Constant false alarm rate detection method and constant false alarm rate detection machine for millimeter wave radar

Abstract

The invention provides a constant false alarm rate detection method and a constant false alarm rate detection machine for a millimeter wave radar, which can cut a large two-dimensional detection plane into a plurality of small detection planes without interrupting a software system (CPU) in a CFAR detection process and participating in operation of a chip hardware system so as to complete partition detection of the two-dimensional plane. The detection process is processed more finely; by automatically adjusting a threshold multiplication factor of constant false alarm rate detection, different noise thresholds in different distance and speed (or angle) ranges are realized, so that the problem of inaccurate noise calculation caused by similar long distance is counteracted, the performance of the millimeter wave radar in the aspect of target detection is improved, and the target detection accuracy is improved. The problems of missing detection and false detection caused by inaccurate long-distance target noise calculation in a millimeter wave radar target detection operation process are solved.

Description

Technical Field

[0001] This invention belongs to the field of radar sensing technology and relates to a constant false alarm rate (CFAR) detection method and a constant CFAR detection machine for millimeter-wave radar. Background Technology

[0002] The radio frequency band between 30-300 GHz (wavelength 1mm-100um) is called the millimeter-wave band. Millimeter-wave radar refers to radar devices operating in the millimeter-wave band. Currently, the 24GHz, 60GHz, 77GHz, and 80GHz radars used in China are all classified as millimeter-wave radars. As a high-frequency product in the radar series, millimeter-wave radar is characterized by its wide bandwidth and short wavelength.

[0003] Frequency-modulated continuous wave (FMCW, chirped wave) is one of the most widely used radar technologies, offering exceptionally good range resolution (able to distinguish targets at close range). The radar transmits a series of chirped waves as a single radar data frame, which is then transformed into a digital domain for digital baseband signal processing to obtain information such as the target's range, velocity, and azimuth.

[0004] In radar chips, range and velocity information can be extracted through fast time-domain Fourier transform and slow time-domain Fourier transform. Targets detected by radar will form prominent peaks in the spatial spectrum of the range-velocity (or velocity-angle) plane; these peaks represent the detected targets. CFAR (Constant False Alarm Rate) detection is the most common method for extracting target information in radar chips. However, in millimeter-wave radar applications, the signal-to-noise ratio (SNR) of distant targets is much lower than that of nearby targets, making zone detection a necessary approach. Furthermore, the mean-based noise estimation method is susceptible to interference from the target itself, which can cause low-SNR targets near high-SNR targets to be missed. Sort to obtain the median can be used as an alternative algorithm, but hardware implementation of sorting algorithms typically has very high computational complexity. Summary of the Invention

[0005] To address the aforementioned issues, this invention discloses a constant false alarm rate (CFAR) detection method and a CFAR detection machine for millimeter-wave radar, which can improve the performance of CFAR detection in complex application scenarios and enhance the detection of distant targets and low signal-to-noise ratio (SNR) targets near high SNR targets.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] A method for detecting the constant false alarm rate (CFAR) of millimeter-wave radar includes the following steps:

[0008] The external device writes the data to be processed into the storage. After the data writing is complete, the external device sends a trigger signal to the CFAR detector, and the state machine of the CFAR detector starts working to perform search detection. In each clock cycle, the input addresser of the CFAR detector sends a complex or real number signal to the CFAR detection calculation unit of the CFAR detector. The CFAR detection calculation unit performs the following detection calculations:

[0009] Each clock cycle, the CFAR detection operation unit sums the specified neighboring cells of the test unit;

[0010] In each clock cycle, the threshold generator in the CFAR detection calculation unit calculates the threshold factor: it generates the threshold factor index through two counters and two comparators, obtains the A-dimensional threshold factor increment and the B-dimensional threshold factor increment through a lookup table method, and then implements the threshold factor for partition detection by configuring different partitions.

[0011] CFAR performs noise estimation on the detection unit, adds histogram statistics to sort the surrounding noise, and selects the real-time noise for CFAR calculation;

[0012] The decision threshold for CFAR is calculated by combining real-time noise and threshold factor, and the CFAR detection result is decided and output based on the threshold.

[0013] Furthermore, the process by which the threshold generator in the CFAR detection calculation unit calculates the threshold factor includes the following steps:

[0014] The A-dimensional counter cntA increments once. When the A-dimensional counter is full, the B-dimensional counter cntB increments by one, and the A-dimensional counter is cleared to zero.

[0015] The A-dimensional and B-dimensional counters are compared to obtain the appropriate A-dimensional incremental list index Aidx and B-dimensional incremental list index Bidx.

[0016] The A-dimensional threshold factor is calculated using the following formula:

[0017] ,

[0018] in, For A-dimensional incremental list;

[0019] The B-dimensional threshold factor is calculated using the following formula:

[0020] ,

[0021] in, For B-dimensional incremental lists;

[0022] The final threshold factor is calculated using the following formula:

[0023] ,

[0024] in, This is the initial value of the threshold factor.

[0025] Furthermore, the process of using histogram statistics to sort the surrounding noise includes:

[0026] Calculated using the following formula:

[0027] ,

[0028] Where hist is the histogram statistics, i is the index of the current test unit, and j is the index of all the noisy training units used to count the histogram.

[0029] Based on the statistical table, a rough sort is obtained, thus yielding the median or data at any position. This is then calculated using the following formula.

[0030] Noise_hist=hist(k),

[0031] Each clock cycle, the statistics module uses a sliding window to obtain two new input data and two data that need to be removed from the statistics. Each clock cycle, it needs to complete four statistical calculations and update the statistical information.

[0032] Get the smallest number in CFAR_HIST_K.

[0033] ,

[0034] Find the group index Nbin of the histogram that satisfies the above formula, and calculate the median of the Nbin-1th and Nbinth groups as the result.

[0035] Furthermore, when selecting real-time noise for CFAR calculation, the selection range includes not only the results of sorting the surrounding noise using histogram statistics, but also the following calculation results:

[0036] ,

[0037] ,

[0038] .

[0039] Furthermore, the decision threshold for CFAR is calculated by combining real-time noise and a threshold factor. The process includes:

[0040] A. Linear mode:

[0041] ,

[0042] B. Logarithmic pattern:

[0043] ,

[0044] in, For real-time noise, This is the threshold factor.

[0045] Furthermore, the process of CFAR detection based on threshold decision includes:

[0046] ,

[0047] TestCellValue is the amplitude or power value of the current test cell;

[0048] If the judgment is successful, the flag is set to 1; otherwise, it is set to 0.

[0049] Furthermore, the CFAR detection result output items include:

[0050] ,

[0051] ,

[0052] ,

[0053] ,

[0054] flag

[0055] When using 16-bit output format, select any two of the output options listed above;

[0056] When outputting in 32-bit format, the A-dimensional and B-dimensional dimensions will be... The energy and noise of the target signal are combined into a single 32-bit character.

[0057] The present invention also provides a constant false alarm rate (CFAR) detection machine, comprising: an input addresser, an output addresser, a data storage SRAM, a state machine, a CFAR detection calculation unit, and a register; the CFAR detection calculation unit is used to implement the detection calculation steps in a constant false alarm rate detection method for millimeter-wave radar.

[0058] The beneficial effects of this invention are as follows:

[0059] 1. The solution provided by this invention can divide a large 2D detection plane into multiple smaller detection planes without interrupting the software system (CPU) during the CFAR detection process or participating in the chip hardware system's calculations. This completes the partitioned detection of the 2D plane and allows for more refined processing of the detection process. By automatically adjusting the threshold multiplication factor of the constant false alarm rate (CFAR) detection, different noise thresholds are achieved within different distance and speed (or angle) ranges. This counteracts problems such as inaccurate noise calculation caused by long distances, improves the performance of millimeter-wave radar in target detection, and solves the problem of missed detections and false detections caused by inaccurate noise calculation of long-distance targets during millimeter-wave radar target detection calculations.

[0060] 2. The present invention can flexibly adapt to the problem of different SNR values ​​for targets at different distances. The hardware-applicable histogram noise ranking estimation algorithm can reduce the noise estimation value of low SNR targets next to high SNR targets. The present invention comprises a CFAR Threshold generator, combined with a lookup table method, which also enhances the programmability of the software. Based on the spatial spectrum diagram of the target signal, the present invention can adapt to constant false alarm rate detection in complex scenarios, thereby improving the performance of millimeter-wave radar in terms of missed detection. Attached Figure Description

[0061] Figure 1 A complete schematic diagram of the constant false alarm rate (CFAR) detector on the on-chip radar;

[0062] Figure 2 This is a schematic diagram of the constant false alarm rate (CFAR) detection and computation unit on an on-chip radar.

[0063] Figure 3 This is a schematic diagram of histogram statistics on the constant false alarm rate (CFAR) detector on the on-chip radar.

[0064] Figure 4 A schematic diagram of the spatial spectrum of a low-SNR target next to a high-SNR target;

[0065] Figure 5 A schematic diagram of the spatial spectrum of a distant target;

[0066] Figure 6 A schematic diagram of a ping-pong mechanism used in the constant false alarm rate (CFAR) detector for on-chip radar.

[0067] Figure 7 This is a schematic diagram of the flow control time for the constant false alarm rate (CFAR) detector on the chip.

[0068] Figure 8 A schematic diagram of the threshold factor partitioning for the constant false alarm rate (CFAR) detector of an on-chip radar.

[0069] Figure 9 This is the output data format of the constant false alarm rate (CFAR) detector for on-chip radar. Detailed Implementation

[0070] The technical solutions provided by the present invention will be described in detail below with reference to specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0071] The constant false alarm rate (CFAR) detection method for millimeter-wave radar provided by this invention is implemented through an adaptive area selection algorithm using the threshold factor of the CFAR detection module.

[0072] Figure 1 This is a schematic diagram of a Constant False Alarm Rate (CFAR) detector on an on-chip radar. As shown in the figure, the CFAR detector includes an input addresser, an output addresser, SRAM data storage (requiring four or more SRAMs), a state machine, a CFAR detector unit, multi-page registers, and static registers. The CFAR detector can extract target information (including target position and signal-to-noise ratio information) from spectral signal data.

[0073] Specifically, the overall steps of the constant false alarm rate (CFAR) detection method for millimeter-wave radar are as follows:

[0074] External devices (such as CPU, DMA, etc.) write the data to be processed into the corresponding memory blocks. After writing the data, the external device sends a trigger signal to the CFAR inspection machine. The CFAR inspection machine's state machine then starts working, and the entire process searches and inspects a 2D plane line by line. When the CFAR inspection machine inspects a portion of the data, such as... Figure 6 As shown, the CPU or other devices can read and write other memory blocks of the CFAR (excluding the memory blocks that the CFAR is currently using for computation and storing results). This allows the CPU's data read and write time to be hidden (without additional time overhead) during continuous system operation.

[0075] The search and detection process is as follows:

[0076] Each clock cycle, the input addresser sends a complex or real number signal to the CFAR detection and processing unit. The CFAR unit first performs operations such as modulo, modulo square, and logarithm calculations to ensure the data stream is converted to real numbers during computation. The CFAR unit also has a buffer to cache data, ensuring that noise near the detection point can be estimated in real-time using a sliding window.

[0077] Each clock cycle, the CFAR detection operation unit sums the values ​​of the specified neighboring cells of the test unit:

[0078] ,

[0079] ,

[0080] Where n is the number of guard units (skipped and not used for summation), k is the number of units used for summation (the length of the sliding window), i is the index corresponding to the data of the current test unit, and j is the index of the training unit used to calculate noise. For each test unit (corresponding to index i), the range of j is determined by the sliding window.

[0081] The following are the index ranges of the noise training units (j) on both sides of the current test unit.

[0082]

[0083] &

[0084]

[0085] The adaptive region selection algorithm is implemented in the CFAR detection computing unit, and this adaptive algorithm is called CFAR Threshold generator.

[0086] The threshold generator calculates CFAR threshold factors every clock cycle. These factors are determined based on the state machine's current state.

[0087] First, the threshold factor consists of three parts: the initial value, the A-dimensional increment, and the B-dimensional increment. The A-dimensional increment is the sample-level count, while the B-dimensional increment is the channel-level count. For example, a 64*512 two-dimensional plane can be viewed as 64 channels, each with 512 sample points. The threshold factor is used to determine the appropriate threshold for different distance ranges and different speed (or angle) ranges. This is achieved by configuring different partitions to implement the threshold factor for partitioned detection. Through software configuration, a large 2D plane for CFAR detection can be divided into 32 smaller planes, achieving partitioned treatment, such as... Figure 8 As shown.

[0088] A-dimensional incremental list index That is, the index is determined by the A-dimensional counter, as follows (t=cntA): T Aidx This is the Aidx-th time decision threshold. This invention supports a total of 8 different thresholds T0, T1, ..., T7 in dimension A, which can also be understood as dividing all sample points of a channel into 8 blocks (not necessarily uniformly segmented). Configured by software. The time decision threshold refers to the index corresponding to the current A-dimensional increment that will be selected when the A-dimensional counter's count reaches this time threshold range.

[0089] During each clock cycle, the A-dimensional counter (cntA) increments once, such as Figure 8 As shown.

[0090] CFAR's state control can find a suitable interval for cntA.

[0091] ,

[0092] When the A-dimensional counter reaches full, the B-dimensional counter (cntB) increments by one, and the A-dimensional counter is reset to 0.

[0093] Similarly,

[0094] ,

[0095] TB0, TB1, ..., TB3 are B-dimensional time decision thresholds, which can be understood as all channels being divided into four parts.

[0096] Using a comparator, the A-dimensional and B-dimensional counters can obtain the appropriate A-dimensional incremental list index Aidx and B-dimensional incremental list index Bidx.

[0097] The A-dimensional threshold factor is calculated as follows:

[0098] ,

[0099] in, This is an A-dimensional incremental list.

[0100] The calculation method for B-dimensional dimensions is the same as that for A-dimensional dimensions:

[0101] ,

[0102] in, This is a B-dimensional incremental list.

[0103] After obtaining the A-dimensional threshold factor increment and the B-dimensional threshold factor increment respectively, the final threshold factor is calculated:

[0104] ,

[0105] By combining initial values ​​and noise estimation for the detection points, zone detection can be achieved. This is the initial value for the threshold calculation factor.

[0106] CFAR performs noise estimation on the detection units. Here, the hardware uses a sliding window to select neighboring training units for noise estimation. The most common approach for CFAR is to average these training units, which is also the simplest for hardware. Since the computational complexity of implementing a ranking algorithm in hardware is very high, this invention provides a new noise estimation method: a hardware-applicable histogram estimation ranking method. This method uses histogram statistics to replace the ranking algorithm, performing ranking operations on surrounding noise to achieve a low-complexity ranking noise estimation algorithm. The histogram statistical method estimates CFAR noise in real time, which can meet the requirements of hardware pipeline design, and can basically complete the detection of one sample point in one clock cycle. The calculation method is as follows:

[0107] ,

[0108] j=ikn, ...,ikn,i+n,i+n+1, ...,i+k+n,

[0109] Here, k is the window length of the sliding window, n is the number of guard units, i is the index corresponding to the current test unit, and j is the index corresponding to all the noisy training units used for counting histograms.

[0110] hist is the statistical result of the histogram, calculated as follows:

[0111] The high-order bit value of the test unit,

[0112] ,

[0113] This involves right-shifting the training units. The number of bits for right shifting, CFAR_HIST_SEL (which determines the class interval of the histogram), is a software-configurable register.

[0114] Then, xh is protected.

[0115] ,

[0116] ,

[0117] The histogram provides 64 read-only registers for calculating the histogram.

[0118] xh (high-order bit value of the test unit) is used to select the group of the histogram.

[0119] ,

[0120] Or

[0121] ,

[0122] The data range used for statistical calculations here is the same as the data used for summation above. A rough sort can be obtained from the statistical table, thus yielding the median or data at any position.

[0123] Noise_hist=hist(k),

[0124] Furthermore, when calculating the noise of the histogram sorting method, specifically, such as Figure 3 As shown, after each clock cycle and each slide, the training units used for noise estimation change as depicted in the figure. Two training units move out of the training window, and two move in. There's no need to re-perform histogram statistics for all training units within the window. The corresponding statistical bin for the units moving in is incremented by one, and the bin for the units moving out is decremented by one.

[0125] Thus, as Figure 7 As shown, there is a time delay when calculating the noise of the first test unit (to allow the sliding window to fill with data, so the delay is equal to the sliding window length). In subsequent cycles (each clock cycle), the sliding window slides once. Histogram statistics do not need to be recalculated; only the training units moving out and into the sliding window on both sides need to be calculated, using the formula above for addition and subtraction. This allows for one CFAR detection to be completed per clock cycle, which is highly efficient.

[0126] Each clock cycle, the statistics module uses sliding windows to obtain two new input data points (one sliding window on each side of the test unit) and two data points to be removed from the statistics. Four statistical calculations need to be completed each clock cycle, and the statistical information needs to be updated.

[0127] Get the smallest number in CFAR_HIST_K

[0128] ,

[0129] Note that CFAR_HIST_K here is a software configuration value.

[0130] Find the Nbin (group index of the histogram) that satisfies the above formula.

[0131] Calculate the median of the Nbin-1th and Nbinth groups as the result (Noise_hist).

[0132] The sum can be processed by summing the sums of both sides, increasing or decreasing the sum, and shifting to the right to obtain the average value.

[0133] ,

[0134] ,

[0135] ,

[0136] The software configuration register controls the selection of one of Noise_hist, CA, CASO, and CAGO as the real-time noise for CFAR calculations.

[0137] like Figure 7 As shown, the calculation of the first noise point will have a delay due to the length of the sliding window, and thereafter, the CFAR detection of the detection unit can be completed once per clock cycle.

[0138] like Figure 4 As shown, at zero velocity, there are three targets awaiting detection. The noise of the smallest target is amplified by neighboring strong targets. In this case, CFAR detection using CA CFAR (derived from the CA formula) makes it difficult to detect this target. Therefore, using histogram statistics, such as selecting the median of 100 nearby points, results in more accurate noise estimation.

[0139] like Figure 5 As shown, there is a target with velocity at a distance. However, as shown in the figure, the noise is amplified at zero speed due to the strong reflection zone on the roadside.

[0140] As mentioned earlier, noise levels can be inflated by the algorithm during estimation for various reasons. On one hand, methods are needed to address this noise inflation problem, such as the hardware-applicable histogram sorting algorithm proposed in this invention. On the other hand, if the distant X signal (the energy or amplitude of the currently detected signal) is too weak, an adaptive threshold calculation factor should be used.

[0141] After obtaining the real-time noise and threshold factor, calculate the decision threshold for CFAR:

[0142] CFAR operations support two data modes:

[0143] C. Linear mode:

[0144] ,

[0145] D. Logarithmic pattern:

[0146] ,

[0147] Where Th is the decision threshold.

[0148] The CFAR test results are determined using the following formula:

[0149] ,

[0150] TestCellValue is the amplitude or power value of the current test cell;

[0151] If the judgment is successful, the flag is set to 1; otherwise, it is set to 0.

[0152] The formatted output items for CFAR test results are as follows:

[0153] ,

[0154] ,

[0155] ,

[0156] ,

[0157] flag

[0158] CFAR supports the following two data output formats:

[0159] 16-bit format:

[0160] You can select any two of the output options listed above.

[0161] 32-bit format:

[0162] Peak index (dimensional A and dimensional B) merges into a 32-bit character

[0163] The energy and noise of the target signal are combined into a 32-bit character.

[0164] The 32-bit output can output the target information's SNR (signal-to-noise ratio) and index information in a 2D plane in one go.

[0165] like Figure 9 As shown, CFAR will confirm whether the test unit is the target value based on the decision result, and then... Figure 9 The output shows the CFAR detection results.

[0166] It should be noted that the above content merely illustrates the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. For those skilled in the art, various improvements and modifications can be made without departing from the principle of the present invention, and all such improvements and modifications fall within the scope of protection of the claims of the present invention.

Claims

1. A constant false alarm rate detection method for a millimeter wave radar, characterized by, The method comprises the following steps: The external device writes the data to be processed into the storage, and after the data is written, the external device sends a trigger signal to the CFAR detection machine, and the state machine of the CFAR detection machine starts to work and performs search detection; in each clock cycle, the input addresser of the CFAR detection machine transmits a complex signal or a real signal into the CFAR detection calculation unit of the CFAR detection machine, and the CFAR detection calculation unit performs the following detection operation: In each clock cycle, the CFAR detection calculation unit sums up the specified adjacent units of the test unit; In each clock cycle, the threshold generator in the CFAR detection calculation unit calculates a threshold factor: the threshold factor index is generated through two counters and two comparators, the A-dimensional threshold factor increment and the B-dimensional threshold factor increment are obtained through the table lookup method, and the threshold factor of the partition detection is obtained through the configuration of different partitions; The CFAR estimates the noise of the test unit, increases the sorting operation of the surrounding noise through the histogram statistics, and selects the real-time noise for the CFAR calculation; The decision threshold of the CFAR is calculated in combination with the real-time noise and the threshold factor, the CFAR detection result is judged based on the threshold, and the CFAR detection result is output.

2. The constant false alarm rate detection method of a millimeter wave radar according to claim 1, characterized by, The process of calculating the threshold factor by the threshold generator in the CFAR detection calculation unit comprises the following steps: The A-dimensional counter cntA is self-accumulated once, and when the A-dimensional counter is full, the B-dimensional counter cntB is increased by one, and the A-dimensional counter is cleared; The A-dimensional and B-dimensional counters obtain the appropriate A-dimensional increment list index Aidx and B-dimensional increment list index Bidx through the comparators; The A-dimensional threshold factor is calculated by the following formula: ， wherein is the A-dimension increment list; The B-dimensional threshold factor is calculated by the following formula: ， wherein is a list of B dimensions; The final threshold factor is calculated by the following formula: ， wherein is the initial value of the threshold factor. 3.The false-alarm-free detection method of the millimeter wave radar according to claim 1, characterized in that, The process of sorting the surrounding noise through the histogram statistics comprises the following steps: The following formula is used for calculation: ， Wherein, hist is the histogram statistics result, i is the index corresponding to the current test unit, and j is the index corresponding to all noise training units for counting the histogram; The rough sorting is obtained according to the statistics table, so that the median or data at any position is obtained, and the following formula is used for calculation Noise_hist=hist(k), In each clock cycle, the sliding window used by the statistics module obtains two new input data and two data to be removed from the statistics, four number statistics calculations are needed in each clock cycle, and the statistics information is updated; The smallest number in the CFAR_HIST_K is obtained, ， CFAR_HIST_K is a software configuration value; The group index Nbin of the histogram satisfying the above formula is found, and the medians of the Nbin-1th group and the Nbinth group are calculated as the calculation results.

4. The false alarm rate constant detection method of the millimeter wave radar according to claim 1, characterized by, When the real-time noise for the CFAR calculation is selected, the selection range includes the following calculation results in addition to the sorting operation result of the surrounding noise through the histogram statistics: ， ， 。 5.The false-alarm-free detection method of the millimeter wave radar according to claim 1, characterized in that, The combination of real-time noise and threshold factor calculates the CFAR decision threshold The process includes: A. Linear mode: ， B. Log mode: ， wherein is the real-time noise, is a threshold factor. 6.The false-alarm-free detection method of the millimeter wave radar according to claim 1, characterized in that, The process of judging the CFAR detection result based on the threshold comprises the following steps: ， TestCellValue is the amplitude or power value of the current test unit; If the judgment is passed, the flag is 1, otherwise, the flag is 0. 7.The false-alarm-free detection method of a millimeter wave radar according to claim 1, characterized in that, The CFAR detection result output items comprise: ， ， ， ， flag, When output in 16-bit format, any two of the above output options are selected; When outputting in 32-bit format, the A-dimensional and B-dimensional dimensions will be... The energy and noise of the target signal are combined into a single 32-bit character.

8. A constant false alarm rate detection machine comprising: The application relates to a radar signal processing unit, which comprises an input addresser, an output addresser, a data storage SRAM, a state machine, a CFAR detection calculation unit and a register; the CFAR detection calculation unit is used to realize the detection operation steps in the constant false alarm rate detection method of the millimeter wave radar in any one of claims 1-7.

Images