposed data slices to perform vector multiplication and sums the multiplication result with an intermediate slice from the second data buffer.

[0006] The second SLC selectively outputs one of the transposed data slice, the multiplication result, and the summation result. The second data buffer stores the output of the second SLC. The third SLC selectively outputs a computed data slice including the transposed summation result. The second address generator generates addresses for the computed data slices such that each computed data slice includes a unique portion of the computed data matrix and provides the address of the computed data slice to the memory interface system.

[0007] The HWA may further include a third address generator, a third data buffer, and a fourth SLC. The third address generator generates addresses of reference slices in the reference matrix such that each reference slice includes a unique portion of the reference matrix and provides the address of the reference slice to the memory interface system. The third data buffer stores the reference slices from the memory interface system. The fourth SLC outputs the reference slices to the MAC, which multiplies the transposed data slice by the reference slice.

[0008] In some implementations, the data matrix includes a range matrix from the radar circuitry, the range matrix corresponding to a range bin and containing data from the radar circuitry indexed by the rate bin and antenna assembly. Each data slice includes a matrix indexed by the rate bin and antenna assembly. The reference matrix includes a manipulation matrix of manipulation vectors indexed by the antenna assembly and assumed angles. Each calculated data slice is stored using a manipulation vector index and a rate bin index, such that a particular calculated data slice is stored at a memory location corresponding to the manipulation vector index of the corresponding particular manipulation vector and the rate bin index of the corresponding particular range slice. Attached Figure Description

[0009] To describe the various examples in detail, reference will now be made to the accompanying drawings, in which:

[0010] Figure 1 The illustration shows a car equipped with multiple radar systems for determining distance and angle of arrival.

[0011] Figure 2 This is a block diagram of an example radar system.

[0012] Figure 3 An exemplary method for processing signals received from a radar system is illustrated in flowchart form.

[0013] Figure 4 This diagram illustrates an example beamforming process for data received from a radar system.

[0014] Figure 5 The diagram illustrates an example data addressing process for inputting data into and outputting data from the beamforming engine.

[0015] Figures 6A to 6B The diagram illustrates the count and index values ​​of the range matrix used as input to the beamforming process.

[0016] Figure 7 This is a block diagram of an example beamforming hardware accelerator.

[0017] Figure 8An exemplary method for processing signals received from a time-division multiplexed (TDM) multiple-input multiple-output (MIMO) radar system is illustrated in flowchart form.

[0018] Figures 9A to 9B The diagram illustrates an exemplary beamforming process of a transmitter that decodes data received from a TDM-MIMO radar system.

[0019] Figure 10 The diagram illustrates an example data addressing process for inputting data into and outputting data from the beamforming engine. Detailed Implementation

[0020] The beamforming HWA described herein performs beamforming analysis of a radar system without reducing its angular resolution or update rate. The beamforming HWA divides data indexed by range, rate, and antenna combination into a set of 2D range matrices, each corresponding to a specific range bin and containing rate data indexed by the antenna combination. Each 2D range matrix is ​​further divided into range slices, a unique partial matrix corresponding to a specific rate bin, and the antenna combination that together constitutes the 2D range matrix. The beamforming HWA stores the range slices in a data buffer and performs beamforming operations on the range slices in the data buffer. The resulting slices of beamforming data are stored as a portion of a larger beamforming matrix corresponding to the 2D range matrix. The beamforming HWA then restarts the process with new range slices from the 2D range matrix until each range slice has been analyzed and a complete beamforming matrix of the 2D range matrix has been generated. By storing partial slices of data and performing beamforming operations sequentially on each partial slice, the beamforming HWA reduces the number of times data is fetched from and re-fetched from memory.

[0021] An address generator generates addresses for range slices and provides these addresses to a memory interface, which loads the range slices at the specified addresses into a data buffer. The address generator ensures that each consecutive range slice is unique and that the entire 2D range matrix is ​​analyzed. Selector logic coupled to the data buffer transposes the range slice data to perform row-by-row and column-by-column operations, thereby reducing the number of operations dedicated to data transposition. Similarly, a second address generator generates addresses for beamforming slices and provides these addresses to a memory interface, which stores the beamforming slices at specified addresses. The second address generator ensures that each consecutive beamforming slice is stored at the appropriate location in memory without overwriting previously calculated beamforming slices and indexing the beamforming slices based on the rate bins of the corresponding range slices.

[0022] Beamforming operations performed on range slices may involve multiplying the range slice by the manipulation matrix corresponding to the assumed angle and summing the result with intermediate slices corresponding to the same rate bin. Beamforming HWA can divide the manipulation matrix into manipulation slices, unique partial matrices corresponding to a specific antenna combination, and assumed angles that together constitute the manipulation matrix. Beamforming HWA stores the manipulation slices in a second data buffer and outputs the manipulation slices and range slices to a multiplier and accumulator unit (MAC). A third address generator generates the address of the manipulation slice and provides the address to a memory interface, which loads the manipulation slice at the specified address into the second data buffer. The third address generator ensures that each consecutive manipulation slice is unique and uses the entire manipulation matrix. Selector logic coupled to the second data buffer transposes the manipulation slice to perform row-by-row and column-by-column operations.

[0023] The MAC receives the range and manipulates the slices, multiplies the two slices together, sums the result with intermediate data corresponding to other antenna combinations with the same specific rate chamber and assumed angle, and generates beamforming data. The beamforming HWA can perform quantitative or dequantization operations on the beamforming data and store the beamforming data in a third data buffer. Selector logic and a second address generator can transpose the finalized beamforming slices and store them as part of the beamforming matrix.

[0024] Figure 1 This is an illustration of a car 100 equipped with multiple radar systems 110 and 112 for determining distance and angle of arrival. High-resolution 77 GHz automotive radar systems have been developed to improve driving comfort and safety by measuring the distances of the vehicle to surrounding objects. These distances can be used to instruct the central controller to react appropriately in various driving scenarios, such as parking assistance or collision avoidance. For example, some vehicles may have a single radar system 110, which can be used for tasks such as adaptive cruise control, object warning, and automatic braking. Some vehicles may have several radar systems, for example, one at the front 110 and one at the rear 112.

[0025] Some vehicles may have several systems at the front, several at the rear, and systems on the left and right sides. For example, such as... Figure 1 As shown, the vehicle may have three systems at the front, one facing forward and the other two tilted to the left and right, and three systems at the rear, one facing backward and the other two tilted to the left and right, one on the left and one on the right. Some radar systems, known as MIMO radar systems, use multiple transmitter antennas and multiple receiver antennas to improve the radar system's resolution when determining the angle of arrival. Some MIMO radar systems use TDM to distinguish between chirped signals from different transmitter antennas.

[0026] To determine the distance and angle of arrival from a radar system to a target, such as for parking assistance, the signal received from the radar system front end is typically processed through a series of signal processing steps, including Doppler correction and beamforming. Doppler correction compensates for the phase difference caused by the movement of the target object between the transmissions of the first and second chirped signals. Without correction, the Doppler shift of the target object's radial velocity introduces errors in distance and angle of arrival determination. Receiver beamforming estimates the angle of arrival by comparing the phase of the received signal with the expected phase of the signal from the object arriving at an assumed angle. Beamforming operations can test a unique set of assumed angles, the number of which is selected based on the radar system's desired angle of arrival resolution and other capabilities.

[0027] Figure 2 This is a block diagram of an example radar system 200. Radar system 200 can be used in vehicles, for example... Figure 1 The vehicle 100. In this example, the radar system 200 includes radar sensor circuitry 210, a central processing unit (CPU) core 230, a display 240, and a storage device 250. The radar sensor circuitry 210 includes a transmitter 220 that drives an antenna array 221 of one or more transmitter (TX) antennas. A receiver 222 receives signals from an antenna array 223 of one or more receiver (RX) antennas. A baseband module 224 amplifies and filters the received signal reflected from an object in the path of the transmitted chirped signal. In this exemplary radar system, the transmitter 220 operates in the 77 GHz region and generates a frequency-modulated continuous wave (FMCW) signal. A local oscillator (LO) 225 is used to frequency-modulate the continuous wave signal to form a series of chirps. In the exemplary radar system 200, the TX antenna array 221 and the RX antenna array 223 are stationary. In other examples, the antenna arrays may be configured to transmit and receive signals across a range, for example, by mechanical movement.

[0028] FMCW radar (also known as continuous wave frequency modulated (CWFM) radar) can determine range, velocity, and angle of arrival. In an FMCW system, the chirped signal transmitted by a continuous wave at a known stable frequency is modulated by a signal that varies the frequency over a fixed time period. The received reflections are then mixed with the transmitted chirped signal to generate a received beat frequency signal, which, after signal processing, provides the range, velocity, and angle of arrival of the target object. The frequency difference between the received reflections and the transmitted chirped signal increases with delay and is therefore proportional to the range.

[0029] The phase difference between received and reflected objects across a series of chirps allows for the calculation of the target's velocity. The phase difference between the received and reflected objects at the first and second receiver antennas allows for the calculation of the target's angle of arrival. Therefore, using an FMCW radar system, the distance between the target and the radar system, the target's relative velocity, the target's relative angle, and so on, can be calculated.

[0030] During normal operation, a linear frequency chirp is transmitted and reflected signals are received. The receiver and transmitter are arranged as a zero-difference system such that the received reflected signal is directly down-converted to baseband using a copy of the transmitted signal from LO 225 in receiver 222. The baseband module 224 then filters and amplifies the baseband signal through a filter and a variable gain amplifier. After converting the baseband signal to the digital domain, time-domain to frequency-domain transformations, such as Fast Fourier Transform (FFT), can be applied, and other signal processing can be performed to determine the distance, velocity, and angle of arrival between the target and the radar system 200.

[0031] CPU core 230 includes one or more CPU cores, digital signal processors, application-specific integrated circuits (ASICs), etc. The term "CPU core" (singular) is used herein to refer to one or more CPU cores and broadly describes central processing units, digital signal processors, ASICs, etc. CPU core 230 includes a chirp timing controller module 231, which receives a data stream from receiver antenna array 223 via analog-to-digital converter (ADC) 237 and performs transmitter chirp generation and control via digital-to-analog converter (DAC) 235. A variable voltage tuning control signal from DAC 235 is used to control LO 225. CPU core 230 also includes a signal processor core 232, which performs signal processing to determine rate, angle of arrival, distance between the target object and radar system 100, etc.

[0032] The signal processor core 232 can provide the determined values ​​to the display 240 and / or communicate with other systems via the network interface 233. For example, the network 233 may include various combinations of local area networks (LANs), wide area networks (WANs), the Internet, and / or other known or later-developed wired or wireless communication mechanisms. The storage device 250 can be used to store instructions and data received from the antenna 223 or the signal processor core 232. The storage device 250 can be any suitable storage medium, such as static random access memory (SRAM).

[0033] Figure 3 An example of beamforming analysis is illustrated in flowchart form 300. Figure 3The process shown is executed by a processing unit that executes instructions stored in a non-transitory computer-readable medium, such as signal processor core 232, which executes instructions stored in a non-transitory computer-readable medium. Figure 2 The instructions in storage device 250 of the radar system 200 shown are illustrated. In some instances, steps 340, 350, and 360 are executed by a beamforming engine or beamforming HWA and instructed to perform beamformer operation 370. First, at step 310, the processing unit performs a range FFT on the sampled data matrix 305, such as data sampled by the ADC 237 from the radar sensor circuit 210. The sampled data matrix 305 is a three-dimensional (3D) matrix containing sampled data values ​​from received beat frequency signals indexed over time by the antenna assembly (antenna assembly × sample × time).

[0034] After the processing unit performs a range FFT, the resulting 3D range matrix 315 contains sampled data values ​​categorized into range bins representing distance ranges between a target object and the radar sensor circuitry, the distance ranges being indexed over time by the antenna assembly (antenna assembly × range bin × time). At step 320, the range matrix 315 is transposed such that the resulting transposed 3D range matrix 325 is indexed by both the antenna assembly and the range bins (antenna assembly × time × range bin). At step 330, the processing unit performs a Doppler FFT on the transposed 3D range matrix 325. The resulting 3D Doppler matrix 335 contains data values ​​categorized into rate bins representing rate ranges of the target object, the rate ranges being indexed by both the antenna assembly and the range bins (antenna assembly × rate bin × range bin).

[0035] At step 340, the Doppler matrix 335 is transposed such that the resulting 3D Doppler matrix 345 contains data values ​​classified by antenna combinations indexed by rate bins and range bins (rate bin × antenna combination × range bin). Therefore, each 2D range matrix contained in the transposed 3D Doppler matrix 345 corresponds to a specific range bin and contains data values ​​classified by the antenna combination and rate bin of that specific range bin (rate bin × antenna combination). At step 350, the processing unit performs beamforming operations on the transposed Doppler matrix 345 to estimate the angle of arrival. At step 360, the processing unit performs a logarithmic operation to obtain a 3D processed data matrix 395, which contains signal peaks in the range bin, rate bin, and angle of arrival bin corresponding to the distance, rate, and angle of arrival to the target object.

[0036] At step 350, the beamforming operation may include comparing the phase of the received signal stored in the transposed Doppler matrix 345 with the expected signal phase of an object arriving from each of a set of hypothetical angles. For example, as described earlier in the background section, the processing unit multiplies each row (i.e., rate bin) of each 2D range matrix contained in the transposed 3D Doppler matrix 345 by each of a set of manipulation vectors, each manipulation vector representing a unique hypothetical angle from the set of hypothetical angles, and sums the resulting spectrum. Therefore, the total number of operations in process 300 can be quite large due to the extraction and re-extraction of each of the set of manipulation vectors and each row of each 2D range matrix in the transposed 3D Doppler matrix 345, data transposition, vector multiplication, etc. Signal processor cores capable of performing so many operations are resource-intensive in terms of semiconductor die surface area and computation time, as well as in compensation components such as memory.

[0037] Figure 4 Diagram illustrating radar systems (e.g.) Figure 2 The example beamforming operation flow 400 of the radar system 200 shown illustrates the received data. To reduce the number of times each row in each 2D range matrix 410 is extracted and re-extracted, the 2D range matrix 410 is divided into range slices 420. Range slice 420 represents a unique portion of the range matrix 410 corresponding to a specific rate compartment and antenna combination. Similarly, to reduce the number of times each manipulation vector in the 2D manipulation matrix 430 is extracted and re-extracted, the 2D manipulation matrix 430 is divided into manipulation slices 440. Manipulation slice 440 represents a unique portion of the manipulation matrix 430 corresponding to a specific antenna combination and assumed angle. In this example, range slice 420 contains the data values ​​of the first four antenna combinations in the first four rate compartments. Manipulation vector slice 440 contains the expected signal phase at the first four antenna combinations at the first four assumed angles. Range slice 420 and manipulation vector slice 440 are stored in their respective local data buffers before being provided to the multiplier and accumulator unit (MAC) 450. When range slice 420 and manipulation vector slice 440 are provided to MAC 450, the corresponding selector logic circuit (SLC) associated with the corresponding local data buffer can transpose each of range slice 420 and manipulation vector slice 440, which corresponds to, for example Figure 3 The transpose operation 340 is shown in the figure.

[0038] MAC 450 generates an intermediate slice 460 corresponding to a specific range slice 420 and a specific manipulation vector slice 440. The intermediate slice 460 contains the multiplied and accumulated values ​​of the first four antenna combinations from the first four rate bins and the first four hypothetical angles, and can be stored in a third local data buffer. Because the intermediate slice 460 does not contain all antenna combinations, the range slice 420 is updated to include the data values ​​of the next four antenna combinations from the first four rate bins, and the manipulation vector slice 440 is updated to include the expected signal phase at the next four antenna combinations from the first four hypothetical angles. MAC 450 multiplies the updated range slice 420 together with the updated manipulation vector slice 440 and accumulates the result with the previously calculated intermediate slice 460 from the third local data buffer for the first four antenna combinations. This process continues until the MAC operation for the last slice of the first four rate bins and the first four hypothetical angles used in this iteration is accumulated onto the intermediate slice in the third local data buffer.

[0039] The updated intermediate slice 460 contains beamforming data values ​​for all eight antenna combinations in the first four rate bins and the first four assumed angles. Since the updated intermediate slice 460 contains all antenna combinations, it is the beamforming slice 470 corresponding to the first four rate bins and the first four assumed angles. Other operations can then be performed on beamforming slice 470, such as... Figure 3 The numerical operation 360 is shown in the diagram. The beamforming slice 470 can then be stored as is or transposed by selector logic circuitry before storage. The corresponding address generators for the range slice 420, manipulation vector slice 440, and beamforming slice 470 ensure that the appropriate range slice and manipulation vector slice are used to calculate a specific beamforming slice, and store each beamforming slice in the appropriate location in memory. The beamforming operation flow 400 can be repeated until each hypothetical angle has been tested in each rate compartment of the 2D range matrix 410.

[0040] Figure 5 The diagram illustrates an example data addressing flow for inputting data into and outputting data from the beamforming engine. A 2D range matrix 510 is contained within a 3D Doppler matrix (not shown) and corresponds to a specific range cell. Each row of the 2D range matrix 510 corresponds to a specific rate cell, and each column corresponds to a specific antenna combination (rate cell × antenna combination). The 2D range matrix 510 contains N rate cells 515 and M antenna combinations 520, resulting in an (N × M) 2D range matrix 510. A range slice 530 contains several rate cells X and several antenna combinations Y, resulting in an (X × Y) range slice 530. In this example, the range slice 530 is a (4 × 4) matrix.

[0041] The 2D manipulation matrix 550 contains a set of manipulation vectors representing the assumed angles of the beamforming engine. Each row of the 2D manipulation matrix 550 corresponds to the expected signal phase at a specific antenna combination, and each column corresponds to a specific manipulation vector and its associated assumed angle (antenna combination × manipulation vector). The 2D manipulation matrix contains M antenna combinations 520 and S manipulation vectors 545, resulting in an (M×S) matrix. The manipulation vector slice 550 contains several antenna combinations Y and several assumed angles Z, resulting in a (Y×Z) matrix.

[0042] As previously referenced in this article Figure 4 As described in the exemplary beamforming operation procedure 400, an appropriate range slice 530 is multiplied and summed with a manipulation vector slice 550 to obtain a beamforming slice 570 corresponding to the specific X rate bins and Z assumed angles represented in the range slice and manipulation vector slice. This process is repeated until the beamforming matrix 580 contains the beamforming slice 570 for each rate bin and each assumed angle. The beamforming matrix 580 contains N rate bins 515 and S manipulation vectors 555, thereby generating an (N×S) matrix.

[0043] Figures 6A to 6B The diagram illustrates the count and index values ​​of the range matrix input into the beamforming process. Figure 6A The diagram illustrates the count values ​​of a 2D range matrix 610, which contains N rate cells 615 and M antenna combinations 620, resulting in an (N×M) matrix. Each range slice 630 from the 2D range matrix 610 contains several rate cells X and several antenna combinations Y, resulting in an (X×Y) matrix. To ensure that each range slice 630 is unique and does not overlap with other range slices, Figure 6B The example demonstrates an instance-based indexing scheme.

[0044] exist Figure 6BIn this context, the N-index 660 ensures that each range slice is taken from a suitable set of rows of the 2D range matrix 610, i.e., rate bins. Similarly, the M-index 650 ensures that each range slice is taken from a suitable set of columns of the 2D range matrix 610, i.e., antenna combinations. For example, the first range slice 630A contains rate bins 0 to 3 and antenna combinations 0 to 3 for each rate bin. The second range slice 630B also contains rate bins 0 to 3, but contains antenna combinations 4 to 7. The X-index 670 is based on the number of antenna combinations Y in the range slice 630 and the number of antenna combinations M 620 in the 2D range matrix 610 to ensure that the N-index 660 does not increment until all M antenna combinations of a particular X rate bin are represented in the beamforming slice. The S-index 680 is based on the number of control vectors and corresponding hypothetical angles in the associated control matrix, and resets the N-index 660, M-index 650 and X-index 670 whenever it increments, so that each rate cell and antenna combination in the 2D range matrix 610 is multiplied and accumulated with each control vector.

[0045] Figure 7 This is a block diagram of an example beamforming HWA 700. The beamforming HWA 700 includes a memory interface 705, a loading unit 710, and a storage unit 790. The loading unit 710 and storage unit 790 load data from and store data in memory via the memory interface 705. The beamforming HWA 700 also includes an address generator 715, a buffer 720, and an SLC 725 for loading range slices from a 2D range matrix in memory, such as... Figure 5 The range slice 530 shown in the figure. Address generator 715, for example, is based on Figure 6B The indexing scheme shown in the diagram generates a lookup address for a specific range slice and provides the lookup address to a loading unit 710. The loading unit 710 obtains the range slice from memory via a memory interface 705 and loads the range slice into a buffer 720. A selection signal provided to the SLC 725 indicates a specific row and column of the buffer 720, and the SLC 725 outputs the data value stored at the specific row and column, allowing the range slice to be transposed before being used in other operations. An address generator 715 generates a second lookup address for another different range slice, thereby ensuring that each range slice is unique and does not overlap with other range slices.

[0046] Similarly, the beamforming HWA 700 also includes an address generator 730, a buffer 735, and an SLC 740 for loading manipulation vector slices from a 2D manipulation matrix in memory, for example... Figure 5 The manipulation vector slice 550 shown in the figure. The address generator 730, for example, uses a similar... Figure 6BThe indexing scheme shown in the diagram generates a lookup address for a specific manipulation vector slice and provides the lookup address to the loading unit 710. The loading unit 710 retrieves the manipulation vector slice from memory via the memory interface 705 and loads the manipulation vector slice into the buffer 735. A selection signal provided to the SLC 740 indicates a specific row and column, and the SLC 740 outputs the data value stored at the specific row and column, allowing the manipulation vector slice to be transposed before being used in other operations. If needed, the address generator 730 generates a second lookup address for another different manipulation vector slice, thereby ensuring that each manipulation vector slice is unique and does not overlap with other manipulation vector slices.

[0047] MAC 750 includes a multiplier 754 and an accumulator 758. Multiplier 754 in MAC 750 multiplies the outputs of SLCs 725 and 740 together and provides the result 744 to accumulator 758 and SLC 760. Accumulator 758 sums the output of multiplier 754 with data 746 in buffer 780 and provides the sum 748 to SLC 760, which also receives the output 742 of SLC 725. A selection signal provided to SLC 760 indicates the operation to be performed, and SLC 760 outputs one of the multiplication result 744, the sum result 748, and the output 742 based on the selection signal. In one example, the selection signal indicates a transpose operation, causing SLC 760 to output the output 742 from SLC 725 and enabling the transpose of a partial range slice stored in buffer 720. In another example, the selection signal indicates a vector multiplication operation, causing the SLC 760 to output a multiplication result 744. In yet another example, the selection signal indicates an accumulation operation, causing the SLC 760 to output a summation result 748.

[0048] The measurement circuit 770 receives the output of the SLC 760 and, based on the operation to be performed, either provides the output unchanged to the buffer 780, or performs a measurement or increment operation on the output. For example, the measurement circuit 770 provides the output unchanged to the buffer 780 for a transpose operation, but performs a measurement or increment operation on the output of an accumulation operation. The buffer 780 stores the measurement circuit 770, intermediate slices (e.g., ...), and intermediate slices. Figure 4 The output of the intermediate slice 460 shown is reused by the accumulator 758, for example, in subsequent operations to multiply and sum other antenna combinations of a specific rate bin and assumed angle represented by the range slice and the manipulation vector slice. In response to the intermediate slice containing beamforming data for all antenna combinations in the specific rate bin and assumed angle, the buffer 780 stores the beamforming slice corresponding to the specific rate bin and assumed angle, which can then be stored more permanently in memory.

[0049] A select signal provided to SLC 785 indicates a specific row and column, and SLC 785 outputs beamforming data values ​​at said specific row and column stored in buffer 780, thereby allowing the beamforming slice to be transposed before being stored in memory. Address generator 790 generates a storage address for the beamforming slice and provides the storage address to storage unit 795, which receives the beamforming slice from selector logic circuit 785 and stores the beamforming slice in memory via memory interface 705. If needed, address generator 790 generates a second storage address for another different beamforming slice, thereby ensuring that each beamforming slice is stored in a unique memory location and does not overwrite other beamforming slices. In this example, SLCs 725, 740, 760, and 785 are multiplexers, but any suitable SLC can be used.

[0050] Figure 8 An exemplary method 800 for processing signals received from a TDM-MIMO radar system is illustrated in flowchart form. In a TDM-MIMO radar system, the RX antenna receives reflected signals but does not distinguish between reflections of chirped signals from different TX antennas. By decoding the TX antennas, the received reflections can be classified into chirped signals from different TX antennas, and the signal separation through antenna combination improves the angle of arrival resolution of the TDM-MIMO radar system compared to other radar systems. Furthermore, the TDM-MIMO radar system performs Doppler correction on the received beat frequency signals to correct for phase differences between different received signals caused by the time difference between the first and second transmitted chirped signals.

[0051] Figure 8 The process shown is executed by a processing unit that executes instructions stored in a non-transitory computer-readable medium, such as signal processor core 232, which executes instructions stored in a non-transitory computer-readable medium. Figure 2 The instructions in storage device 250 of the radar system 200 shown are illustrated. In some instances, steps 830, 840, 850, 860, and 870 are executed by a beamforming engine or a beamforming HWA, for example... Figure 7 The beamforming HWA 700 is shown in the figure and indicated as beamformer operation 880. In some embodiments, the beamforming HWA performing beamforming operation 880 performs steps 830, 840 and 850 in a single step 885, and steps 860 and 870 in a single step 890, as further discussed herein with reference to Figure 9.

[0052] First, at step 810, the processing unit performs a Doppler FFT on the range data 805 to generate a Doppler matrix 815. The Doppler matrix 815 contains data values ​​categorized into rate bins indexed by the RX antenna and range bin, (RX antenna × rate bin × range bin). At step 820, the processing unit performs Doppler correction to compensate for the phase difference in the received signal caused by the movement of the target object between the transmission of the first chirp signal and the transmission of the second chirp signal. At step 830, the obtained corrected Doppler matrix 825 is transposed to obtain a transposed corrected Doppler matrix 835. The transposed corrected Doppler matrix 835 contains data values ​​categorized by the RX antenna indexed by the rate bin and range bin, (rate bin × RX antenna × range bin).

[0053] At step 840, the processing unit performs TX antenna decoding on the transposed calibrated Doppler matrix 835. This uses a set of demodulation vectors to separate the received beat frequency signal from each RX antenna into its component TX chirp signal. For example, a TDM-MIMO radar system includes two RX antennas RX0 and RX1 and three TX antennas TX0, TX1, and TX2. The transposed calibrated Doppler matrix 835 contains sampled data values ​​classified by RX0 and RX1, but does not distinguish between TX0, TX1, and TX2. After TX antenna decoding, the decoded Doppler matrix 845 contains sampled data values ​​classified by combinations of RX and TX antennas, such as by RX0 and TX0, RX0 and TX1, RX0 and TX2, etc.

[0054] At step 850, the decoded Doppler matrix 845 is transposed to generate a transposed decoded Doppler matrix 855, which contains sampled data values ​​classified by the TX antenna. For example, the transposed decoded Doppler matrix 855 is sorted such that the received beat frequency signals of TX0, TX0 and RX0, and TX0 and RX1 are stored first. Next, the received beat frequency signals of TX1, TX1 and RX0, and TX1 and RX1 are stored. At step 860, the processing unit performs beamforming operations on the transposed decoded Doppler matrix 855 to estimate the angle of arrival. (Referring to...) Figure 3 As discussed, the beamforming operation may involve multiplying each row of each range matrix contained in the transposed decoded Doppler matrix 855 by a set of manipulation vectors and summing the resulting spectrum to produce a beamforming data matrix 865. At step 870, the processing unit performs a logarithmic operation to obtain a processed data matrix 895, which contains signal peaks in the range bins, velocity bins, and angle-of-arrival bins corresponding to the distance, velocity, and angle of arrival relative to the target object.

[0055] Figure 9 illustrates an exemplary beamforming process for steps 885 and 890 of data received from a TDM-MIMO radar system. For clarity, refer to... Figure 8 Instance methods 800 and Figure 7 Figure 9 is illustrated using beamforming HWA 700. To illustrate step 885, address generator 715, buffer 720, and SLC 725 load range slice 905 from the corrected Doppler matrix 825 in memory. In this example, range slice 905 contains the received beat frequency signal corresponding to a specific RX antenna RX0. Address generator 715 generates a lookup address for the received beat frequency signal corresponding to the specific range slice, buffer 720 stores the lookup address, and SLC 725 enables the lookup address to be transposed, thereby completing transpose step 830.

[0056] Similarly, address generator 730, buffer 735, and SLC 740 can load the TX demodulation matrix 910 from memory. Multiplier 754 in MAC 750 multiplies the outputs of SLC 725 and 740 together to complete TX decoding step 840 and obtain decoded range slice 915, which is stored unchanged in buffer 780. Address generator 790 generates the storage address of decoded range slice 915 in memory, and at step 880, before decoded range slice 915 is stored in memory as part of the transposed decoded Doppler matrix 855, SLC 785 enables the transposition of decoded range slice 915.

[0057] Step 890 has been discussed in more detail previously with reference to the operation of beamforming HWA 700. To briefly illustrate step 890, address generator 715, buffer 720, and SLC 725 load range slices from a transposed decoded Doppler matrix 855 in memory. Address generator 715 generates a lookup address for a specific range slice, buffer 720 stores the lookup address, and SLC 725 outputs the lookup address to multiplier 754 in MAC 750. Similarly, address generator 730, buffer 735, and SLC 740 load manipulation slices from a 2D manipulation matrix in memory.

[0058] In MAC 750, multiplier 754 multiplies the outputs of SLC 725 and 740 together and provides the result to accumulator 758. Accumulator 758 sums the result with, for example, an intermediate slice stored in buffer 780 and completes beamforming step 860 to obtain an intermediate slice, or, in response to an intermediate slice representing all antenna combinations for a specific rate chamber and assumed angle, obtains beamforming data matrix 865. SLC 760 receives beamforming data matrix 865 from accumulator 758 and provides it to measurement circuit 770, which performs measurement or auto-measurement operations on beamforming data matrix 865 and completes auto-measurement step 870. The resulting processed data matrix 895 is stored in buffer 780. Address generator 790 generates a storage address for processed data matrix 895 in memory, and SLC 785 enables processed data matrix 895 to be transposed before it is stored in memory.

[0059] Figure 10 The diagram illustrates an example data addressing process used to input data into and output data from a beamforming engine. Table 1010 illustrates the data addressing process included in the 3D transposed Doppler matrix (e.g., Figure 3 The data points in the 2D range matrix (345) of the 3D transposed Doppler matrix are stored in the correct order, and include data values ​​classified by antenna combinations indexed by rate bins and range bins (rate bin × antenna combination × range bin). The Doppler FFT output does not store data values ​​in order. (See reference...) Figure 6B The discussion, for example, uses the sample vector index of X index 670 to indicate which vectors should be included in a specific range slice. This is not as... Figure 6B Instead of starting from zero and incrementing by 1 as shown in Table 1010, the X index 670 can be modified based on the data point storage order shown in Table 1010 to start with a different number and increment, causing the beamformer to read data in the order shown in Table 1020. The beamformer performs beamforming operations and stores the resulting beamformed slices in local data buffer 1030. Selector logic circuitry 1040 allows the beamformed slices in buffer 1030 to be transposed before being stored in memory in the order shown in Table 1050.

[0060] In this example, each range slice is a (4×4) matrix. The first range slice 1025 read from memory by the beamformer contains vectors 1012, 1014, 1016, and 1018. In this example, the X index is equal to the bit inverse variable divided by 4, i.e., the number of vectors contained in each range slice. This causes range slice 1025 to group the data index into columns. The resulting beamforming slices in buffer 1030 are indexed in the order range slice 1025 is read, but selector logic circuitry 1040 can transpose them when the beamforming slices are stored, such that the stored beamforming slices group the data index into rows instead of columns.

[0061] In this specification, the term "couple" (or "couples") means an indirect or direct wired or wireless connection. Therefore, if a first device is coupled to a second device, the connection can be a direct connection or an indirect connection via other devices and connections. The statement "based on" means "at least partially based on". Therefore, if X is based on Y, then X can be a function of Y and any number of other factors. Examples are described herein with reference to the FT and FFT, but can be extended to other types of time-domain to frequency-domain transforms, such as sine and cosine transforms, etc.

[0062] Modifications are possible in the described embodiments, and other embodiments are possible within the scope of the claims.

Claims

1. A non-transitory computer-readable medium storing instructions for: An N×M range matrix with multiple antennas, including data from radar circuitry, is obtained. The range matrix is ​​indexed by N rate bins and M antenna combinations. Multiple unique X×Y range slices are extracted from the N×M range matrix, where X is a multiple of N and Y is a multiple of M. An M×S manipulation matrix including the expected phase is obtained, the manipulation matrix being indexed by the M antenna combinations and S assumed angles, and multiple Y×Z manipulation vector slices are extracted from the M×S manipulation matrix, where Z is a multiple of S. Iteratively, the process continues until the data for all range slices corresponding to a specific set of X rate bays are incorporated into the intermediate slice: Store a specific range slice corresponding to the specific set of X rate chambers and a specific set of Y antenna combinations from the M antenna combinations; and Iteratively: The expected phase of all manipulator vector slices corresponding to a specific set of S assumed angles is incorporated into the intermediate slice. Store specific manipulation vector slices corresponding to a specific set of Z hypothetical angles and a specific set of Y antenna combinations among the M antenna combinations; and A merging operation is performed on the first X×Y range slice and the first Y×Z manipulation vector slice to form the first version of the intermediate slice. The merging operation is then iteratively performed on the subsequent corresponding X×Y range slices and the corresponding subsequent corresponding Y×Z manipulation vector slices to generate a merging result and update the intermediate slice until the data of the X×Y range slices and all Y×Z manipulation vector slices are incorporated into the intermediate slice. The intermediate slice is stored as a beamforming slice, which incorporates data for the specific set of X rate chambers and the specific set of Z assumed angles of M antenna combinations. Beamforming is performed on the beamforming slice to obtain the result; as well as Store the results.

2. The non-transitory computer-readable medium of claim 1, wherein the instructions for performing the merging operation include instructions for multiplying the particular range slice with the particular manipulation vector slice.

3. The non-transitory computer-readable medium of claim 2, wherein the instructions for updating the intermediate slice comprise instructions for: The result of multiplying the specific range slice with the specific manipulation vector slice is summed with the intermediate slice to obtain the updated intermediate slice.

4. The non-transitory computer-readable medium of claim 3, further comprising instructions for generating a plurality of beamforming slices corresponding to all N rate cells and all S assumed angles, and for generating an N×S beamforming matrix including the plurality of beamforming slices.

5. The non-transitory computer-readable medium of claim 1, further comprising instructions for storing each of the plurality of X×Y range slices in a first data buffer, including instructions for: The address of a specific X×Y range slice is generated based on the N index and the M index, wherein the N index represents the specific set of X rate modules to be included in the specific range slice, and wherein the M index represents the specific set of Y antenna combinations to be included in the specific range slice. In response to storing the result as an intermediate slice corresponding to the specific set of Y antenna combinations contained in the specific range slice in a second data buffer, the M index is incremented by Y; and In response to storing the intermediate slice as the beamforming slice, the N index is incremented by X.

6. The non-transitory computer-readable medium of claim 5, further comprising instructions for storing each of the plurality of Y×Z manipulation vector slices in a third data buffer, including instructions for: The address of a specific Y×Z manipulation vector slice is generated based on the M-index and S-index, wherein the S-index represents the specific set of Z hypothetical angles to be included in the specific manipulation vector slice; and In response to storing the result as the intermediate slice in the second data buffer, the S index is incremented by Z.

7. The non-transitory computer-readable medium of claim 6, further comprising instructions for storing the intermediate slice as the beamforming slice including instructions for generating an address of the beamforming slice based on the N index.

8. An apparatus comprising: Sensor circuitry configured to receive reflected signals; One or more processor cores coupled to the sensor circuitry; and A non-transitory computer-readable medium storing machine instructions, which, when executed by the one or more processor cores, cause the one or more processor cores to iteratively: Extract data slices from a first data matrix consisting of multiple data slices; Extract data slices from a second data matrix consisting of multiple data slices; Multiplication operations are performed on corresponding data slices of the first data matrix and the second data matrix to produce a multiplication result, wherein the multiplication operation performed on the first data slice of the first data matrix and the first data slice of the second data matrix produces a first version of an intermediate slice; and An accumulation operation is performed on each of the second and subsequent multiplication results to update the current version of the intermediate slice until all of the plurality of data slices of the first data matrix and the second data matrix have been processed to produce the final intermediate slice; The machine instructions further include instructions for storing the final intermediate slice.

9. The apparatus of claim 8, further comprising a first data buffer for storing each data slice extracted from the first data matrix, a second data buffer for storing each data slice extracted from the second data matrix, and a third buffer for storing each intermediate slice.

10. The device of claim 9, further comprising a hardware accelerator (HWA), wherein the HWA comprises: The first address generator is configured to: Generate a unique address for each of the plurality of data slices in the first data matrix; The corresponding unique addresses of the plurality of data slices of the first data matrix are provided to the memory interface system for extraction from the first data matrix; A first selector logic circuit, which is coupled to the first data buffer; The second address generator is configured to: Generate a unique address for each of the plurality of data slices in the second data matrix; and The corresponding unique addresses of the plurality of data slices of the second data matrix are provided to the memory interface system for extraction from the second data matrix; and The second selector logic circuit is coupled to the second data buffer.

11. The device of claim 10, wherein the HWA further comprises a multiplier and accumulator MAC coupled to the first selector logic circuit and the second selector logic circuit.

12. The apparatus of claim 10, wherein the first selector logic circuit is further configured to perform a transpose operation on each data slice extracted from the first data matrix.

13. The device of claim 12, wherein the first data matrix includes a range matrix from the radar circuitry, wherein the range matrix corresponds to a range bin and includes data from the radar circuitry indexed by the rate bin and antenna combination, wherein each data slice extracted from the first data matrix includes a matrix indexed by the rate bin and antenna combination, wherein the second data matrix includes a manipulation matrix, and wherein the manipulation matrix includes a plurality of manipulation vectors indexed by the antenna combination and assumed angles.

14. The apparatus of claim 13, wherein each computed data slice is stored using a manipulation vector index and a rate bin index, such that a particular computed data slice is stored at a memory location corresponding to the manipulation vector index of a particular manipulation vector and the rate bin index of a particular range slice.

15. The device of claim 13, further comprising the radar circuitry, wherein the machine instructions for obtaining the data matrix cause the one or more processor cores to: This causes the radar circuit to transmit multiple chirped signals; This causes the radar circuit to generate a three-dimensional 3D sampled data matrix, wherein the 3D sampled data matrix includes received beat frequency signals corresponding to the plurality of chirped signals indexed by the antenna and sampling time; A range Fast Fourier Transform (FFT) is performed on the 3D sampled data matrix to obtain a 3D range matrix, wherein the 3D range matrix includes data classified by range bins indexed by the antenna and sampling time; Transpose the 3D extent matrix to obtain the transposed 3D extent matrix; Perform a Doppler FFT on the transposed 3D range matrix to obtain a 3D Doppler matrix, wherein the 3D Doppler matrix includes data classified by rate bins indexed by antenna and range bins; The 3D Doppler matrix is ​​transposed to obtain a transposed 3D Doppler matrix, wherein the transposed 3D Doppler matrix includes rate data of the antenna through range bin classification; and The range matrix is ​​obtained from the transposed 3D Doppler matrix.

16. The device of claim 15, wherein the radar circuitry comprises: An oscillator configured to generate the plurality of chirped signals; A transmitter TX antenna element, coupled to the oscillator and configured to transmit the plurality of chirped signals; The receiver RX antenna element is configured to receive reflections of the transmitted plurality of chirped signals; A baseband module coupled to the RX antenna element and configured to generate the received beat frequency signal based on the received reflections of the transmitted plurality of chirped signals; and An analog-to-digital converter coupled to the baseband module and configured to sample the received beat frequency signal to generate the 3D sampled data matrix.

17. The device of claim 15, wherein the radar circuitry comprises time-division multiplexed multiple-input multiple-output radar circuitry, wherein the antenna indexes comprise RX antenna indexes, wherein the antenna combination comprises RX antennas and TX antennas, and wherein the machine instructions for obtaining the range matrix from the transposed 3D Doppler matrix further cause the one or more processor cores to: The transposed 3D Doppler matrix is ​​multiplied by the demodulation matrix to obtain a modified transposed 3D Doppler matrix, wherein the modified transposed 3D Doppler matrix includes rate data of the antenna assembly classified by range bins; and The range matrix is ​​obtained from the modified transposed 3D Doppler matrix.

18. The apparatus of claim 11, wherein the first selector logic circuit is configured to receive a first select signal and, in response, provide row and column values ​​of the first data buffer as outputs, and the second selector logic circuit is configured to receive a second select signal and, in response, provide row and column values ​​of the second data buffer as outputs.

19. The device of claim 18, wherein the MAC includes a multiplier having an input and an output configured to receive the outputs of the first selector logic circuit and the second selector logic circuit, respectively, and the MAC further includes an accumulator having an input coupled to the output of the multiplier.

20. A circuit comprising: The first address generator is configured to: Generate the address of a specific X×Y range slice from an N×M range matrix such that the specific range slice includes a unique part of the N×M range matrix, where N represents the number of rate bins, X is a multiple of N, M represents the number of antenna combinations, and Y is a multiple of M. The address of the specific range slice is provided to the memory interface system; A first data buffer is configured to store the specific range slice from the memory interface system; A first selector logic circuit coupled to the first data buffer and configured to output a first selector logic circuit output, the first selector logic circuit output selectively including the specific range slice or a transposed range slice. The multiplier and accumulator MAC are coupled to the first selector logic circuit and configured as follows: The vector multiplication is performed using the output of the first selector logic circuit; and The result of the multiplication is summed with the data stored in the second data buffer; A second selector logic circuit is coupled to the first selector logic circuit and the MAC, wherein the second selector logic circuit is configured to output one of the transposed range slice, the multiplication result, and the summation result; The second data buffer is coupled to the output of the second selector logic circuit and configured to store the output of the second selector logic circuit; A third selector logic circuit is coupled to the second data buffer and configured to output a beamforming slice, wherein the beamforming slice includes a transposed summation result; and The second address generator is configured to: The address of the beamforming slice is generated such that the beamforming slice includes a unique portion of the beamforming matrix; and The address of the beamforming slice is provided to the memory interface system.

21. The circuit of claim 20, wherein the N×M range matrix corresponds to a range bin and includes rate data of the antenna combination in the radar circuit.

22. The circuit of claim 20, further comprising a magnitude circuit coupled between the second selector logic circuit and the second data buffer, and said magnitude circuit being configured to selectively: Perform a magnitude operation on the summation result to obtain beamforming data, wherein the beamforming slice includes transposed beamforming data; or The output of the second selector logic circuit remains unchanged.

23. The circuit of claim 20, wherein the MAC is further configured to multiply the transposed range slice by an M×S manipulation matrix, wherein the M×S manipulation matrix comprises a plurality of manipulation vectors.

24. The circuit of claim 23, further comprising: A third address generator, configured to: Generate the address of a specific X×Z manipulation slice such that each specific manipulation slice includes a unique portion of the M×S manipulation matrix; Provide the address of the specific manipulation slice to the memory interface system; A third data buffer is configured to store the specific manipulation slice from the memory interface system; and A fourth selector logic circuit coupled to the third data buffer, wherein the fourth selector logic circuit is configured to selectively output the specific manipulation slice, and wherein the MAC is further coupled to the fourth selector logic circuit and configured to multiply the output of the first selector logic circuit by the specific manipulation slice.

25. The circuit of claim 24, wherein the address of the beamforming slice includes a manipulation vector index and a rate bin index, such that the beamforming slice is stored at a memory location corresponding to the manipulation vector index of the particular manipulation slice and the rate bin index of the transposed range slice.

Citation Information

Patent Citations

  • Verifying data integrity utilizing dispersed storage

    US20130325820A1